Improve synthetic_size json output:

- add new fields 'parent_lsn' and 'wal_from_parent' to segments;
- output 'timeline_inputs' lsns as u64 to be consistent with other places

.dockerignore

@@ -15,8 +15,10 @@
 !proxy/
 !safekeeper/
 !storage_broker/
+!trace/
 !vendor/postgres-v14/
 !vendor/postgres-v15/
 !workspace_hack/
 !neon_local/
 !scripts/ninstall.sh
+!vm-cgconfig.conf

.github/actions/run-python-test-set/action.yml

@@ -123,8 +123,8 @@ runs:
           exit 1
         fi
         if [[ "${{ inputs.run_in_parallel }}" == "true" ]]; then
-          # -n4 uses four processes to run tests via pytest-xdist
-          EXTRA_PARAMS="-n4 $EXTRA_PARAMS"
+          # -n16 uses sixteen processes to run tests via pytest-xdist
+          EXTRA_PARAMS="-n16 $EXTRA_PARAMS"
 
           # --dist=loadgroup points tests marked with @pytest.mark.xdist_group
           # to the same worker to make @pytest.mark.order work with xdist

.github/ansible/prod.ap-southeast-1.hosts.yaml

@@ -2,11 +2,11 @@ storage:
   vars:
     bucket_name: neon-prod-storage-ap-southeast-1
     bucket_region: ap-southeast-1
-    console_mgmt_base_url: http://console-release.local
+    console_mgmt_base_url: http://neon-internal-api.aws.neon.tech
     broker_endpoint: http://storage-broker-lb.epsilon.ap-southeast-1.internal.aws.neon.tech:50051
     pageserver_config_stub:
       pg_distrib_dir: /usr/local
-      metric_collection_endpoint: http://console-release.local/billing/api/v1/usage_events
+      metric_collection_endpoint: http://neon-internal-api.aws.neon.tech/billing/api/v1/usage_events
       metric_collection_interval: 10min
       remote_storage:
         bucket_name: "{{ bucket_name }}"
@@ -32,7 +32,7 @@ storage:
       hosts:
         safekeeper-0.ap-southeast-1.aws.neon.tech:
           ansible_host:  i-0d6f1dc5161eef894
-        safekeeper-1.ap-southeast-1.aws.neon.tech:
-          ansible_host:  i-0e338adda8eb2d19f
         safekeeper-2.ap-southeast-1.aws.neon.tech:
           ansible_host:  i-04fb63634e4679eb9
+        safekeeper-3.ap-southeast-1.aws.neon.tech:
+          ansible_host:  i-05481f3bc88cfc2d4

.github/ansible/prod.eu-central-1.hosts.yaml

@@ -2,11 +2,11 @@ storage:
   vars:
     bucket_name: neon-prod-storage-eu-central-1
     bucket_region: eu-central-1
-    console_mgmt_base_url: http://console-release.local
+    console_mgmt_base_url: http://neon-internal-api.aws.neon.tech
     broker_endpoint: http://storage-broker-lb.gamma.eu-central-1.internal.aws.neon.tech:50051
     pageserver_config_stub:
       pg_distrib_dir: /usr/local
-      metric_collection_endpoint: http://console-release.local/billing/api/v1/usage_events
+      metric_collection_endpoint: http://neon-internal-api.aws.neon.tech/billing/api/v1/usage_events
       metric_collection_interval: 10min
       remote_storage:
         bucket_name: "{{ bucket_name }}"

.github/ansible/prod.us-east-2.hosts.yaml

@@ -2,11 +2,11 @@ storage:
   vars:
     bucket_name: neon-prod-storage-us-east-2
     bucket_region: us-east-2
-    console_mgmt_base_url: http://console-release.local
+    console_mgmt_base_url: http://neon-internal-api.aws.neon.tech
     broker_endpoint: http://storage-broker-lb.delta.us-east-2.internal.aws.neon.tech:50051
     pageserver_config_stub:
       pg_distrib_dir: /usr/local
-      metric_collection_endpoint: http://console-release.local/billing/api/v1/usage_events
+      metric_collection_endpoint: http://neon-internal-api.aws.neon.tech/billing/api/v1/usage_events
       metric_collection_interval: 10min
       remote_storage:
         bucket_name: "{{ bucket_name }}"
@@ -27,6 +27,8 @@ storage:
           ansible_host:  i-062227ba7f119eb8c
         pageserver-1.us-east-2.aws.neon.tech:
           ansible_host:  i-0b3ec0afab5968938
+        pageserver-2.us-east-2.aws.neon.tech:
+          ansible_host:  i-0d7a1c4325e71421d
 
     safekeepers:
       hosts:

.github/ansible/prod.us-west-2.hosts.yaml

@@ -2,11 +2,11 @@ storage:
   vars:
     bucket_name: neon-prod-storage-us-west-2
     bucket_region: us-west-2
-    console_mgmt_base_url: http://console-release.local
+    console_mgmt_base_url: http://neon-internal-api.aws.neon.tech
     broker_endpoint: http://storage-broker-lb.eta.us-west-2.internal.aws.neon.tech:50051
     pageserver_config_stub:
       pg_distrib_dir: /usr/local
-      metric_collection_endpoint: http://console-release.local/billing/api/v1/usage_events
+      metric_collection_endpoint: http://neon-internal-api.aws.neon.tech/billing/api/v1/usage_events
       metric_collection_interval: 10min
       remote_storage:
         bucket_name: "{{ bucket_name }}"
@@ -29,6 +29,8 @@ storage:
           ansible_host: i-0c834be1dddba8b3f
         pageserver-2.us-west-2.aws.neon.tech:
           ansible_host: i-051642d372c0a4f32
+        pageserver-3.us-west-2.aws.neon.tech:
+          ansible_host: i-00c3844beb9ad1c6b
 
     safekeepers:
       hosts:

.github/ansible/production.hosts.yaml

@@ -1,40 +0,0 @@
----
-storage:
-  vars:
-    console_mgmt_base_url: http://console-release.local
-    bucket_name: zenith-storage-oregon
-    bucket_region: us-west-2
-    broker_endpoint: http://storage-broker.prod.local:50051
-    pageserver_config_stub:
-      pg_distrib_dir: /usr/local
-      metric_collection_endpoint: http://console-release.local/billing/api/v1/usage_events
-      metric_collection_interval: 10min
-      remote_storage:
-        bucket_name: "{{ bucket_name }}"
-        bucket_region: "{{ bucket_region }}"
-        prefix_in_bucket: "{{ inventory_hostname }}"
-    safekeeper_s3_prefix: prod-1/wal
-    hostname_suffix: ".local"
-    remote_user: admin
-    sentry_environment: production
-
-  children:
-    pageservers:
-      hosts:
-        zenith-1-ps-2:
-          console_region_id: aws-us-west-2
-        zenith-1-ps-3:
-          console_region_id: aws-us-west-2
-        zenith-1-ps-4:
-          console_region_id: aws-us-west-2
-        zenith-1-ps-5:
-          console_region_id: aws-us-west-2
-
-    safekeepers:
-      hosts:
-        zenith-1-sk-1:
-          console_region_id: aws-us-west-2
-        zenith-1-sk-2:
-          console_region_id: aws-us-west-2
-        zenith-1-sk-4:
-          console_region_id: aws-us-west-2

.github/ansible/staging.eu-west-1.hosts.yaml

@@ -2,12 +2,17 @@ storage:
   vars:
     bucket_name: neon-dev-storage-eu-west-1
     bucket_region: eu-west-1
-    console_mgmt_base_url: http://console-staging.local
+    console_mgmt_base_url: http://neon-internal-api.aws.neon.build
     broker_endpoint: http://storage-broker-lb.zeta.eu-west-1.internal.aws.neon.build:50051
     pageserver_config_stub:
       pg_distrib_dir: /usr/local
-      metric_collection_endpoint: http://console-staging.local/billing/api/v1/usage_events
+      metric_collection_endpoint: http://neon-internal-api.aws.neon.build/billing/api/v1/usage_events
       metric_collection_interval: 10min
+      tenant_config:
+        eviction_policy:
+          kind: "LayerAccessThreshold"
+          period: "20m"
+          threshold: "20m"
       remote_storage:
         bucket_name: "{{ bucket_name }}"
         bucket_region: "{{ bucket_region }}"

.github/ansible/staging.us-east-2.hosts.yaml

@@ -2,12 +2,17 @@ storage:
   vars:
     bucket_name: neon-staging-storage-us-east-2
     bucket_region: us-east-2
-    console_mgmt_base_url: http://console-staging.local
+    console_mgmt_base_url: http://neon-internal-api.aws.neon.build
     broker_endpoint: http://storage-broker-lb.beta.us-east-2.internal.aws.neon.build:50051
     pageserver_config_stub:
       pg_distrib_dir: /usr/local
-      metric_collection_endpoint: http://console-staging.local/billing/api/v1/usage_events
+      metric_collection_endpoint: http://neon-internal-api.aws.neon.build/billing/api/v1/usage_events
       metric_collection_interval: 10min
+      tenant_config:
+        eviction_policy:
+          kind: "LayerAccessThreshold"
+          period: "20m"
+          threshold: "20m"
       remote_storage:
         bucket_name: "{{ bucket_name }}"
         bucket_region: "{{ bucket_region }}"
@@ -31,6 +36,8 @@ storage:
           ansible_host: i-01e31cdf7e970586a
         pageserver-3.us-east-2.aws.neon.build:
           ansible_host: i-0602a0291365ef7cc
+        pageserver-99.us-east-2.aws.neon.build:
+          ansible_host: i-0c39491109bb88824
 
     safekeepers:
       hosts:
@@ -40,3 +47,5 @@ storage:
           ansible_host: i-0171efc3604a7b907
         safekeeper-2.us-east-2.aws.neon.build:
           ansible_host: i-0de0b03a51676a6ce
+        safekeeper-99.us-east-2.aws.neon.build:
+          ansible_host: i-0d61b6a2ea32028d5

.github/helm-values/dev-eu-west-1-zeta.neon-proxy-scram.yaml

@@ -1,16 +1,31 @@
 # Helm chart values for neon-proxy-scram.
 # This is a YAML-formatted file.
 
+deploymentStrategy:
+  type: RollingUpdate
+  rollingUpdate:
+    maxSurge: 100%
+    maxUnavailable: 50%
+
+# Delay the kill signal by 7 days (7 * 24 * 60 * 60)
+# The pod(s) will stay in Terminating, keeps the existing connections
+# but doesn't receive new ones
+containerLifecycle:
+  preStop:
+    exec:
+      command: ["/bin/sh", "-c", "sleep 604800"]
+terminationGracePeriodSeconds: 604800
+
 image:
   repository: neondatabase/neon
 
 settings:
   authBackend: "console"
-  authEndpoint: "http://console-staging.local/management/api/v2"
+  authEndpoint: "http://neon-internal-api.aws.neon.build/management/api/v2"
   domain: "*.eu-west-1.aws.neon.build"
   sentryEnvironment: "staging"
   wssPort: 8443
-  metricCollectionEndpoint: "http://console-staging.local/billing/api/v1/usage_events"
+  metricCollectionEndpoint: "http://neon-internal-api.aws.neon.build/billing/api/v1/usage_events"
   metricCollectionInterval: "1min"
 
 # -- Additional labels for neon-proxy pods

.github/helm-values/dev-us-east-2-beta.neon-proxy-link.yaml

@@ -8,8 +8,9 @@ settings:
   authBackend: "link"
   authEndpoint: "https://console.stage.neon.tech/authenticate_proxy_request/"
   uri: "https://console.stage.neon.tech/psql_session/"
+  domain: "pg.neon.build"
   sentryEnvironment: "staging"
-  metricCollectionEndpoint: "http://console-staging.local/billing/api/v1/usage_events"
+  metricCollectionEndpoint: "http://neon-internal-api.aws.neon.build/billing/api/v1/usage_events"
   metricCollectionInterval: "1min"
 
 # -- Additional labels for neon-proxy-link pods

.github/helm-values/dev-us-east-2-beta.neon-proxy-scram-legacy.yaml

@@ -6,11 +6,11 @@ image:
 
 settings:
   authBackend: "console"
-  authEndpoint: "http://console-staging.local/management/api/v2"
+  authEndpoint: "http://neon-internal-api.aws.neon.build/management/api/v2"
   domain: "*.cloud.stage.neon.tech"
   sentryEnvironment: "staging"
   wssPort: 8443
-  metricCollectionEndpoint: "http://console-staging.local/billing/api/v1/usage_events"
+  metricCollectionEndpoint: "http://neon-internal-api.aws.neon.build/billing/api/v1/usage_events"
   metricCollectionInterval: "1min"
 
 # -- Additional labels for neon-proxy pods

.github/helm-values/dev-us-east-2-beta.neon-proxy-scram.yaml

@@ -1,16 +1,31 @@
 # Helm chart values for neon-proxy-scram.
 # This is a YAML-formatted file.
 
+deploymentStrategy:
+  type: RollingUpdate
+  rollingUpdate:
+    maxSurge: 100%
+    maxUnavailable: 50%
+
+# Delay the kill signal by 7 days (7 * 24 * 60 * 60)
+# The pod(s) will stay in Terminating, keeps the existing connections
+# but doesn't receive new ones
+containerLifecycle:
+  preStop:
+    exec:
+      command: ["/bin/sh", "-c", "sleep 604800"]
+terminationGracePeriodSeconds: 604800
+
 image:
   repository: neondatabase/neon
 
 settings:
   authBackend: "console"
-  authEndpoint: "http://console-staging.local/management/api/v2"
+  authEndpoint: "http://neon-internal-api.aws.neon.build/management/api/v2"
   domain: "*.us-east-2.aws.neon.build"
   sentryEnvironment: "staging"
   wssPort: 8443
-  metricCollectionEndpoint: "http://console-staging.local/billing/api/v1/usage_events"
+  metricCollectionEndpoint: "http://neon-internal-api.aws.neon.build/billing/api/v1/usage_events"
   metricCollectionInterval: "1min"
 
 # -- Additional labels for neon-proxy pods

.github/helm-values/prod-ap-southeast-1-epsilon.neon-proxy-scram.yaml

@@ -1,16 +1,32 @@
 # Helm chart values for neon-proxy-scram.
 # This is a YAML-formatted file.
 
+deploymentStrategy:
+  type: RollingUpdate
+  rollingUpdate:
+    maxSurge: 100%
+    maxUnavailable: 50%
+
+# Delay the kill signal by 7 days (7 * 24 * 60 * 60)
+# The pod(s) will stay in Terminating, keeps the existing connections
+# but doesn't receive new ones
+containerLifecycle:
+  preStop:
+    exec:
+      command: ["/bin/sh", "-c", "sleep 604800"]
+terminationGracePeriodSeconds: 604800
+
+
 image:
   repository: neondatabase/neon
 
 settings:
   authBackend: "console"
-  authEndpoint: "http://console-release.local/management/api/v2"
+  authEndpoint: "http://neon-internal-api.aws.neon.tech/management/api/v2"
   domain: "*.ap-southeast-1.aws.neon.tech"
   sentryEnvironment: "production"
   wssPort: 8443
-  metricCollectionEndpoint: "http://console-release.local/billing/api/v1/usage_events"
+  metricCollectionEndpoint: "http://neon-internal-api.aws.neon.tech/billing/api/v1/usage_events"
   metricCollectionInterval: "10min"
 
 # -- Additional labels for neon-proxy pods

.github/helm-values/prod-eu-central-1-gamma.neon-proxy-scram.yaml

@@ -1,16 +1,32 @@
 # Helm chart values for neon-proxy-scram.
 # This is a YAML-formatted file.
 
+deploymentStrategy:
+  type: RollingUpdate
+  rollingUpdate:
+    maxSurge: 100%
+    maxUnavailable: 50%
+
+# Delay the kill signal by 7 days (7 * 24 * 60 * 60)
+# The pod(s) will stay in Terminating, keeps the existing connections
+# but doesn't receive new ones
+containerLifecycle:
+  preStop:
+    exec:
+      command: ["/bin/sh", "-c", "sleep 604800"]
+terminationGracePeriodSeconds: 604800
+
+
 image:
   repository: neondatabase/neon
 
 settings:
   authBackend: "console"
-  authEndpoint: "http://console-release.local/management/api/v2"
+  authEndpoint: "http://neon-internal-api.aws.neon.tech/management/api/v2"
   domain: "*.eu-central-1.aws.neon.tech"
   sentryEnvironment: "production"
   wssPort: 8443
-  metricCollectionEndpoint: "http://console-release.local/billing/api/v1/usage_events"
+  metricCollectionEndpoint: "http://neon-internal-api.aws.neon.tech/billing/api/v1/usage_events"
   metricCollectionInterval: "10min"
 
 # -- Additional labels for neon-proxy pods

.github/helm-values/prod-us-east-2-delta.neon-proxy-link.yaml

@@ -1,37 +1,37 @@
+# Helm chart values for neon-proxy-link.
+# This is a YAML-formatted file.
+
+image:
+  repository: neondatabase/neon
+
 settings:
   authBackend: "link"
   authEndpoint: "https://console.neon.tech/authenticate_proxy_request/"
   uri: "https://console.neon.tech/psql_session/"
+  domain: "pg.neon.tech"
   sentryEnvironment: "production"
 
 # -- Additional labels for zenith-proxy pods
 podLabels:
   zenith_service: proxy
   zenith_env: production
-  zenith_region: us-west-2
-  zenith_region_slug: oregon
+  zenith_region: us-east-2
+  zenith_region_slug: us-east-2
 
 service:
+  type: LoadBalancer
   annotations:
     service.beta.kubernetes.io/aws-load-balancer-type: external
     service.beta.kubernetes.io/aws-load-balancer-nlb-target-type: ip
     service.beta.kubernetes.io/aws-load-balancer-scheme: internal
-    external-dns.alpha.kubernetes.io/hostname: proxy-release.local
-  type: LoadBalancer
+    external-dns.alpha.kubernetes.io/hostname: neon-proxy-link-mgmt.delta.us-east-2.aws.neon.tech
 
 exposedService:
   annotations:
     service.beta.kubernetes.io/aws-load-balancer-type: external
     service.beta.kubernetes.io/aws-load-balancer-nlb-target-type: ip
     service.beta.kubernetes.io/aws-load-balancer-scheme: internet-facing
-    external-dns.alpha.kubernetes.io/hostname: connect.neon.tech,pg.neon.tech
-
-metrics:
-  enabled: true
-  serviceMonitor:
-    enabled: true
-    selector:
-      release: kube-prometheus-stack
+    external-dns.alpha.kubernetes.io/hostname: neon-proxy-link.delta.us-east-2.aws.neon.tech
 
 extraManifests:
   - apiVersion: operator.victoriametrics.com/v1beta1

.github/helm-values/prod-us-east-2-delta.neon-proxy-scram.yaml

@@ -1,16 +1,32 @@
 # Helm chart values for neon-proxy-scram.
 # This is a YAML-formatted file.
 
+deploymentStrategy:
+  type: RollingUpdate
+  rollingUpdate:
+    maxSurge: 100%
+    maxUnavailable: 50%
+
+# Delay the kill signal by 7 days (7 * 24 * 60 * 60)
+# The pod(s) will stay in Terminating, keeps the existing connections
+# but doesn't receive new ones
+containerLifecycle:
+  preStop:
+    exec:
+      command: ["/bin/sh", "-c", "sleep 604800"]
+terminationGracePeriodSeconds: 604800
+
+
 image:
   repository: neondatabase/neon
 
 settings:
   authBackend: "console"
-  authEndpoint: "http://console-release.local/management/api/v2"
+  authEndpoint: "http://neon-internal-api.aws.neon.tech/management/api/v2"
   domain: "*.us-east-2.aws.neon.tech"
   sentryEnvironment: "production"
   wssPort: 8443
-  metricCollectionEndpoint: "http://console-release.local/billing/api/v1/usage_events"
+  metricCollectionEndpoint: "http://neon-internal-api.aws.neon.tech/billing/api/v1/usage_events"
   metricCollectionInterval: "10min"
 
 # -- Additional labels for neon-proxy pods

.github/helm-values/prod-us-west-2-eta.neon-proxy-scram-legacy.yaml

@@ -1,32 +1,39 @@
+# Helm chart values for neon-proxy-scram.
+# This is a YAML-formatted file.
+
+image:
+  repository: neondatabase/neon
+
 settings:
   authBackend: "console"
-  authEndpoint: "http://console-release.local/management/api/v2"
+  authEndpoint: "http://neon-internal-api.aws.neon.tech/management/api/v2"
   domain: "*.cloud.neon.tech"
   sentryEnvironment: "production"
   wssPort: 8443
-  metricCollectionEndpoint: "http://console-release.local/billing/api/v1/usage_events"
+  metricCollectionEndpoint: "http://neon-internal-api.aws.neon.tech/billing/api/v1/usage_events"
   metricCollectionInterval: "10min"
 
+# -- Additional labels for neon-proxy pods
 podLabels:
   zenith_service: proxy-scram
-  zenith_env: production
+  zenith_env: prod
   zenith_region: us-west-2
-  zenith_region_slug: oregon
+  zenith_region_slug: us-west-2
 
 exposedService:
   annotations:
     service.beta.kubernetes.io/aws-load-balancer-type: external
     service.beta.kubernetes.io/aws-load-balancer-nlb-target-type: ip
     service.beta.kubernetes.io/aws-load-balancer-scheme: internet-facing
-    external-dns.alpha.kubernetes.io/hostname: '*.cloud.neon.tech'
+    external-dns.alpha.kubernetes.io/hostname: neon-proxy-scram-legacy.eta.us-west-2.aws.neon.tech
   httpsPort: 443
 
-metrics:
-  enabled: true
-  serviceMonitor:
-    enabled: true
-    selector:
-      release: kube-prometheus-stack
+#metrics:
+#  enabled: true
+#  serviceMonitor:
+#    enabled: true
+#    selector:
+#      release: kube-prometheus-stack
 
 extraManifests:
   - apiVersion: operator.victoriametrics.com/v1beta1

.github/helm-values/prod-us-west-2-eta.neon-proxy-scram.yaml

@@ -1,16 +1,32 @@
 # Helm chart values for neon-proxy-scram.
 # This is a YAML-formatted file.
 
+deploymentStrategy:
+  type: RollingUpdate
+  rollingUpdate:
+    maxSurge: 100%
+    maxUnavailable: 50%
+
+# Delay the kill signal by 7 days (7 * 24 * 60 * 60)
+# The pod(s) will stay in Terminating, keeps the existing connections
+# but doesn't receive new ones
+containerLifecycle:
+  preStop:
+    exec:
+      command: ["/bin/sh", "-c", "sleep 604800"]
+terminationGracePeriodSeconds: 604800
+
+
 image:
   repository: neondatabase/neon
 
 settings:
   authBackend: "console"
-  authEndpoint: "http://console-release.local/management/api/v2"
+  authEndpoint: "http://neon-internal-api.aws.neon.tech/management/api/v2"
   domain: "*.us-west-2.aws.neon.tech"
   sentryEnvironment: "production"
   wssPort: 8443
-  metricCollectionEndpoint: "http://console-release.local/billing/api/v1/usage_events"
+  metricCollectionEndpoint: "http://neon-internal-api.aws.neon.tech/billing/api/v1/usage_events"
   metricCollectionInterval: "10min"
 
 # -- Additional labels for neon-proxy pods

.github/helm-values/production.neon-storage-broker.yaml

@@ -1,56 +0,0 @@
-# Helm chart values for neon-storage-broker
-podLabels:
-  neon_env: production
-  neon_service: storage-broker
-
-# Use L4 LB
-service:
-  # service.annotations -- Annotations to add to the service
-  annotations:
-    service.beta.kubernetes.io/aws-load-balancer-type: external  # use newer AWS Load Balancer Controller
-    service.beta.kubernetes.io/aws-load-balancer-nlb-target-type: ip
-    service.beta.kubernetes.io/aws-load-balancer-scheme: internal  # deploy LB to private subnet
-    # assign service to this name at external-dns
-    external-dns.alpha.kubernetes.io/hostname: storage-broker.prod.local
-  # service.type -- Service type
-  type: LoadBalancer
-  # service.port -- broker listen port
-  port: 50051
-
-ingress:
-  enabled: false
-
-metrics:
-  enabled: true
-  serviceMonitor:
-    enabled: true
-    selector:
-      release: kube-prometheus-stack
-
-extraManifests:
-  - apiVersion: operator.victoriametrics.com/v1beta1
-    kind: VMServiceScrape
-    metadata:
-      name: "{{ include \"neon-storage-broker.fullname\" . }}"
-      labels:
-        helm.sh/chart: neon-storage-broker-{{ .Chart.Version }}
-        app.kubernetes.io/name: neon-storage-broker
-        app.kubernetes.io/instance: neon-storage-broker
-        app.kubernetes.io/version: "{{ .Chart.AppVersion }}"
-        app.kubernetes.io/managed-by: Helm
-      namespace: "{{ .Release.Namespace }}"
-    spec:
-      selector:
-        matchLabels:
-          app.kubernetes.io/name: "neon-storage-broker"
-      endpoints:
-        - port: broker
-          path: /metrics
-          interval: 10s
-          scrapeTimeout: 10s
-      namespaceSelector:
-        matchNames:
-          - "{{ .Release.Namespace }}"
-
-settings:
-  sentryEnvironment: "production"

.github/workflows/build_and_test.yml

@@ -1,4 +1,4 @@
-name: Test and Deploy
+name: Build and Test
 
 on:
   push:
@@ -19,10 +19,12 @@ concurrency:
 env:
   RUST_BACKTRACE: 1
   COPT: '-Werror'
+  AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
+  AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
 
 jobs:
   tag:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, small ]
     container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/base:pinned
     outputs:
       build-tag: ${{steps.build-tag.outputs.tag}}
@@ -50,9 +52,9 @@ jobs:
         id: build-tag
 
   check-codestyle-python:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, small ]
     container:
-      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cloud:pinned
+      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
 
     steps:
@@ -85,7 +87,7 @@ jobs:
         run: poetry run mypy .
 
   check-codestyle-rust:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, large ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
@@ -97,16 +99,16 @@ jobs:
           submodules: true
           fetch-depth: 1
 
-      - name: Restore cargo deps cache
-        id: cache_cargo
-        uses: actions/cache@v3
-        with:
-          path: |
-            ~/.cargo/registry/
-            !~/.cargo/registry/src
-            ~/.cargo/git/
-            target/
-          key: v1-${{ runner.os }}-cargo-clippy-${{ hashFiles('rust-toolchain.toml') }}-${{ hashFiles('Cargo.lock') }}
+#      Disabled for now
+#      - name: Restore cargo deps cache
+#        id: cache_cargo
+#        uses: actions/cache@v3
+#        with:
+#          path: |
+#            !~/.cargo/registry/src
+#            ~/.cargo/git/
+#            target/
+#          key: v1-${{ runner.os }}-cargo-clippy-${{ hashFiles('rust-toolchain.toml') }}-${{ hashFiles('Cargo.lock') }}
 
       # Some of our rust modules use FFI and need those to be checked
       - name: Get postgres headers
@@ -133,7 +135,7 @@ jobs:
         run: cargo deny check
 
   build-neon:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, large ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
@@ -141,7 +143,6 @@ jobs:
       fail-fast: false
       matrix:
         build_type: [ debug, release ]
-
     env:
       BUILD_TYPE: ${{ matrix.build_type }}
       GIT_VERSION: ${{ github.sha }}
@@ -194,24 +195,26 @@ jobs:
           echo "cov_prefix=${cov_prefix}" >> $GITHUB_ENV
           echo "CARGO_FEATURES=${CARGO_FEATURES}" >> $GITHUB_ENV
           echo "CARGO_FLAGS=${CARGO_FLAGS}" >> $GITHUB_ENV
+          echo "CARGO_HOME=${GITHUB_WORKSPACE}/.cargo" >> $GITHUB_ENV
 
+      # Disabled for now
       # Don't include the ~/.cargo/registry/src directory. It contains just
       # uncompressed versions of the crates in ~/.cargo/registry/cache
       # directory, and it's faster to let 'cargo' to rebuild it from the
       # compressed crates.
-      - name: Cache cargo deps
-        id: cache_cargo
-        uses: actions/cache@v3
-        with:
-          path: |
-            ~/.cargo/registry/
-            !~/.cargo/registry/src
-            ~/.cargo/git/
-            target/
-          # Fall back to older versions of the key, if no cache for current Cargo.lock was found
-          key: |
-            v1-${{ runner.os }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-${{ hashFiles('Cargo.lock') }}
-            v1-${{ runner.os }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-
+#      - name: Cache cargo deps
+#        id: cache_cargo
+#        uses: actions/cache@v3
+#        with:
+#          path: |
+#            ~/.cargo/registry/
+#            !~/.cargo/registry/src
+#            ~/.cargo/git/
+#            target/
+#          # Fall back to older versions of the key, if no cache for current Cargo.lock was found
+#          key: |
+#            v1-${{ runner.os }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-${{ hashFiles('Cargo.lock') }}
+#            v1-${{ runner.os }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-
 
       - name: Cache postgres v14 build
         id: cache_pg_14
@@ -301,7 +304,7 @@ jobs:
         uses: ./.github/actions/save-coverage-data
 
   regress-tests:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, large ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
@@ -334,7 +337,7 @@ jobs:
         uses: ./.github/actions/save-coverage-data
 
   benchmarks:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, small ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
@@ -365,7 +368,7 @@ jobs:
       # while coverage is currently collected for the debug ones
 
   merge-allure-report:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, small ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
@@ -402,7 +405,7 @@ jobs:
           DATABASE_URL="$TEST_RESULT_CONNSTR" poetry run python3 scripts/ingest_regress_test_result.py --revision ${SHA} --reference ${GITHUB_REF} --build-type ${BUILD_TYPE} --ingest suites.json
 
   coverage-report:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, small ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
@@ -418,16 +421,17 @@ jobs:
           submodules: true
           fetch-depth: 1
 
-      - name: Restore cargo deps cache
-        id: cache_cargo
-        uses: actions/cache@v3
-        with:
-          path: |
-            ~/.cargo/registry/
-            !~/.cargo/registry/src
-            ~/.cargo/git/
-            target/
-          key: v1-${{ runner.os }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-${{ hashFiles('Cargo.lock') }}
+#      Disabled for now
+#      - name: Restore cargo deps cache
+#        id: cache_cargo
+#        uses: actions/cache@v3
+#        with:
+#          path: |
+#            ~/.cargo/registry/
+#            !~/.cargo/registry/src
+#            ~/.cargo/git/
+#            target/
+#          key: v1-${{ runner.os }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-${{ hashFiles('Cargo.lock') }}
 
       - name: Get Neon artifact
         uses: ./.github/actions/download
@@ -477,7 +481,7 @@ jobs:
             }"
 
   trigger-e2e-tests:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, small ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/base:pinned
       options: --init
@@ -522,9 +526,10 @@ jobs:
             }"
 
   neon-image:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, large ]
     needs: [ tag ]
-    container: gcr.io/kaniko-project/executor:v1.9.0-debug
+    # https://github.com/GoogleContainerTools/kaniko/issues/2005
+    container: gcr.io/kaniko-project/executor:v1.7.0-debug
     defaults:
       run:
         shell: sh -eu {0}
@@ -540,12 +545,16 @@ jobs:
         run: echo "{\"credsStore\":\"ecr-login\"}" > /kaniko/.docker/config.json
 
       - name: Kaniko build neon
-        run: /kaniko/executor --snapshotMode=redo --cache=true --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache --snapshotMode=redo --context . --build-arg GIT_VERSION=${{ github.sha }} --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:${{needs.tag.outputs.build-tag}}
+        run: /kaniko/executor --reproducible --snapshotMode=redo --skip-unused-stages --cache=true --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache --context . --build-arg GIT_VERSION=${{ github.sha }} --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:${{needs.tag.outputs.build-tag}}
+
+      # Cleanup script fails otherwise - rm: cannot remove '/nvme/actions-runner/_work/_temp/_github_home/.ecr': Permission denied
+      - name: Cleanup ECR folder
+        run: rm -rf ~/.ecr
 
   compute-tools-image:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, large ]
     needs: [ tag ]
-    container: gcr.io/kaniko-project/executor:v1.9.0-debug
+    container: gcr.io/kaniko-project/executor:v1.7.0-debug
     defaults:
       run:
         shell: sh -eu {0}
@@ -558,11 +567,14 @@ jobs:
         run: echo "{\"credsStore\":\"ecr-login\"}" > /kaniko/.docker/config.json
 
       - name: Kaniko build compute tools
-        run: /kaniko/executor --snapshotMode=redo --cache=true --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache --snapshotMode=redo --context . --build-arg GIT_VERSION=${{ github.sha }} --dockerfile Dockerfile.compute-tools --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{needs.tag.outputs.build-tag}}
+        run: /kaniko/executor --reproducible --snapshotMode=redo --skip-unused-stages --cache=true --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache --context . --build-arg GIT_VERSION=${{ github.sha }} --dockerfile Dockerfile.compute-tools --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{needs.tag.outputs.build-tag}}
+
+      - name: Cleanup ECR folder
+        run: rm -rf ~/.ecr
 
   compute-node-image:
-    runs-on: [ self-hosted, dev, x64 ]
-    container: gcr.io/kaniko-project/executor:v1.9.0-debug
+    runs-on: [ self-hosted, gen3, large ]
+    container: gcr.io/kaniko-project/executor:v1.7.0-debug
     needs: [ tag ]
     strategy:
       fail-fast: false
@@ -583,10 +595,13 @@ jobs:
         run: echo "{\"credsStore\":\"ecr-login\"}" > /kaniko/.docker/config.json
 
       - name: Kaniko build compute node with extensions
-        run: /kaniko/executor --skip-unused-stages  --snapshotMode=redo --cache=true --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache  --context . --build-arg GIT_VERSION=${{ github.sha }} --dockerfile Dockerfile.compute-node-${{ matrix.version }} --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
+        run: /kaniko/executor --reproducible --snapshotMode=redo --skip-unused-stages --cache=true --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache  --context . --build-arg GIT_VERSION=${{ github.sha }} --build-arg PG_VERSION=${{ matrix.version }} --dockerfile Dockerfile.compute-node --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
+
+      - name: Cleanup ECR folder
+        run: rm -rf ~/.ecr
 
   vm-compute-node-image:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, large ]
     needs: [ tag, compute-node-image ]
     strategy:
       fail-fast: false
@@ -596,34 +611,31 @@ jobs:
       run:
         shell: sh -eu {0}
     env:
-      VM_INFORMANT_VERSION: 0.1.1
+      VM_BUILDER_VERSION: v0.4.6
 
     steps:
-      - name: Downloading latest vm-builder
+      - name: Checkout
+        uses: actions/checkout@v1
+        with:
+          fetch-depth: 0
+
+      - name: Downloading vm-builder
         run: |
-          curl -L https://github.com/neondatabase/neonvm/releases/latest/download/vm-builder -o vm-builder
+          curl -L https://github.com/neondatabase/neonvm/releases/download/$VM_BUILDER_VERSION/vm-builder -o vm-builder
           chmod +x vm-builder
 
       - name: Pulling compute-node image
         run: |
           docker pull 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
 
-      - name: Downloading VM informant version ${{ env.VM_INFORMANT_VERSION }}
+      - name: Building VM compute-node rootfs
         run: |
-          curl -fL https://github.com/neondatabase/autoscaling/releases/download/${{ env.VM_INFORMANT_VERSION }}/vm-informant -o vm-informant
-          chmod +x vm-informant
-
-      - name: Adding VM informant to compute-node image
-        run: |
-          ID=$(docker create 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}})
-          docker cp vm-informant $ID:/bin/vm-informant
-          docker commit $ID temp-vm-compute-node
-          docker rm -f $ID
+          docker build -t temp-vm-compute-node --build-arg SRC_IMAGE=369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}} -f Dockerfile.vm-compute-node .
 
       - name: Build vm image
         run: |
           # note: as of 2023-01-12, vm-builder requires a trailing ":latest" for local images
-          ./vm-builder -src=temp-vm-compute-node:latest -dst=369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
+          ./vm-builder -use-inittab -src=temp-vm-compute-node:latest -dst=369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
 
       - name: Pushing vm-compute-node image
         run: |
@@ -631,7 +643,7 @@ jobs:
 
   test-images:
     needs: [ tag, neon-image, compute-node-image, compute-tools-image ]
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, small ]
 
     steps:
       - name: Checkout
@@ -673,20 +685,39 @@ jobs:
           docker compose -f ./docker-compose/docker-compose.yml down
 
   promote-images:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, small ]
     needs: [ tag, test-images, vm-compute-node-image ]
+    container: golang:1.19-bullseye
     if: github.event_name != 'workflow_dispatch'
-    container: amazon/aws-cli
-    strategy:
-      fail-fast: false
-      matrix:
-        name: [ neon, compute-node-v14, vm-compute-node-v14, compute-node-v15, vm-compute-node-v15, compute-tools]
 
     steps:
-      - name: Promote image to latest
+      - name: Install Crane & ECR helper
+        if: |
+          (github.ref_name == 'main' || github.ref_name == 'release') &&
+          github.event_name != 'workflow_dispatch'
         run: |
-          export MANIFEST=$(aws ecr batch-get-image --repository-name ${{ matrix.name }} --image-ids imageTag=${{needs.tag.outputs.build-tag}} --query 'images[].imageManifest' --output text)
-          aws ecr put-image --repository-name ${{ matrix.name }} --image-tag latest --image-manifest "$MANIFEST"
+          go install github.com/google/go-containerregistry/cmd/crane@31786c6cbb82d6ec4fb8eb79cd9387905130534e # v0.11.0
+          go install github.com/awslabs/amazon-ecr-credential-helper/ecr-login/cli/docker-credential-ecr-login@69c85dc22db6511932bbf119e1a0cc5c90c69a7f # v0.6.0
+
+      - name: Configure ECR login
+        run: |
+          mkdir /github/home/.docker/
+          echo "{\"credsStore\":\"ecr-login\"}" > /github/home/.docker/config.json
+
+      - name: Add latest tag to images
+        if: |
+          (github.ref_name == 'main' || github.ref_name == 'release') &&
+          github.event_name != 'workflow_dispatch'
+        run: |
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v14:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v14:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v15:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v15:${{needs.tag.outputs.build-tag}} latest
+
+      - name: Cleanup ECR folder
+        run: rm -rf ~/.ecr
 
   push-docker-hub:
     runs-on: [ self-hosted, dev, x64 ]
@@ -776,114 +807,11 @@ jobs:
           crane tag neondatabase/compute-node-v15:${{needs.tag.outputs.build-tag}} latest
           crane tag neondatabase/vm-compute-node-v15:${{needs.tag.outputs.build-tag}} latest
 
-  calculate-deploy-targets:
-    runs-on: [ self-hosted, dev, x64 ]
-    if: |
-      github.ref_name == 'release' &&
-      github.event_name != 'workflow_dispatch'
-    outputs:
-      matrix-include: ${{ steps.set-matrix.outputs.include }}
-    steps:
-      - id: set-matrix
-        run: |
-          if [[ "$GITHUB_REF_NAME" == "release" ]]; then
-            PRODUCTION='{"env_name": "production", "proxy_job": "neon-proxy", "proxy_config": "production.proxy", "storage_broker_ns": "neon-storage-broker", "storage_broker_config": "production.neon-storage-broker", "kubeconfig_secret": "PRODUCTION_KUBECONFIG_DATA", "console_api_key_secret": "NEON_PRODUCTION_API_KEY"}'
-            echo "include=[$PRODUCTION]" >> $GITHUB_OUTPUT
-          else
-            echo "GITHUB_REF_NAME (value '$GITHUB_REF_NAME') is not set to 'release'"
-            exit 1
-          fi
-
-  deploy:
-    runs-on: [ self-hosted, dev, x64 ]
-    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:pinned
-    # We need both storage **and** compute images for deploy, because control plane picks the compute version based on the storage version.
-    # If it notices a fresh storage it may bump the compute version. And if compute image failed to build it may break things badly
-    needs: [ push-docker-hub, calculate-deploy-targets, tag, regress-tests ]
-    if: |
-      github.ref_name == 'release' &&
-      github.event_name != 'workflow_dispatch'
-    defaults:
-      run:
-        shell: bash
-    strategy:
-      matrix:
-        include: ${{fromJSON(needs.calculate-deploy-targets.outputs.matrix-include)}}
-    environment:
-      name: prod-old
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-        with:
-          submodules: true
-          fetch-depth: 0
-
-      - name: Redeploy
-        run: |
-          export DOCKER_TAG=${{needs.tag.outputs.build-tag}}
-          cd "$(pwd)/.github/ansible"
-
-          if [[ "$GITHUB_REF_NAME" == "main" ]]; then
-            ./get_binaries.sh
-          elif [[ "$GITHUB_REF_NAME" == "release" ]]; then
-            RELEASE=true ./get_binaries.sh
-          else
-            echo "GITHUB_REF_NAME (value '$GITHUB_REF_NAME') is not set to either 'main' or 'release'"
-            exit 1
-          fi
-
-          eval $(ssh-agent)
-          echo "${{ secrets.TELEPORT_SSH_KEY }}"  | tr -d '\n'| base64 --decode >ssh-key
-          echo "${{ secrets.TELEPORT_SSH_CERT }}" | tr -d '\n'| base64 --decode >ssh-key-cert.pub
-          chmod 0600 ssh-key
-          ssh-add ssh-key
-          rm -f ssh-key ssh-key-cert.pub
-          ANSIBLE_CONFIG=./ansible.cfg ansible-galaxy collection install sivel.toiletwater
-          ANSIBLE_CONFIG=./ansible.cfg ansible-playbook deploy.yaml -i ${{ matrix.env_name }}.hosts.yaml -e CONSOLE_API_TOKEN=${{ secrets[matrix.console_api_key_secret] }} -e SENTRY_URL_PAGESERVER=${{ secrets.SENTRY_URL_PAGESERVER }} -e SENTRY_URL_SAFEKEEPER=${{ secrets.SENTRY_URL_SAFEKEEPER }}
-          rm -f neon_install.tar.gz .neon_current_version
-
-  deploy-new:
-    runs-on: [ self-hosted, dev, x64 ]
-    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:pinned
-    # We need both storage **and** compute images for deploy, because control plane picks the compute version based on the storage version.
-    # If it notices a fresh storage it may bump the compute version. And if compute image failed to build it may break things badly
-    needs: [ push-docker-hub, tag, regress-tests ]
-    if: |
-      (github.ref_name == 'main') &&
-      github.event_name != 'workflow_dispatch'
-    defaults:
-      run:
-        shell: bash
-    strategy:
-      matrix:
-        target_region: [ eu-west-1, us-east-2 ]
-    environment:
-      name: dev-${{ matrix.target_region }}
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-        with:
-          submodules: true
-          fetch-depth: 0
-
-      - name: Redeploy
-        run: |
-          export DOCKER_TAG=${{needs.tag.outputs.build-tag}}
-          cd "$(pwd)/.github/ansible"
-          if [[ "$GITHUB_REF_NAME" == "main" ]]; then
-            ./get_binaries.sh
-          elif [[ "$GITHUB_REF_NAME" == "release" ]]; then
-            RELEASE=true ./get_binaries.sh
-          else
-            echo "GITHUB_REF_NAME (value '$GITHUB_REF_NAME') is not set to either 'main' or 'release'"
-            exit 1
-          fi
-          ansible-galaxy collection install sivel.toiletwater
-          ansible-playbook deploy.yaml -i staging.${{ matrix.target_region }}.hosts.yaml -e @ssm_config -e CONSOLE_API_TOKEN=${{ secrets.NEON_STAGING_API_KEY }} -e SENTRY_URL_PAGESERVER=${{ secrets.SENTRY_URL_PAGESERVER }} -e SENTRY_URL_SAFEKEEPER=${{ secrets.SENTRY_URL_SAFEKEEPER }}
-          rm -f neon_install.tar.gz .neon_current_version
+      - name: Cleanup ECR folder
+        run: rm -rf ~/.ecr
 
   deploy-pr-test-new:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, small ]
     container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:pinned
     # We need both storage **and** compute images for deploy, because control plane picks the compute version based on the storage version.
     # If it notices a fresh storage it may bump the compute version. And if compute image failed to build it may break things badly
@@ -915,311 +843,40 @@ jobs:
           ansible-playbook deploy.yaml -i staging.${{ matrix.target_region }}.hosts.yaml -e @ssm_config -e CONSOLE_API_TOKEN=${{ secrets.NEON_STAGING_API_KEY }} -e SENTRY_URL_PAGESERVER=${{ secrets.SENTRY_URL_PAGESERVER }} -e SENTRY_URL_SAFEKEEPER=${{ secrets.SENTRY_URL_SAFEKEEPER }}
           rm -f neon_install.tar.gz .neon_current_version
 
-  deploy-prod-new:
-    runs-on: prod
-    container: 093970136003.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
-    # We need both storage **and** compute images for deploy, because control plane picks the compute version based on the storage version.
-    # If it notices a fresh storage it may bump the compute version. And if compute image failed to build it may break things badly
+      - name: Cleanup ansible folder
+        run: rm -rf ~/.ansible
+
+  deploy:
+    runs-on: [ self-hosted, gen3, small ]
+    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
     needs: [ push-docker-hub, tag, regress-tests ]
-    if: |
-      (github.ref_name == 'release') &&
-      github.event_name != 'workflow_dispatch'
-    defaults:
-      run:
-        shell: bash
-    strategy:
-      matrix:
-        target_region: [ us-east-2, us-west-2, eu-central-1, ap-southeast-1 ]
-    environment:
-      name: prod-${{ matrix.target_region }}
+    if: ( github.ref_name == 'main' || github.ref_name == 'release' ) && github.event_name != 'workflow_dispatch'
     steps:
       - name: Checkout
         uses: actions/checkout@v3
         with:
-          submodules: true
+          submodules: false
           fetch-depth: 0
 
-      - name: Redeploy
+      - name: Trigger deploy workflow
+        env:
+          GH_TOKEN: ${{ github.token }}
         run: |
-          export DOCKER_TAG=${{needs.tag.outputs.build-tag}}
-          cd "$(pwd)/.github/ansible"
-
           if [[ "$GITHUB_REF_NAME" == "main" ]]; then
-            ./get_binaries.sh
+            gh workflow run deploy-dev.yml --ref main -f branch=${{ github.sha }} -f dockerTag=${{needs.tag.outputs.build-tag}}
           elif [[ "$GITHUB_REF_NAME" == "release" ]]; then
-            RELEASE=true ./get_binaries.sh
+            gh workflow run deploy-prod.yml --ref release -f branch=${{ github.sha }} -f dockerTag=${{needs.tag.outputs.build-tag}} -f disclamerAcknowledged=true
           else
             echo "GITHUB_REF_NAME (value '$GITHUB_REF_NAME') is not set to either 'main' or 'release'"
             exit 1
           fi
 
-          ansible-galaxy collection install sivel.toiletwater
-          ansible-playbook deploy.yaml -i prod.${{ matrix.target_region }}.hosts.yaml -e @ssm_config -e CONSOLE_API_TOKEN=${{ secrets.NEON_PRODUCTION_API_KEY }} -e SENTRY_URL_PAGESERVER=${{ secrets.SENTRY_URL_PAGESERVER }} -e SENTRY_URL_SAFEKEEPER=${{ secrets.SENTRY_URL_SAFEKEEPER }}
-          rm -f neon_install.tar.gz .neon_current_version
-
-  deploy-proxy:
-    runs-on: [ self-hosted, dev, x64 ]
-    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/base:pinned
-    # Compute image isn't strictly required for proxy deploy, but let's still wait for it to run all deploy jobs consistently.
-    needs: [ push-docker-hub, calculate-deploy-targets, tag, regress-tests ]
-    if: |
-      github.ref_name == 'release' &&
-      github.event_name != 'workflow_dispatch'
-    defaults:
-      run:
-        shell: bash
-    strategy:
-      matrix:
-        include: ${{fromJSON(needs.calculate-deploy-targets.outputs.matrix-include)}}
-    environment:
-      name: prod-old
-    env:
-      KUBECONFIG: .kubeconfig
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-        with:
-          submodules: true
-          fetch-depth: 0
-
-      - name: Add curl
-        run: apt update && apt install curl -y
-
-      - name: Store kubeconfig file
-        run: |
-          echo "${{ secrets[matrix.kubeconfig_secret] }}" | base64 --decode > ${KUBECONFIG}
-          chmod 0600 ${KUBECONFIG}
-
-      - name: Setup helm v3
-        run: |
-          curl -s https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash
-          helm repo add neondatabase https://neondatabase.github.io/helm-charts
-
-      - name: Re-deploy proxy
-        run: |
-          DOCKER_TAG=${{needs.tag.outputs.build-tag}}
-          helm upgrade ${{ matrix.proxy_job }}       neondatabase/neon-proxy --namespace neon-proxy --install --atomic -f .github/helm-values/${{ matrix.proxy_config }}.yaml       --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
-          helm upgrade ${{ matrix.proxy_job }}-scram neondatabase/neon-proxy --namespace neon-proxy --install --atomic -f .github/helm-values/${{ matrix.proxy_config }}-scram.yaml --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
-
-  deploy-storage-broker:
-    name: deploy storage broker on old staging and old prod
-    runs-on: [ self-hosted, dev, x64 ]
-    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/base:pinned
-    # Compute image isn't strictly required for proxy deploy, but let's still wait for it to run all deploy jobs consistently.
-    needs: [ push-docker-hub, calculate-deploy-targets, tag, regress-tests ]
-    if: |
-      github.ref_name == 'release' &&
-      github.event_name != 'workflow_dispatch'
-    defaults:
-      run:
-        shell: bash
-    strategy:
-      matrix:
-        include: ${{fromJSON(needs.calculate-deploy-targets.outputs.matrix-include)}}
-    environment:
-      name: prod-old
-    env:
-      KUBECONFIG: .kubeconfig
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-        with:
-          submodules: true
-          fetch-depth: 0
-
-      - name: Add curl
-        run: apt update && apt install curl -y
-
-      - name: Store kubeconfig file
-        run: |
-          echo "${{ secrets[matrix.kubeconfig_secret] }}" | base64 --decode > ${KUBECONFIG}
-          chmod 0600 ${KUBECONFIG}
-
-      - name: Setup helm v3
-        run: |
-          curl -s https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash
-          helm repo add neondatabase https://neondatabase.github.io/helm-charts
-
-      - name: Deploy storage-broker
-        run:
-          helm upgrade neon-storage-broker neondatabase/neon-storage-broker --namespace ${{ matrix.storage_broker_ns }} --create-namespace --install --atomic -f .github/helm-values/${{ matrix.storage_broker_config }}.yaml --set image.tag=${{ needs.tag.outputs.build-tag }} --set settings.sentryUrl=${{ secrets.SENTRY_URL_BROKER }} --wait --timeout 5m0s
-
-  deploy-proxy-new:
-    runs-on: [ self-hosted, dev, x64 ]
-    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:pinned
-    # Compute image isn't strictly required for proxy deploy, but let's still wait for it to run all deploy jobs consistently.
-    needs: [ push-docker-hub, tag, regress-tests ]
-    if: |
-      (github.ref_name == 'main') &&
-      github.event_name != 'workflow_dispatch'
-    defaults:
-      run:
-        shell: bash
-    strategy:
-      matrix:
-        include:
-          - target_region:  us-east-2
-            target_cluster: dev-us-east-2-beta
-            deploy_link_proxy: true
-            deploy_legacy_scram_proxy: true
-          - target_region:  eu-west-1
-            target_cluster: dev-eu-west-1-zeta
-            deploy_link_proxy: false
-            deploy_legacy_scram_proxy: false
-    environment:
-      name: dev-${{ matrix.target_region }}
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-        with:
-          submodules: true
-          fetch-depth: 0
-
-      - name: Configure environment
-        run: |
-          helm repo add neondatabase https://neondatabase.github.io/helm-charts
-          aws --region ${{ matrix.target_region }} eks update-kubeconfig --name  ${{ matrix.target_cluster }}
-
-      - name: Re-deploy scram proxy
-        run: |
-          DOCKER_TAG=${{needs.tag.outputs.build-tag}}
-          helm upgrade neon-proxy-scram neondatabase/neon-proxy --namespace neon-proxy --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-proxy-scram.yaml --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
-
-      - name: Re-deploy link proxy
-        if: matrix.deploy_link_proxy
-        run: |
-          DOCKER_TAG=${{needs.tag.outputs.build-tag}}
-          helm upgrade neon-proxy-link neondatabase/neon-proxy --namespace neon-proxy --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-proxy-link.yaml --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
-
-      - name: Re-deploy legacy scram proxy
-        if: matrix.deploy_legacy_scram_proxy
-        run: |
-          DOCKER_TAG=${{needs.tag.outputs.build-tag}}
-          helm upgrade neon-proxy-scram-legacy neondatabase/neon-proxy --namespace neon-proxy --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-proxy-scram-legacy.yaml --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
-
-  deploy-storage-broker-dev-new:
-    runs-on: [ self-hosted, dev, x64 ]
-    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:pinned
-    # Compute image isn't strictly required for proxy deploy, but let's still wait for it to run all deploy jobs consistently.
-    needs: [ push-docker-hub, tag, regress-tests ]
-    if: |
-      (github.ref_name == 'main') &&
-      github.event_name != 'workflow_dispatch'
-    defaults:
-      run:
-        shell: bash
-    strategy:
-      matrix:
-        include:
-          - target_region:  us-east-2
-            target_cluster: dev-us-east-2-beta
-          - target_region:  eu-west-1
-            target_cluster: dev-eu-west-1-zeta
-    environment:
-      name: dev-${{ matrix.target_region }}
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-        with:
-          submodules: true
-          fetch-depth: 0
-
-      - name: Configure environment
-        run: |
-          helm repo add neondatabase https://neondatabase.github.io/helm-charts
-          aws --region ${{ matrix.target_region }} eks update-kubeconfig --name  ${{ matrix.target_cluster }}
-
-      - name: Deploy storage-broker
-        run:
-          helm upgrade neon-storage-broker-lb neondatabase/neon-storage-broker --namespace neon-storage-broker-lb --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-storage-broker.yaml --set image.tag=${{ needs.tag.outputs.build-tag }} --set settings.sentryUrl=${{ secrets.SENTRY_URL_BROKER }} --wait --timeout 5m0s
-
-  deploy-proxy-prod-new:
-    runs-on: prod
-    container: 093970136003.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
-    # Compute image isn't strictly required for proxy deploy, but let's still wait for it to run all deploy jobs consistently.
-    needs: [ push-docker-hub, tag, regress-tests ]
-    if: |
-      (github.ref_name == 'release') &&
-      github.event_name != 'workflow_dispatch'
-    defaults:
-      run:
-        shell: bash
-    strategy:
-      matrix:
-        include:
-          - target_region:  us-east-2
-            target_cluster: prod-us-east-2-delta
-          - target_region:  us-west-2
-            target_cluster: prod-us-west-2-eta
-          - target_region: eu-central-1
-            target_cluster: prod-eu-central-1-gamma
-          - target_region: ap-southeast-1
-            target_cluster: prod-ap-southeast-1-epsilon
-    environment:
-      name: prod-${{ matrix.target_region }}
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-        with:
-          submodules: true
-          fetch-depth: 0
-
-      - name: Configure environment
-        run: |
-          helm repo add neondatabase https://neondatabase.github.io/helm-charts
-          aws --region ${{ matrix.target_region }} eks update-kubeconfig --name  ${{ matrix.target_cluster }}
-
-      - name: Re-deploy proxy
-        run: |
-          DOCKER_TAG=${{needs.tag.outputs.build-tag}}
-          helm upgrade neon-proxy-scram neondatabase/neon-proxy --namespace neon-proxy --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-proxy-scram.yaml --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
-
-  deploy-storage-broker-prod-new:
-    runs-on: prod
-    container: 093970136003.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
-    # Compute image isn't strictly required for proxy deploy, but let's still wait for it to run all deploy jobs consistently.
-    needs: [ push-docker-hub, tag, regress-tests ]
-    if: |
-      (github.ref_name == 'release') &&
-      github.event_name != 'workflow_dispatch'
-    defaults:
-      run:
-        shell: bash
-    strategy:
-      matrix:
-        include:
-          - target_region:  us-east-2
-            target_cluster: prod-us-east-2-delta
-          - target_region:  us-west-2
-            target_cluster: prod-us-west-2-eta
-          - target_region: eu-central-1
-            target_cluster: prod-eu-central-1-gamma
-          - target_region: ap-southeast-1
-            target_cluster: prod-ap-southeast-1-epsilon
-    environment:
-      name: prod-${{ matrix.target_region }}
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-        with:
-          submodules: true
-          fetch-depth: 0
-
-      - name: Configure environment
-        run: |
-          helm repo add neondatabase https://neondatabase.github.io/helm-charts
-          aws --region ${{ matrix.target_region }} eks update-kubeconfig --name  ${{ matrix.target_cluster }}
-
-      - name: Deploy storage-broker
-        run:
-          helm upgrade neon-storage-broker-lb neondatabase/neon-storage-broker --namespace neon-storage-broker-lb --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-storage-broker.yaml --set image.tag=${{ needs.tag.outputs.build-tag }} --set settings.sentryUrl=${{ secrets.SENTRY_URL_BROKER }} --wait --timeout 5m0s
-
   promote-compatibility-data:
-    runs-on: [ self-hosted, dev, x64 ]
+    runs-on: [ self-hosted, gen3, small ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
-    needs: [ deploy, deploy-proxy ]
+    needs: [ push-docker-hub, tag, regress-tests ]
     if: github.ref_name == 'release' && github.event_name != 'workflow_dispatch'
     steps:
       - name: Promote compatibility snapshot for the release

.github/workflows/deploy-dev.yml

@@ -0,0 +1,179 @@
+name: Neon Deploy dev
+
+on:
+  workflow_dispatch:
+    inputs:
+      dockerTag:
+        description: 'Docker tag to deploy'
+        required: true
+        type: string
+      branch:
+        description: 'Branch or commit used for deploy scripts and configs'
+        required: true
+        type: string
+        default: 'main'
+      deployStorage:
+        description: 'Deploy storage'
+        required: true
+        type: boolean
+        default: true
+      deployProxy:
+        description: 'Deploy proxy'
+        required: true
+        type: boolean
+        default: true
+      deployStorageBroker:
+        description: 'Deploy storage-broker'
+        required: true
+        type: boolean
+        default: true
+
+env:
+  AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
+  AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
+
+concurrency:
+  group: deploy-dev
+  cancel-in-progress: false
+
+jobs:
+  deploy-storage-new:
+    runs-on: [ self-hosted, gen3, small ]
+    container:
+      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:pinned
+      options: --user root --privileged
+    if: inputs.deployStorage
+    defaults:
+      run:
+        shell: bash
+    strategy:
+      matrix:
+        target_region: [ eu-west-1, us-east-2 ]
+    environment:
+      name: dev-${{ matrix.target_region }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+        with:
+          submodules: true
+          fetch-depth: 0
+          ref: ${{ inputs.branch }}
+
+      - name: Redeploy
+        run: |
+          export DOCKER_TAG=${{ inputs.dockerTag }}
+          cd "$(pwd)/.github/ansible"
+
+          ./get_binaries.sh
+
+          ansible-galaxy collection install sivel.toiletwater
+          ansible-playbook -v deploy.yaml -i staging.${{ matrix.target_region }}.hosts.yaml -e @ssm_config -e CONSOLE_API_TOKEN=${{ secrets.NEON_STAGING_API_KEY }} -e SENTRY_URL_PAGESERVER=${{ secrets.SENTRY_URL_PAGESERVER }} -e SENTRY_URL_SAFEKEEPER=${{ secrets.SENTRY_URL_SAFEKEEPER }}
+          rm -f neon_install.tar.gz .neon_current_version
+
+      - name: Cleanup ansible folder
+        run: rm -rf ~/.ansible
+
+  deploy-proxy-new:
+    runs-on: [ self-hosted, gen3, small ]
+    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:pinned
+    if: inputs.deployProxy
+    defaults:
+      run:
+        shell: bash
+    strategy:
+      matrix:
+        include:
+          - target_region:  us-east-2
+            target_cluster: dev-us-east-2-beta
+            deploy_link_proxy: true
+            deploy_legacy_scram_proxy: true
+          - target_region:  eu-west-1
+            target_cluster: dev-eu-west-1-zeta
+            deploy_link_proxy: false
+            deploy_legacy_scram_proxy: false
+    environment:
+      name: dev-${{ matrix.target_region }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+        with:
+          submodules: true
+          fetch-depth: 0
+          ref: ${{ inputs.branch }}
+  
+      - name: Configure AWS Credentials
+        uses: aws-actions/configure-aws-credentials@v1-node16
+        with:
+          role-to-assume: arn:aws:iam::369495373322:role/github-runner
+          aws-region: eu-central-1
+          role-skip-session-tagging: true
+          role-duration-seconds: 1800
+  
+      - name: Configure environment
+        run: |
+          helm repo add neondatabase https://neondatabase.github.io/helm-charts
+          aws --region ${{ matrix.target_region }} eks update-kubeconfig --name  ${{ matrix.target_cluster }}
+  
+      - name: Re-deploy scram proxy
+        run: |
+          DOCKER_TAG=${{ inputs.dockerTag }}
+          helm upgrade neon-proxy-scram neondatabase/neon-proxy --namespace neon-proxy --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-proxy-scram.yaml --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
+  
+      - name: Re-deploy link proxy
+        if: matrix.deploy_link_proxy
+        run: |
+          DOCKER_TAG=${{ inputs.dockerTag }}
+          helm upgrade neon-proxy-link neondatabase/neon-proxy --namespace neon-proxy --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-proxy-link.yaml --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
+  
+      - name: Re-deploy legacy scram proxy
+        if: matrix.deploy_legacy_scram_proxy
+        run: |
+          DOCKER_TAG=${{ inputs.dockerTag }}
+          helm upgrade neon-proxy-scram-legacy neondatabase/neon-proxy --namespace neon-proxy --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-proxy-scram-legacy.yaml --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
+  
+      - name: Cleanup helm folder
+        run: rm -rf ~/.cache
+  
+  deploy-storage-broker-new:
+    runs-on: [ self-hosted, gen3, small ]
+    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:pinned
+    if: inputs.deployStorageBroker
+    defaults:
+      run:
+        shell: bash
+    strategy:
+      matrix:
+        include:
+          - target_region:  us-east-2
+            target_cluster: dev-us-east-2-beta
+          - target_region:  eu-west-1
+            target_cluster: dev-eu-west-1-zeta
+    environment:
+      name: dev-${{ matrix.target_region }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+        with:
+          submodules: true
+          fetch-depth: 0
+          ref: ${{ inputs.branch }}
+  
+      - name: Configure AWS Credentials
+        uses: aws-actions/configure-aws-credentials@v1-node16
+        with:
+          role-to-assume: arn:aws:iam::369495373322:role/github-runner
+          aws-region: eu-central-1
+          role-skip-session-tagging: true
+          role-duration-seconds: 1800
+  
+      - name: Configure environment
+        run: |
+          helm repo add neondatabase https://neondatabase.github.io/helm-charts
+          aws --region ${{ matrix.target_region }} eks update-kubeconfig --name  ${{ matrix.target_cluster }}
+  
+      - name: Deploy storage-broker
+        run:
+          helm upgrade neon-storage-broker-lb neondatabase/neon-storage-broker --namespace neon-storage-broker-lb --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-storage-broker.yaml --set image.tag=${{ inputs.dockerTag }} --set settings.sentryUrl=${{ secrets.SENTRY_URL_BROKER }} --wait --timeout 5m0s
+  
+      - name: Cleanup helm folder
+        run: rm -rf ~/.cache

.github/workflows/deploy-prod.yml

@@ -0,0 +1,167 @@
+name: Neon Deploy prod
+
+on:
+  workflow_dispatch:
+    inputs:
+      dockerTag:
+        description: 'Docker tag to deploy'
+        required: true
+        type: string
+      branch:
+        description: 'Branch or commit used for deploy scripts and configs'
+        required: true
+        type: string
+        default: 'release'
+      deployStorage:
+        description: 'Deploy storage'
+        required: true
+        type: boolean
+        default: true
+      deployProxy:
+        description: 'Deploy proxy'
+        required: true
+        type: boolean
+        default: true
+      deployStorageBroker:
+        description: 'Deploy storage-broker'
+        required: true
+        type: boolean
+        default: true
+      disclamerAcknowledged:
+        description: 'I confirm that there is an emergency and I can not use regular release workflow'
+        required: true
+        type: boolean
+        default: false
+
+concurrency:
+  group: deploy-prod
+  cancel-in-progress: false
+
+jobs:
+  deploy-prod-new:
+    runs-on: prod
+    container:
+      image: 093970136003.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
+      options: --user root --privileged
+    if: inputs.deployStorage && inputs.disclamerAcknowledged
+    defaults:
+      run:
+        shell: bash
+    strategy:
+      matrix:
+        target_region: [ us-east-2, us-west-2, eu-central-1, ap-southeast-1 ]
+    environment:
+      name: prod-${{ matrix.target_region }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+        with:
+          submodules: true
+          fetch-depth: 0
+          ref: ${{ inputs.branch }}
+
+      - name: Redeploy
+        run: |
+          export DOCKER_TAG=${{ inputs.dockerTag }}
+          cd "$(pwd)/.github/ansible"
+
+          ./get_binaries.sh
+
+          ansible-galaxy collection install sivel.toiletwater
+          ansible-playbook -v deploy.yaml -i prod.${{ matrix.target_region }}.hosts.yaml -e @ssm_config -e CONSOLE_API_TOKEN=${{ secrets.NEON_PRODUCTION_API_KEY }} -e SENTRY_URL_PAGESERVER=${{ secrets.SENTRY_URL_PAGESERVER }} -e SENTRY_URL_SAFEKEEPER=${{ secrets.SENTRY_URL_SAFEKEEPER }}
+          rm -f neon_install.tar.gz .neon_current_version
+
+  deploy-proxy-prod-new:
+    runs-on: prod
+    container: 093970136003.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
+    if: inputs.deployProxy && inputs.disclamerAcknowledged
+    defaults:
+      run:
+        shell: bash
+    strategy:
+      matrix:
+        include:
+          - target_region:  us-east-2
+            target_cluster: prod-us-east-2-delta
+            deploy_link_proxy: true
+            deploy_legacy_scram_proxy: false
+          - target_region:  us-west-2
+            target_cluster: prod-us-west-2-eta
+            deploy_link_proxy: false
+            deploy_legacy_scram_proxy: true
+          - target_region: eu-central-1
+            target_cluster: prod-eu-central-1-gamma
+            deploy_link_proxy: false
+            deploy_legacy_scram_proxy: false
+          - target_region: ap-southeast-1
+            target_cluster: prod-ap-southeast-1-epsilon
+            deploy_link_proxy: false
+            deploy_legacy_scram_proxy: false
+    environment:
+      name: prod-${{ matrix.target_region }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+        with:
+          submodules: true
+          fetch-depth: 0
+          ref: ${{ inputs.branch }}
+
+      - name: Configure environment
+        run: |
+          helm repo add neondatabase https://neondatabase.github.io/helm-charts
+          aws --region ${{ matrix.target_region }} eks update-kubeconfig --name  ${{ matrix.target_cluster }}
+
+      - name: Re-deploy scram proxy
+        run: |
+          DOCKER_TAG=${{ inputs.dockerTag }}
+          helm upgrade neon-proxy-scram neondatabase/neon-proxy --namespace neon-proxy --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-proxy-scram.yaml --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
+
+      - name: Re-deploy link proxy
+        if: matrix.deploy_link_proxy
+        run: |
+          DOCKER_TAG=${{ inputs.dockerTag }}
+          helm upgrade neon-proxy-link neondatabase/neon-proxy --namespace neon-proxy --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-proxy-link.yaml --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
+
+      - name: Re-deploy legacy scram proxy
+        if: matrix.deploy_legacy_scram_proxy
+        run: |
+          DOCKER_TAG=${{ inputs.dockerTag }}
+          helm upgrade neon-proxy-scram-legacy neondatabase/neon-proxy --namespace neon-proxy --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-proxy-scram-legacy.yaml --set image.tag=${DOCKER_TAG} --set settings.sentryUrl=${{ secrets.SENTRY_URL_PROXY }} --wait --timeout 15m0s
+
+  deploy-storage-broker-prod-new:
+    runs-on: prod
+    container: 093970136003.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
+    if: inputs.deployStorageBroker && inputs.disclamerAcknowledged
+    defaults:
+      run:
+        shell: bash
+    strategy:
+      matrix:
+        include:
+          - target_region:  us-east-2
+            target_cluster: prod-us-east-2-delta
+          - target_region:  us-west-2
+            target_cluster: prod-us-west-2-eta
+          - target_region: eu-central-1
+            target_cluster: prod-eu-central-1-gamma
+          - target_region: ap-southeast-1
+            target_cluster: prod-ap-southeast-1-epsilon
+    environment:
+      name: prod-${{ matrix.target_region }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+        with:
+          submodules: true
+          fetch-depth: 0
+          ref: ${{ inputs.branch }}
+
+      - name: Configure environment
+        run: |
+          helm repo add neondatabase https://neondatabase.github.io/helm-charts
+          aws --region ${{ matrix.target_region }} eks update-kubeconfig --name  ${{ matrix.target_cluster }}
+
+      - name: Deploy storage-broker
+        run:
+          helm upgrade neon-storage-broker-lb neondatabase/neon-storage-broker --namespace neon-storage-broker-lb --create-namespace --install --atomic -f .github/helm-values/${{ matrix.target_cluster }}.neon-storage-broker.yaml --set image.tag=${{ inputs.dockerTag }} --set settings.sentryUrl=${{ secrets.SENTRY_URL_BROKER }} --wait --timeout 5m0s

.github/workflows/neon_extra_builds.yml

@@ -4,6 +4,7 @@ on:
   push:
     branches:
     - main
+  pull_request:
 
 defaults:
   run:
@@ -20,6 +21,7 @@ env:
 
 jobs:
   check-macos-build:
+    if: github.ref_name == 'main' || contains(github.event.pull_request.labels.*.name, 'run-extra-build-macos')
     timeout-minutes: 90
     runs-on: macos-latest
 
@@ -93,11 +95,16 @@ jobs:
         run: ./run_clippy.sh
 
   gather-rust-build-stats:
-    timeout-minutes: 90
-    runs-on: ubuntu-latest
+    if: github.ref_name == 'main' || contains(github.event.pull_request.labels.*.name, 'run-extra-build-stats')
+    runs-on: [ self-hosted, gen3, large ]
+    container:
+      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
+      options: --init
 
     env:
       BUILD_TYPE: release
+      # remove the cachepot wrapper and build without crate caches
+      RUSTC_WRAPPER: ""
       # build with incremental compilation produce partial results
       # so do not attempt to cache this build, also disable the incremental compilation
       CARGO_INCREMENTAL: 0
@@ -109,11 +116,6 @@ jobs:
           submodules: true
           fetch-depth: 1
 
-      - name: Install Ubuntu postgres dependencies
-        run: |
-          sudo apt update
-          sudo apt install build-essential libreadline-dev zlib1g-dev flex bison libseccomp-dev libssl-dev protobuf-compiler
-
       # Some of our rust modules use FFI and need those to be checked
       - name: Get postgres headers
         run: make postgres-headers -j$(nproc)
@@ -122,7 +124,31 @@ jobs:
         run: cargo build --all --release --timings
 
       - name: Upload the build stats
-        uses: actions/upload-artifact@v3
+        id: upload-stats
+        env:
+          BUCKET: neon-github-public-dev
+          SHA: ${{ github.event.pull_request.head.sha || github.sha }}
+          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
+          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
+        run: |
+          REPORT_URL=https://${BUCKET}.s3.amazonaws.com/build-stats/${SHA}/${GITHUB_RUN_ID}/cargo-timing.html
+          aws s3 cp --only-show-errors ./target/cargo-timings/cargo-timing.html "s3://${BUCKET}/build-stats/${SHA}/${GITHUB_RUN_ID}/"
+          echo "report-url=${REPORT_URL}" >> $GITHUB_OUTPUT
+
+      - name: Publish build stats report
+        uses: actions/github-script@v6
+        env:
+          REPORT_URL: ${{ steps.upload-stats.outputs.report-url }}
+          SHA: ${{ github.event.pull_request.head.sha || github.sha }}
         with:
-          name: neon-${{ runner.os }}-release-build-stats
-          path: ./target/cargo-timings/
+          script: |
+            const { REPORT_URL, SHA } = process.env
+
+            await github.rest.repos.createCommitStatus({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              sha: `${SHA}`,
+              state: 'success',
+              target_url: `${REPORT_URL}`,
+              context: `Build stats (release)`,
+            })

.github/workflows/release.yml

@@ -0,0 +1,33 @@
+name: Create Release Branch
+
+on:
+  schedule:
+    - cron: '0 10 * * 2'
+
+jobs:
+  create_release_branch:
+    runs-on: [ubuntu-latest]
+
+    steps:
+    - name: Check out code
+      uses: actions/checkout@v3
+      with:
+        ref: main
+
+    - name: Get current date
+      id: date
+      run: echo "date=$(date +'%Y-%m-%d')" >> $GITHUB_OUTPUT
+
+    - name: Create release branch
+      run: git checkout -b releases/${{ steps.date.outputs.date }}
+
+    - name: Push new branch
+      run: git push origin releases/${{ steps.date.outputs.date }}
+
+    - name: Create pull request into release
+      uses: thomaseizinger/create-pull-request@e3972219c86a56550fb70708d96800d8e24ba862 # 1.3.0
+      with:
+        GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        head: releases/${{ steps.date.outputs.date }}
+        base: release
+        title: Release ${{ steps.date.outputs.date }}

Cargo.toml

@@ -7,6 +7,7 @@ members = [
     "safekeeper",
     "storage_broker",
     "workspace_hack",
+    "trace",
     "libs/*",
 ]
 
@@ -31,12 +32,15 @@ bstr = "1.0"
 byteorder = "1.4"
 bytes = "1.0"
 chrono = { version = "0.4", default-features = false, features = ["clock"] }
-clap = "4.0"
+clap = { version = "4.0", features = ["derive"] }
 close_fds = "0.3.2"
 comfy-table = "6.1"
 const_format = "0.2"
 crc32c = "0.6"
 crossbeam-utils = "0.8.5"
+either = "1.8"
+enum-map = "2.4.2"
+enumset = "1.0.12"
 fail = "0.5.0"
 fs2 = "0.4.3"
 futures = "0.3"
@@ -44,6 +48,7 @@ futures-core = "0.3"
 futures-util = "0.3"
 git-version = "0.3"
 hashbrown = "0.13"
+hashlink = "0.8.1"
 hex = "0.4"
 hex-literal = "0.3"
 hmac = "0.12.1"
@@ -61,6 +66,9 @@ nix = "0.26"
 notify = "5.0.0"
 num-traits = "0.2.15"
 once_cell = "1.13"
+opentelemetry = "0.18.0"
+opentelemetry-otlp = { version = "0.11.0", default_features=false, features = ["http-proto", "trace", "http", "reqwest-client"] }
+opentelemetry-semantic-conventions = "0.10.0"
 parking_lot = "0.12"
 pin-project-lite = "0.2"
 prometheus = {version = "0.13", default_features=false, features = ["process"]} # removes protobuf dependency
@@ -68,6 +76,8 @@ prost = "0.11"
 rand = "0.8"
 regex = "1.4"
 reqwest = { version = "0.11", default-features = false, features = ["rustls-tls"] }
+reqwest-tracing = { version = "0.4.0", features = ["opentelemetry_0_18"] }
+reqwest-middleware = "0.2.0"
 routerify = "3"
 rpds = "0.12.0"
 rustls = "0.20"
@@ -84,6 +94,7 @@ socket2 = "0.4.4"
 strum = "0.24"
 strum_macros = "0.24"
 svg_fmt = "0.4.1"
+sync_wrapper = "0.1.2"
 tar = "0.4"
 thiserror = "1.0"
 tls-listener = { version = "0.6", features = ["rustls", "hyper-h1"] }
@@ -96,6 +107,7 @@ toml = "0.5"
 toml_edit = { version = "0.17", features = ["easy"] }
 tonic = {version = "0.8", features = ["tls", "tls-roots"]}
 tracing = "0.1"
+tracing-opentelemetry = "0.18.0"
 tracing-subscriber = { version = "0.3", features = ["env-filter"] }
 url = "2.2"
 uuid = { version = "1.2", features = ["v4", "serde"] }
@@ -114,6 +126,9 @@ postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", re
 tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="43e6db254a97fdecbce33d8bc0890accfd74495e" }
 tokio-tar = { git = "https://github.com/neondatabase/tokio-tar.git", rev="404df61437de0feef49ba2ccdbdd94eb8ad6e142" }
 
+## Other git libraries
+heapless = { default-features=false, features=[], git = "https://github.com/japaric/heapless.git", rev = "644653bf3b831c6bb4963be2de24804acf5e5001" } # upstream release pending
+
 ## Local libraries
 consumption_metrics = { version = "0.1", path = "./libs/consumption_metrics/" }
 metrics = { version = "0.1", path = "./libs/metrics/" }
@@ -125,6 +140,7 @@ remote_storage = { version = "0.1", path = "./libs/remote_storage/" }
 safekeeper_api = { version = "0.1", path = "./libs/safekeeper_api" }
 storage_broker = { version = "0.1", path = "./storage_broker/" } # Note: main broker code is inside the binary crate, so linking with the library shouldn't be heavy.
 tenant_size_model = { version = "0.1", path = "./libs/tenant_size_model/" }
+tracing-utils = { version = "0.1", path = "./libs/tracing-utils/" }
 utils = { version = "0.1", path = "./libs/utils/" }
 
 ## Common library dependency
@@ -134,7 +150,7 @@ workspace_hack = { version = "0.1", path = "./workspace_hack/" }
 criterion = "0.4"
 rcgen = "0.10"
 rstest = "0.16"
-tempfile = "3.2"
+tempfile = "3.4"
 tonic-build = "0.8"
 
 # This is only needed for proxy's tests.

Dockerfile.compute-node

@@ -0,0 +1,389 @@
+ARG PG_VERSION
+ARG REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com
+ARG IMAGE=rust
+ARG TAG=pinned
+
+#########################################################################################
+#
+# Layer "build-deps"
+#
+#########################################################################################
+FROM debian:bullseye-slim AS build-deps
+RUN apt update &&  \
+    apt install -y git autoconf automake libtool build-essential bison flex libreadline-dev \
+    zlib1g-dev libxml2-dev libcurl4-openssl-dev libossp-uuid-dev wget pkg-config libssl-dev \
+    libicu-dev libxslt1-dev
+
+#########################################################################################
+#
+# Layer "pg-build"
+# Build Postgres from the neon postgres repository.
+#
+#########################################################################################
+FROM build-deps AS pg-build
+ARG PG_VERSION
+COPY vendor/postgres-${PG_VERSION} postgres
+RUN cd postgres && \
+    ./configure CFLAGS='-O2 -g3' --enable-debug --with-openssl --with-uuid=ossp --with-icu \
+    --with-libxml --with-libxslt && \
+    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s install && \
+    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C contrib/ install && \
+    # Install headers
+    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C src/include install && \
+    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C src/interfaces/libpq install && \
+    # Enable some of contrib extensions
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/autoinc.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/bloom.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/earthdistance.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/insert_username.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/intagg.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/moddatetime.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgrowlocks.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgstattuple.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/refint.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/xml2.control
+
+#########################################################################################
+#
+# Layer "postgis-build"
+# Build PostGIS from the upstream PostGIS mirror.
+#
+#########################################################################################
+FROM build-deps AS postgis-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+RUN apt update && \
+    apt install -y cmake gdal-bin libboost-dev libboost-thread-dev libboost-filesystem-dev \
+    libboost-system-dev libboost-iostreams-dev libboost-program-options-dev libboost-timer-dev \
+    libcgal-dev libgdal-dev libgmp-dev libmpfr-dev libopenscenegraph-dev libprotobuf-c-dev \
+    protobuf-c-compiler xsltproc
+
+# SFCGAL > 1.3 requires CGAL > 5.2, Bullseye's libcgal-dev is 5.2
+RUN wget https://gitlab.com/Oslandia/SFCGAL/-/archive/v1.3.10/SFCGAL-v1.3.10.tar.gz -O SFCGAL.tar.gz && \
+    mkdir sfcgal-src && cd sfcgal-src && tar xvzf ../SFCGAL.tar.gz --strip-components=1 -C . && \
+    cmake . && make -j $(getconf _NPROCESSORS_ONLN) && \
+    DESTDIR=/sfcgal make install -j $(getconf _NPROCESSORS_ONLN) && \
+    make clean && cp -R /sfcgal/* /
+
+ENV PATH "/usr/local/pgsql/bin:$PATH"
+
+RUN wget https://download.osgeo.org/postgis/source/postgis-3.3.2.tar.gz -O postgis.tar.gz && \
+    mkdir postgis-src && cd postgis-src && tar xvzf ../postgis.tar.gz --strip-components=1 -C . && \
+    ./autogen.sh && \
+    ./configure --with-sfcgal=/usr/local/bin/sfcgal-config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    cd extensions/postgis && \
+    make clean && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis_raster.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis_sfcgal.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis_tiger_geocoder.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis_topology.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/address_standardizer.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/address_standardizer_data_us.control
+
+RUN wget https://github.com/pgRouting/pgrouting/archive/v3.4.2.tar.gz -O pgrouting.tar.gz && \
+    mkdir pgrouting-src && cd pgrouting-src && tar xvzf ../pgrouting.tar.gz --strip-components=1 -C . && \
+    mkdir build && \
+    cd build && \
+    cmake .. && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgrouting.control
+
+#########################################################################################
+#
+# Layer "plv8-build"
+# Build plv8
+#
+#########################################################################################
+FROM build-deps AS plv8-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+RUN apt update && \
+    apt install -y ninja-build python3-dev libncurses5 binutils clang
+
+RUN wget https://github.com/plv8/plv8/archive/refs/tags/v3.1.5.tar.gz -O plv8.tar.gz && \
+    mkdir plv8-src && cd plv8-src && tar xvzf ../plv8.tar.gz --strip-components=1 -C . && \
+    export PATH="/usr/local/pgsql/bin:$PATH" && \
+    make DOCKER=1 -j $(getconf _NPROCESSORS_ONLN) install && \
+    rm -rf /plv8-* && \
+    find /usr/local/pgsql/ -name "plv8-*.so" | xargs strip && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/plv8.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/plcoffee.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/plls.control
+
+#########################################################################################
+#
+# Layer "h3-pg-build"
+# Build h3_pg
+#
+#########################################################################################
+FROM build-deps AS h3-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+# packaged cmake is too old
+RUN wget https://github.com/Kitware/CMake/releases/download/v3.24.2/cmake-3.24.2-linux-x86_64.sh \
+      -q -O /tmp/cmake-install.sh \
+      && chmod u+x /tmp/cmake-install.sh \
+      && /tmp/cmake-install.sh --skip-license --prefix=/usr/local/ \
+      && rm /tmp/cmake-install.sh
+
+RUN wget https://github.com/uber/h3/archive/refs/tags/v4.1.0.tar.gz -O h3.tar.gz && \
+    mkdir h3-src && cd h3-src && tar xvzf ../h3.tar.gz --strip-components=1 -C . && \
+    mkdir build && cd build && \
+    cmake .. -DCMAKE_BUILD_TYPE=Release && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    DESTDIR=/h3 make install && \
+    cp -R /h3/usr / && \
+    rm -rf build
+
+RUN wget https://github.com/zachasme/h3-pg/archive/refs/tags/v4.1.2.tar.gz -O h3-pg.tar.gz && \
+    mkdir h3-pg-src && cd h3-pg-src && tar xvzf ../h3-pg.tar.gz --strip-components=1 -C . && \
+    export PATH="/usr/local/pgsql/bin:$PATH" && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/h3.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/h3_postgis.control
+
+#########################################################################################
+#
+# Layer "unit-pg-build"
+# compile unit extension
+#
+#########################################################################################
+FROM build-deps AS unit-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.7.tar.gz -O postgresql-unit.tar.gz && \
+    mkdir postgresql-unit-src && cd postgresql-unit-src && tar xvzf ../postgresql-unit.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    # unit extension's "create extension" script relies on absolute install path to fill some reference tables.
+    # We move the extension from '/usr/local/pgsql/' to '/usr/local/'  after it is build. So we need to adjust the path.
+    # This one-liner removes pgsql/ part of the path.
+    # NOTE: Other extensions that rely on MODULEDIR variable after building phase will need the same fix.
+    find /usr/local/pgsql/share/extension/ -name "unit*.sql" -print0 | xargs -0 sed -i "s|pgsql/||g" && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/unit.control
+
+#########################################################################################
+#
+# Layer "vector-pg-build"
+# compile pgvector extension
+#
+#########################################################################################
+FROM build-deps AS vector-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.4.0.tar.gz -O pgvector.tar.gz && \
+    mkdir pgvector-src && cd pgvector-src && tar xvzf ../pgvector.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/vector.control
+
+#########################################################################################
+#
+# Layer "pgjwt-pg-build"
+# compile pgjwt extension
+#
+#########################################################################################
+FROM build-deps AS pgjwt-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+# 9742dab1b2f297ad3811120db7b21451bca2d3c9 made on 13/11/2021
+RUN wget https://github.com/michelp/pgjwt/archive/9742dab1b2f297ad3811120db7b21451bca2d3c9.tar.gz -O pgjwt.tar.gz && \
+    mkdir pgjwt-src && cd pgjwt-src && tar xvzf ../pgjwt.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgjwt.control
+
+#########################################################################################
+#
+# Layer "hypopg-pg-build"
+# compile hypopg extension
+#
+#########################################################################################
+FROM build-deps AS hypopg-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN wget https://github.com/HypoPG/hypopg/archive/refs/tags/1.3.1.tar.gz -O hypopg.tar.gz && \
+    mkdir hypopg-src && cd hypopg-src && tar xvzf ../hypopg.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/hypopg.control
+
+#########################################################################################
+#
+# Layer "pg-hashids-pg-build"
+# compile pg_hashids extension
+#
+#########################################################################################
+FROM build-deps AS pg-hashids-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN wget https://github.com/iCyberon/pg_hashids/archive/refs/tags/v1.2.1.tar.gz -O pg_hashids.tar.gz && \
+    mkdir pg_hashids-src && cd pg_hashids-src && tar xvzf ../pg_hashids.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pg_hashids.control
+
+#########################################################################################
+# 
+# Layer "rust extensions"
+# This layer is used to build `pgx` deps
+#
+#########################################################################################
+FROM build-deps AS rust-extensions-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN apt-get update && \
+    apt-get install -y curl libclang-dev cmake && \
+    useradd -ms /bin/bash nonroot -b /home
+
+ENV HOME=/home/nonroot
+ENV PATH="/home/nonroot/.cargo/bin:/usr/local/pgsql/bin/:$PATH"
+USER nonroot
+WORKDIR /home/nonroot
+ARG PG_VERSION
+
+RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && \
+    chmod +x rustup-init && \
+    ./rustup-init -y --no-modify-path --profile minimal --default-toolchain stable && \
+    rm rustup-init && \
+    cargo install --git https://github.com/vadim2404/pgx --branch neon_abi_v0.6.1 --locked cargo-pgx && \
+    /bin/bash -c 'cargo pgx init --pg${PG_VERSION:1}=/usr/local/pgsql/bin/pg_config'
+
+USER root
+
+#########################################################################################
+# 
+# Layer "pg-jsonschema-pg-build"
+# Compile "pg_jsonschema" extension
+#
+#########################################################################################
+
+FROM rust-extensions-build AS pg-jsonschema-pg-build
+
+RUN git clone --depth=1 --single-branch --branch neon_abi_v0.1.4 https://github.com/vadim2404/pg_jsonschema/ && \
+    cd pg_jsonschema && \
+    cargo pgx install --release && \
+    # it's needed to enable extension because it uses untrusted C language
+    sed -i 's/superuser = false/superuser = true/g' /usr/local/pgsql/share/extension/pg_jsonschema.control && \
+    echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_jsonschema.control
+
+#########################################################################################
+# 
+# Layer "pg-graphql-pg-build"
+# Compile "pg_graphql" extension
+#
+#########################################################################################
+
+FROM rust-extensions-build AS pg-graphql-pg-build
+
+RUN git clone --depth=1 --single-branch --branch neon_abi_v1.1.0 https://github.com/vadim2404/pg_graphql && \
+    cd pg_graphql && \  
+    cargo pgx install --release && \
+    # it's needed to enable extension because it uses untrusted C language
+    sed -i 's/superuser = false/superuser = true/g' /usr/local/pgsql/share/extension/pg_graphql.control && \
+    echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_graphql.control
+
+#########################################################################################
+#
+# Layer "neon-pg-ext-build"
+# compile neon extensions
+#
+#########################################################################################
+FROM build-deps AS neon-pg-ext-build
+COPY --from=postgis-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=postgis-build /sfcgal/* /
+COPY --from=plv8-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=h3-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=h3-pg-build /h3/usr /
+COPY --from=unit-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=vector-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=pgjwt-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=pg-jsonschema-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=pg-graphql-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=hypopg-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=pg-hashids-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY pgxn/ pgxn/
+
+RUN make -j $(getconf _NPROCESSORS_ONLN) \
+        PG_CONFIG=/usr/local/pgsql/bin/pg_config \
+        -C pgxn/neon \
+        -s install
+
+#########################################################################################
+#
+# Compile and run the Neon-specific `compute_ctl` binary
+#
+#########################################################################################
+FROM $REPOSITORY/$IMAGE:$TAG AS compute-tools
+USER nonroot
+# Copy entire project to get Cargo.* files with proper dependencies for the whole project
+COPY --chown=nonroot . .
+RUN cd compute_tools && cargo build --locked --profile release-line-debug-size-lto
+
+#########################################################################################
+#
+# Clean up postgres folder before inclusion
+#
+#########################################################################################
+FROM neon-pg-ext-build AS postgres-cleanup-layer
+COPY --from=neon-pg-ext-build /usr/local/pgsql /usr/local/pgsql
+
+# Remove binaries from /bin/ that we won't use (or would manually copy & install otherwise)
+RUN cd /usr/local/pgsql/bin && rm ecpg raster2pgsql shp2pgsql pgtopo_export pgtopo_import pgsql2shp
+
+# Remove headers that we won't need anymore - we've completed installation of all extensions
+RUN rm -r /usr/local/pgsql/include
+
+# Remove static postgresql libraries - all compilation is finished, so we
+# can now remove these files - they must be included in other binaries by now
+# if they were to be used by other libraries.
+RUN rm /usr/local/pgsql/lib/lib*.a
+
+#########################################################################################
+#
+# Final layer
+# Put it all together into the final image
+#
+#########################################################################################
+FROM debian:bullseye-slim
+# Add user postgres
+RUN mkdir /var/db && useradd -m -d /var/db/postgres postgres && \
+    echo "postgres:test_console_pass" | chpasswd && \
+    mkdir /var/db/postgres/compute && mkdir /var/db/postgres/specs && \
+    chown -R postgres:postgres /var/db/postgres && \
+    chmod 0750 /var/db/postgres/compute && \
+    echo '/usr/local/lib' >> /etc/ld.so.conf && /sbin/ldconfig && \
+    # create folder for file cache
+    mkdir -p -m 777 /neon/cache
+
+COPY --from=postgres-cleanup-layer --chown=postgres /usr/local/pgsql /usr/local
+COPY --from=compute-tools --chown=postgres /home/nonroot/target/release-line-debug-size-lto/compute_ctl /usr/local/bin/compute_ctl
+
+# Install:
+# libreadline8 for psql
+# libicu67, locales for collations (including ICU)
+# libossp-uuid16 for extension ossp-uuid
+# libgeos, libgdal, libsfcgal1, libproj and libprotobuf-c1 for PostGIS
+# libxml2, libxslt1.1 for xml2
+RUN apt update &&  \
+    apt install --no-install-recommends -y \
+        locales \
+        libicu67 \
+        libreadline8 \
+        libossp-uuid16 \
+        libgeos-c1v5 \
+        libgdal28 \
+        libproj19 \
+        libprotobuf-c1 \
+        libsfcgal1 \
+        libxml2 \
+        libxslt1.1 \
+        gdb && \
+    rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* && \
+    localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8
+
+ENV LANG en_US.utf8
+USER postgres
+ENTRYPOINT ["/usr/local/bin/compute_ctl"]

Dockerfile.compute-node-v14

@@ -1,220 +0,0 @@
-#
-# This file is identical to the Dockerfile.compute-node-v15 file
-# except for the version of Postgres that is built.
-#
-
-ARG TAG=pinned
-
-#########################################################################################
-#
-# Layer "build-deps"
-#
-#########################################################################################
-FROM debian:bullseye-slim AS build-deps
-RUN apt update &&  \
-    apt install -y git autoconf automake libtool build-essential bison flex libreadline-dev \
-    zlib1g-dev libxml2-dev libcurl4-openssl-dev libossp-uuid-dev wget pkg-config libssl-dev
-
-#########################################################################################
-#
-# Layer "pg-build"
-# Build Postgres from the neon postgres repository.
-#
-#########################################################################################
-FROM build-deps AS pg-build
-COPY vendor/postgres-v14 postgres
-RUN cd postgres && \
-    ./configure CFLAGS='-O2 -g3' --enable-debug --with-openssl --with-uuid=ossp && \
-    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s install && \
-    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C contrib/ install && \
-    # Install headers
-    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C src/include install && \
-    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C src/interfaces/libpq install && \
-    # Enable some of contrib extensions
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/bloom.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgrowlocks.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/intagg.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgstattuple.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/earthdistance.control
-
-#########################################################################################
-#
-# Layer "postgis-build"
-# Build PostGIS from the upstream PostGIS mirror.
-#
-#########################################################################################
-FROM build-deps AS postgis-build
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
-RUN apt update && \
-    apt install -y gdal-bin libgdal-dev libprotobuf-c-dev protobuf-c-compiler xsltproc
-
-RUN wget https://download.osgeo.org/postgis/source/postgis-3.3.1.tar.gz && \
-    tar xvzf postgis-3.3.1.tar.gz && \
-    cd postgis-3.3.1 && \
-    ./autogen.sh && \
-    export PATH="/usr/local/pgsql/bin:$PATH" && \
-    ./configure && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
-    cd extensions/postgis && \
-    make clean && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis_raster.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis_tiger_geocoder.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis_topology.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/address_standardizer.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/address_standardizer_data_us.control
-
-#########################################################################################
-#
-# Layer "plv8-build"
-# Build plv8
-#
-#########################################################################################
-FROM build-deps AS plv8-build
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
-RUN apt update && \
-    apt install -y ninja-build python3-dev libc++-dev libc++abi-dev libncurses5 binutils
-
-# https://github.com/plv8/plv8/issues/475:
-#   v8 uses gold for linking and sets `--thread-count=4` which breaks
-#   gold version <= 1.35 (https://sourceware.org/bugzilla/show_bug.cgi?id=23607)
-# Install newer gold version manually as debian-testing binutils version updates
-# libc version, which in turn breaks other extension built against non-testing libc.
-RUN wget https://ftp.gnu.org/gnu/binutils/binutils-2.38.tar.gz && \
-    tar xvzf binutils-2.38.tar.gz && \
-    cd binutils-2.38 && \
-    cd libiberty && ./configure && make -j $(getconf _NPROCESSORS_ONLN) && \
-    cd ../bfd && ./configure && make bfdver.h && \
-    cd ../gold && ./configure && make -j $(getconf _NPROCESSORS_ONLN) && make install && \
-    cp /usr/local/bin/ld.gold /usr/bin/gold
-
-# Sed is used to patch for https://github.com/plv8/plv8/issues/503
-RUN wget https://github.com/plv8/plv8/archive/refs/tags/v3.1.4.tar.gz && \
-    tar xvzf v3.1.4.tar.gz && \
-    cd plv8-3.1.4 && \
-    export PATH="/usr/local/pgsql/bin:$PATH" && \
-    sed -i 's/MemoryContextAlloc(/MemoryContextAllocZero(/' plv8.cc && \
-    make DOCKER=1 -j $(getconf _NPROCESSORS_ONLN) install && \
-    rm -rf /plv8-* && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/plv8.control
-
-#########################################################################################
-#
-# Layer "h3-pg-build"
-# Build h3_pg
-#
-#########################################################################################
-FROM build-deps AS h3-pg-build
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
-
-# packaged cmake is too old
-RUN wget https://github.com/Kitware/CMake/releases/download/v3.24.2/cmake-3.24.2-linux-x86_64.sh \
-      -q -O /tmp/cmake-install.sh \
-      && chmod u+x /tmp/cmake-install.sh \
-      && /tmp/cmake-install.sh --skip-license --prefix=/usr/local/ \
-      && rm /tmp/cmake-install.sh
-
-RUN wget https://github.com/uber/h3/archive/refs/tags/v4.0.1.tar.gz -O h3.tgz && \
-    tar xvzf h3.tgz  && \
-    cd h3-4.0.1 && \
-    mkdir build && \
-    cd build && \
-    cmake .. -DCMAKE_BUILD_TYPE=Release && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    DESTDIR=/h3 make install && \
-    cp -R /h3/usr / && \
-    rm -rf build
-
-RUN wget https://github.com/zachasme/h3-pg/archive/refs/tags/v4.0.1.tar.gz -O h3-pg.tgz && \
-    tar xvzf h3-pg.tgz && \
-    cd h3-pg-4.0.1 && \
-    export PATH="/usr/local/pgsql/bin:$PATH" && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/h3.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/h3_postgis.control
-
-#########################################################################################
-#
-# Layer "neon-pg-ext-build"
-# compile neon extensions
-#
-#########################################################################################
-FROM build-deps AS neon-pg-ext-build
-COPY --from=postgis-build /usr/local/pgsql/ /usr/local/pgsql/
-COPY --from=plv8-build /usr/local/pgsql/ /usr/local/pgsql/
-COPY --from=h3-pg-build /usr/local/pgsql/ /usr/local/pgsql/
-COPY --from=h3-pg-build /h3/usr /
-COPY pgxn/ pgxn/
-
-RUN make -j $(getconf _NPROCESSORS_ONLN) \
-        PG_CONFIG=/usr/local/pgsql/bin/pg_config \
-        -C pgxn/neon \
-        -s install
-
-#########################################################################################
-#
-# Compile and run the Neon-specific `compute_ctl` binary
-#
-#########################################################################################
-FROM 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:$TAG AS compute-tools
-USER nonroot
-# Copy entire project to get Cargo.* files with proper dependencies for the whole project
-COPY --chown=nonroot . .
-RUN cd compute_tools && cargo build --locked --profile release-line-debug-size-lto
-
-#########################################################################################
-#
-# Clean up postgres folder before inclusion
-#
-#########################################################################################
-FROM neon-pg-ext-build AS postgres-cleanup-layer
-COPY --from=neon-pg-ext-build /usr/local/pgsql /usr/local/pgsql
-
-# Remove binaries from /bin/ that we won't use (or would manually copy & install otherwise)
-RUN cd /usr/local/pgsql/bin && rm ecpg raster2pgsql shp2pgsql pgtopo_export pgtopo_import pgsql2shp
-
-# Remove headers that we won't need anymore - we've completed installation of all extensions
-RUN rm -r /usr/local/pgsql/include
-
-# Remove static postgresql libraries - all compilation is finished, so we
-# can now remove these files - they must be included in other binaries by now
-# if they were to be used by other libraries.
-RUN rm /usr/local/pgsql/lib/lib*.a
-
-#########################################################################################
-#
-# Final layer
-# Put it all together into the final image
-#
-#########################################################################################
-FROM debian:bullseye-slim
-# Add user postgres
-RUN mkdir /var/db && useradd -m -d /var/db/postgres postgres && \
-    echo "postgres:test_console_pass" | chpasswd && \
-    mkdir /var/db/postgres/compute && mkdir /var/db/postgres/specs && \
-    chown -R postgres:postgres /var/db/postgres && \
-    chmod 0750 /var/db/postgres/compute && \
-    echo '/usr/local/lib' >> /etc/ld.so.conf && /sbin/ldconfig
-
-COPY --from=postgres-cleanup-layer --chown=postgres /usr/local/pgsql /usr/local
-COPY --from=compute-tools --chown=postgres /home/nonroot/target/release-line-debug-size-lto/compute_ctl /usr/local/bin/compute_ctl
-
-# Install:
-# libreadline8 for psql
-# libossp-uuid16 for extension ossp-uuid
-# libgeos, libgdal, libproj and libprotobuf-c1 for PostGIS
-RUN apt update &&  \
-    apt install --no-install-recommends -y \
-        libreadline8 \
-        libossp-uuid16 \
-        libgeos-c1v5 \
-        libgdal28 \
-        libproj19 \
-        libprotobuf-c1 \
-        gdb && \
-    rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
-
-USER postgres
-ENTRYPOINT ["/usr/local/bin/compute_ctl"]

Dockerfile.compute-node-v15

@@ -1,220 +0,0 @@
-#
-# This file is identical to the Dockerfile.compute-node-v14 file
-# except for the version of Postgres that is built.
-#
-
-ARG TAG=pinned
-
-#########################################################################################
-#
-# Layer "build-deps"
-#
-#########################################################################################
-FROM debian:bullseye-slim AS build-deps
-RUN apt update &&  \
-    apt install -y git autoconf automake libtool build-essential bison flex libreadline-dev \
-    zlib1g-dev libxml2-dev libcurl4-openssl-dev libossp-uuid-dev wget pkg-config libssl-dev
-
-#########################################################################################
-#
-# Layer "pg-build"
-# Build Postgres from the neon postgres repository.
-#
-#########################################################################################
-FROM build-deps AS pg-build
-COPY vendor/postgres-v15 postgres
-RUN cd postgres && \
-    ./configure CFLAGS='-O2 -g3' --enable-debug --with-openssl --with-uuid=ossp && \
-    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s install && \
-    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C contrib/ install && \
-    # Install headers
-    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C src/include install && \
-    make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C src/interfaces/libpq install && \
-    # Enable some of contrib extensions
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/bloom.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgrowlocks.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/intagg.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgstattuple.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/earthdistance.control
-
-#########################################################################################
-#
-# Layer "postgis-build"
-# Build PostGIS from the upstream PostGIS mirror.
-#
-#########################################################################################
-FROM build-deps AS postgis-build
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
-RUN apt update && \
-    apt install -y gdal-bin libgdal-dev libprotobuf-c-dev protobuf-c-compiler xsltproc
-
-RUN wget https://download.osgeo.org/postgis/source/postgis-3.3.1.tar.gz && \
-    tar xvzf postgis-3.3.1.tar.gz && \
-    cd postgis-3.3.1 && \
-    ./autogen.sh && \
-    export PATH="/usr/local/pgsql/bin:$PATH" && \
-    ./configure && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
-    cd extensions/postgis && \
-    make clean && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis_raster.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis_tiger_geocoder.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/postgis_topology.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/address_standardizer.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/address_standardizer_data_us.control
-
-#########################################################################################
-#
-# Layer "plv8-build"
-# Build plv8
-#
-#########################################################################################
-FROM build-deps AS plv8-build
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
-RUN apt update && \
-    apt install -y ninja-build python3-dev libc++-dev libc++abi-dev libncurses5 binutils
-
-# https://github.com/plv8/plv8/issues/475:
-#   v8 uses gold for linking and sets `--thread-count=4` which breaks
-#   gold version <= 1.35 (https://sourceware.org/bugzilla/show_bug.cgi?id=23607)
-# Install newer gold version manually as debian-testing binutils version updates
-# libc version, which in turn breaks other extension built against non-testing libc.
-RUN wget https://ftp.gnu.org/gnu/binutils/binutils-2.38.tar.gz && \
-    tar xvzf binutils-2.38.tar.gz && \
-    cd binutils-2.38 && \
-    cd libiberty && ./configure && make -j $(getconf _NPROCESSORS_ONLN) && \
-    cd ../bfd && ./configure && make bfdver.h && \
-    cd ../gold && ./configure && make -j $(getconf _NPROCESSORS_ONLN) && make install && \
-    cp /usr/local/bin/ld.gold /usr/bin/gold
-
-# Sed is used to patch for https://github.com/plv8/plv8/issues/503
-RUN wget https://github.com/plv8/plv8/archive/refs/tags/v3.1.4.tar.gz && \
-    tar xvzf v3.1.4.tar.gz && \
-    cd plv8-3.1.4 && \
-    export PATH="/usr/local/pgsql/bin:$PATH" && \
-    sed -i 's/MemoryContextAlloc(/MemoryContextAllocZero(/' plv8.cc && \
-    make DOCKER=1 -j $(getconf _NPROCESSORS_ONLN) install && \
-    rm -rf /plv8-* && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/plv8.control
-
-#########################################################################################
-#
-# Layer "h3-pg-build"
-# Build h3_pg
-#
-#########################################################################################
-FROM build-deps AS h3-pg-build
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
-
-# packaged cmake is too old
-RUN wget https://github.com/Kitware/CMake/releases/download/v3.24.2/cmake-3.24.2-linux-x86_64.sh \
-      -q -O /tmp/cmake-install.sh \
-      && chmod u+x /tmp/cmake-install.sh \
-      && /tmp/cmake-install.sh --skip-license --prefix=/usr/local/ \
-      && rm /tmp/cmake-install.sh
-
-RUN wget https://github.com/uber/h3/archive/refs/tags/v4.0.1.tar.gz -O h3.tgz && \
-    tar xvzf h3.tgz  && \
-    cd h3-4.0.1 && \
-    mkdir build && \
-    cd build && \
-    cmake .. -DCMAKE_BUILD_TYPE=Release && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    DESTDIR=/h3 make install && \
-    cp -R /h3/usr / && \
-    rm -rf build
-
-RUN wget https://github.com/zachasme/h3-pg/archive/refs/tags/v4.0.1.tar.gz -O h3-pg.tgz && \
-    tar xvzf h3-pg.tgz && \
-    cd h3-pg-4.0.1 && \
-    export PATH="/usr/local/pgsql/bin:$PATH" && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/h3.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/h3_postgis.control
-
-#########################################################################################
-#
-# Layer "neon-pg-ext-build"
-# compile neon extensions
-#
-#########################################################################################
-FROM build-deps AS neon-pg-ext-build
-COPY --from=postgis-build /usr/local/pgsql/ /usr/local/pgsql/
-COPY --from=plv8-build /usr/local/pgsql/ /usr/local/pgsql/
-COPY --from=h3-pg-build /usr/local/pgsql/ /usr/local/pgsql/
-COPY --from=h3-pg-build /h3/usr /
-COPY pgxn/ pgxn/
-
-RUN make -j $(getconf _NPROCESSORS_ONLN) \
-        PG_CONFIG=/usr/local/pgsql/bin/pg_config \
-        -C pgxn/neon \
-        -s install
-
-#########################################################################################
-#
-# Compile and run the Neon-specific `compute_ctl` binary
-#
-#########################################################################################
-FROM 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:$TAG AS compute-tools
-USER nonroot
-# Copy entire project to get Cargo.* files with proper dependencies for the whole project
-COPY --chown=nonroot . .
-RUN cd compute_tools && cargo build --locked --profile release-line-debug-size-lto
-
-#########################################################################################
-#
-# Clean up postgres folder before inclusion
-#
-#########################################################################################
-FROM neon-pg-ext-build AS postgres-cleanup-layer
-COPY --from=neon-pg-ext-build /usr/local/pgsql /usr/local/pgsql
-
-# Remove binaries from /bin/ that we won't use (or would manually copy & install otherwise)
-RUN cd /usr/local/pgsql/bin && rm ecpg raster2pgsql shp2pgsql pgtopo_export pgtopo_import pgsql2shp
-
-# Remove headers that we won't need anymore - we've completed installation of all extensions
-RUN rm -r /usr/local/pgsql/include
-
-# Remove static postgresql libraries - all compilation is finished, so we
-# can now remove these files - they must be included in other binaries by now
-# if they were to be used by other libraries.
-RUN rm /usr/local/pgsql/lib/lib*.a
-
-#########################################################################################
-#
-# Final layer
-# Put it all together into the final image
-#
-#########################################################################################
-FROM debian:bullseye-slim
-# Add user postgres
-RUN mkdir /var/db && useradd -m -d /var/db/postgres postgres && \
-    echo "postgres:test_console_pass" | chpasswd && \
-    mkdir /var/db/postgres/compute && mkdir /var/db/postgres/specs && \
-    chown -R postgres:postgres /var/db/postgres && \
-    chmod 0750 /var/db/postgres/compute && \
-    echo '/usr/local/lib' >> /etc/ld.so.conf && /sbin/ldconfig
-
-COPY --from=postgres-cleanup-layer --chown=postgres /usr/local/pgsql /usr/local
-COPY --from=compute-tools --chown=postgres /home/nonroot/target/release-line-debug-size-lto/compute_ctl /usr/local/bin/compute_ctl
-
-# Install:
-# libreadline8 for psql
-# libossp-uuid16 for extension ossp-uuid
-# libgeos, libgdal, libproj and libprotobuf-c1 for PostGIS
-RUN apt update &&  \
-    apt install --no-install-recommends -y \
-        libreadline8 \
-        libossp-uuid16 \
-        libgeos-c1v5 \
-        libgdal28 \
-        libproj19 \
-        libprotobuf-c1 \
-        gdb && \
-    rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
-
-USER postgres
-ENTRYPOINT ["/usr/local/bin/compute_ctl"]

Dockerfile.vm-compute-node

@@ -0,0 +1,32 @@
+# Note: this file *mostly* just builds on Dockerfile.compute-node
+
+ARG SRC_IMAGE
+ARG VM_INFORMANT_VERSION=v0.1.6
+
+# Pull VM informant and set up inittab
+FROM neondatabase/vm-informant:$VM_INFORMANT_VERSION as informant
+
+RUN set -e \
+	&& rm -f /etc/inittab \
+	&& touch /etc/inittab
+
+RUN set -e \
+	&& echo "::sysinit:cgconfigparser -l /etc/cgconfig.conf -s 1664" >> /etc/inittab \
+	&& echo "::respawn:su vm-informant -c '/usr/local/bin/vm-informant --auto-restart --cgroup=neon-postgres'" >> /etc/inittab
+
+# Combine, starting from non-VM compute node image.
+FROM $SRC_IMAGE as base
+
+# Temporarily set user back to root so we can run apt update and adduser
+USER root
+RUN apt update && \
+	apt install --no-install-recommends -y \
+        cgroup-tools
+RUN adduser vm-informant --disabled-password --no-create-home
+USER postgres
+
+ADD vm-cgconfig.conf /etc/cgconfig.conf
+COPY --from=informant /etc/inittab /etc/inittab
+COPY --from=informant /usr/bin/vm-informant /usr/local/bin/vm-informant
+
+ENTRYPOINT ["/usr/sbin/cgexec", "-g", "*:neon-postgres", "/usr/local/bin/compute_ctl"]

Makefile

@@ -136,9 +136,15 @@ neon-pg-ext-%: postgres-%
 
 .PHONY: neon-pg-ext-clean-%
 neon-pg-ext-clean-%:
-	$(MAKE) -C $(POSTGRES_INSTALL_DIR)/pgxn/neon-$* -f $(ROOT_PROJECT_DIR)/pgxn/neon/Makefile clean
-	$(MAKE) -C $(POSTGRES_INSTALL_DIR)/pgxn/neon_walredo-$* -f $(ROOT_PROJECT_DIR)/pgxn/neon_walredo/Makefile clean
-	$(MAKE) -C $(POSTGRES_INSTALL_DIR)/pgxn/neon_test_utils-$* -f $(ROOT_PROJECT_DIR)/pgxn/neon_test_utils/Makefile clean
+	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/$*/bin/pg_config \
+	-C $(POSTGRES_INSTALL_DIR)/build/neon-$* \
+	-f $(ROOT_PROJECT_DIR)/pgxn/neon/Makefile clean
+	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/$*/bin/pg_config \
+	-C $(POSTGRES_INSTALL_DIR)/build/neon-walredo-$* \
+	-f $(ROOT_PROJECT_DIR)/pgxn/neon_walredo/Makefile clean
+	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/$*/bin/pg_config \
+	-C $(POSTGRES_INSTALL_DIR)/build/neon-test-utils-$* \
+	-f $(ROOT_PROJECT_DIR)/pgxn/neon_test_utils/Makefile clean
 
 .PHONY: neon-pg-ext
 neon-pg-ext: \

README.md

@@ -34,6 +34,11 @@ dnf install flex bison readline-devel zlib-devel openssl-devel \
   libseccomp-devel perl clang cmake postgresql postgresql-contrib protobuf-compiler \
   protobuf-devel
 ```
+* On Arch based systems, these packages are needed:
+```bash
+pacman -S base-devel readline zlib libseccomp openssl clang \
+postgresql-libs cmake postgresql protobuf
+```
 
 2. [Install Rust](https://www.rust-lang.org/tools/install)
 ```
@@ -83,9 +88,10 @@ cd neon
 
 # The preferred and default is to make a debug build. This will create a
 # demonstrably slower build than a release build. For a release build,
-# use "BUILD_TYPE=release make -j`nproc`"
+# use "BUILD_TYPE=release make -j`nproc` -s"
+# Remove -s for the verbose build log
 
-make -j`nproc`
+make -j`nproc` -s
 ```
 
 #### Building on OSX
@@ -99,16 +105,17 @@ cd neon
 
 # The preferred and default is to make a debug build. This will create a
 # demonstrably slower build than a release build. For a release build,
-# use "BUILD_TYPE=release make -j`sysctl -n hw.logicalcpu`"
+# use "BUILD_TYPE=release make -j`sysctl -n hw.logicalcpu` -s"
+# Remove -s for the verbose build log
 
-make -j`sysctl -n hw.logicalcpu`
+make -j`sysctl -n hw.logicalcpu` -s
 ```
 
 #### Dependency installation notes
 To run the `psql` client, install the `postgresql-client` package or modify `PATH` and `LD_LIBRARY_PATH` to include `pg_install/bin` and `pg_install/lib`, respectively.
 
 To run the integration tests or Python scripts (not required to use the code), install
-Python (3.9 or higher), and install python3 packages using `./scripts/pysync` (requires [poetry](https://python-poetry.org/)) in the project directory.
+Python (3.9 or higher), and install python3 packages using `./scripts/pysync` (requires [poetry>=1.3](https://python-poetry.org/)) in the project directory.
 
 
 #### Running neon database

compute_tools/Cargo.toml

@@ -11,15 +11,19 @@ clap.workspace = true
 futures.workspace = true
 hyper = { workspace = true, features = ["full"] }
 notify.workspace = true
+opentelemetry.workspace = true
 postgres.workspace = true
 regex.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 tar.workspace = true
+reqwest = { workspace = true, features = ["json"] }
 tokio = { workspace = true, features = ["rt", "rt-multi-thread"] }
 tokio-postgres.workspace = true
 tracing.workspace = true
+tracing-opentelemetry.workspace = true
 tracing-subscriber.workspace = true
+tracing-utils.workspace = true
 url.workspace = true
 
 workspace_hack.workspace = true

compute_tools/src/bin/compute_ctl.rs

@@ -44,7 +44,6 @@ use tracing::{error, info};
 
 use compute_tools::compute::{ComputeMetrics, ComputeNode, ComputeState, ComputeStatus};
 use compute_tools::http::api::launch_http_server;
-use compute_tools::informant::spawn_vm_informant_if_present;
 use compute_tools::logger::*;
 use compute_tools::monitor::launch_monitor;
 use compute_tools::params::*;
@@ -53,7 +52,7 @@ use compute_tools::spec::*;
 use url::Url;
 
 fn main() -> Result<()> {
-    init_logger(DEFAULT_LOG_LEVEL)?;
+    init_tracing_and_logging(DEFAULT_LOG_LEVEL)?;
 
     let matches = cli().get_matches();
 
@@ -66,6 +65,9 @@ fn main() -> Result<()> {
     let spec = matches.get_one::<String>("spec");
     let spec_path = matches.get_one::<String>("spec-path");
 
+    let compute_id = matches.get_one::<String>("compute-id");
+    let control_plane_uri = matches.get_one::<String>("control-plane-uri");
+
     // Try to use just 'postgres' if no path is provided
     let pgbin = matches.get_one::<String>("pgbin").unwrap();
 
@@ -78,12 +80,54 @@ fn main() -> Result<()> {
                 let path = Path::new(sp);
                 let file = File::open(path)?;
                 serde_json::from_reader(file)?
+            } else if let Some(id) = compute_id {
+                if let Some(cp_base) = control_plane_uri {
+                    let cp_uri = format!("{cp_base}/management/api/v1/{id}/spec");
+                    let jwt: String = match std::env::var("NEON_CONSOLE_JWT") {
+                        Ok(v) => v,
+                        Err(_) => "".to_string(),
+                    };
+
+                    reqwest::blocking::Client::new()
+                        .get(cp_uri)
+                        .header("Authorization", jwt)
+                        .send()?
+                        .json()?
+                } else {
+                    panic!(
+                        "must specify --control-plane-uri \"{:#?}\" and --compute-id \"{:#?}\"",
+                        control_plane_uri, compute_id
+                    );
+                }
             } else {
-                panic!("cluster spec should be provided via --spec or --spec-path argument");
+                panic!("compute spec should be provided via --spec or --spec-path argument");
             }
         }
     };
 
+    // Extract OpenTelemetry context for the startup actions from the spec, and
+    // attach it to the current tracing context.
+    //
+    // This is used to propagate the context for the 'start_compute' operation
+    // from the neon control plane. This allows linking together the wider
+    // 'start_compute' operation that creates the compute container, with the
+    // startup actions here within the container.
+    //
+    // Switch to the startup context here, and exit it once the startup has
+    // completed and Postgres is up and running.
+    //
+    // NOTE: This is supposed to only cover the *startup* actions. Once
+    // postgres is configured and up-and-running, we exit this span. Any other
+    // actions that are performed on incoming HTTP requests, for example, are
+    // performed in separate spans.
+    let startup_context_guard = if let Some(ref carrier) = spec.startup_tracing_context {
+        use opentelemetry::propagation::TextMapPropagator;
+        use opentelemetry::sdk::propagation::TraceContextPropagator;
+        Some(TraceContextPropagator::new().extract(carrier).attach())
+    } else {
+        None
+    };
+
     let pageserver_connstr = spec
         .cluster
         .settings
@@ -118,8 +162,6 @@ fn main() -> Result<()> {
     // requests, while configuration is still in progress.
     let _http_handle = launch_http_server(&compute).expect("cannot launch http endpoint thread");
     let _monitor_handle = launch_monitor(&compute).expect("cannot launch compute monitor thread");
-    // Also spawn the thread responsible for handling the VM informant -- if it's present
-    let _vm_informant_handle = spawn_vm_informant_if_present().expect("cannot launch VM informant");
 
     // Start Postgres
     let mut delay_exit = false;
@@ -140,6 +182,9 @@ fn main() -> Result<()> {
     // Wait for the child Postgres process forever. In this state Ctrl+C will
     // propagate to Postgres and it will be shut down as well.
     if let Some(mut pg) = pg {
+        // Startup is finished, exit the startup tracing span
+        drop(startup_context_guard);
+
         let ecode = pg
             .wait()
             .expect("failed to start waiting on Postgres process");
@@ -159,6 +204,10 @@ fn main() -> Result<()> {
         info!("shutting down");
     }
 
+    // Shutdown trace pipeline gracefully, so that it has a chance to send any
+    // pending traces before we exit.
+    tracing_utils::shutdown_tracing();
+
     exit(exit_code.unwrap_or(1))
 }
 
@@ -200,6 +249,18 @@ fn cli() -> clap::Command {
                 .long("spec-path")
                 .value_name("SPEC_PATH"),
         )
+        .arg(
+            Arg::new("compute-id")
+                .short('i')
+                .long("compute-id")
+                .value_name("COMPUTE_ID"),
+        )
+        .arg(
+            Arg::new("control-plane-uri")
+                .short('p')
+                .long("control-plane-uri")
+                .value_name("CONTROL_PLANE"),
+        )
 }
 
 #[test]

compute_tools/src/http/api.rs

@@ -3,16 +3,21 @@ use std::net::SocketAddr;
 use std::sync::Arc;
 use std::thread;
 
+use crate::compute::ComputeNode;
 use anyhow::Result;
 use hyper::service::{make_service_fn, service_fn};
 use hyper::{Body, Method, Request, Response, Server, StatusCode};
 use serde_json;
 use tracing::{error, info};
-
-use crate::compute::ComputeNode;
+use tracing_utils::http::OtelName;
 
 // Service function to handle all available routes.
-async fn routes(req: Request<Body>, compute: Arc<ComputeNode>) -> Response<Body> {
+async fn routes(req: Request<Body>, compute: &Arc<ComputeNode>) -> Response<Body> {
+    //
+    // NOTE: The URI path is currently included in traces. That's OK because
+    // it doesn't contain any variable parts or sensitive information. But
+    // please keep that in mind if you change the routing here.
+    //
     match (req.method(), req.uri().path()) {
         // Serialized compute state.
         (&Method::GET, "/status") => {
@@ -30,7 +35,7 @@ async fn routes(req: Request<Body>, compute: Arc<ComputeNode>) -> Response<Body>
 
         (&Method::POST, "/check_writability") => {
             info!("serving /check_writability POST request");
-            let res = crate::checker::check_writability(&compute).await;
+            let res = crate::checker::check_writability(compute).await;
             match res {
                 Ok(_) => Response::new(Body::from("true")),
                 Err(e) => Response::new(Body::from(e.to_string())),
@@ -56,7 +61,19 @@ async fn serve(state: Arc<ComputeNode>) {
         async move {
             Ok::<_, Infallible>(service_fn(move |req: Request<Body>| {
                 let state = state.clone();
-                async move { Ok::<_, Infallible>(routes(req, state).await) }
+                async move {
+                    Ok::<_, Infallible>(
+                        // NOTE: We include the URI path in the string. It
+                        // doesn't contain any variable parts or sensitive
+                        // information in this API.
+                        tracing_utils::http::tracing_handler(
+                            req,
+                            |req| routes(req, &state),
+                            OtelName::UriPath,
+                        )
+                        .await,
+                    )
+                }
             }))
         }
     });

compute_tools/src/informant.rs

@@ -1,50 +0,0 @@
-use std::path::Path;
-use std::process;
-use std::thread;
-use std::time::Duration;
-use tracing::{info, warn};
-
-use anyhow::{Context, Result};
-
-const VM_INFORMANT_PATH: &str = "/bin/vm-informant";
-const RESTART_INFORMANT_AFTER_MILLIS: u64 = 5000;
-
-/// Launch a thread to start the VM informant if it's present (and restart, on failure)
-pub fn spawn_vm_informant_if_present() -> Result<Option<thread::JoinHandle<()>>> {
-    let exists = Path::new(VM_INFORMANT_PATH)
-        .try_exists()
-        .context("could not check if path exists")?;
-
-    if !exists {
-        return Ok(None);
-    }
-
-    Ok(Some(
-        thread::Builder::new()
-            .name("run-vm-informant".into())
-            .spawn(move || run_informant())?,
-    ))
-}
-
-fn run_informant() -> ! {
-    let restart_wait = Duration::from_millis(RESTART_INFORMANT_AFTER_MILLIS);
-
-    info!("starting VM informant");
-
-    loop {
-        let mut cmd = process::Command::new(VM_INFORMANT_PATH);
-        // Block on subprocess:
-        let result = cmd.status();
-
-        match result {
-            Err(e) => warn!("failed to run VM informant at {VM_INFORMANT_PATH:?}: {e}"),
-            Ok(status) if !status.success() => {
-                warn!("{VM_INFORMANT_PATH} exited with code {status:?}, retrying")
-            }
-            Ok(_) => info!("{VM_INFORMANT_PATH} ended gracefully (unexpectedly). Retrying"),
-        }
-
-        // Wait before retrying
-        thread::sleep(restart_wait);
-    }
-}

compute_tools/src/lib.rs

@@ -8,7 +8,6 @@ pub mod http;
 #[macro_use]
 pub mod logger;
 pub mod compute;
-pub mod informant;
 pub mod monitor;
 pub mod params;
 pub mod pg_helpers;

compute_tools/src/logger.rs

@@ -1,21 +1,37 @@
-use anyhow::Result;
+use tracing_opentelemetry::OpenTelemetryLayer;
 use tracing_subscriber::layer::SubscriberExt;
 use tracing_subscriber::prelude::*;
 
-/// Initialize `env_logger` using either `default_level` or
+/// Initialize logging to stderr, and OpenTelemetry tracing and exporter.
+///
+/// Logging is configured using either `default_log_level` or
 /// `RUST_LOG` environment variable as default log level.
-pub fn init_logger(default_level: &str) -> Result<()> {
+///
+/// OpenTelemetry is configured with OTLP/HTTP exporter. It picks up
+/// configuration from environment variables. For example, to change the destination,
+/// set `OTEL_EXPORTER_OTLP_ENDPOINT=http://jaeger:4318`. See
+/// `tracing-utils` package description.
+///
+pub fn init_tracing_and_logging(default_log_level: &str) -> anyhow::Result<()> {
+    // Initialize Logging
     let env_filter = tracing_subscriber::EnvFilter::try_from_default_env()
-        .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new(default_level));
+        .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new(default_log_level));
 
     let fmt_layer = tracing_subscriber::fmt::layer()
         .with_target(false)
         .with_writer(std::io::stderr);
 
+    // Initialize OpenTelemetry
+    let otlp_layer =
+        tracing_utils::init_tracing_without_runtime("compute_ctl").map(OpenTelemetryLayer::new);
+
+    // Put it all together
     tracing_subscriber::registry()
         .with(env_filter)
+        .with(otlp_layer)
         .with(fmt_layer)
         .init();
+    tracing::info!("logging and tracing started");
 
     Ok(())
 }

compute_tools/src/spec.rs

@@ -1,3 +1,4 @@
+use std::collections::HashMap;
 use std::path::Path;
 use std::str::FromStr;
 
@@ -22,6 +23,8 @@ pub struct ComputeSpec {
     /// Expected cluster state at the end of transition process.
     pub cluster: Cluster,
     pub delta_operations: Option<Vec<DeltaOp>>,
+
+    pub startup_tracing_context: Option<HashMap<String, String>>,
 }
 
 /// Cluster state seen from the perspective of the external tools
@@ -384,13 +387,13 @@ pub fn handle_databases(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
                     name.pg_quote(),
                     db.owner.pg_quote()
                 );
-                let _ = info_span!("executing", query).entered();
+                let _guard = info_span!("executing", query).entered();
                 client.execute(query.as_str(), &[])?;
             }
             DatabaseAction::Create => {
                 let mut query: String = format!("CREATE DATABASE {} ", name.pg_quote());
                 query.push_str(&db.to_pg_options());
-                let _ = info_span!("executing", query).entered();
+                let _guard = info_span!("executing", query).entered();
                 client.execute(query.as_str(), &[])?;
             }
         };

control_plane/Cargo.toml

@@ -15,6 +15,7 @@ postgres.workspace = true
 regex.workspace = true
 reqwest = { workspace = true, features = ["blocking", "json"] }
 serde.workspace = true
+serde_json.workspace = true
 serde_with.workspace = true
 tar.workspace = true
 thiserror.workspace = true

control_plane/src/pageserver.rs

@@ -419,6 +419,11 @@ impl PageServerNode {
                     .map(|x| x.parse::<bool>())
                     .transpose()
                     .context("Failed to parse 'trace_read_requests' as bool")?,
+                eviction_policy: settings
+                    .get("eviction_policy")
+                    .map(|x| serde_json::from_str(x))
+                    .transpose()
+                    .context("Failed to parse 'eviction_policy' json")?,
             })
             .send()?
             .error_from_body()?;

docs/settings.md

@@ -16,7 +16,7 @@ listen_http_addr = '127.0.0.1:9898'
 checkpoint_distance = '268435456' # in bytes
 checkpoint_timeout = '10m'
 
-gc_period = '100 s'
+gc_period = '1 hour'
 gc_horizon = '67108864'
 
 max_file_descriptors = '100'
@@ -101,7 +101,7 @@ away.
 
 #### gc_period
 
-Interval at which garbage collection is triggered. Default is 100 s.
+Interval at which garbage collection is triggered. Default is 1 hour.
 
 #### image_creation_threshold
 
@@ -109,7 +109,7 @@ L0 delta layer threshold for L1 image layer creation. Default is 3.
 
 #### pitr_interval
 
-WAL retention duration for PITR branching. Default is 30 days.
+WAL retention duration for PITR branching. Default is 7 days.
 
 #### walreceiver_connect_timeout
 

docs/synthetic-size.md

@@ -0,0 +1,335 @@
+# Synthetic size
+
+Neon storage has copy-on-write branching, which makes it difficult to
+answer the question "how large is my database"? To give one reasonable
+answer, we calculate _synthetic size_ for a project.
+
+The calculation is called "synthetic", because it is based purely on
+the user-visible logical size, which is the size that you would see on
+a standalone PostgreSQL installation, and the amount of WAL, which is
+also the same as what you'd see on a standalone PostgreSQL, for the
+same set of updates.
+
+The synthetic size does *not* depend on the actual physical size
+consumed in the storage, or implementation details of the Neon storage
+like garbage collection, compaction and compression.  There is a
+strong *correlation* between the physical size and the synthetic size,
+but the synthetic size is designed to be independent of the
+implementation details, so that any improvements we make in the
+storage system simply reduce our COGS. And vice versa: any bugs or bad
+implementation where we keep more data than we would need to, do not
+change the synthetic size or incur any costs to the user.
+
+The synthetic size is calculated for the whole project. It is not
+straighforward to attribute size to individual branches. See "What is
+the size of an individual branch?" for discussion on those
+difficulties.
+
+The synthetic size is designed to:
+
+- Take into account the copy-on-write nature of the storage. For
+  example, if you create a branch, it doesn't immediately add anything
+  to the synthetic size. It starts to affect the synthetic size only
+  as it diverges from the parent branch.
+
+- Be independent of any implementation details of the storage, like
+  garbage collection, remote storage, or compression.
+
+## Terms & assumptions
+
+- logical size is the size of a branch *at a given point in
+  time*. It's the total size of all tables in all databases, as you
+  see with "\l+" in psql for example, plus the Postgres SLRUs and some
+  small amount of metadata. NOTE that currently, Neon does not include
+  the SLRUs and metadata in the logical size. See comment to `get_current_logical_size_non_incremental()`.
+
+- a "point in time" is defined as an LSN value. You can convert a
+  timestamp to an LSN, but the storage internally works with LSNs.
+
+- PITR horizon can be set per-branch.
+
+- PITR horizon can be set as a time interval, e.g. 5 days or hours, or
+  as amount of WAL, in bytes.  If it's given as a time interval, it's
+  converted to an LSN for the calculation.
+
+- PITR horizon can be set to 0, if you don't want to retain any history.
+
+## Calculation
+
+Inputs to the calculation are:
+- logical size of the database at different points in time,
+- amount of WAL generated, and
+- the PITR horizon settings
+
+The synthetic size is based on an idealistic model of the storage
+system, where we pretend that the storage consists of two things:
+- snapshots, containing a full snapshot of the database, at a given
+  point in time, and
+- WAL.
+
+In the simple case that the project contains just one branch (main),
+and a fixed PITR horizon, the synthetic size is the sum of:
+
+- the logical size of the branch *at the beginning of the PITR
+  horizon*, i.e. at the oldest point that you can still recover to, and
+- the size of the WAL covering the PITR horizon.
+
+The snapshot allows you to recover to the beginning of the PITR
+horizon, and the WAL allows you to recover from that point to any
+point within the horizon.
+
+```
+                             WAL
+   -----------------------#########>
+                          ^
+                       snapshot
+
+Legend:
+  ##### PITR horizon. This is the region that you can still access
+        with Point-in-time query and you can still create branches
+        from.
+  ----- history that has fallen out of the PITR horizon, and can no
+        longer be accessed
+```
+
+NOTE: This is not how the storage system actually works! The actual
+implementation is also based on snapshots and WAL, but the snapshots
+are taken for individual database pages and ranges of pages rather
+than the whole database, and it is much more complicated. This model
+is a reasonable approximation, however, to make the synthetic size a
+useful proxy for the actual storage consumption.
+
+
+## Example: Data is INSERTed
+
+For example, let's assume that your database contained 10 GB of data
+at the beginning of the PITR horizon, and you have since then inserted
+5 GB of additional data into it. The additional insertions of 5 GB of
+data consume roughly 5 GB of WAL. In that case, the synthetic size is:
+
+> 10 GB (snapshot) +  5 GB (WAL) = 15 GB
+
+If you now set the PITR horizon on the project to 0, so that no
+historical data is retained, then the beginning PITR horizon would be
+at the end of the branch, so the size of the snapshot would be
+calculated at the end of the branch, after the insertions. Then the
+synthetic size is:
+
+> 15 GB (snapshot) + 0 GB (WAL) = 15 GB.
+
+In this case, the synthetic size is the same, regardless of the PITR horizon,
+because all the history consists of inserts. The newly inserted data takes
+up the same amount of space, whether it's stored as part of the logical
+snapshot, or as WAL. (*)
+
+(*) This is a rough approximation. In reality, the WAL contains
+headers and other overhead, and on the other hand, the logical
+snapshot includes empty space on pages, so the size of insertions in
+WAL can be smaller or greater than the size of the final table after
+the insertions. But in most cases, it's in the same ballpark.
+
+## Example: Data is DELETEd
+
+Let's look at another example:
+
+Let's start again with a database that contains 10 GB of data. Then,
+you DELETE 5 GB of the data, and run VACUUM to free up the space, so
+that the logical size of the database is now only 5 GB.
+
+Let's assume that the WAL for the deletions and the vacuum take up
+100 MB of space. In that case, the synthetic size of the project is:
+
+> 10 GB (snapshot) + 100 MB (WAL) = 10.1 GB
+
+This is much larger than the logical size of the database after the
+deletions (5 GB). That's because the system still needs to retain the
+deleted data, because it's still accessible to queries and branching
+in the PITR window.
+
+If you now set the PITR horizon to 0 or just wait for time to pass so
+that the data falls out of the PITR horizon, making the deleted data
+inaccessible, the synthetic size shrinks:
+
+> 5 GB (snapshot) + 0 GB (WAL) = 5 GB
+
+
+# Branching
+
+Things get more complicated with branching. Branches in Neon are
+copy-on-write, which is also reflected in the synthetic size.
+
+When you create a branch, it doesn't immediately change the synthetic
+size at all. The branch point is within the PITR horizon, and all the
+data needed to recover to that point in time needs to be retained
+anyway.
+
+However, if you make modifications on the branch, the system needs to
+keep the WAL of those modifications. The WAL is included in the
+synthetic size.
+
+## Example: branch and INSERT
+
+Let's assume that you again start with a 10 GB database.
+On the main branch, you insert 2 GB of data. Then you create
+a branch at that point, and insert another 3 GB of data on the
+main branch, and 1 GB of data on the child branch
+
+```
+  child                 +#####>
+                        |
+                        |    WAL
+  main    ---------###############>
+                   ^
+                snapshot
+```
+
+In this case, the synthetic size consists of:
+- the snapshot at the beginning of the PITR horizon (10 GB)
+- the WAL on the main branch (2 GB + 3 GB = 5 GB)
+- the WAL on the child branch (1 GB)
+
+Total: 16 GB
+
+# Diverging branches
+
+If there is only a small amount of changes in the database on the
+different branches, as in the previous example, the synthetic size
+consists of a snapshot before the branch point, containing all the
+shared data, and the WAL on both branches. However, if the branches
+diverge a lot, it is more efficient to store a separate snapshot of
+branches.
+
+## Example: diverging branches
+
+You start with a 10 GB database. You insert 5 GB of data on the main
+branch. Then you create a branch, and immediately delete all the data
+on the child branch and insert 5 GB of new data to it. Then you do the
+same on the main branch. Let's assume
+that the PITR horizon requires keeping the last 1 GB of WAL on the
+both branches.
+
+```
+                              snapshot
+                                  v     WAL
+  child                 +---------##############>
+                        |
+                        |
+  main     -------------+---------##############>
+                                  ^     WAL
+                              snapshot
+```
+
+In this case, the synthetic size consists of:
+- snapshot at the beginning of the PITR horizon on the main branch (4 GB)
+- WAL on the main branch (1 GB)
+- snapshot at the beginning of the PITR horizon on the child branch (4 GB)
+- last 1 GB of WAL on the child branch (1 GB)
+
+Total: 10 GB
+
+The alternative way to store this would be to take only one snapshot
+at the beginning of branch point, and keep all the WAL on both
+branches.  However, the size with that method would be larger, as it
+would require one 10 GB snapshot, and 5 GB + 5 GB of WAL. It depends
+on the amount of changes (WAL) on both branches, and the logical size
+at the branch point, which method would result in a smaller synthetic
+size. On each branch point, the system performs the calculation with
+both methods, and uses the method that is cheaper, i.e. the one that
+results in a smaller synthetic size.
+
+One way to think about this is that when you create a branch, it
+starts out as a thin branch that only stores the WAL since the branch
+point.  As you modify it, and the amount of WAL grows, at some point
+it becomes cheaper to store a completely new snapshot of the branch
+and truncate the WAL.
+
+
+# What is the size of an individual branch?
+
+Synthetic size is calculated for the whole project, and includes all
+branches. There is no such thing as the size of a branch, because it
+is not straighforward to attribute the parts of size to individual
+branches.
+
+## Example: attributing size to branches
+
+(copied from https://github.com/neondatabase/neon/pull/2884#discussion_r1029365278)
+
+Imagine that you create two branches, A and B, at the same point from
+main branch, and do a couple of small updates on both branches. Then
+six months pass, and during those six months the data on the main
+branch churns over completely multiple times. The retention period is,
+say 1 month.
+
+```
+                      +------> A
+                     /
+--------------------*-------------------------------> main
+                     \
+                      +--------> B
+```
+
+In that situation, the synthetic tenant size would be calculated based
+on a "logical snapshot" at the branch point, that is, the logical size
+of the database at that point. Plus the WAL on branches A and B. Let's
+say that the snapshot size is 10 GB, and the WAL is 1 MB on both
+branches A and B. So the total synthetic storage size is 10002
+MB. (Let's ignore the main branch for now, that would be just added to
+the sum)
+
+How would you break that down per branch? I can think of three
+different ways to do it, and all of them have their own problems:
+
+### Subtraction method
+
+For each branch, calculate how much smaller the total synthetic size
+would be, if that branch didn't exist. In other words, how much would
+you save if you dropped the branch. With this method, the size of
+branches A and B is 1 MB.
+
+With this method, the 10 GB shared logical snapshot is not included
+for A nor B. So the size of all branches is not equal to the total
+synthetic size of the tenant. If you drop branch A, you save 1 MB as
+you'd expect, but also the size of B suddenly jumps from 1 MB to 10001
+MB, which might feel surprising.
+
+### Division method
+
+Divide the common parts evenly across all branches that need
+them. With this method, the size of branches A and B would be 5001 MB.
+
+With this method, the sum of all branches adds up to the total
+synthetic size. But it's surprising in other ways: if you drop branch
+A, you might think that you save 5001 MB, but in reality you only save
+1 MB, and the size of branch B suddenly grows from 5001 to 10001 MB.
+
+### Addition method
+
+For each branch, include all the snapshots and WAL that it depends on,
+even if some of them are shared by other branches. With this method,
+the size of branches A and B would be 10001 MB.
+
+The surprise with this method is that the sum of all the branches is
+larger than the total synthetic size. And if you drop branch A, the
+total synthetic size doesn't fall by 10001 MB as you might think.
+
+# Alternatives
+
+A sort of cop-out method would be to show the whole tree of branches
+graphically, and for each section of WAL or logical snapshot, display
+the size of that section. You can then see which branches depend on
+which sections, which sections are shared etc. That would be good to
+have in the UI anyway.
+
+Or perhaps calculate per-branch numbers using the subtraction method,
+and in addition to that, one more number for "shared size" that
+includes all the data that is needed by more than one branch.
+
+## Which is the right method?
+
+The bottom line is that it's not straightforward to attribute the
+synthetic size to individual branches. There are things we can do, and
+all of those methods are pretty straightforward to implement, but they
+all have their own problems. What makes sense depends a lot on what
+you want to do with the number, what question you are trying to
+answer.

libs/metrics/Cargo.toml

@@ -8,5 +8,6 @@ license.workspace = true
 prometheus.workspace = true
 libc.workspace = true
 once_cell.workspace = true
+chrono.workspace = true
 
 workspace_hack.workspace = true

libs/metrics/src/launch_timestamp.rs

@@ -0,0 +1,34 @@
+//! A timestamp captured at process startup to identify restarts of the process, e.g., in logs and metrics.
+
+use chrono::Utc;
+
+use super::register_uint_gauge;
+use std::fmt::Display;
+
+pub struct LaunchTimestamp(chrono::DateTime<Utc>);
+
+impl LaunchTimestamp {
+    pub fn generate() -> Self {
+        LaunchTimestamp(Utc::now())
+    }
+}
+
+impl Display for LaunchTimestamp {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}", self.0)
+    }
+}
+
+pub fn set_launch_timestamp_metric(launch_ts: &'static LaunchTimestamp) {
+    let millis_since_epoch: u64 = launch_ts
+        .0
+        .timestamp_millis()
+        .try_into()
+        .expect("we're after the epoch, this should be positive");
+    let metric = register_uint_gauge!(
+        "libmetrics_launch_timestamp",
+        "Timestamp (millis since epoch) at wich the process launched."
+    )
+    .unwrap();
+    metric.set(millis_since_epoch);
+}

libs/metrics/src/lib.rs

@@ -8,6 +8,7 @@ pub use prometheus::opts;
 pub use prometheus::register;
 pub use prometheus::{core, default_registry, proto};
 pub use prometheus::{exponential_buckets, linear_buckets};
+pub use prometheus::{register_counter_vec, Counter, CounterVec};
 pub use prometheus::{register_gauge, Gauge};
 pub use prometheus::{register_gauge_vec, GaugeVec};
 pub use prometheus::{register_histogram, Histogram};
@@ -19,6 +20,7 @@ pub use prometheus::{register_int_gauge_vec, IntGaugeVec};
 pub use prometheus::{Encoder, TextEncoder};
 use prometheus::{Registry, Result};
 
+pub mod launch_timestamp;
 mod wrappers;
 pub use wrappers::{CountedReader, CountedWriter};
 
@@ -33,6 +35,14 @@ macro_rules! register_uint_gauge_vec {
     }};
 }
 
+#[macro_export]
+macro_rules! register_uint_gauge {
+    ($NAME:expr, $HELP:expr $(,)?) => {{
+        let gauge = $crate::UIntGauge::new($NAME, $HELP).unwrap();
+        $crate::register(Box::new(gauge.clone())).map(|_| gauge)
+    }};
+}
+
 /// Special internal registry, to collect metrics independently from the default registry.
 /// Was introduced to fix deadlock with lazy registration of metrics in the default registry.
 static INTERNAL_REGISTRY: Lazy<Registry> = Lazy::new(Registry::new);

libs/pageserver_api/Cargo.toml

@@ -13,5 +13,7 @@ bytes.workspace = true
 byteorder.workspace = true
 utils.workspace = true
 postgres_ffi.workspace = true
+enum-map.workspace = true
+serde_json.workspace = true
 
 workspace_hack.workspace = true

libs/pageserver_api/src/models.rs

@@ -1,9 +1,14 @@
-use std::num::{NonZeroU64, NonZeroUsize};
+use std::{
+    collections::HashMap,
+    num::{NonZeroU64, NonZeroUsize},
+    time::SystemTime,
+};
 
 use byteorder::{BigEndian, ReadBytesExt};
 use serde::{Deserialize, Serialize};
 use serde_with::{serde_as, DisplayFromStr};
 use utils::{
+    history_buffer::HistoryBufferWithDropCounter,
     id::{NodeId, TenantId, TimelineId},
     lsn::Lsn,
 };
@@ -29,6 +34,14 @@ pub enum TenantState {
     Broken,
 }
 
+pub mod state {
+    pub const LOADING: &str = "loading";
+    pub const ATTACHING: &str = "attaching";
+    pub const ACTIVE: &str = "active";
+    pub const STOPPING: &str = "stopping";
+    pub const BROKEN: &str = "broken";
+}
+
 impl TenantState {
     pub fn has_in_progress_downloads(&self) -> bool {
         match self {
@@ -39,6 +52,16 @@ impl TenantState {
             Self::Broken => false,
         }
     }
+
+    pub fn as_str(&self) -> &'static str {
+        match self {
+            TenantState::Loading => state::LOADING,
+            TenantState::Attaching => state::ATTACHING,
+            TenantState::Active => state::ACTIVE,
+            TenantState::Stopping => state::STOPPING,
+            TenantState::Broken => state::BROKEN,
+        }
+    }
 }
 
 /// A state of a timeline in pageserver's memory.
@@ -119,7 +142,6 @@ pub struct TenantConfigRequest {
     #[serde_as(as = "DisplayFromStr")]
     pub tenant_id: TenantId,
     #[serde(default)]
-    #[serde_as(as = "Option<DisplayFromStr>")]
     pub checkpoint_distance: Option<u64>,
     pub checkpoint_timeout: Option<String>,
     pub compaction_target_size: Option<u64>,
@@ -133,6 +155,11 @@ pub struct TenantConfigRequest {
     pub lagging_wal_timeout: Option<String>,
     pub max_lsn_wal_lag: Option<NonZeroU64>,
     pub trace_read_requests: Option<bool>,
+    // We defer the parsing of the eviction_policy field to the request handler.
+    // Otherwise we'd have to move the types for eviction policy into this package.
+    // We might do that once the eviction feature has stabilizied.
+    // For now, this field is not even documented in the openapi_spec.yml.
+    pub eviction_policy: Option<serde_json::Value>,
 }
 
 impl TenantConfigRequest {
@@ -152,6 +179,7 @@ impl TenantConfigRequest {
             lagging_wal_timeout: None,
             max_lsn_wal_lag: None,
             trace_read_requests: None,
+            eviction_policy: None,
         }
     }
 }
@@ -209,6 +237,130 @@ pub struct TimelineInfo {
     pub state: TimelineState,
 }
 
+#[derive(Debug, Clone, Serialize)]
+pub struct LayerMapInfo {
+    pub in_memory_layers: Vec<InMemoryLayerInfo>,
+    pub historic_layers: Vec<HistoricLayerInfo>,
+}
+
+#[derive(Debug, Hash, PartialEq, Eq, Clone, Copy, Serialize, Deserialize, enum_map::Enum)]
+#[repr(usize)]
+pub enum LayerAccessKind {
+    GetValueReconstructData,
+    Iter,
+    KeyIter,
+    Dump,
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct LayerAccessStatFullDetails {
+    pub when_millis_since_epoch: u64,
+    pub task_kind: &'static str,
+    pub access_kind: LayerAccessKind,
+}
+
+/// An event that impacts the layer's residence status.
+#[serde_as]
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct LayerResidenceEvent {
+    /// The time when the event occurred.
+    /// NB: this timestamp is captured while the residence status changes.
+    /// So, it might be behind/ahead of the actual residence change by a short amount of time.
+    ///
+    #[serde(rename = "timestamp_millis_since_epoch")]
+    #[serde_as(as = "serde_with::TimestampMilliSeconds")]
+    pub timestamp: SystemTime,
+    /// The new residence status of the layer.
+    pub status: LayerResidenceStatus,
+    /// The reason why we had to record this event.
+    pub reason: LayerResidenceEventReason,
+}
+
+/// The reason for recording a given [`ResidenceEvent`].
+#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
+pub enum LayerResidenceEventReason {
+    /// The layer map is being populated, e.g. during timeline load or attach.
+    /// This includes [`RemoteLayer`] objects created in [`reconcile_with_remote`].
+    /// We need to record such events because there is no persistent storage for the events.
+    LayerLoad,
+    /// We just created the layer (e.g., freeze_and_flush or compaction).
+    /// Such layers are always [`LayerResidenceStatus::Resident`].
+    LayerCreate,
+    /// We on-demand downloaded or evicted the given layer.
+    ResidenceChange,
+}
+
+/// The residence status of the layer, after the given [`LayerResidenceEvent`].
+#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
+pub enum LayerResidenceStatus {
+    /// Residence status for a layer file that exists locally.
+    /// It may also exist on the remote, we don't care here.
+    Resident,
+    /// Residence status for a layer file that only exists on the remote.
+    Evicted,
+}
+
+impl LayerResidenceEvent {
+    pub fn new(status: LayerResidenceStatus, reason: LayerResidenceEventReason) -> Self {
+        Self {
+            status,
+            reason,
+            timestamp: SystemTime::now(),
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize)]
+pub struct LayerAccessStats {
+    pub access_count_by_access_kind: HashMap<LayerAccessKind, u64>,
+    pub task_kind_access_flag: Vec<&'static str>,
+    pub first: Option<LayerAccessStatFullDetails>,
+    pub accesses_history: HistoryBufferWithDropCounter<LayerAccessStatFullDetails, 16>,
+    pub residence_events_history: HistoryBufferWithDropCounter<LayerResidenceEvent, 16>,
+}
+
+#[serde_as]
+#[derive(Debug, Clone, Serialize)]
+#[serde(tag = "kind")]
+pub enum InMemoryLayerInfo {
+    Open {
+        #[serde_as(as = "DisplayFromStr")]
+        lsn_start: Lsn,
+    },
+    Frozen {
+        #[serde_as(as = "DisplayFromStr")]
+        lsn_start: Lsn,
+        #[serde_as(as = "DisplayFromStr")]
+        lsn_end: Lsn,
+    },
+}
+
+#[serde_as]
+#[derive(Debug, Clone, Serialize)]
+#[serde(tag = "kind")]
+pub enum HistoricLayerInfo {
+    Delta {
+        layer_file_name: String,
+        layer_file_size: Option<u64>,
+
+        #[serde_as(as = "DisplayFromStr")]
+        lsn_start: Lsn,
+        #[serde_as(as = "DisplayFromStr")]
+        lsn_end: Lsn,
+        remote: bool,
+        access_stats: LayerAccessStats,
+    },
+    Image {
+        layer_file_name: String,
+        layer_file_size: Option<u64>,
+
+        #[serde_as(as = "DisplayFromStr")]
+        lsn_start: Lsn,
+        remote: bool,
+        access_stats: LayerAccessStats,
+    },
+}
+
 #[derive(Debug, Serialize, Deserialize)]
 pub struct DownloadRemoteLayersTaskSpawnRequest {
     pub max_concurrent_downloads: NonZeroUsize,
@@ -249,7 +401,7 @@ pub struct TimelineGcRequest {
 }
 
 // Wrapped in libpq CopyData
-#[derive(PartialEq, Eq)]
+#[derive(PartialEq, Eq, Debug)]
 pub enum PagestreamFeMessage {
     Exists(PagestreamExistsRequest),
     Nblocks(PagestreamNblocksRequest),

libs/pageserver_api/src/reltag.rs

@@ -98,6 +98,15 @@ impl RelTag {
 
         name
     }
+
+    pub fn with_forknum(&self, forknum: u8) -> Self {
+        RelTag {
+            forknum,
+            spcnode: self.spcnode,
+            dbnode: self.dbnode,
+            relnode: self.relnode,
+        }
+    }
 }
 
 ///

libs/pq_proto/src/lib.rs

@@ -75,27 +75,36 @@ impl StartupMessageParams {
     /// taking into account all escape sequences but leaving them as-is.
     /// [`None`] means that there's no `options` in [`Self`].
     pub fn options_raw(&self) -> Option<impl Iterator<Item = &str>> {
-        // See `postgres: pg_split_opts`.
-        let mut last_was_escape = false;
-        let iter = self
-            .get("options")?
-            .split(move |c: char| {
-                // We split by non-escaped whitespace symbols.
-                let should_split = c.is_ascii_whitespace() && !last_was_escape;
-                last_was_escape = c == '\\' && !last_was_escape;
-                should_split
-            })
-            .filter(|s| !s.is_empty());
-
-        Some(iter)
+        self.get("options").map(Self::parse_options_raw)
     }
 
     /// Split command-line options according to PostgreSQL's logic,
     /// applying all escape sequences (using owned strings as needed).
     /// [`None`] means that there's no `options` in [`Self`].
     pub fn options_escaped(&self) -> Option<impl Iterator<Item = Cow<'_, str>>> {
+        self.get("options").map(Self::parse_options_escaped)
+    }
+
+    /// Split command-line options according to PostgreSQL's logic,
+    /// taking into account all escape sequences but leaving them as-is.
+    pub fn parse_options_raw(input: &str) -> impl Iterator<Item = &str> {
         // See `postgres: pg_split_opts`.
-        let iter = self.options_raw()?.map(|s| {
+        let mut last_was_escape = false;
+        input
+            .split(move |c: char| {
+                // We split by non-escaped whitespace symbols.
+                let should_split = c.is_ascii_whitespace() && !last_was_escape;
+                last_was_escape = c == '\\' && !last_was_escape;
+                should_split
+            })
+            .filter(|s| !s.is_empty())
+    }
+
+    /// Split command-line options according to PostgreSQL's logic,
+    /// applying all escape sequences (using owned strings as needed).
+    pub fn parse_options_escaped(input: &str) -> impl Iterator<Item = Cow<'_, str>> {
+        // See `postgres: pg_split_opts`.
+        Self::parse_options_raw(input).map(|s| {
             let mut preserve_next_escape = false;
             let escape = |c| {
                 // We should remove '\\' unless it's preceded by '\\'.
@@ -108,9 +117,12 @@ impl StartupMessageParams {
                 true => Cow::Owned(s.replace(escape, "")),
                 false => Cow::Borrowed(s),
             }
-        });
+        })
+    }
 
-        Some(iter)
+    /// Iterate through key-value pairs in an arbitrary order.
+    pub fn iter(&self) -> impl Iterator<Item = (&str, &str)> {
+        self.params.iter().map(|(k, v)| (k.as_str(), v.as_str()))
     }
 
     // This function is mostly useful in tests.

libs/remote_storage/Cargo.toml

@@ -21,7 +21,7 @@ toml_edit.workspace = true
 tracing.workspace = true
 metrics.workspace = true
 utils.workspace = true
-
+pin-project-lite.workspace = true
 workspace_hack.workspace = true
 
 [dev-dependencies]

libs/remote_storage/src/s3_bucket.rs

@@ -20,7 +20,10 @@ use aws_sdk_s3::{
 };
 use aws_smithy_http::body::SdkBody;
 use hyper::Body;
-use tokio::{io, sync::Semaphore};
+use tokio::{
+    io::{self, AsyncRead},
+    sync::Semaphore,
+};
 use tokio_util::io::ReaderStream;
 use tracing::debug;
 
@@ -102,7 +105,7 @@ pub struct S3Bucket {
     // Every request to S3 can be throttled or cancelled, if a certain number of requests per second is exceeded.
     // Same goes to IAM, which is queried before every S3 request, if enabled. IAM has even lower RPS threshold.
     // The helps to ensure we don't exceed the thresholds.
-    concurrency_limiter: Semaphore,
+    concurrency_limiter: Arc<Semaphore>,
 }
 
 #[derive(Default)]
@@ -162,7 +165,7 @@ impl S3Bucket {
             client,
             bucket_name: aws_config.bucket_name.clone(),
             prefix_in_bucket,
-            concurrency_limiter: Semaphore::new(aws_config.concurrency_limit.get()),
+            concurrency_limiter: Arc::new(Semaphore::new(aws_config.concurrency_limit.get())),
         })
     }
 
@@ -194,9 +197,10 @@ impl S3Bucket {
     }
 
     async fn download_object(&self, request: GetObjectRequest) -> Result<Download, DownloadError> {
-        let _guard = self
+        let permit = self
             .concurrency_limiter
-            .acquire()
+            .clone()
+            .acquire_owned()
             .await
             .context("Concurrency limiter semaphore got closed during S3 download")
             .map_err(DownloadError::Other)?;
@@ -217,9 +221,10 @@ impl S3Bucket {
                 let metadata = object_output.metadata().cloned().map(StorageMetadata);
                 Ok(Download {
                     metadata,
-                    download_stream: Box::pin(io::BufReader::new(
+                    download_stream: Box::pin(io::BufReader::new(RatelimitedAsyncRead::new(
+                        permit,
                         object_output.body.into_async_read(),
-                    )),
+                    ))),
                 })
             }
             Err(SdkError::ServiceError {
@@ -240,6 +245,32 @@ impl S3Bucket {
     }
 }
 
+pin_project_lite::pin_project! {
+    /// An `AsyncRead` adapter which carries a permit for the lifetime of the value.
+    struct RatelimitedAsyncRead<S> {
+        permit: tokio::sync::OwnedSemaphorePermit,
+        #[pin]
+        inner: S,
+    }
+}
+
+impl<S: AsyncRead> RatelimitedAsyncRead<S> {
+    fn new(permit: tokio::sync::OwnedSemaphorePermit, inner: S) -> Self {
+        RatelimitedAsyncRead { permit, inner }
+    }
+}
+
+impl<S: AsyncRead> AsyncRead for RatelimitedAsyncRead<S> {
+    fn poll_read(
+        self: std::pin::Pin<&mut Self>,
+        cx: &mut std::task::Context<'_>,
+        buf: &mut io::ReadBuf<'_>,
+    ) -> std::task::Poll<std::io::Result<()>> {
+        let this = self.project();
+        this.inner.poll_read(cx, buf)
+    }
+}
+
 #[async_trait::async_trait]
 impl RemoteStorage for S3Bucket {
     async fn list(&self) -> anyhow::Result<Vec<RemotePath>> {

libs/tenant_size_model/Cargo.toml

@@ -7,5 +7,7 @@ license.workspace = true
 
 [dependencies]
 anyhow.workspace = true
+serde.workspace = true
+serde_json.workspace = true
 
 workspace_hack.workspace = true

libs/tenant_size_model/src/calculation.rs

@@ -0,0 +1,219 @@
+use crate::{SegmentMethod, SegmentSizeResult, SizeResult, StorageModel};
+
+//
+//                 *-g--*---D--->
+//                /
+//               /
+//              /                 *---b----*-B--->
+//             /                 /
+//            /                 /
+//      -----*--e---*-----f----* C
+//           E                  \
+//                               \
+//                                *--a---*---A-->
+//
+// If A and B need to be retained, is it cheaper to store
+// snapshot at C+a+b, or snapshots at A and B ?
+//
+// If D also needs to be retained, which is cheaper:
+//
+// 1. E+g+e+f+a+b
+// 2. D+C+a+b
+// 3. D+A+B
+
+/// [`Segment`] which has had it's size calculated.
+#[derive(Clone, Debug)]
+struct SegmentSize {
+    method: SegmentMethod,
+
+    // calculated size of this subtree, using this method
+    accum_size: u64,
+
+    seg_id: usize,
+    children: Vec<SegmentSize>,
+}
+
+struct SizeAlternatives {
+    // cheapest alternative if parent is available.
+    incremental: SegmentSize,
+
+    // cheapest alternative if parent node is not available
+    non_incremental: Option<SegmentSize>,
+}
+
+impl StorageModel {
+    pub fn calculate(&self) -> SizeResult {
+        // Build adjacency list. 'child_list' is indexed by segment id. Each entry
+        // contains a list of all child segments of the segment.
+        let mut roots: Vec<usize> = Vec::new();
+        let mut child_list: Vec<Vec<usize>> = Vec::new();
+        child_list.resize(self.segments.len(), Vec::new());
+
+        for (seg_id, seg) in self.segments.iter().enumerate() {
+            if let Some(parent_id) = seg.parent {
+                child_list[parent_id].push(seg_id);
+            } else {
+                roots.push(seg_id);
+            }
+        }
+
+        let mut segment_results = Vec::new();
+        segment_results.resize(
+            self.segments.len(),
+            SegmentSizeResult {
+                method: SegmentMethod::Skipped,
+                accum_size: 0,
+            },
+        );
+
+        let mut total_size = 0;
+        for root in roots {
+            if let Some(selected) = self.size_here(root, &child_list).non_incremental {
+                StorageModel::fill_selected_sizes(&selected, &mut segment_results);
+                total_size += selected.accum_size;
+            } else {
+                // Couldn't find any way to get this root. Error?
+            }
+        }
+
+        SizeResult {
+            total_size,
+            segments: segment_results,
+        }
+    }
+
+    fn fill_selected_sizes(selected: &SegmentSize, result: &mut Vec<SegmentSizeResult>) {
+        result[selected.seg_id] = SegmentSizeResult {
+            method: selected.method,
+            accum_size: selected.accum_size,
+        };
+        // recurse to children
+        for child in selected.children.iter() {
+            StorageModel::fill_selected_sizes(child, result);
+        }
+    }
+
+    //
+    // This is the core of the sizing calculation.
+    //
+    // This is a recursive function, that for each Segment calculates the best way
+    // to reach all the Segments that are marked as needed in this subtree, under two
+    // different conditions:
+    // a) when the parent of this segment is available (as a snaphot or through WAL), and
+    // b) when the parent of this segment is not available.
+    //
+    fn size_here(&self, seg_id: usize, child_list: &Vec<Vec<usize>>) -> SizeAlternatives {
+        let seg = &self.segments[seg_id];
+        // First figure out the best way to get each child
+        let mut children = Vec::new();
+        for child_id in &child_list[seg_id] {
+            children.push(self.size_here(*child_id, child_list))
+        }
+
+        // Method 1. If this node is not needed, we can skip it as long as we
+        // take snapshots later in each sub-tree
+        let snapshot_later = if !seg.needed {
+            let mut snapshot_later = SegmentSize {
+                seg_id,
+                method: SegmentMethod::Skipped,
+                accum_size: 0,
+                children: Vec::new(),
+            };
+
+            let mut possible = true;
+            for child in children.iter() {
+                if let Some(non_incremental) = &child.non_incremental {
+                    snapshot_later.accum_size += non_incremental.accum_size;
+                    snapshot_later.children.push(non_incremental.clone())
+                } else {
+                    possible = false;
+                    break;
+                }
+            }
+            if possible {
+                Some(snapshot_later)
+            } else {
+                None
+            }
+        } else {
+            None
+        };
+
+        // Method 2. Get a snapshot here. This assumed to be possible, if the 'size' of
+        // this Segment was given.
+        let snapshot_here = if !seg.needed || seg.parent.is_none() {
+            if let Some(snapshot_size) = seg.size {
+                let mut snapshot_here = SegmentSize {
+                    seg_id,
+                    method: SegmentMethod::SnapshotHere,
+                    accum_size: snapshot_size,
+                    children: Vec::new(),
+                };
+                for child in children.iter() {
+                    snapshot_here.accum_size += child.incremental.accum_size;
+                    snapshot_here.children.push(child.incremental.clone())
+                }
+                Some(snapshot_here)
+            } else {
+                None
+            }
+        } else {
+            None
+        };
+
+        // Method 3. Use WAL to get here from parent
+        let wal_here = {
+            let mut wal_here = SegmentSize {
+                seg_id,
+                method: SegmentMethod::Wal,
+                accum_size: if let Some(parent_id) = seg.parent {
+                    seg.lsn - self.segments[parent_id].lsn
+                } else {
+                    0
+                },
+                children: Vec::new(),
+            };
+            for child in children {
+                wal_here.accum_size += child.incremental.accum_size;
+                wal_here.children.push(child.incremental)
+            }
+            wal_here
+        };
+
+        // If the parent is not available, what's the cheapest method involving
+        // a snapshot here or later?
+        let mut cheapest_non_incremental: Option<SegmentSize> = None;
+        if let Some(snapshot_here) = snapshot_here {
+            cheapest_non_incremental = Some(snapshot_here);
+        }
+        if let Some(snapshot_later) = snapshot_later {
+            // Use <=, to prefer skipping if the size is equal
+            if let Some(parent) = &cheapest_non_incremental {
+                if snapshot_later.accum_size <= parent.accum_size {
+                    cheapest_non_incremental = Some(snapshot_later);
+                }
+            } else {
+                cheapest_non_incremental = Some(snapshot_later);
+            }
+        }
+
+        // And what's the cheapest method, if the parent is available?
+        let cheapest_incremental = if let Some(cheapest_non_incremental) = &cheapest_non_incremental
+        {
+            // Is it cheaper to use a snapshot here or later, anyway?
+            // Use <, to prefer Wal over snapshot if the cost is the same
+            if wal_here.accum_size < cheapest_non_incremental.accum_size {
+                wal_here
+            } else {
+                cheapest_non_incremental.clone()
+            }
+        } else {
+            wal_here
+        };
+
+        SizeAlternatives {
+            incremental: cheapest_incremental,
+            non_incremental: cheapest_non_incremental,
+        }
+    }
+}

libs/tenant_size_model/src/lib.rs

@@ -1,401 +1,75 @@
-use std::borrow::Cow;
-use std::collections::HashMap;
+//! Synthetic size calculation
 
-use anyhow::Context;
+mod calculation;
+pub mod svg;
 
-/// Pricing model or history size builder.
+/// StorageModel is the input to the synthetic size calculation. It represents
+/// a tree of timelines, with just the information that's needed for the
+/// calculation. This doesn't track timeline names or where each timeline
+/// begins and ends, for example. Instead, it consists of "points of interest"
+/// on the timelines. A point of interest could be the timeline start or end point,
+/// the oldest point on a timeline that needs to be retained because of PITR
+/// cutoff, or snapshot points named by the user. For each such point, and the
+/// edge connecting the points (implicit in Segment), we store information about
+/// whether we need to be able to recover to the point, and if known, the logical
+/// size at the point.
 ///
-/// Maintains knowledge of the branches and their modifications. Generic over the branch name key
-/// type.
-pub struct Storage<K: 'static> {
-    segments: Vec<Segment>,
-
-    /// Mapping from the branch name to the index of a segment describing it's latest state.
-    branches: HashMap<K, usize>,
+/// The segments must form a well-formed tree, with no loops.
+#[derive(serde::Serialize)]
+pub struct StorageModel {
+    pub segments: Vec<Segment>,
 }
 
-/// Snapshot of a branch.
-#[derive(Clone, Debug, Eq, PartialEq)]
+/// Segment represents one point in the tree of branches, *and* the edge that leads
+/// to it (if any). We don't need separate structs for points and edges, because each
+/// point can have only one parent.
+///
+/// When 'needed' is true, it means that we need to be able to reconstruct
+/// any version between 'parent.lsn' and 'lsn'. If you want to represent that only
+/// a single point is needed, create two Segments with the same lsn, and mark only
+/// the child as needed.
+///
+#[derive(Clone, Debug, Eq, PartialEq, serde::Serialize, serde::Deserialize)]
 pub struct Segment {
     /// Previous segment index into ['Storage::segments`], if any.
-    parent: Option<usize>,
+    pub parent: Option<usize>,
 
-    /// Description of how did we get to this state.
-    ///
-    /// Mainly used in the original scenarios 1..=4 with insert, delete and update. Not used when
-    /// modifying a branch directly.
-    pub op: Cow<'static, str>,
+    /// LSN at this point
+    pub lsn: u64,
 
-    /// LSN before this state
-    start_lsn: u64,
+    pub parent_lsn: Option<u64>,
 
-    /// LSN at this state
-    pub end_lsn: u64,
+    pub wal_from_parent: Option<u64>,
 
-    /// Logical size before this state
-    start_size: u64,
+    /// Logical size at this node, if known.
+    pub size: Option<u64>,
 
-    /// Logical size at this state. Can be None in the last Segment of a branch.
-    pub end_size: Option<u64>,
-
-    /// Indices to [`Storage::segments`]
-    ///
-    /// FIXME: this could be an Option<usize>
-    children_after: Vec<usize>,
-
-    /// Determined by `retention_period` given to [`Storage::calculate`]
+    /// If true, the segment from parent to this node is needed by `retention_period`
+    /// if wal_from_parent is needed by `retention_period`?
     pub needed: bool,
 }
 
-//
-//
-//
-//
-//                 *-g--*---D--->
-//                /
-//               /
-//              /                 *---b----*-B--->
-//             /                 /
-//            /                 /
-//      -----*--e---*-----f----* C
-//           E                  \
-//                               \
-//                                *--a---*---A-->
-//
-// If A and B need to be retained, is it cheaper to store
-// snapshot at C+a+b, or snapshots at A and B ?
-//
-// If D also needs to be retained, which is cheaper:
-//
-// 1. E+g+e+f+a+b
-// 2. D+C+a+b
-// 3. D+A+B
+/// Result of synthetic size calculation. Returned by StorageModel::calculate()
+pub struct SizeResult {
+    pub total_size: u64,
 
-/// [`Segment`] which has had it's size calculated.
-pub struct SegmentSize {
-    pub seg_id: usize,
-
-    pub method: SegmentMethod,
-
-    this_size: u64,
-
-    pub children: Vec<SegmentSize>,
+    // This has same length as the StorageModel::segments vector in the input.
+    // Each entry in this array corresponds to the entry with same index in
+    // StorageModel::segments.
+    pub segments: Vec<SegmentSizeResult>,
 }
 
-impl SegmentSize {
-    fn total(&self) -> u64 {
-        self.this_size + self.children.iter().fold(0, |acc, x| acc + x.total())
-    }
-
-    pub fn total_children(&self) -> u64 {
-        if self.method == SnapshotAfter {
-            self.this_size + self.children.iter().fold(0, |acc, x| acc + x.total())
-        } else {
-            self.children.iter().fold(0, |acc, x| acc + x.total())
-        }
-    }
+#[derive(Clone, Debug, Eq, PartialEq, serde::Serialize, serde::Deserialize)]
+pub struct SegmentSizeResult {
+    pub method: SegmentMethod,
+    // calculated size of this subtree, using this method
+    pub accum_size: u64,
 }
 
 /// Different methods to retain history from a particular state
-#[derive(Clone, Copy, Debug, Eq, PartialEq)]
+#[derive(Clone, Copy, Debug, Eq, PartialEq, serde::Serialize, serde::Deserialize)]
 pub enum SegmentMethod {
-    SnapshotAfter,
-    Wal,
-    WalNeeded,
+    SnapshotHere, // A logical snapshot is needed after this segment
+    Wal,          // Keep WAL leading up to this node
     Skipped,
 }
-
-use SegmentMethod::*;
-
-impl<K: std::hash::Hash + Eq + 'static> Storage<K> {
-    /// Creates a new storage with the given default branch name.
-    pub fn new(initial_branch: K) -> Storage<K> {
-        let init_segment = Segment {
-            op: "".into(),
-            needed: false,
-            parent: None,
-            start_lsn: 0,
-            end_lsn: 0,
-            start_size: 0,
-            end_size: Some(0),
-            children_after: Vec::new(),
-        };
-
-        Storage {
-            segments: vec![init_segment],
-            branches: HashMap::from([(initial_branch, 0)]),
-        }
-    }
-
-    /// Advances the branch with a new point, at given LSN.
-    pub fn insert_point<Q: ?Sized>(
-        &mut self,
-        branch: &Q,
-        op: Cow<'static, str>,
-        lsn: u64,
-        size: Option<u64>,
-    ) -> anyhow::Result<()>
-    where
-        K: std::borrow::Borrow<Q>,
-        Q: std::hash::Hash + Eq + std::fmt::Debug,
-    {
-        let Some(lastseg_id) = self.branches.get(branch).copied() else { anyhow::bail!("branch not found: {branch:?}") };
-        let newseg_id = self.segments.len();
-        let lastseg = &mut self.segments[lastseg_id];
-
-        assert!(lsn > lastseg.end_lsn);
-
-        let Some(start_size) = lastseg.end_size else { anyhow::bail!("no end_size on latest segment for {branch:?}") };
-
-        let newseg = Segment {
-            op,
-            parent: Some(lastseg_id),
-            start_lsn: lastseg.end_lsn,
-            end_lsn: lsn,
-            start_size,
-            end_size: size,
-            children_after: Vec::new(),
-            needed: false,
-        };
-        lastseg.children_after.push(newseg_id);
-
-        self.segments.push(newseg);
-        *self.branches.get_mut(branch).expect("read already") = newseg_id;
-
-        Ok(())
-    }
-
-    /// Advances the branch with the named operation, by the relative LSN and logical size bytes.
-    pub fn modify_branch<Q: ?Sized>(
-        &mut self,
-        branch: &Q,
-        op: Cow<'static, str>,
-        lsn_bytes: u64,
-        size_bytes: i64,
-    ) -> anyhow::Result<()>
-    where
-        K: std::borrow::Borrow<Q>,
-        Q: std::hash::Hash + Eq + std::fmt::Debug,
-    {
-        let Some(lastseg_id) = self.branches.get(branch).copied() else { anyhow::bail!("branch not found: {branch:?}") };
-        let newseg_id = self.segments.len();
-        let lastseg = &mut self.segments[lastseg_id];
-
-        let Some(last_end_size) = lastseg.end_size else { anyhow::bail!("no end_size on latest segment for {branch:?}") };
-
-        let newseg = Segment {
-            op,
-            parent: Some(lastseg_id),
-            start_lsn: lastseg.end_lsn,
-            end_lsn: lastseg.end_lsn + lsn_bytes,
-            start_size: last_end_size,
-            end_size: Some((last_end_size as i64 + size_bytes) as u64),
-            children_after: Vec::new(),
-            needed: false,
-        };
-        lastseg.children_after.push(newseg_id);
-
-        self.segments.push(newseg);
-        *self.branches.get_mut(branch).expect("read already") = newseg_id;
-        Ok(())
-    }
-
-    pub fn insert<Q: ?Sized>(&mut self, branch: &Q, bytes: u64) -> anyhow::Result<()>
-    where
-        K: std::borrow::Borrow<Q>,
-        Q: std::hash::Hash + Eq + std::fmt::Debug,
-    {
-        self.modify_branch(branch, "insert".into(), bytes, bytes as i64)
-    }
-
-    pub fn update<Q: ?Sized>(&mut self, branch: &Q, bytes: u64) -> anyhow::Result<()>
-    where
-        K: std::borrow::Borrow<Q>,
-        Q: std::hash::Hash + Eq + std::fmt::Debug,
-    {
-        self.modify_branch(branch, "update".into(), bytes, 0i64)
-    }
-
-    pub fn delete<Q: ?Sized>(&mut self, branch: &Q, bytes: u64) -> anyhow::Result<()>
-    where
-        K: std::borrow::Borrow<Q>,
-        Q: std::hash::Hash + Eq + std::fmt::Debug,
-    {
-        self.modify_branch(branch, "delete".into(), bytes, -(bytes as i64))
-    }
-
-    pub fn branch<Q: ?Sized>(&mut self, parent: &Q, name: K) -> anyhow::Result<()>
-    where
-        K: std::borrow::Borrow<Q> + std::fmt::Debug,
-        Q: std::hash::Hash + Eq + std::fmt::Debug,
-    {
-        // Find the right segment
-        let branchseg_id = *self.branches.get(parent).with_context(|| {
-            format!(
-                "should had found the parent {:?} by key. in branches {:?}",
-                parent, self.branches
-            )
-        })?;
-
-        let _branchseg = &mut self.segments[branchseg_id];
-
-        // Create branch name for it
-        self.branches.insert(name, branchseg_id);
-        Ok(())
-    }
-
-    pub fn calculate(&mut self, retention_period: u64) -> anyhow::Result<SegmentSize> {
-        // Phase 1: Mark all the segments that need to be retained
-        for (_branch, &last_seg_id) in self.branches.iter() {
-            let last_seg = &self.segments[last_seg_id];
-            let cutoff_lsn = last_seg.start_lsn.saturating_sub(retention_period);
-            let mut seg_id = last_seg_id;
-            loop {
-                let seg = &mut self.segments[seg_id];
-                if seg.end_lsn < cutoff_lsn {
-                    break;
-                }
-                seg.needed = true;
-                if let Some(prev_seg_id) = seg.parent {
-                    seg_id = prev_seg_id;
-                } else {
-                    break;
-                }
-            }
-        }
-
-        // Phase 2: For each oldest segment in a chain that needs to be retained,
-        // calculate if we should store snapshot or WAL
-        self.size_from_snapshot_later(0)
-    }
-
-    fn size_from_wal(&self, seg_id: usize) -> anyhow::Result<SegmentSize> {
-        let seg = &self.segments[seg_id];
-
-        let this_size = seg.end_lsn - seg.start_lsn;
-
-        let mut children = Vec::new();
-
-        // try both ways
-        for &child_id in seg.children_after.iter() {
-            // try each child both ways
-            let child = &self.segments[child_id];
-            let p1 = self.size_from_wal(child_id)?;
-
-            let p = if !child.needed {
-                let p2 = self.size_from_snapshot_later(child_id)?;
-                if p1.total() < p2.total() {
-                    p1
-                } else {
-                    p2
-                }
-            } else {
-                p1
-            };
-            children.push(p);
-        }
-        Ok(SegmentSize {
-            seg_id,
-            method: if seg.needed { WalNeeded } else { Wal },
-            this_size,
-            children,
-        })
-    }
-
-    fn size_from_snapshot_later(&self, seg_id: usize) -> anyhow::Result<SegmentSize> {
-        // If this is needed, then it's time to do the snapshot and continue
-        // with wal method.
-        let seg = &self.segments[seg_id];
-        //eprintln!("snap: seg{}: {} needed: {}", seg_id, seg.children_after.len(), seg.needed);
-        if seg.needed {
-            let mut children = Vec::new();
-
-            for &child_id in seg.children_after.iter() {
-                // try each child both ways
-                let child = &self.segments[child_id];
-                let p1 = self.size_from_wal(child_id)?;
-
-                let p = if !child.needed {
-                    let p2 = self.size_from_snapshot_later(child_id)?;
-                    if p1.total() < p2.total() {
-                        p1
-                    } else {
-                        p2
-                    }
-                } else {
-                    p1
-                };
-                children.push(p);
-            }
-            Ok(SegmentSize {
-                seg_id,
-                method: WalNeeded,
-                this_size: seg.start_size,
-                children,
-            })
-        } else {
-            // If any of the direct children are "needed", need to be able to reconstruct here
-            let mut children_needed = false;
-            for &child in seg.children_after.iter() {
-                let seg = &self.segments[child];
-                if seg.needed {
-                    children_needed = true;
-                    break;
-                }
-            }
-
-            let method1 = if !children_needed {
-                let mut children = Vec::new();
-                for child in seg.children_after.iter() {
-                    children.push(self.size_from_snapshot_later(*child)?);
-                }
-                Some(SegmentSize {
-                    seg_id,
-                    method: Skipped,
-                    this_size: 0,
-                    children,
-                })
-            } else {
-                None
-            };
-
-            // If this a junction, consider snapshotting here
-            let method2 = if children_needed || seg.children_after.len() >= 2 {
-                let mut children = Vec::new();
-                for child in seg.children_after.iter() {
-                    children.push(self.size_from_wal(*child)?);
-                }
-                let Some(this_size) = seg.end_size else { anyhow::bail!("no end_size at junction {seg_id}") };
-                Some(SegmentSize {
-                    seg_id,
-                    method: SnapshotAfter,
-                    this_size,
-                    children,
-                })
-            } else {
-                None
-            };
-
-            Ok(match (method1, method2) {
-                (None, None) => anyhow::bail!(
-                    "neither method was applicable: children_after={}, children_needed={}",
-                    seg.children_after.len(),
-                    children_needed
-                ),
-                (Some(method), None) => method,
-                (None, Some(method)) => method,
-                (Some(method1), Some(method2)) => {
-                    if method1.total() < method2.total() {
-                        method1
-                    } else {
-                        method2
-                    }
-                }
-            })
-        }
-    }
-
-    pub fn into_segments(self) -> Vec<Segment> {
-        self.segments
-    }
-}

libs/tenant_size_model/src/main.rs

@@ -1,269 +0,0 @@
-//! Tenant size model testing ground.
-//!
-//! Has a number of scenarios and a `main` for invoking these by number, calculating the history
-//! size, outputs graphviz graph. Makefile in directory shows how to use graphviz to turn scenarios
-//! into pngs.
-
-use tenant_size_model::{Segment, SegmentSize, Storage};
-
-// Main branch only. Some updates on it.
-fn scenario_1() -> anyhow::Result<(Vec<Segment>, SegmentSize)> {
-    // Create main branch
-    let mut storage = Storage::new("main");
-
-    // Bulk load 5 GB of data to it
-    storage.insert("main", 5_000)?;
-
-    // Stream of updates
-    for _ in 0..5 {
-        storage.update("main", 1_000)?;
-    }
-
-    let size = storage.calculate(1000)?;
-
-    Ok((storage.into_segments(), size))
-}
-
-// Main branch only. Some updates on it.
-fn scenario_2() -> anyhow::Result<(Vec<Segment>, SegmentSize)> {
-    // Create main branch
-    let mut storage = Storage::new("main");
-
-    // Bulk load 5 GB of data to it
-    storage.insert("main", 5_000)?;
-
-    // Stream of updates
-    for _ in 0..5 {
-        storage.update("main", 1_000)?;
-    }
-
-    // Branch
-    storage.branch("main", "child")?;
-    storage.update("child", 1_000)?;
-
-    // More updates on parent
-    storage.update("main", 1_000)?;
-
-    let size = storage.calculate(1000)?;
-
-    Ok((storage.into_segments(), size))
-}
-
-// Like 2, but more updates on main
-fn scenario_3() -> anyhow::Result<(Vec<Segment>, SegmentSize)> {
-    // Create main branch
-    let mut storage = Storage::new("main");
-
-    // Bulk load 5 GB of data to it
-    storage.insert("main", 5_000)?;
-
-    // Stream of updates
-    for _ in 0..5 {
-        storage.update("main", 1_000)?;
-    }
-
-    // Branch
-    storage.branch("main", "child")?;
-    storage.update("child", 1_000)?;
-
-    // More updates on parent
-    for _ in 0..5 {
-        storage.update("main", 1_000)?;
-    }
-
-    let size = storage.calculate(1000)?;
-
-    Ok((storage.into_segments(), size))
-}
-
-// Diverged branches
-fn scenario_4() -> anyhow::Result<(Vec<Segment>, SegmentSize)> {
-    // Create main branch
-    let mut storage = Storage::new("main");
-
-    // Bulk load 5 GB of data to it
-    storage.insert("main", 5_000)?;
-
-    // Stream of updates
-    for _ in 0..5 {
-        storage.update("main", 1_000)?;
-    }
-
-    // Branch
-    storage.branch("main", "child")?;
-    storage.update("child", 1_000)?;
-
-    // More updates on parent
-    for _ in 0..8 {
-        storage.update("main", 1_000)?;
-    }
-
-    let size = storage.calculate(1000)?;
-
-    Ok((storage.into_segments(), size))
-}
-
-fn scenario_5() -> anyhow::Result<(Vec<Segment>, SegmentSize)> {
-    let mut storage = Storage::new("a");
-    storage.insert("a", 5000)?;
-    storage.branch("a", "b")?;
-    storage.update("b", 4000)?;
-    storage.update("a", 2000)?;
-    storage.branch("a", "c")?;
-    storage.insert("c", 4000)?;
-    storage.insert("a", 2000)?;
-
-    let size = storage.calculate(5000)?;
-
-    Ok((storage.into_segments(), size))
-}
-
-fn scenario_6() -> anyhow::Result<(Vec<Segment>, SegmentSize)> {
-    use std::borrow::Cow;
-
-    const NO_OP: Cow<'static, str> = Cow::Borrowed("");
-
-    let branches = [
-        Some(0x7ff1edab8182025f15ae33482edb590a_u128),
-        Some(0xb1719e044db05401a05a2ed588a3ad3f),
-        Some(0xb68d6691c895ad0a70809470020929ef),
-    ];
-
-    // compared to other scenarios, this one uses bytes instead of kB
-
-    let mut storage = Storage::new(None);
-
-    storage.branch(&None, branches[0])?; // at 0
-    storage.modify_branch(&branches[0], NO_OP, 108951064, 43696128)?; // at 108951064
-    storage.branch(&branches[0], branches[1])?; // at 108951064
-    storage.modify_branch(&branches[1], NO_OP, 15560408, -1851392)?; // at 124511472
-    storage.modify_branch(&branches[0], NO_OP, 174464360, -1531904)?; // at 283415424
-    storage.branch(&branches[0], branches[2])?; // at 283415424
-    storage.modify_branch(&branches[2], NO_OP, 15906192, 8192)?; // at 299321616
-    storage.modify_branch(&branches[0], NO_OP, 18909976, 32768)?; // at 302325400
-
-    let size = storage.calculate(100_000)?;
-
-    Ok((storage.into_segments(), size))
-}
-
-fn main() {
-    let args: Vec<String> = std::env::args().collect();
-
-    let scenario = if args.len() < 2 { "1" } else { &args[1] };
-
-    let (segments, size) = match scenario {
-        "1" => scenario_1(),
-        "2" => scenario_2(),
-        "3" => scenario_3(),
-        "4" => scenario_4(),
-        "5" => scenario_5(),
-        "6" => scenario_6(),
-        other => {
-            eprintln!("invalid scenario {}", other);
-            std::process::exit(1);
-        }
-    }
-    .unwrap();
-
-    graphviz_tree(&segments, &size);
-}
-
-fn graphviz_recurse(segments: &[Segment], node: &SegmentSize) {
-    use tenant_size_model::SegmentMethod::*;
-
-    let seg_id = node.seg_id;
-    let seg = segments.get(seg_id).unwrap();
-    let lsn = seg.end_lsn;
-    let size = seg.end_size.unwrap_or(0);
-    let method = node.method;
-
-    println!("  {{");
-    println!("    node [width=0.1 height=0.1 shape=oval]");
-
-    let tenant_size = node.total_children();
-
-    let penwidth = if seg.needed { 6 } else { 3 };
-    let x = match method {
-        SnapshotAfter =>
-            format!("label=\"lsn: {lsn}\\nsize: {size}\\ntenant_size: {tenant_size}\" style=filled penwidth={penwidth}"),
-        Wal =>
-            format!("label=\"lsn: {lsn}\\nsize: {size}\\ntenant_size: {tenant_size}\" color=\"black\" penwidth={penwidth}"),
-        WalNeeded =>
-            format!("label=\"lsn: {lsn}\\nsize: {size}\\ntenant_size: {tenant_size}\" color=\"black\" penwidth={penwidth}"),
-        Skipped =>
-            format!("label=\"lsn: {lsn}\\nsize: {size}\\ntenant_size: {tenant_size}\" color=\"gray\" penwidth={penwidth}"),
-    };
-
-    println!("    \"seg{seg_id}\" [{x}]");
-    println!("  }}");
-
-    // Recurse. Much of the data is actually on the edge
-    for child in node.children.iter() {
-        let child_id = child.seg_id;
-        graphviz_recurse(segments, child);
-
-        let edge_color = match child.method {
-            SnapshotAfter => "gray",
-            Wal => "black",
-            WalNeeded => "black",
-            Skipped => "gray",
-        };
-
-        println!("  {{");
-        println!("    edge [] ");
-        print!("    \"seg{seg_id}\" -> \"seg{child_id}\" [");
-        print!("color={edge_color}");
-        if child.method == WalNeeded {
-            print!(" penwidth=6");
-        }
-        if child.method == Wal {
-            print!(" penwidth=3");
-        }
-
-        let next = segments.get(child_id).unwrap();
-
-        if next.op.is_empty() {
-            print!(
-                " label=\"{} / {}\"",
-                next.end_lsn - seg.end_lsn,
-                (next.end_size.unwrap_or(0) as i128 - seg.end_size.unwrap_or(0) as i128)
-            );
-        } else {
-            print!(" label=\"{}: {}\"", next.op, next.end_lsn - seg.end_lsn);
-        }
-        println!("]");
-        println!("  }}");
-    }
-}
-
-fn graphviz_tree(segments: &[Segment], tree: &SegmentSize) {
-    println!("digraph G {{");
-    println!("  fontname=\"Helvetica,Arial,sans-serif\"");
-    println!("  node [fontname=\"Helvetica,Arial,sans-serif\"]");
-    println!("  edge [fontname=\"Helvetica,Arial,sans-serif\"]");
-    println!("  graph [center=1 rankdir=LR]");
-    println!("  edge [dir=none]");
-
-    graphviz_recurse(segments, tree);
-
-    println!("}}");
-}
-
-#[test]
-fn scenarios_return_same_size() {
-    type ScenarioFn = fn() -> anyhow::Result<(Vec<Segment>, SegmentSize)>;
-    let truths: &[(u32, ScenarioFn, _)] = &[
-        (line!(), scenario_1, 8000),
-        (line!(), scenario_2, 9000),
-        (line!(), scenario_3, 13000),
-        (line!(), scenario_4, 16000),
-        (line!(), scenario_5, 17000),
-        (line!(), scenario_6, 333_792_000),
-    ];
-
-    for (line, scenario, expected) in truths {
-        let (_, size) = scenario().unwrap();
-        assert_eq!(*expected, size.total_children(), "scenario on line {line}");
-    }
-}

libs/tenant_size_model/src/svg.rs

@@ -0,0 +1,193 @@
+use crate::{SegmentMethod, SegmentSizeResult, SizeResult, StorageModel};
+use std::fmt::Write;
+
+const SVG_WIDTH: f32 = 500.0;
+
+struct SvgDraw<'a> {
+    storage: &'a StorageModel,
+    branches: &'a [String],
+    seg_to_branch: &'a [usize],
+    sizes: &'a [SegmentSizeResult],
+
+    // layout
+    xscale: f32,
+    min_lsn: u64,
+    seg_coordinates: Vec<(f32, f32)>,
+}
+
+fn draw_legend(result: &mut String) -> anyhow::Result<()> {
+    writeln!(
+        result,
+        "<circle cx=\"10\" cy=\"10\" r=\"5\" stroke=\"red\"/>"
+    )?;
+    writeln!(result, "<text x=\"20\" y=\"15\">logical snapshot</text>")?;
+    writeln!(
+        result,
+        "<line x1=\"5\" y1=\"30\" x2=\"15\" y2=\"30\" stroke-width=\"6\" stroke=\"black\" />"
+    )?;
+    writeln!(
+        result,
+        "<text x=\"20\" y=\"35\">WAL within retention period</text>"
+    )?;
+    writeln!(
+        result,
+        "<line x1=\"5\" y1=\"50\" x2=\"15\" y2=\"50\" stroke-width=\"3\" stroke=\"black\" />"
+    )?;
+    writeln!(
+        result,
+        "<text x=\"20\" y=\"55\">WAL retained to avoid copy</text>"
+    )?;
+    writeln!(
+        result,
+        "<line x1=\"5\" y1=\"70\" x2=\"15\" y2=\"70\" stroke-width=\"1\" stroke=\"gray\" />"
+    )?;
+    writeln!(result, "<text x=\"20\" y=\"75\">WAL not retained</text>")?;
+    Ok(())
+}
+
+pub fn draw_svg(
+    storage: &StorageModel,
+    branches: &[String],
+    seg_to_branch: &[usize],
+    sizes: &SizeResult,
+) -> anyhow::Result<String> {
+    let mut draw = SvgDraw {
+        storage,
+        branches,
+        seg_to_branch,
+        sizes: &sizes.segments,
+
+        xscale: 0.0,
+        min_lsn: 0,
+        seg_coordinates: Vec::new(),
+    };
+
+    let mut result = String::new();
+
+    writeln!(result, "<svg xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\" height=\"300\" width=\"500\">")?;
+
+    draw.calculate_svg_layout();
+
+    // Draw the tree
+    for (seg_id, _seg) in storage.segments.iter().enumerate() {
+        draw.draw_seg_phase1(seg_id, &mut result)?;
+    }
+
+    // Draw snapshots
+    for (seg_id, _seg) in storage.segments.iter().enumerate() {
+        draw.draw_seg_phase2(seg_id, &mut result)?;
+    }
+
+    draw_legend(&mut result)?;
+
+    write!(result, "</svg>")?;
+
+    Ok(result)
+}
+
+impl<'a> SvgDraw<'a> {
+    fn calculate_svg_layout(&mut self) {
+        // Find x scale
+        let segments = &self.storage.segments;
+        let min_lsn = segments.iter().map(|s| s.lsn).fold(u64::MAX, std::cmp::min);
+        let max_lsn = segments.iter().map(|s| s.lsn).fold(0, std::cmp::max);
+
+        // Start with 1 pixel = 1 byte. Double the scale until it fits into the image
+        let mut xscale = 1.0;
+        while (max_lsn - min_lsn) as f32 / xscale > SVG_WIDTH {
+            xscale *= 2.0;
+        }
+
+        // Layout the timelines on Y dimension.
+        // TODO
+        let mut y = 100.0;
+        let mut branch_y_coordinates = Vec::new();
+        for _branch in self.branches {
+            branch_y_coordinates.push(y);
+            y += 40.0;
+        }
+
+        // Calculate coordinates for each point
+        let seg_coordinates = std::iter::zip(segments, self.seg_to_branch)
+            .map(|(seg, branch_id)| {
+                let x = (seg.lsn - min_lsn) as f32 / xscale;
+                let y = branch_y_coordinates[*branch_id];
+                (x, y)
+            })
+            .collect();
+
+        self.xscale = xscale;
+        self.min_lsn = min_lsn;
+        self.seg_coordinates = seg_coordinates;
+    }
+
+    /// Draws lines between points
+    fn draw_seg_phase1(&self, seg_id: usize, result: &mut String) -> anyhow::Result<()> {
+        let seg = &self.storage.segments[seg_id];
+
+        let wal_bytes = if let Some(parent_id) = seg.parent {
+            seg.lsn - self.storage.segments[parent_id].lsn
+        } else {
+            0
+        };
+
+        let style = match self.sizes[seg_id].method {
+            SegmentMethod::SnapshotHere => "stroke-width=\"1\" stroke=\"gray\"",
+            SegmentMethod::Wal if seg.needed && wal_bytes > 0 => {
+                "stroke-width=\"6\" stroke=\"black\""
+            }
+            SegmentMethod::Wal => "stroke-width=\"3\" stroke=\"black\"",
+            SegmentMethod::Skipped => "stroke-width=\"1\" stroke=\"gray\"",
+        };
+        if let Some(parent_id) = seg.parent {
+            let (x1, y1) = self.seg_coordinates[parent_id];
+            let (x2, y2) = self.seg_coordinates[seg_id];
+
+            writeln!(
+                result,
+                "<line x1=\"{x1}\" y1=\"{y1}\" x2=\"{x2}\" y2=\"{y2}\" {style}>",
+            )?;
+            writeln!(
+                result,
+                "  <title>{wal_bytes} bytes of WAL (seg {seg_id})</title>"
+            )?;
+            writeln!(result, "</line>")?;
+        } else {
+            // draw a little dash to mark the starting point of this branch
+            let (x, y) = self.seg_coordinates[seg_id];
+            let (x1, y1) = (x, y - 5.0);
+            let (x2, y2) = (x, y + 5.0);
+
+            writeln!(
+                result,
+                "<line x1=\"{x1}\" y1=\"{y1}\" x2=\"{x2}\" y2=\"{y2}\" {style}>",
+            )?;
+            writeln!(result, "  <title>(seg {seg_id})</title>")?;
+            writeln!(result, "</line>")?;
+        }
+
+        Ok(())
+    }
+
+    /// Draw circles where snapshots are taken
+    fn draw_seg_phase2(&self, seg_id: usize, result: &mut String) -> anyhow::Result<()> {
+        let seg = &self.storage.segments[seg_id];
+
+        // draw a snapshot point if it's needed
+        let (coord_x, coord_y) = self.seg_coordinates[seg_id];
+        if self.sizes[seg_id].method == SegmentMethod::SnapshotHere {
+            writeln!(
+                result,
+                "<circle cx=\"{coord_x}\" cy=\"{coord_y}\" r=\"5\" stroke=\"red\">",
+            )?;
+            writeln!(
+                result,
+                "  <title>logical size {}</title>",
+                seg.size.unwrap()
+            )?;
+            write!(result, "</circle>")?;
+        }
+
+        Ok(())
+    }
+}

libs/tenant_size_model/tests/tests.rs

@@ -0,0 +1,317 @@
+//! Tenant size model tests.
+
+use tenant_size_model::{Segment, SizeResult, StorageModel};
+
+use std::collections::HashMap;
+
+struct ScenarioBuilder {
+    segments: Vec<Segment>,
+
+    /// Mapping from the branch name to the index of a segment describing its latest state.
+    branches: HashMap<String, usize>,
+}
+
+impl ScenarioBuilder {
+    /// Creates a new storage with the given default branch name.
+    pub fn new(initial_branch: &str) -> ScenarioBuilder {
+        let init_segment = Segment {
+            parent: None,
+            lsn: 0,
+            parent_lsn: None,
+            wal_from_parent: None,
+            size: Some(0),
+            needed: false, // determined later
+        };
+
+        ScenarioBuilder {
+            segments: vec![init_segment],
+            branches: HashMap::from([(initial_branch.into(), 0)]),
+        }
+    }
+
+    /// Advances the branch with the named operation, by the relative LSN and logical size bytes.
+    pub fn modify_branch(&mut self, branch: &str, lsn_bytes: u64, size_bytes: i64) {
+        let lastseg_id = *self.branches.get(branch).unwrap();
+        let newseg_id = self.segments.len();
+        let lastseg = &mut self.segments[lastseg_id];
+
+        let newseg = Segment {
+            parent: Some(lastseg_id),
+            lsn: lastseg.lsn + lsn_bytes,
+            parent_lsn: Some(lastseg.lsn),
+            wal_from_parent: Some(lsn_bytes),
+            size: Some((lastseg.size.unwrap() as i64 + size_bytes) as u64),
+            needed: false,
+        };
+
+        self.segments.push(newseg);
+        *self.branches.get_mut(branch).expect("read already") = newseg_id;
+    }
+
+    pub fn insert(&mut self, branch: &str, bytes: u64) {
+        self.modify_branch(branch, bytes, bytes as i64);
+    }
+
+    pub fn update(&mut self, branch: &str, bytes: u64) {
+        self.modify_branch(branch, bytes, 0i64);
+    }
+
+    pub fn _delete(&mut self, branch: &str, bytes: u64) {
+        self.modify_branch(branch, bytes, -(bytes as i64));
+    }
+
+    /// Panics if the parent branch cannot be found.
+    pub fn branch(&mut self, parent: &str, name: &str) {
+        // Find the right segment
+        let branchseg_id = *self
+            .branches
+            .get(parent)
+            .expect("should had found the parent by key");
+        let _branchseg = &mut self.segments[branchseg_id];
+
+        // Create branch name for it
+        self.branches.insert(name.to_string(), branchseg_id);
+    }
+
+    pub fn calculate(&mut self, retention_period: u64) -> (StorageModel, SizeResult) {
+        // Phase 1: Mark all the segments that need to be retained
+        for (_branch, &last_seg_id) in self.branches.iter() {
+            let last_seg = &self.segments[last_seg_id];
+            let cutoff_lsn = last_seg.lsn.saturating_sub(retention_period);
+            let mut seg_id = last_seg_id;
+            loop {
+                let seg = &mut self.segments[seg_id];
+                if seg.lsn <= cutoff_lsn {
+                    break;
+                }
+                seg.needed = true;
+                if let Some(prev_seg_id) = seg.parent {
+                    seg_id = prev_seg_id;
+                } else {
+                    break;
+                }
+            }
+        }
+
+        // Perform the calculation
+        let storage_model = StorageModel {
+            segments: self.segments.clone(),
+        };
+        let size_result = storage_model.calculate();
+        (storage_model, size_result)
+    }
+}
+
+// Main branch only. Some updates on it.
+#[test]
+fn scenario_1() {
+    // Create main branch
+    let mut scenario = ScenarioBuilder::new("main");
+
+    // Bulk load 5 GB of data to it
+    scenario.insert("main", 5_000);
+
+    // Stream of updates
+    for _ in 0..5 {
+        scenario.update("main", 1_000);
+    }
+
+    // Calculate the synthetic size with retention horizon 1000
+    let (_model, result) = scenario.calculate(1000);
+
+    // The end of the branch is at LSN 10000. Need to retain
+    // a logical snapshot at LSN 9000, plus the WAL between 9000-10000.
+    // The logical snapshot has size 5000.
+    assert_eq!(result.total_size, 5000 + 1000);
+}
+
+// Main branch only. Some updates on it.
+#[test]
+fn scenario_2() {
+    // Create main branch
+    let mut scenario = ScenarioBuilder::new("main");
+
+    // Bulk load 5 GB of data to it
+    scenario.insert("main", 5_000);
+
+    // Stream of updates
+    for _ in 0..5 {
+        scenario.update("main", 1_000);
+    }
+
+    // Branch
+    scenario.branch("main", "child");
+    scenario.update("child", 1_000);
+
+    // More updates on parent
+    scenario.update("main", 1_000);
+
+    //
+    // The history looks like this now:
+    //
+    //         10000          11000
+    // *----*----*--------------*    main
+    //           |
+    //           |            11000
+    //           +--------------     child
+    //
+    //
+    // With retention horizon 1000, we need to retain logical snapshot
+    // at the branch point, size 5000, and the WAL from 10000-11000 on
+    // both branches.
+    let (_model, result) = scenario.calculate(1000);
+
+    assert_eq!(result.total_size, 5000 + 1000 + 1000);
+}
+
+// Like 2, but more updates on main
+#[test]
+fn scenario_3() {
+    // Create main branch
+    let mut scenario = ScenarioBuilder::new("main");
+
+    // Bulk load 5 GB of data to it
+    scenario.insert("main", 5_000);
+
+    // Stream of updates
+    for _ in 0..5 {
+        scenario.update("main", 1_000);
+    }
+
+    // Branch
+    scenario.branch("main", "child");
+    scenario.update("child", 1_000);
+
+    // More updates on parent
+    for _ in 0..5 {
+        scenario.update("main", 1_000);
+    }
+
+    //
+    // The history looks like this now:
+    //
+    //         10000                                 15000
+    // *----*----*------------------------------------*    main
+    //           |
+    //           |            11000
+    //           +--------------     child
+    //
+    //
+    // With retention horizon 1000, it's still cheapest to retain
+    // - snapshot at branch point (size 5000)
+    // - WAL on child between 10000-11000
+    // - WAL on main between 10000-15000
+    //
+    // This is in total 5000 + 1000 + 5000
+    //
+    let (_model, result) = scenario.calculate(1000);
+
+    assert_eq!(result.total_size, 5000 + 1000 + 5000);
+}
+
+// Diverged branches
+#[test]
+fn scenario_4() {
+    // Create main branch
+    let mut scenario = ScenarioBuilder::new("main");
+
+    // Bulk load 5 GB of data to it
+    scenario.insert("main", 5_000);
+
+    // Stream of updates
+    for _ in 0..5 {
+        scenario.update("main", 1_000);
+    }
+
+    // Branch
+    scenario.branch("main", "child");
+    scenario.update("child", 1_000);
+
+    // More updates on parent
+    for _ in 0..8 {
+        scenario.update("main", 1_000);
+    }
+
+    //
+    // The history looks like this now:
+    //
+    //         10000                                 18000
+    // *----*----*------------------------------------*    main
+    //           |
+    //           |            11000
+    //           +--------------     child
+    //
+    //
+    // With retention horizon 1000, it's now cheapest to retain
+    // separate snapshots on both branches:
+    // - snapshot on main branch at LSN 17000 (size 5000)
+    // - WAL on main between 17000-18000
+    // - snapshot on child branch at LSN 10000 (size 5000)
+    // - WAL on child between 10000-11000
+    //
+    // This is in total 5000 + 1000 + 5000 + 1000 = 12000
+    //
+    // (If we used the the method from the previous scenario, and
+    // kept only snapshot at the branch point, we'd need to keep
+    // all the WAL between 10000-18000 on the main branch, so
+    // the total size would be 5000 + 1000 + 8000 = 14000. The
+    // calculation always picks the cheapest alternative)
+
+    let (_model, result) = scenario.calculate(1000);
+
+    assert_eq!(result.total_size, 5000 + 1000 + 5000 + 1000);
+}
+
+#[test]
+fn scenario_5() {
+    let mut scenario = ScenarioBuilder::new("a");
+    scenario.insert("a", 5000);
+    scenario.branch("a", "b");
+    scenario.update("b", 4000);
+    scenario.update("a", 2000);
+    scenario.branch("a", "c");
+    scenario.insert("c", 4000);
+    scenario.insert("a", 2000);
+
+    let (_model, result) = scenario.calculate(1000);
+
+    assert_eq!(result.total_size, 17000);
+}
+
+#[test]
+fn scenario_6() {
+    let branches = [
+        "7ff1edab8182025f15ae33482edb590a",
+        "b1719e044db05401a05a2ed588a3ad3f",
+        "0xb68d6691c895ad0a70809470020929ef",
+    ];
+
+    // compared to other scenarios, this one uses bytes instead of kB
+
+    let mut scenario = ScenarioBuilder::new("");
+
+    scenario.branch("", branches[0]); // at 0
+    scenario.modify_branch(branches[0], 108951064, 43696128); // at 108951064
+    scenario.branch(branches[0], branches[1]); // at 108951064
+    scenario.modify_branch(branches[1], 15560408, -1851392); // at 124511472
+    scenario.modify_branch(branches[0], 174464360, -1531904); // at 283415424
+    scenario.branch(branches[0], branches[2]); // at 283415424
+    scenario.modify_branch(branches[2], 15906192, 8192); // at 299321616
+    scenario.modify_branch(branches[0], 18909976, 32768); // at 302325400
+
+    let (model, result) = scenario.calculate(100_000);
+
+    // FIXME: We previously calculated 333_792_000. But with this PR, we get
+    // a much lower number. At a quick look at the model output and the
+    // calculations here, the new result seems correct to me.
+    eprintln!(
+        " MODEL: {}",
+        serde_json::to_string(&model.segments).unwrap()
+    );
+    eprintln!(
+        "RESULT: {}",
+        serde_json::to_string(&result.segments).unwrap()
+    );
+
+    assert_eq!(result.total_size, 136_236_928);
+}

libs/tracing-utils/Cargo.toml

@@ -0,0 +1,17 @@
+[package]
+name = "tracing-utils"
+version = "0.1.0"
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+hyper.workspace = true
+opentelemetry = { workspace = true, features=["rt-tokio"] }
+opentelemetry-otlp = { workspace = true, default_features=false, features = ["http-proto", "trace", "http", "reqwest-client"] }
+opentelemetry-semantic-conventions.workspace = true
+reqwest = { workspace = true, default-features = false, features = ["rustls-tls"] }
+tokio = { workspace = true, features = ["rt", "rt-multi-thread"] }
+tracing.workspace = true
+tracing-opentelemetry.workspace = true
+tracing-subscriber.workspace = true
+workspace_hack = { version = "0.1", path = "../../workspace_hack" }

libs/tracing-utils/src/http.rs

@@ -0,0 +1,96 @@
+//! Tracing wrapper for Hyper HTTP server
+
+use hyper::HeaderMap;
+use hyper::{Body, Request, Response};
+use std::future::Future;
+use tracing::Instrument;
+use tracing_opentelemetry::OpenTelemetrySpanExt;
+
+/// Configuration option for what to use as the "otel.name" field in the traces.
+pub enum OtelName<'a> {
+    /// Use a constant string
+    Constant(&'a str),
+
+    /// Use the path from the request.
+    ///
+    /// That's very useful information, but is not appropriate if the
+    /// path contains parameters that differ on ever request, or worse,
+    /// sensitive information like usernames or email addresses.
+    ///
+    /// See <https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/semantic_conventions/http.md#name>
+    UriPath,
+}
+
+/// Handle an incoming HTTP request using the given handler function,
+/// with OpenTelemetry tracing.
+///
+/// This runs 'handler' on the request in a new span, with fields filled in
+/// from the request. Notably, if the request contains tracing information,
+/// it is propagated to the span, so that this request is traced as part of
+/// the same trace.
+///
+/// XXX: Usually, this is handled by existing libraries, or built
+/// directly into HTTP servers. However, I couldn't find one for Hyper,
+/// so I had to write our own. OpenTelemetry website has a registry of
+/// instrumentation libraries at:
+/// https://opentelemetry.io/registry/?language=rust&component=instrumentation
+/// If a Hyper crate appears, consider switching to that.
+pub async fn tracing_handler<F, R>(
+    req: Request<Body>,
+    handler: F,
+    otel_name: OtelName<'_>,
+) -> Response<Body>
+where
+    F: Fn(Request<Body>) -> R,
+    R: Future<Output = Response<Body>>,
+{
+    // Create a tracing span, with context propagated from the incoming
+    // request if any.
+    //
+    // See list of standard fields defined for HTTP requests at
+    // https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/semantic_conventions/http.md
+    // We only fill in a few of the most useful ones here.
+    let otel_name = match otel_name {
+        OtelName::Constant(s) => s,
+        OtelName::UriPath => req.uri().path(),
+    };
+
+    let span = tracing::info_span!(
+        "http request",
+        otel.name= %otel_name,
+        http.method = %req.method(),
+        http.status_code = tracing::field::Empty,
+    );
+    let parent_ctx = extract_remote_context(req.headers());
+    span.set_parent(parent_ctx);
+
+    // Handle the request within the span
+    let response = handler(req).instrument(span.clone()).await;
+
+    // Fill in the fields from the response code
+    let status = response.status();
+    span.record("http.status_code", status.as_str());
+    span.record(
+        "otel.status_code",
+        if status.is_success() { "OK" } else { "ERROR" },
+    );
+
+    response
+}
+
+// Extract remote tracing context from the HTTP headers
+fn extract_remote_context(headers: &HeaderMap) -> opentelemetry::Context {
+    struct HeaderExtractor<'a>(&'a HeaderMap);
+
+    impl<'a> opentelemetry::propagation::Extractor for HeaderExtractor<'a> {
+        fn get(&self, key: &str) -> Option<&str> {
+            self.0.get(key).and_then(|value| value.to_str().ok())
+        }
+
+        fn keys(&self) -> Vec<&str> {
+            self.0.keys().map(|value| value.as_str()).collect()
+        }
+    }
+    let extractor = HeaderExtractor(headers);
+    opentelemetry::global::get_text_map_propagator(|propagator| propagator.extract(&extractor))
+}

libs/tracing-utils/src/lib.rs

@@ -0,0 +1,168 @@
+//! Helper functions to set up OpenTelemetry tracing.
+//!
+//! This comes in two variants, depending on whether you have a Tokio runtime available.
+//! If you do, call `init_tracing()`. It sets up the trace processor and exporter to use
+//! the current tokio runtime. If you don't have a runtime available, or you don't want
+//! to share the runtime with the tracing tasks, call `init_tracing_without_runtime()`
+//! instead. It sets up a dedicated single-threaded Tokio runtime for the tracing tasks.
+//!
+//! Example:
+//!
+//! ```rust,no_run
+//! use tracing_subscriber::prelude::*;
+//! use tracing_opentelemetry::OpenTelemetryLayer;
+//!
+//! #[tokio::main]
+//! async fn main() {
+//!     // Set up logging to stderr
+//!     let env_filter = tracing_subscriber::EnvFilter::try_from_default_env()
+//!         .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new("info"));
+//!     let fmt_layer = tracing_subscriber::fmt::layer()
+//!         .with_target(false)
+//!         .with_writer(std::io::stderr);
+//!
+//!     // Initialize OpenTelemetry. Exports tracing spans as OpenTelemetry traces
+//!     let otlp_layer = tracing_utils::init_tracing("my_application").await.map(OpenTelemetryLayer::new);
+//!
+//!     // Put it all together
+//!     tracing_subscriber::registry()
+//!         .with(env_filter)
+//!         .with(otlp_layer)
+//!         .with(fmt_layer)
+//!         .init();
+//! }
+//! ```
+
+use opentelemetry::sdk::Resource;
+use opentelemetry::KeyValue;
+use opentelemetry_otlp::WithExportConfig;
+use opentelemetry_otlp::{OTEL_EXPORTER_OTLP_ENDPOINT, OTEL_EXPORTER_OTLP_TRACES_ENDPOINT};
+
+pub use tracing_opentelemetry::OpenTelemetryLayer;
+
+pub mod http;
+
+/// Set up OpenTelemetry exporter, using configuration from environment variables.
+///
+/// `service_name` is set as the OpenTelemetry 'service.name' resource (see
+/// <https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/resource/semantic_conventions/README.md#service>)
+///
+/// We try to follow the conventions for the environment variables specified in
+/// <https://opentelemetry.io/docs/reference/specification/sdk-environment-variables/>
+///
+/// However, we only support a subset of those options:
+///
+/// - OTEL_SDK_DISABLED is supported. The default is "false", meaning tracing
+///   is enabled by default. Set it to "true" to disable.
+///
+/// - We use the OTLP exporter, with HTTP protocol. Most of the OTEL_EXPORTER_OTLP_*
+///   settings specified in
+///   <https://opentelemetry.io/docs/reference/specification/protocol/exporter/>
+///   are supported, as they are handled by the `opentelemetry-otlp` crate.
+///   Settings related to other exporters have no effect.
+///
+/// - Some other settings are supported by the `opentelemetry` crate.
+///
+/// If you need some other setting, please test if it works first. And perhaps
+/// add a comment in the list above to save the effort of testing for the next
+/// person.
+///
+/// This doesn't block, but is marked as 'async' to hint that this must be called in
+/// asynchronous execution context.
+pub async fn init_tracing(service_name: &str) -> Option<opentelemetry::sdk::trace::Tracer> {
+    if std::env::var("OTEL_SDK_DISABLED") == Ok("true".to_string()) {
+        return None;
+    };
+    Some(init_tracing_internal(service_name.to_string()))
+}
+
+/// Like `init_tracing`, but creates a separate tokio Runtime for the tracing
+/// tasks.
+pub fn init_tracing_without_runtime(
+    service_name: &str,
+) -> Option<opentelemetry::sdk::trace::Tracer> {
+    if std::env::var("OTEL_SDK_DISABLED") == Ok("true".to_string()) {
+        return None;
+    };
+
+    // The opentelemetry batch processor and the OTLP exporter needs a Tokio
+    // runtime. Create a dedicated runtime for them. One thread should be
+    // enough.
+    //
+    // (Alternatively, instead of batching, we could use the "simple
+    // processor", which doesn't need Tokio, and use "reqwest-blocking"
+    // feature for the OTLP exporter, which also doesn't need Tokio.  However,
+    // batching is considered best practice, and also I have the feeling that
+    // the non-Tokio codepaths in the opentelemetry crate are less used and
+    // might be more buggy, so better to stay on the well-beaten path.)
+    //
+    // We leak the runtime so that it keeps running after we exit the
+    // function.
+    let runtime = Box::leak(Box::new(
+        tokio::runtime::Builder::new_multi_thread()
+            .enable_all()
+            .thread_name("otlp runtime thread")
+            .worker_threads(1)
+            .build()
+            .unwrap(),
+    ));
+    let _guard = runtime.enter();
+
+    Some(init_tracing_internal(service_name.to_string()))
+}
+
+fn init_tracing_internal(service_name: String) -> opentelemetry::sdk::trace::Tracer {
+    // Set up exporter from the OTEL_EXPORTER_* environment variables
+    let mut exporter = opentelemetry_otlp::new_exporter().http().with_env();
+
+    // XXX opentelemetry-otlp v0.18.0 has a bug in how it uses the
+    // OTEL_EXPORTER_OTLP_ENDPOINT env variable. According to the
+    // OpenTelemetry spec at
+    // <https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/protocol/exporter.md#endpoint-urls-for-otlphttp>,
+    // the full exporter URL is formed by appending "/v1/traces" to the value
+    // of OTEL_EXPORTER_OTLP_ENDPOINT. However, opentelemetry-otlp only does
+    // that with the grpc-tonic exporter. Other exporters, like the HTTP
+    // exporter, use the URL from OTEL_EXPORTER_OTLP_ENDPOINT as is, without
+    // appending "/v1/traces".
+    //
+    // See https://github.com/open-telemetry/opentelemetry-rust/pull/950
+    //
+    // Work around that by checking OTEL_EXPORTER_OTLP_ENDPOINT, and setting
+    // the endpoint url with the "/v1/traces" path ourselves. If the bug is
+    // fixed in a later version, we can remove this code. But if we don't
+    // remember to remove this, it won't do any harm either, as the crate will
+    // just ignore the OTEL_EXPORTER_OTLP_ENDPOINT setting when the endpoint
+    // is set directly with `with_endpoint`.
+    if std::env::var(OTEL_EXPORTER_OTLP_TRACES_ENDPOINT).is_err() {
+        if let Ok(mut endpoint) = std::env::var(OTEL_EXPORTER_OTLP_ENDPOINT) {
+            if !endpoint.ends_with('/') {
+                endpoint.push('/');
+            }
+            endpoint.push_str("v1/traces");
+            exporter = exporter.with_endpoint(endpoint);
+        }
+    }
+
+    // Propagate trace information in the standard W3C TraceContext format.
+    opentelemetry::global::set_text_map_propagator(
+        opentelemetry::sdk::propagation::TraceContextPropagator::new(),
+    );
+
+    opentelemetry_otlp::new_pipeline()
+        .tracing()
+        .with_exporter(exporter)
+        .with_trace_config(
+            opentelemetry::sdk::trace::config().with_resource(Resource::new(vec![KeyValue::new(
+                opentelemetry_semantic_conventions::resource::SERVICE_NAME,
+                service_name,
+            )])),
+        )
+        .install_batch(opentelemetry::runtime::Tokio)
+        .expect("could not initialize opentelemetry exporter")
+}
+
+// Shutdown trace pipeline gracefully, so that it has a chance to send any
+// pending traces before we exit.
+pub fn shutdown_tracing() {
+    opentelemetry::global::shutdown_tracer_provider();
+}

libs/utils/Cargo.toml

@@ -11,6 +11,7 @@ async-trait.workspace = true
 anyhow.workspace = true
 bincode.workspace = true
 bytes.workspace = true
+heapless.workspace = true
 hyper = { workspace = true, features = ["full"] }
 routerify.workspace = true
 serde.workspace = true
@@ -37,6 +38,7 @@ metrics.workspace = true
 pq_proto.workspace = true
 
 workspace_hack.workspace = true
+url.workspace = true
 
 [dev-dependencies]
 byteorder.workspace = true

libs/utils/src/history_buffer.rs

@@ -0,0 +1,161 @@
+//! A heapless buffer for events of sorts.
+
+use std::ops;
+
+use heapless::HistoryBuffer;
+
+#[derive(Debug, Clone)]
+pub struct HistoryBufferWithDropCounter<T, const L: usize> {
+    buffer: HistoryBuffer<T, L>,
+    drop_count: u64,
+}
+
+impl<T, const L: usize> HistoryBufferWithDropCounter<T, L> {
+    pub fn write(&mut self, data: T) {
+        let len_before = self.buffer.len();
+        self.buffer.write(data);
+        let len_after = self.buffer.len();
+        self.drop_count += u64::from(len_before == len_after);
+    }
+    pub fn drop_count(&self) -> u64 {
+        self.drop_count
+    }
+    pub fn map<U, F: Fn(&T) -> U>(&self, f: F) -> HistoryBufferWithDropCounter<U, L> {
+        let mut buffer = HistoryBuffer::new();
+        buffer.extend(self.buffer.oldest_ordered().map(f));
+        HistoryBufferWithDropCounter::<U, L> {
+            buffer,
+            drop_count: self.drop_count,
+        }
+    }
+}
+
+impl<T, const L: usize> Default for HistoryBufferWithDropCounter<T, L> {
+    fn default() -> Self {
+        Self {
+            buffer: HistoryBuffer::default(),
+            drop_count: 0,
+        }
+    }
+}
+
+impl<T, const L: usize> ops::Deref for HistoryBufferWithDropCounter<T, L> {
+    type Target = HistoryBuffer<T, L>;
+
+    fn deref(&self) -> &Self::Target {
+        &self.buffer
+    }
+}
+
+#[derive(serde::Serialize)]
+struct SerdeRepr<T> {
+    buffer: Vec<T>,
+    drop_count: u64,
+}
+
+impl<'a, T, const L: usize> From<&'a HistoryBufferWithDropCounter<T, L>> for SerdeRepr<T>
+where
+    T: Clone + serde::Serialize,
+{
+    fn from(value: &'a HistoryBufferWithDropCounter<T, L>) -> Self {
+        let HistoryBufferWithDropCounter { buffer, drop_count } = value;
+        SerdeRepr {
+            buffer: buffer.iter().cloned().collect(),
+            drop_count: *drop_count,
+        }
+    }
+}
+
+impl<T, const L: usize> serde::Serialize for HistoryBufferWithDropCounter<T, L>
+where
+    T: Clone + serde::Serialize,
+{
+    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
+    where
+        S: serde::Serializer,
+    {
+        SerdeRepr::from(self).serialize(serializer)
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::HistoryBufferWithDropCounter;
+
+    #[test]
+    fn test_basics() {
+        let mut b = HistoryBufferWithDropCounter::<_, 2>::default();
+        b.write(1);
+        b.write(2);
+        b.write(3);
+        assert!(b.iter().any(|e| *e == 2));
+        assert!(b.iter().any(|e| *e == 3));
+        assert!(!b.iter().any(|e| *e == 1));
+    }
+
+    #[test]
+    fn test_drop_count_works() {
+        let mut b = HistoryBufferWithDropCounter::<_, 2>::default();
+        b.write(1);
+        assert_eq!(b.drop_count(), 0);
+        b.write(2);
+        assert_eq!(b.drop_count(), 0);
+        b.write(3);
+        assert_eq!(b.drop_count(), 1);
+        b.write(4);
+        assert_eq!(b.drop_count(), 2);
+    }
+
+    #[test]
+    fn test_clone_works() {
+        let mut b = HistoryBufferWithDropCounter::<_, 2>::default();
+        b.write(1);
+        b.write(2);
+        b.write(3);
+        assert_eq!(b.drop_count(), 1);
+        let mut c = b.clone();
+        assert_eq!(c.drop_count(), 1);
+        assert!(c.iter().any(|e| *e == 2));
+        assert!(c.iter().any(|e| *e == 3));
+        assert!(!c.iter().any(|e| *e == 1));
+
+        c.write(4);
+        assert!(c.iter().any(|e| *e == 4));
+        assert!(!b.iter().any(|e| *e == 4));
+    }
+
+    #[test]
+    fn test_map() {
+        let mut b = HistoryBufferWithDropCounter::<_, 2>::default();
+
+        b.write(1);
+        assert_eq!(b.drop_count(), 0);
+        {
+            let c = b.map(|i| i + 10);
+            assert_eq!(c.oldest_ordered().cloned().collect::<Vec<_>>(), vec![11]);
+            assert_eq!(c.drop_count(), 0);
+        }
+
+        b.write(2);
+        assert_eq!(b.drop_count(), 0);
+        {
+            let c = b.map(|i| i + 10);
+            assert_eq!(
+                c.oldest_ordered().cloned().collect::<Vec<_>>(),
+                vec![11, 12]
+            );
+            assert_eq!(c.drop_count(), 0);
+        }
+
+        b.write(3);
+        assert_eq!(b.drop_count(), 1);
+        {
+            let c = b.map(|i| i + 10);
+            assert_eq!(
+                c.oldest_ordered().cloned().collect::<Vec<_>>(),
+                vec![12, 13]
+            );
+            assert_eq!(c.drop_count(), 1);
+        }
+    }
+}

libs/utils/src/http/endpoint.rs

@@ -1,18 +1,21 @@
 use crate::auth::{Claims, JwtAuth};
 use crate::http::error;
-use anyhow::anyhow;
-use hyper::header::AUTHORIZATION;
+use anyhow::{anyhow, Context};
+use hyper::header::{HeaderName, AUTHORIZATION};
+use hyper::http::HeaderValue;
 use hyper::{header::CONTENT_TYPE, Body, Request, Response, Server};
+use hyper::{Method, StatusCode};
 use metrics::{register_int_counter, Encoder, IntCounter, TextEncoder};
 use once_cell::sync::Lazy;
 use routerify::ext::RequestExt;
 use routerify::RequestInfo;
 use routerify::{Middleware, Router, RouterBuilder, RouterService};
 use tokio::task::JoinError;
-use tracing::info;
+use tracing;
 
 use std::future::Future;
 use std::net::TcpListener;
+use std::str::FromStr;
 
 use super::error::ApiError;
 
@@ -25,7 +28,14 @@ static SERVE_METRICS_COUNT: Lazy<IntCounter> = Lazy::new(|| {
 });
 
 async fn logger(res: Response<Body>, info: RequestInfo) -> Result<Response<Body>, ApiError> {
-    info!("{} {} {}", info.method(), info.uri().path(), res.status(),);
+    // cannot factor out the Level to avoid the repetition
+    // because tracing can only work with const Level
+    // which is not the case here
+    if info.method() == Method::GET && res.status() == StatusCode::OK {
+        tracing::debug!("{} {} {}", info.method(), info.uri().path(), res.status());
+    } else {
+        tracing::info!("{} {} {}", info.method(), info.uri().path(), res.status());
+    }
     Ok(res)
 }
 
@@ -143,6 +153,38 @@ pub fn auth_middleware<B: hyper::body::HttpBody + Send + Sync + 'static>(
     })
 }
 
+pub fn add_response_header_middleware<B>(
+    header: &str,
+    value: &str,
+) -> anyhow::Result<Middleware<B, ApiError>>
+where
+    B: hyper::body::HttpBody + Send + Sync + 'static,
+{
+    let name =
+        HeaderName::from_str(header).with_context(|| format!("invalid header name: {header}"))?;
+    let value =
+        HeaderValue::from_str(value).with_context(|| format!("invalid header value: {value}"))?;
+    Ok(Middleware::post_with_info(
+        move |mut response, request_info| {
+            let name = name.clone();
+            let value = value.clone();
+            async move {
+                let headers = response.headers_mut();
+                if headers.contains_key(&name) {
+                    tracing::warn!(
+                        "{} response already contains header {:?}",
+                        request_info.uri(),
+                        &name,
+                    );
+                } else {
+                    headers.insert(name, value);
+                }
+                Ok(response)
+            }
+        },
+    ))
+}
+
 pub fn check_permission_with(
     req: &Request<Body>,
     check_permission: impl Fn(&Claims) -> Result<(), anyhow::Error>,
@@ -169,7 +211,7 @@ pub fn serve_thread_main<S>(
 where
     S: Future<Output = ()> + Send + Sync,
 {
-    info!("Starting an HTTP endpoint at {}", listener.local_addr()?);
+    tracing::info!("Starting an HTTP endpoint at {}", listener.local_addr()?);
 
     // Create a Service from the router above to handle incoming requests.
     let service = RouterService::new(router_builder.build().map_err(|err| anyhow!(err))?).unwrap();

libs/utils/src/http/error.rs

@@ -1,6 +1,7 @@
 use hyper::{header, Body, Response, StatusCode};
 use serde::{Deserialize, Serialize};
 use thiserror::Error;
+use tracing::error;
 
 #[derive(Debug, Error)]
 pub enum ApiError {
@@ -76,8 +77,16 @@ impl HttpErrorBody {
 }
 
 pub async fn handler(err: routerify::RouteError) -> Response<Body> {
-    tracing::error!("Error processing HTTP request: {:?}", err);
-    err.downcast::<ApiError>()
-        .expect("handler should always return api error")
-        .into_response()
+    let api_error = err
+        .downcast::<ApiError>()
+        .expect("handler should always return api error");
+
+    // Print a stack trace for Internal Server errors
+    if let ApiError::InternalServerError(_) = api_error.as_ref() {
+        error!("Error processing HTTP request: {api_error:?}");
+    } else {
+        error!("Error processing HTTP request: {api_error:#}");
+    }
+
+    api_error.into_response()
 }

libs/utils/src/http/request.rs

@@ -1,4 +1,5 @@
-use std::str::FromStr;
+use core::fmt;
+use std::{borrow::Cow, str::FromStr};
 
 use super::error::ApiError;
 use anyhow::anyhow;
@@ -29,6 +30,50 @@ pub fn parse_request_param<T: FromStr>(
     }
 }
 
+fn get_query_param<'a>(
+    request: &'a Request<Body>,
+    param_name: &str,
+) -> Result<Option<Cow<'a, str>>, ApiError> {
+    let query = match request.uri().query() {
+        Some(q) => q,
+        None => return Ok(None),
+    };
+    let mut values = url::form_urlencoded::parse(query.as_bytes())
+        .filter_map(|(k, v)| if k == param_name { Some(v) } else { None })
+        // we call .next() twice below. If it's None the first time, .fuse() ensures it's None afterwards
+        .fuse();
+
+    let value1 = values.next();
+    if values.next().is_some() {
+        return Err(ApiError::BadRequest(anyhow!(
+            "param {param_name} specified more than once"
+        )));
+    }
+    Ok(value1)
+}
+
+pub fn must_get_query_param<'a>(
+    request: &'a Request<Body>,
+    param_name: &str,
+) -> Result<Cow<'a, str>, ApiError> {
+    get_query_param(request, param_name)?.ok_or_else(|| {
+        ApiError::BadRequest(anyhow!("no {param_name} specified in query parameters"))
+    })
+}
+
+pub fn parse_query_param<E: fmt::Display, T: FromStr<Err = E>>(
+    request: &Request<Body>,
+    param_name: &str,
+) -> Result<Option<T>, ApiError> {
+    get_query_param(request, param_name)?
+        .map(|v| {
+            v.parse().map_err(|e| {
+                ApiError::BadRequest(anyhow!("cannot parse query param {param_name}: {e}"))
+            })
+        })
+        .transpose()
+}
+
 pub async fn ensure_no_body(request: &mut Request<Body>) -> Result<(), ApiError> {
     match request.body_mut().data().await {
         Some(_) => Err(ApiError::BadRequest(anyhow!("Unexpected request body"))),

libs/utils/src/lib.rs

@@ -52,6 +52,8 @@ pub mod signals;
 
 pub mod fs_ext;
 
+pub mod history_buffer;
+
 /// use with fail::cfg("$name", "return(2000)")
 #[macro_export]
 macro_rules! failpoint_sleep_millis_async {

libs/utils/src/logging.rs

@@ -45,3 +45,115 @@ pub fn init(log_format: LogFormat) -> anyhow::Result<()> {
 
     Ok(())
 }
+
+/// Disable the default rust panic hook by using `set_hook`.
+///
+/// For neon binaries, the assumption is that tracing is configured before with [`init`], after
+/// that sentry is configured (if needed). sentry will install it's own on top of this, always
+/// processing the panic before we log it.
+///
+/// When the return value is dropped, the hook is reverted to std default hook (prints to stderr).
+/// If the assumptions about the initialization order are not held, use
+/// [`TracingPanicHookGuard::disarm`] but keep in mind, if tracing is stopped, then panics will be
+/// lost.
+#[must_use]
+pub fn replace_panic_hook_with_tracing_panic_hook() -> TracingPanicHookGuard {
+    std::panic::set_hook(Box::new(tracing_panic_hook));
+    TracingPanicHookGuard::new()
+}
+
+/// Drop guard which restores the std panic hook on drop.
+///
+/// Tracing should not be used when it's not configured, but we cannot really latch on to any
+/// imaginary lifetime of tracing.
+pub struct TracingPanicHookGuard {
+    act: bool,
+}
+
+impl TracingPanicHookGuard {
+    fn new() -> Self {
+        TracingPanicHookGuard { act: true }
+    }
+
+    /// Make this hook guard not do anything when dropped.
+    pub fn forget(&mut self) {
+        self.act = false;
+    }
+}
+
+impl Drop for TracingPanicHookGuard {
+    fn drop(&mut self) {
+        if self.act {
+            let _ = std::panic::take_hook();
+        }
+    }
+}
+
+/// Named symbol for our panic hook, which logs the panic.
+fn tracing_panic_hook(info: &std::panic::PanicInfo) {
+    // following rust 1.66.1 std implementation:
+    // https://github.com/rust-lang/rust/blob/90743e7298aca107ddaa0c202a4d3604e29bfeb6/library/std/src/panicking.rs#L235-L288
+    let location = info.location();
+
+    let msg = match info.payload().downcast_ref::<&'static str>() {
+        Some(s) => *s,
+        None => match info.payload().downcast_ref::<String>() {
+            Some(s) => &s[..],
+            None => "Box<dyn Any>",
+        },
+    };
+
+    let thread = std::thread::current();
+    let thread = thread.name().unwrap_or("<unnamed>");
+    let backtrace = std::backtrace::Backtrace::capture();
+
+    let _entered = if let Some(location) = location {
+        tracing::error_span!("panic", %thread, location = %PrettyLocation(location))
+    } else {
+        // very unlikely to hit here, but the guarantees of std could change
+        tracing::error_span!("panic", %thread)
+    }
+    .entered();
+
+    if backtrace.status() == std::backtrace::BacktraceStatus::Captured {
+        // this has an annoying extra '\n' in the end which anyhow doesn't do, but we cannot really
+        // get rid of it as we cannot get in between of std::fmt::Formatter<'_>; we could format to
+        // string, maybe even to a TLS one but tracing already does that.
+        tracing::error!("{msg}\n\nStack backtrace:\n{backtrace}");
+    } else {
+        tracing::error!("{msg}");
+    }
+
+    // ensure that we log something on the panic if this hook is left after tracing has been
+    // unconfigured. worst case when teardown is racing the panic is to log the panic twice.
+    tracing::dispatcher::get_default(|d| {
+        if let Some(_none) = d.downcast_ref::<tracing::subscriber::NoSubscriber>() {
+            let location = location.map(PrettyLocation);
+            log_panic_to_stderr(thread, msg, location, &backtrace);
+        }
+    });
+}
+
+#[cold]
+fn log_panic_to_stderr(
+    thread: &str,
+    msg: &str,
+    location: Option<PrettyLocation<'_, '_>>,
+    backtrace: &std::backtrace::Backtrace,
+) {
+    eprintln!("panic while tracing is unconfigured: thread '{thread}' panicked at '{msg}', {location:?}\nStack backtrace:\n{backtrace}");
+}
+
+struct PrettyLocation<'a, 'b>(&'a std::panic::Location<'b>);
+
+impl std::fmt::Display for PrettyLocation<'_, '_> {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}:{}:{}", self.0.file(), self.0.line(), self.0.column())
+    }
+}
+
+impl std::fmt::Debug for PrettyLocation<'_, '_> {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        <Self as std::fmt::Display>::fmt(self, f)
+    }
+}

pageserver/Cargo.toml

@@ -23,6 +23,7 @@ const_format.workspace = true
 consumption_metrics.workspace = true
 crc32c.workspace = true
 crossbeam-utils.workspace = true
+either.workspace = true
 fail.workspace = true
 futures.workspace = true
 git-version.workspace = true
@@ -51,7 +52,7 @@ thiserror.workspace = true
 tokio = { workspace = true, features = ["process", "sync", "fs", "rt", "io-util", "time"] }
 tokio-postgres.workspace = true
 tokio-util.workspace = true
-toml_edit.workspace = true
+toml_edit = { workspace = true, features = [ "serde" ] }
 tracing.workspace = true
 url.workspace = true
 walkdir.workspace = true
@@ -67,6 +68,10 @@ utils.workspace = true
 workspace_hack.workspace = true
 reqwest.workspace = true
 rpds.workspace = true
+enum-map.workspace = true
+enumset.workspace = true
+strum.workspace = true
+strum_macros.workspace = true
 
 [dev-dependencies]
 criterion.workspace = true

pageserver/benches/bench_layer_map.rs

@@ -1,8 +1,7 @@
 use pageserver::keyspace::{KeyPartitioning, KeySpace};
 use pageserver::repository::Key;
 use pageserver::tenant::layer_map::LayerMap;
-use pageserver::tenant::storage_layer::Layer;
-use pageserver::tenant::storage_layer::{DeltaFileName, ImageFileName, LayerDescriptor};
+use pageserver::tenant::storage_layer::{Layer, LayerDescriptor, LayerFileName};
 use rand::prelude::{SeedableRng, SliceRandom, StdRng};
 use std::cmp::{max, min};
 use std::fs::File;
@@ -26,30 +25,15 @@ fn build_layer_map(filename_dump: PathBuf) -> LayerMap<LayerDescriptor> {
 
     let mut updates = layer_map.batch_update();
     for fname in filenames {
-        let fname = &fname.unwrap();
-        if let Some(imgfilename) = ImageFileName::parse_str(fname) {
-            let layer = LayerDescriptor {
-                key: imgfilename.key_range,
-                lsn: imgfilename.lsn..(imgfilename.lsn + 1),
-                is_incremental: false,
-                short_id: fname.to_string(),
-            };
-            updates.insert_historic(Arc::new(layer));
-            min_lsn = min(min_lsn, imgfilename.lsn);
-            max_lsn = max(max_lsn, imgfilename.lsn);
-        } else if let Some(deltafilename) = DeltaFileName::parse_str(fname) {
-            let layer = LayerDescriptor {
-                key: deltafilename.key_range.clone(),
-                lsn: deltafilename.lsn_range.clone(),
-                is_incremental: true,
-                short_id: fname.to_string(),
-            };
-            updates.insert_historic(Arc::new(layer));
-            min_lsn = min(min_lsn, deltafilename.lsn_range.start);
-            max_lsn = max(max_lsn, deltafilename.lsn_range.end);
-        } else {
-            panic!("unexpected filename {fname}");
-        }
+        let fname = fname.unwrap();
+        let fname = LayerFileName::from_str(&fname).unwrap();
+        let layer = LayerDescriptor::from(fname);
+
+        let lsn_range = layer.get_lsn_range();
+        min_lsn = min(min_lsn, lsn_range.start);
+        max_lsn = max(max_lsn, Lsn(lsn_range.end.0 - 1));
+
+        updates.insert_historic(Arc::new(layer));
     }
 
     println!("min: {min_lsn}, max: {max_lsn}");

pageserver/src/basebackup.rs

@@ -27,11 +27,13 @@ use tracing::*;
 ///
 use tokio_tar::{Builder, EntryType, Header};
 
+use crate::context::RequestContext;
 use crate::tenant::Timeline;
 use pageserver_api::reltag::{RelTag, SlruKind};
 
 use postgres_ffi::pg_constants::{DEFAULTTABLESPACE_OID, GLOBALTABLESPACE_OID};
 use postgres_ffi::pg_constants::{PGDATA_SPECIAL_FILES, PGDATA_SUBDIRS, PG_HBA};
+use postgres_ffi::relfile_utils::{INIT_FORKNUM, MAIN_FORKNUM};
 use postgres_ffi::TransactionId;
 use postgres_ffi::XLogFileName;
 use postgres_ffi::PG_TLI;
@@ -52,6 +54,7 @@ pub async fn send_basebackup_tarball<'a, W>(
     req_lsn: Option<Lsn>,
     prev_lsn: Option<Lsn>,
     full_backup: bool,
+    ctx: &'a RequestContext,
 ) -> anyhow::Result<()>
 where
     W: AsyncWrite + Send + Sync + Unpin,
@@ -110,6 +113,7 @@ where
         lsn: backup_lsn,
         prev_record_lsn: prev_lsn,
         full_backup,
+        ctx,
     };
     basebackup
         .send_tarball()
@@ -129,6 +133,7 @@ where
     lsn: Lsn,
     prev_record_lsn: Lsn,
     full_backup: bool,
+    ctx: &'a RequestContext,
 }
 
 impl<'a, W> Basebackup<'a, W>
@@ -171,23 +176,54 @@ where
             SlruKind::MultiXactOffsets,
             SlruKind::MultiXactMembers,
         ] {
-            for segno in self.timeline.list_slru_segments(kind, self.lsn).await? {
+            for segno in self
+                .timeline
+                .list_slru_segments(kind, self.lsn, self.ctx)
+                .await?
+            {
                 self.add_slru_segment(kind, segno).await?;
             }
         }
 
         // Create tablespace directories
-        for ((spcnode, dbnode), has_relmap_file) in self.timeline.list_dbdirs(self.lsn).await? {
+        for ((spcnode, dbnode), has_relmap_file) in
+            self.timeline.list_dbdirs(self.lsn, self.ctx).await?
+        {
             self.add_dbdir(spcnode, dbnode, has_relmap_file).await?;
 
-            // Gather and send relational files in each database if full backup is requested.
-            if self.full_backup {
-                for rel in self.timeline.list_rels(spcnode, dbnode, self.lsn).await? {
-                    self.add_rel(rel).await?;
+            // If full backup is requested, include all relation files.
+            // Otherwise only include init forks of unlogged relations.
+            let rels = self
+                .timeline
+                .list_rels(spcnode, dbnode, self.lsn, self.ctx)
+                .await?;
+            for &rel in rels.iter() {
+                // Send init fork as main fork to provide well formed empty
+                // contents of UNLOGGED relations. Postgres copies it in
+                // `reinit.c` during recovery.
+                if rel.forknum == INIT_FORKNUM {
+                    // I doubt we need _init fork itself, but having it at least
+                    // serves as a marker relation is unlogged.
+                    self.add_rel(rel, rel).await?;
+                    self.add_rel(rel, rel.with_forknum(MAIN_FORKNUM)).await?;
+                    continue;
+                }
+
+                if self.full_backup {
+                    if rel.forknum == MAIN_FORKNUM && rels.contains(&rel.with_forknum(INIT_FORKNUM))
+                    {
+                        // skip this, will include it when we reach the init fork
+                        continue;
+                    }
+                    self.add_rel(rel, rel).await?;
                 }
             }
         }
-        for xid in self.timeline.list_twophase_files(self.lsn).await? {
+        for xid in self
+            .timeline
+            .list_twophase_files(self.lsn, self.ctx)
+            .await?
+        {
             self.add_twophase_file(xid).await?;
         }
 
@@ -202,12 +238,16 @@ where
         Ok(())
     }
 
-    async fn add_rel(&mut self, tag: RelTag) -> anyhow::Result<()> {
-        let nblocks = self.timeline.get_rel_size(tag, self.lsn, false).await?;
+    /// Add contents of relfilenode `src`, naming it as `dst`.
+    async fn add_rel(&mut self, src: RelTag, dst: RelTag) -> anyhow::Result<()> {
+        let nblocks = self
+            .timeline
+            .get_rel_size(src, self.lsn, false, self.ctx)
+            .await?;
 
         // If the relation is empty, create an empty file
         if nblocks == 0 {
-            let file_name = tag.to_segfile_name(0);
+            let file_name = dst.to_segfile_name(0);
             let header = new_tar_header(&file_name, 0)?;
             self.ar.append(&header, &mut io::empty()).await?;
             return Ok(());
@@ -223,12 +263,12 @@ where
             for blknum in startblk..endblk {
                 let img = self
                     .timeline
-                    .get_rel_page_at_lsn(tag, blknum, self.lsn, false)
+                    .get_rel_page_at_lsn(src, blknum, self.lsn, false, self.ctx)
                     .await?;
                 segment_data.extend_from_slice(&img[..]);
             }
 
-            let file_name = tag.to_segfile_name(seg as u32);
+            let file_name = dst.to_segfile_name(seg as u32);
             let header = new_tar_header(&file_name, segment_data.len() as u64)?;
             self.ar.append(&header, segment_data.as_slice()).await?;
 
@@ -245,14 +285,14 @@ where
     async fn add_slru_segment(&mut self, slru: SlruKind, segno: u32) -> anyhow::Result<()> {
         let nblocks = self
             .timeline
-            .get_slru_segment_size(slru, segno, self.lsn)
+            .get_slru_segment_size(slru, segno, self.lsn, self.ctx)
             .await?;
 
         let mut slru_buf: Vec<u8> = Vec::with_capacity(nblocks as usize * BLCKSZ as usize);
         for blknum in 0..nblocks {
             let img = self
                 .timeline
-                .get_slru_page_at_lsn(slru, segno, blknum, self.lsn)
+                .get_slru_page_at_lsn(slru, segno, blknum, self.lsn, self.ctx)
                 .await?;
 
             if slru == SlruKind::Clog {
@@ -287,7 +327,7 @@ where
         let relmap_img = if has_relmap_file {
             let img = self
                 .timeline
-                .get_relmap_file(spcnode, dbnode, self.lsn)
+                .get_relmap_file(spcnode, dbnode, self.lsn, self.ctx)
                 .await?;
             ensure!(img.len() == 512);
             Some(img)
@@ -323,7 +363,7 @@ where
             if !has_relmap_file
                 && self
                     .timeline
-                    .list_rels(spcnode, dbnode, self.lsn)
+                    .list_rels(spcnode, dbnode, self.lsn, self.ctx)
                     .await?
                     .is_empty()
             {
@@ -356,7 +396,10 @@ where
     // Extract twophase state files
     //
     async fn add_twophase_file(&mut self, xid: TransactionId) -> anyhow::Result<()> {
-        let img = self.timeline.get_twophase_file(xid, self.lsn).await?;
+        let img = self
+            .timeline
+            .get_twophase_file(xid, self.lsn, self.ctx)
+            .await?;
 
         let mut buf = BytesMut::new();
         buf.extend_from_slice(&img[..]);
@@ -394,12 +437,12 @@ where
 
         let checkpoint_bytes = self
             .timeline
-            .get_checkpoint(self.lsn)
+            .get_checkpoint(self.lsn, self.ctx)
             .await
             .context("failed to get checkpoint bytes")?;
         let pg_control_bytes = self
             .timeline
-            .get_control_file(self.lsn)
+            .get_control_file(self.lsn, self.ctx)
             .await
             .context("failed get control bytes")?;
 

pageserver/src/bin/layer_map_analyzer.rs

@@ -0,0 +1,230 @@
+//! Tool for extracting content-dependent metadata about layers. Useful for scanning real project layer files and evaluating the effectiveness of different heuristics on them.
+//!
+//! Currently it only analyzes holes, which are regions within the layer range that the layer contains no updates for. In the future it might do more analysis (maybe key quantiles?) but it should never return sensitive data.
+
+use anyhow::Result;
+use std::cmp::Ordering;
+use std::collections::BinaryHeap;
+use std::ops::Range;
+use std::{env, fs, path::Path, path::PathBuf, str, str::FromStr};
+
+use pageserver::page_cache::PAGE_SZ;
+use pageserver::repository::{Key, KEY_SIZE};
+use pageserver::tenant::block_io::{BlockReader, FileBlockReader};
+use pageserver::tenant::disk_btree::{DiskBtreeReader, VisitDirection};
+use pageserver::tenant::storage_layer::delta_layer::{Summary, DELTA_KEY_SIZE};
+use pageserver::tenant::storage_layer::range_overlaps;
+use pageserver::virtual_file::VirtualFile;
+
+use utils::{bin_ser::BeSer, lsn::Lsn};
+
+const MIN_HOLE_LENGTH: i128 = (128 * 1024 * 1024 / PAGE_SZ) as i128;
+const DEFAULT_MAX_HOLES: usize = 10;
+
+/// Wrapper for key range to provide reverse ordering by range length for BinaryHeap
+#[derive(PartialEq, Eq)]
+struct Hole(Range<Key>);
+
+impl Ord for Hole {
+    fn cmp(&self, other: &Self) -> Ordering {
+        let other_len = other.0.end.to_i128() - other.0.start.to_i128();
+        let self_len = self.0.end.to_i128() - self.0.start.to_i128();
+        other_len.cmp(&self_len)
+    }
+}
+
+impl PartialOrd for Hole {
+    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
+        Some(self.cmp(other))
+    }
+}
+
+struct LayerFile {
+    key_range: Range<Key>,
+    lsn_range: Range<Lsn>,
+    is_delta: bool,
+    holes: Vec<Hole>,
+}
+
+impl LayerFile {
+    fn skips(&self, key_range: &Range<Key>) -> bool {
+        if !range_overlaps(&self.key_range, key_range) {
+            return false;
+        }
+        let start = match self
+            .holes
+            .binary_search_by_key(&key_range.start, |hole| hole.0.start)
+        {
+            Ok(index) => index,
+            Err(index) => {
+                if index == 0 {
+                    return false;
+                }
+                index - 1
+            }
+        };
+        self.holes[start].0.end >= key_range.end
+    }
+}
+
+fn parse_filename(name: &str) -> Option<LayerFile> {
+    let split: Vec<&str> = name.split("__").collect();
+    if split.len() != 2 {
+        return None;
+    }
+    let keys: Vec<&str> = split[0].split('-').collect();
+    let mut lsns: Vec<&str> = split[1].split('-').collect();
+    let is_delta = if lsns.len() == 1 {
+        lsns.push(lsns[0]);
+        false
+    } else {
+        true
+    };
+
+    let key_range = Key::from_hex(keys[0]).unwrap()..Key::from_hex(keys[1]).unwrap();
+    let lsn_range = Lsn::from_hex(lsns[0]).unwrap()..Lsn::from_hex(lsns[1]).unwrap();
+    let holes = Vec::new();
+    Some(LayerFile {
+        key_range,
+        lsn_range,
+        is_delta,
+        holes,
+    })
+}
+
+// Finds the max_holes largest holes, ignoring any that are smaller than MIN_HOLE_LENGTH"
+fn get_holes(path: &Path, max_holes: usize) -> Result<Vec<Hole>> {
+    let file = FileBlockReader::new(VirtualFile::open(path)?);
+    let summary_blk = file.read_blk(0)?;
+    let actual_summary = Summary::des_prefix(summary_blk.as_ref())?;
+    let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
+        actual_summary.index_start_blk,
+        actual_summary.index_root_blk,
+        file,
+    );
+    // min-heap (reserve space for one more element added before eviction)
+    let mut heap: BinaryHeap<Hole> = BinaryHeap::with_capacity(max_holes + 1);
+    let mut prev_key: Option<Key> = None;
+    tree_reader.visit(
+        &[0u8; DELTA_KEY_SIZE],
+        VisitDirection::Forwards,
+        |key, _value| {
+            let curr = Key::from_slice(&key[..KEY_SIZE]);
+            if let Some(prev) = prev_key {
+                if curr.to_i128() - prev.to_i128() >= MIN_HOLE_LENGTH {
+                    heap.push(Hole(prev..curr));
+                    if heap.len() > max_holes {
+                        heap.pop(); // remove smallest hole
+                    }
+                }
+            }
+            prev_key = Some(curr.next());
+            true
+        },
+    )?;
+    let mut holes = heap.into_vec();
+    holes.sort_by_key(|hole| hole.0.start);
+    Ok(holes)
+}
+
+fn main() -> Result<()> {
+    let args: Vec<String> = env::args().collect();
+    if args.len() < 2 {
+        println!("Usage: layer_map_analyzer PAGESERVER_DATA_DIR [MAX_HOLES]");
+        return Ok(());
+    }
+    let storage_path = PathBuf::from_str(&args[1])?;
+    let max_holes = if args.len() > 2 {
+        args[2].parse::<usize>().unwrap()
+    } else {
+        DEFAULT_MAX_HOLES
+    };
+
+    // Initialize virtual_file (file desriptor cache) and page cache which are needed to access layer persistent B-Tree.
+    pageserver::virtual_file::init(10);
+    pageserver::page_cache::init(100);
+
+    let mut total_delta_layers = 0usize;
+    let mut total_image_layers = 0usize;
+    let mut total_excess_layers = 0usize;
+    for tenant in fs::read_dir(storage_path.join("tenants"))? {
+        let tenant = tenant?;
+        if !tenant.file_type()?.is_dir() {
+            continue;
+        }
+        for timeline in fs::read_dir(tenant.path().join("timelines"))? {
+            let timeline = timeline?;
+            if !timeline.file_type()?.is_dir() {
+                continue;
+            }
+            // Collect sorted vec of layers and count deltas
+            let mut layers = Vec::new();
+            let mut n_deltas = 0usize;
+
+            for layer in fs::read_dir(timeline.path())? {
+                let layer = layer?;
+                if let Some(mut layer_file) =
+                    parse_filename(&layer.file_name().into_string().unwrap())
+                {
+                    if layer_file.is_delta {
+                        layer_file.holes = get_holes(&layer.path(), max_holes)?;
+                        n_deltas += 1;
+                    }
+                    layers.push(layer_file);
+                }
+            }
+            layers.sort_by_key(|layer| layer.lsn_range.end);
+
+            // Count the number of holes and number of excess layers.
+            // Excess layer is image layer generated when holes in delta layers are not considered.
+            let mut n_excess_layers = 0usize;
+            let mut n_holes = 0usize;
+
+            for i in 0..layers.len() {
+                if !layers[i].is_delta {
+                    let mut n_deltas_since_last_image = 0usize;
+                    let mut n_skipped = 0usize;
+                    let img_key_range = &layers[i].key_range;
+                    for j in (0..i).rev() {
+                        if range_overlaps(img_key_range, &layers[j].key_range) {
+                            if layers[j].is_delta {
+                                n_deltas_since_last_image += 1;
+                                if layers[j].skips(img_key_range) {
+                                    n_skipped += 1;
+                                }
+                            } else {
+                                // Image layer is always dense, despite to the fact that it doesn't contain all possible
+                                // key values in the specified range: there are may be no keys in the storage belonging
+                                // to the image layer range but not present in the image layer.
+                                break;
+                            }
+                        }
+                    }
+                    if n_deltas_since_last_image >= 3 && n_deltas_since_last_image - n_skipped < 3 {
+                        // It is just approximation: it doesn't take in account all image coverage.
+                        // Moreover the new layer map doesn't count total deltas, but the max stack of overlapping deltas.
+                        n_excess_layers += 1;
+                    }
+                    n_holes += n_skipped;
+                }
+            }
+            println!(
+                "Tenant {} timeline {} delta layers {} image layers {} excess layers {} holes {}",
+                tenant.file_name().into_string().unwrap(),
+                timeline.file_name().into_string().unwrap(),
+                n_deltas,
+                layers.len() - n_deltas,
+                n_excess_layers,
+                n_holes
+            );
+            total_delta_layers += n_deltas;
+            total_image_layers += layers.len() - n_deltas;
+            total_excess_layers += n_excess_layers;
+        }
+    }
+    println!(
+        "Total delta layers {} image layers {} excess layers {}",
+        total_delta_layers, total_image_layers, total_excess_layers
+    );
+    Ok(())
+}

pageserver/src/bin/pageserver.rs

@@ -7,12 +7,14 @@ use std::{env, ops::ControlFlow, path::Path, str::FromStr};
 use anyhow::{anyhow, Context};
 use clap::{Arg, ArgAction, Command};
 use fail::FailScenario;
+use metrics::launch_timestamp::{set_launch_timestamp_metric, LaunchTimestamp};
 use remote_storage::GenericRemoteStorage;
 use tracing::*;
 
 use metrics::set_build_info_metric;
 use pageserver::{
     config::{defaults::*, PageServerConf},
+    context::{DownloadBehavior, RequestContext},
     http, page_cache, page_service, task_mgr,
     task_mgr::TaskKind,
     task_mgr::{
@@ -26,7 +28,7 @@ use utils::{
     logging,
     postgres_backend::AuthType,
     project_git_version,
-    sentry_init::{init_sentry, release_name},
+    sentry_init::init_sentry,
     signals::{self, Signal},
     tcp_listener,
 };
@@ -51,6 +53,8 @@ fn version() -> String {
 }
 
 fn main() -> anyhow::Result<()> {
+    let launch_ts = Box::leak(Box::new(LaunchTimestamp::generate()));
+
     let arg_matches = cli().get_matches();
 
     if arg_matches.get_flag("enabled-features") {
@@ -84,8 +88,18 @@ fn main() -> anyhow::Result<()> {
         }
     };
 
+    // Initialize logging, which must be initialized before the custom panic hook is installed.
+    logging::init(conf.log_format)?;
+
+    // mind the order required here: 1. logging, 2. panic_hook, 3. sentry.
+    // disarming this hook on pageserver, because we never tear down tracing.
+    logging::replace_panic_hook_with_tracing_panic_hook().forget();
+
     // initialize sentry if SENTRY_DSN is provided
-    let _sentry_guard = init_sentry(release_name!(), &[("node_id", &conf.id.to_string())]);
+    let _sentry_guard = init_sentry(
+        Some(GIT_VERSION.into()),
+        &[("node_id", &conf.id.to_string())],
+    );
 
     let tenants_path = conf.tenants_path();
     if !tenants_path.exists() {
@@ -104,7 +118,7 @@ fn main() -> anyhow::Result<()> {
     virtual_file::init(conf.max_file_descriptors);
     page_cache::init(conf.page_cache_size);
 
-    start_pageserver(conf).context("Failed to start pageserver")?;
+    start_pageserver(launch_ts, conf).context("Failed to start pageserver")?;
 
     scenario.teardown();
     Ok(())
@@ -199,13 +213,21 @@ fn initialize_config(
     })
 }
 
-fn start_pageserver(conf: &'static PageServerConf) -> anyhow::Result<()> {
-    // Initialize logging
-    logging::init(conf.log_format)?;
-
-    // Print version to the log, and expose it as a prometheus metric too.
-    info!("version: {}", version());
+fn start_pageserver(
+    launch_ts: &'static LaunchTimestamp,
+    conf: &'static PageServerConf,
+) -> anyhow::Result<()> {
+    // Print version and launch timestamp to the log,
+    // and expose them as prometheus metrics.
+    // A changed version string indicates changed software.
+    // A changed launch timestamp indicates a pageserver restart.
+    info!(
+        "version: {} launch_timestamp: {}",
+        version(),
+        launch_ts.to_string()
+    );
     set_build_info_metric(GIT_VERSION);
+    set_launch_timestamp_metric(launch_ts);
 
     // If any failpoints were set from FAILPOINTS environment variable,
     // print them to the log for debugging purposes
@@ -246,7 +268,7 @@ fn start_pageserver(conf: &'static PageServerConf) -> anyhow::Result<()> {
     let signals = signals::install_shutdown_handlers()?;
 
     // Launch broker client
-    WALRECEIVER_RUNTIME.block_on(pageserver::walreceiver::init_broker_client(conf))?;
+    WALRECEIVER_RUNTIME.block_on(pageserver::broker_client::init_broker_client(conf))?;
 
     // Initialize authentication for incoming connections
     let auth = match &conf.auth_type {
@@ -303,7 +325,7 @@ fn start_pageserver(conf: &'static PageServerConf) -> anyhow::Result<()> {
     {
         let _rt_guard = MGMT_REQUEST_RUNTIME.enter();
 
-        let router = http::make_router(conf, auth.clone(), remote_storage)?
+        let router = http::make_router(conf, launch_ts, auth.clone(), remote_storage)?
             .build()
             .map_err(|err| anyhow!(err))?;
         let service = utils::http::RouterService::new(router).unwrap();
@@ -325,6 +347,13 @@ fn start_pageserver(conf: &'static PageServerConf) -> anyhow::Result<()> {
         );
 
         if let Some(metric_collection_endpoint) = &conf.metric_collection_endpoint {
+            let metrics_ctx = RequestContext::todo_child(
+                TaskKind::MetricsCollection,
+                // This task itself shouldn't download anything.
+                // The actual size calculation does need downloads, and
+                // creates a child context with the right DownloadBehavior.
+                DownloadBehavior::Error,
+            );
             task_mgr::spawn(
                 MGMT_REQUEST_RUNTIME.handle(),
                 TaskKind::MetricsCollection,
@@ -336,8 +365,10 @@ fn start_pageserver(conf: &'static PageServerConf) -> anyhow::Result<()> {
                     pageserver::consumption_metrics::collect_metrics(
                         metric_collection_endpoint,
                         conf.metric_collection_interval,
+                        conf.cached_metric_collection_interval,
                         conf.synthetic_size_calculation_interval,
                         conf.id,
+                        metrics_ctx,
                     )
                     .instrument(info_span!("metrics_collection"))
                     .await?;
@@ -349,17 +380,34 @@ fn start_pageserver(conf: &'static PageServerConf) -> anyhow::Result<()> {
 
     // Spawn a task to listen for libpq connections. It will spawn further tasks
     // for each connection. We created the listener earlier already.
-    task_mgr::spawn(
-        COMPUTE_REQUEST_RUNTIME.handle(),
-        TaskKind::LibpqEndpointListener,
-        None,
-        None,
-        "libpq endpoint listener",
-        true,
-        async move {
-            page_service::libpq_listener_main(conf, auth, pageserver_listener, conf.auth_type).await
-        },
-    );
+    {
+        let libpq_ctx = RequestContext::todo_child(
+            TaskKind::LibpqEndpointListener,
+            // listener task shouldn't need to download anything. (We will
+            // create a separate sub-contexts for each connection, with their
+            // own download behavior. This context is used only to listen and
+            // accept connections.)
+            DownloadBehavior::Error,
+        );
+        task_mgr::spawn(
+            COMPUTE_REQUEST_RUNTIME.handle(),
+            TaskKind::LibpqEndpointListener,
+            None,
+            None,
+            "libpq endpoint listener",
+            true,
+            async move {
+                page_service::libpq_listener_main(
+                    conf,
+                    auth,
+                    pageserver_listener,
+                    conf.auth_type,
+                    libpq_ctx,
+                )
+                .await
+            },
+        );
+    }
 
     // All started up! Now just sit and wait for shutdown signal.
     signals.handle(|signal| match signal {

pageserver/src/bin/pageserver_binutils.rs

@@ -12,7 +12,9 @@ use anyhow::Context;
 use clap::{value_parser, Arg, Command};
 
 use pageserver::{
+    context::{DownloadBehavior, RequestContext},
     page_cache,
+    task_mgr::TaskKind,
     tenant::{dump_layerfile_from_path, metadata::TimelineMetadata},
     virtual_file,
 };
@@ -75,7 +77,8 @@ fn print_layerfile(path: &Path) -> anyhow::Result<()> {
     // Basic initialization of things that don't change after startup
     virtual_file::init(10);
     page_cache::init(100);
-    dump_layerfile_from_path(path, true)
+    let ctx = RequestContext::new(TaskKind::DebugTool, DownloadBehavior::Error);
+    dump_layerfile_from_path(path, true, &ctx)
 }
 
 fn handle_metadata(path: &Path, arg_matches: &clap::ArgMatches) -> Result<(), anyhow::Error> {

pageserver/src/broker_client.rs

@@ -0,0 +1,48 @@
+//! The broker client instance of the pageserver, created during pageserver startup.
+//! Used by each timelines' [`walreceiver`].
+
+use crate::config::PageServerConf;
+
+use anyhow::Context;
+use once_cell::sync::OnceCell;
+use storage_broker::BrokerClientChannel;
+use tracing::*;
+
+static BROKER_CLIENT: OnceCell<BrokerClientChannel> = OnceCell::new();
+
+///
+/// Initialize the broker client. This must be called once at page server startup.
+///
+pub async fn init_broker_client(conf: &'static PageServerConf) -> anyhow::Result<()> {
+    let broker_endpoint = conf.broker_endpoint.clone();
+
+    // Note: we do not attempt connecting here (but validate endpoints sanity).
+    let broker_client =
+        storage_broker::connect(broker_endpoint.clone(), conf.broker_keepalive_interval).context(
+            format!(
+                "Failed to create broker client to {}",
+                &conf.broker_endpoint
+            ),
+        )?;
+
+    if BROKER_CLIENT.set(broker_client).is_err() {
+        panic!("broker already initialized");
+    }
+
+    info!(
+        "Initialized broker client with endpoints: {}",
+        broker_endpoint
+    );
+    Ok(())
+}
+
+///
+/// Get a handle to the broker client
+///
+pub fn get_broker_client() -> &'static BrokerClientChannel {
+    BROKER_CLIENT.get().expect("broker client not initialized")
+}
+
+pub fn is_broker_client_initialized() -> bool {
+    BROKER_CLIENT.get().is_some()
+}

pageserver/src/config.rs

@@ -58,6 +58,7 @@ pub mod defaults {
         super::ConfigurableSemaphore::DEFAULT_INITIAL.get();
 
     pub const DEFAULT_METRIC_COLLECTION_INTERVAL: &str = "10 min";
+    pub const DEFAULT_CACHED_METRIC_COLLECTION_INTERVAL: &str = "1 hour";
     pub const DEFAULT_METRIC_COLLECTION_ENDPOINT: Option<reqwest::Url> = None;
     pub const DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL: &str = "10 min";
 
@@ -85,6 +86,7 @@ pub mod defaults {
 #concurrent_tenant_size_logical_size_queries = '{DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES}'
 
 #metric_collection_interval = '{DEFAULT_METRIC_COLLECTION_INTERVAL}'
+#cached_metric_collection_interval = '{DEFAULT_CACHED_METRIC_COLLECTION_INTERVAL}'
 #synthetic_size_calculation_interval = '{DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL}'
 
 # [tenant_config]
@@ -154,10 +156,14 @@ pub struct PageServerConf {
 
     // How often to collect metrics and send them to the metrics endpoint.
     pub metric_collection_interval: Duration,
+    // How often to send unchanged cached metrics to the metrics endpoint.
+    pub cached_metric_collection_interval: Duration,
     pub metric_collection_endpoint: Option<Url>,
     pub synthetic_size_calculation_interval: Duration,
 
     pub test_remote_failures: u64,
+
+    pub ondemand_download_behavior_treat_error_as_warn: bool,
 }
 
 /// We do not want to store this in a PageServerConf because the latter may be logged
@@ -218,10 +224,13 @@ struct PageServerConfigBuilder {
     concurrent_tenant_size_logical_size_queries: BuilderValue<ConfigurableSemaphore>,
 
     metric_collection_interval: BuilderValue<Duration>,
+    cached_metric_collection_interval: BuilderValue<Duration>,
     metric_collection_endpoint: BuilderValue<Option<Url>>,
     synthetic_size_calculation_interval: BuilderValue<Duration>,
 
     test_remote_failures: BuilderValue<u64>,
+
+    ondemand_download_behavior_treat_error_as_warn: BuilderValue<bool>,
 }
 
 impl Default for PageServerConfigBuilder {
@@ -260,6 +269,10 @@ impl Default for PageServerConfigBuilder {
                 DEFAULT_METRIC_COLLECTION_INTERVAL,
             )
             .expect("cannot parse default metric collection interval")),
+            cached_metric_collection_interval: Set(humantime::parse_duration(
+                DEFAULT_CACHED_METRIC_COLLECTION_INTERVAL,
+            )
+            .expect("cannot parse default cached_metric_collection_interval")),
             synthetic_size_calculation_interval: Set(humantime::parse_duration(
                 DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL,
             )
@@ -267,6 +280,8 @@ impl Default for PageServerConfigBuilder {
             metric_collection_endpoint: Set(DEFAULT_METRIC_COLLECTION_ENDPOINT),
 
             test_remote_failures: Set(0),
+
+            ondemand_download_behavior_treat_error_as_warn: Set(false),
         }
     }
 }
@@ -347,6 +362,14 @@ impl PageServerConfigBuilder {
         self.metric_collection_interval = BuilderValue::Set(metric_collection_interval)
     }
 
+    pub fn cached_metric_collection_interval(
+        &mut self,
+        cached_metric_collection_interval: Duration,
+    ) {
+        self.cached_metric_collection_interval =
+            BuilderValue::Set(cached_metric_collection_interval)
+    }
+
     pub fn metric_collection_endpoint(&mut self, metric_collection_endpoint: Option<Url>) {
         self.metric_collection_endpoint = BuilderValue::Set(metric_collection_endpoint)
     }
@@ -363,6 +386,14 @@ impl PageServerConfigBuilder {
         self.test_remote_failures = BuilderValue::Set(fail_first);
     }
 
+    pub fn ondemand_download_behavior_treat_error_as_warn(
+        &mut self,
+        ondemand_download_behavior_treat_error_as_warn: bool,
+    ) {
+        self.ondemand_download_behavior_treat_error_as_warn =
+            BuilderValue::Set(ondemand_download_behavior_treat_error_as_warn);
+    }
+
     pub fn build(self) -> anyhow::Result<PageServerConf> {
         Ok(PageServerConf {
             listen_pg_addr: self
@@ -413,6 +444,9 @@ impl PageServerConfigBuilder {
             metric_collection_interval: self
                 .metric_collection_interval
                 .ok_or(anyhow!("missing metric_collection_interval"))?,
+            cached_metric_collection_interval: self
+                .cached_metric_collection_interval
+                .ok_or(anyhow!("missing cached_metric_collection_interval"))?,
             metric_collection_endpoint: self
                 .metric_collection_endpoint
                 .ok_or(anyhow!("missing metric_collection_endpoint"))?,
@@ -422,6 +456,11 @@ impl PageServerConfigBuilder {
             test_remote_failures: self
                 .test_remote_failures
                 .ok_or(anyhow!("missing test_remote_failuers"))?,
+            ondemand_download_behavior_treat_error_as_warn: self
+                .ondemand_download_behavior_treat_error_as_warn
+                .ok_or(anyhow!(
+                    "missing ondemand_download_behavior_treat_error_as_warn"
+                ))?,
         })
     }
 }
@@ -593,6 +632,7 @@ impl PageServerConf {
                     ConfigurableSemaphore::new(permits)
                 }),
                 "metric_collection_interval" => builder.metric_collection_interval(parse_toml_duration(key, item)?),
+                "cached_metric_collection_interval" => builder.cached_metric_collection_interval(parse_toml_duration(key, item)?),
                 "metric_collection_endpoint" => {
                     let endpoint = parse_toml_string(key, item)?.parse().context("failed to parse metric_collection_endpoint")?;
                     builder.metric_collection_endpoint(Some(endpoint));
@@ -600,6 +640,7 @@ impl PageServerConf {
                 "synthetic_size_calculation_interval" =>
                     builder.synthetic_size_calculation_interval(parse_toml_duration(key, item)?),
                 "test_remote_failures" => builder.test_remote_failures(parse_toml_u64(key, item)?),
+                "ondemand_download_behavior_treat_error_as_warn" => builder.ondemand_download_behavior_treat_error_as_warn(parse_toml_bool(key, item)?),
                 _ => bail!("unrecognized pageserver option '{key}'"),
             }
         }
@@ -690,6 +731,13 @@ impl PageServerConf {
                 })?);
         }
 
+        if let Some(eviction_policy) = item.get("eviction_policy") {
+            t_conf.eviction_policy = Some(
+                toml_edit::de::from_item(eviction_policy.clone())
+                    .context("parse eviction_policy")?,
+            );
+        }
+
         Ok(t_conf)
     }
 
@@ -721,9 +769,11 @@ impl PageServerConf {
             log_format: LogFormat::from_str(defaults::DEFAULT_LOG_FORMAT).unwrap(),
             concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore::default(),
             metric_collection_interval: Duration::from_secs(60),
+            cached_metric_collection_interval: Duration::from_secs(60 * 60),
             metric_collection_endpoint: defaults::DEFAULT_METRIC_COLLECTION_ENDPOINT,
             synthetic_size_calculation_interval: Duration::from_secs(60),
             test_remote_failures: 0,
+            ondemand_download_behavior_treat_error_as_warn: false,
         }
     }
 }
@@ -749,6 +799,11 @@ fn parse_toml_u64(name: &str, item: &Item) -> Result<u64> {
     Ok(i as u64)
 }
 
+fn parse_toml_bool(name: &str, item: &Item) -> Result<bool> {
+    item.as_bool()
+        .with_context(|| format!("configure option {name} is not a bool"))
+}
+
 fn parse_toml_duration(name: &str, item: &Item) -> Result<Duration> {
     let s = item
         .as_str()
@@ -855,6 +910,7 @@ initial_superuser_name = 'zzzz'
 id = 10
 
 metric_collection_interval = '222 s'
+cached_metric_collection_interval = '22200 s'
 metric_collection_endpoint = 'http://localhost:80/metrics'
 synthetic_size_calculation_interval = '333 s'
 log_format = 'json'
@@ -902,11 +958,15 @@ log_format = 'json'
                 metric_collection_interval: humantime::parse_duration(
                     defaults::DEFAULT_METRIC_COLLECTION_INTERVAL
                 )?,
+                cached_metric_collection_interval: humantime::parse_duration(
+                    defaults::DEFAULT_CACHED_METRIC_COLLECTION_INTERVAL
+                )?,
                 metric_collection_endpoint: defaults::DEFAULT_METRIC_COLLECTION_ENDPOINT,
                 synthetic_size_calculation_interval: humantime::parse_duration(
                     defaults::DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL
                 )?,
                 test_remote_failures: 0,
+                ondemand_download_behavior_treat_error_as_warn: false,
             },
             "Correct defaults should be used when no config values are provided"
         );
@@ -951,9 +1011,11 @@ log_format = 'json'
                 log_format: LogFormat::Json,
                 concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore::default(),
                 metric_collection_interval: Duration::from_secs(222),
+                cached_metric_collection_interval: Duration::from_secs(22200),
                 metric_collection_endpoint: Some(Url::parse("http://localhost:80/metrics")?),
                 synthetic_size_calculation_interval: Duration::from_secs(333),
                 test_remote_failures: 0,
+                ondemand_download_behavior_treat_error_as_warn: false,
             },
             "Should be able to parse all basic config values correctly"
         );

pageserver/src/consumption_metrics.rs

@@ -3,6 +3,7 @@
 //! and push them to a HTTP endpoint.
 //! Cache metrics to send only the updated ones.
 //!
+use crate::context::{DownloadBehavior, RequestContext};
 use crate::task_mgr::{self, TaskKind, BACKGROUND_RUNTIME};
 use crate::tenant::mgr;
 use anyhow;
@@ -24,7 +25,7 @@ const REMOTE_STORAGE_SIZE: &str = "remote_storage_size";
 const TIMELINE_LOGICAL_SIZE: &str = "timeline_logical_size";
 
 #[serde_as]
-#[derive(Serialize)]
+#[derive(Serialize, Debug)]
 struct Ids {
     #[serde_as(as = "DisplayFromStr")]
     tenant_id: TenantId,
@@ -45,14 +46,17 @@ pub struct PageserverConsumptionMetricsKey {
 pub async fn collect_metrics(
     metric_collection_endpoint: &Url,
     metric_collection_interval: Duration,
+    cached_metric_collection_interval: Duration,
     synthetic_size_calculation_interval: Duration,
     node_id: NodeId,
+    ctx: RequestContext,
 ) -> anyhow::Result<()> {
     let mut ticker = tokio::time::interval(metric_collection_interval);
-
     info!("starting collect_metrics");
 
     // spin up background worker that caclulates tenant sizes
+    let worker_ctx =
+        ctx.detached_child(TaskKind::CalculateSyntheticSize, DownloadBehavior::Download);
     task_mgr::spawn(
         BACKGROUND_RUNTIME.handle(),
         TaskKind::CalculateSyntheticSize,
@@ -61,7 +65,7 @@ pub async fn collect_metrics(
         "synthetic size calculation",
         false,
         async move {
-            calculate_synthetic_size_worker(synthetic_size_calculation_interval)
+            calculate_synthetic_size_worker(synthetic_size_calculation_interval, &worker_ctx)
                 .instrument(info_span!("synthetic_size_worker"))
                 .await?;
             Ok(())
@@ -71,6 +75,7 @@ pub async fn collect_metrics(
     // define client here to reuse it for all requests
     let client = reqwest::Client::new();
     let mut cached_metrics: HashMap<PageserverConsumptionMetricsKey, u64> = HashMap::new();
+    let mut prev_iteration_time: std::time::Instant = std::time::Instant::now();
 
     loop {
         tokio::select! {
@@ -79,10 +84,15 @@ pub async fn collect_metrics(
                 return Ok(());
             },
             _ = ticker.tick() => {
-                if let Err(err) = collect_metrics_iteration(&client, &mut cached_metrics, metric_collection_endpoint, node_id).await
-                {
-                    error!("metrics collection failed: {err:?}");
+
+                // send cached metrics every cached_metric_collection_interval
+                let send_cached = prev_iteration_time.elapsed() >= cached_metric_collection_interval;
+
+                if send_cached {
+                    prev_iteration_time = std::time::Instant::now();
                 }
+
+                collect_metrics_iteration(&client, &mut cached_metrics, metric_collection_endpoint, node_id, &ctx, send_cached).await;
             }
         }
     }
@@ -93,16 +103,19 @@ pub async fn collect_metrics(
 /// Gather per-tenant and per-timeline metrics and send them to the `metric_collection_endpoint`.
 /// Cache metrics to avoid sending the same metrics multiple times.
 ///
+/// This function handles all errors internally
+/// and doesn't break iteration if just one tenant fails.
+///
 /// TODO
 /// - refactor this function (chunking+sending part) to reuse it in proxy module;
-/// - improve error handling. Now if one tenant fails to collect metrics,
-/// the whole iteration fails and metrics for other tenants are not collected.
 pub async fn collect_metrics_iteration(
     client: &reqwest::Client,
     cached_metrics: &mut HashMap<PageserverConsumptionMetricsKey, u64>,
     metric_collection_endpoint: &reqwest::Url,
     node_id: NodeId,
-) -> anyhow::Result<()> {
+    ctx: &RequestContext,
+    send_cached: bool,
+) {
     let mut current_metrics: Vec<(PageserverConsumptionMetricsKey, u64)> = Vec::new();
     trace!(
         "starting collect_metrics_iteration. metric_collection_endpoint: {}",
@@ -110,7 +123,13 @@ pub async fn collect_metrics_iteration(
     );
 
     // get list of tenants
-    let tenants = mgr::list_tenants().await;
+    let tenants = match mgr::list_tenants().await {
+        Ok(tenants) => tenants,
+        Err(err) => {
+            error!("failed to list tenants: {:?}", err);
+            return;
+        }
+    };
 
     // iterate through list of Active tenants and collect metrics
     for (tenant_id, tenant_state) in tenants {
@@ -118,7 +137,15 @@ pub async fn collect_metrics_iteration(
             continue;
         }
 
-        let tenant = mgr::get_tenant(tenant_id, true).await?;
+        let tenant = match mgr::get_tenant(tenant_id, true).await {
+            Ok(tenant) => tenant,
+            Err(err) => {
+                // It is possible that tenant was deleted between
+                // `list_tenants` and `get_tenant`, so just warn about it.
+                warn!("failed to get tenant {tenant_id:?}: {err:?}");
+                continue;
+            }
+        };
 
         let mut tenant_resident_size = 0;
 
@@ -137,29 +164,51 @@ pub async fn collect_metrics_iteration(
                     timeline_written_size,
                 ));
 
-                let (timeline_logical_size, is_exact) = timeline.get_current_logical_size()?;
-                // Only send timeline logical size when it is fully calculated.
-                if is_exact {
-                    current_metrics.push((
-                        PageserverConsumptionMetricsKey {
-                            tenant_id,
-                            timeline_id: Some(timeline.timeline_id),
-                            metric: TIMELINE_LOGICAL_SIZE,
-                        },
-                        timeline_logical_size,
-                    ));
-                }
+                match timeline.get_current_logical_size(ctx) {
+                    // Only send timeline logical size when it is fully calculated.
+                    Ok((size, is_exact)) if is_exact => {
+                        current_metrics.push((
+                            PageserverConsumptionMetricsKey {
+                                tenant_id,
+                                timeline_id: Some(timeline.timeline_id),
+                                metric: TIMELINE_LOGICAL_SIZE,
+                            },
+                            size,
+                        ));
+                    }
+                    Ok((_, _)) => {}
+                    Err(err) => {
+                        error!(
+                            "failed to get current logical size for timeline {}: {err:?}",
+                            timeline.timeline_id
+                        );
+                        continue;
+                    }
+                };
             }
 
             let timeline_resident_size = timeline.get_resident_physical_size();
             tenant_resident_size += timeline_resident_size;
         }
 
-        let tenant_remote_size = tenant.get_remote_size().await?;
-        debug!(
-            "collected current metrics for tenant: {}: state={:?} resident_size={} remote_size={}",
-            tenant_id, tenant_state, tenant_resident_size, tenant_remote_size
-        );
+        match tenant.get_remote_size().await {
+            Ok(tenant_remote_size) => {
+                current_metrics.push((
+                    PageserverConsumptionMetricsKey {
+                        tenant_id,
+                        timeline_id: None,
+                        metric: REMOTE_STORAGE_SIZE,
+                    },
+                    tenant_remote_size,
+                ));
+            }
+            Err(err) => {
+                error!(
+                    "failed to get remote size for tenant {}: {err:?}",
+                    tenant_id
+                );
+            }
+        }
 
         current_metrics.push((
             PageserverConsumptionMetricsKey {
@@ -170,15 +219,6 @@ pub async fn collect_metrics_iteration(
             tenant_resident_size,
         ));
 
-        current_metrics.push((
-            PageserverConsumptionMetricsKey {
-                tenant_id,
-                timeline_id: None,
-                metric: REMOTE_STORAGE_SIZE,
-            },
-            tenant_remote_size,
-        ));
-
         // Note that this metric is calculated in a separate bgworker
         // Here we only use cached value, which may lag behind the real latest one
         let tenant_synthetic_size = tenant.get_cached_synthetic_size();
@@ -192,15 +232,18 @@ pub async fn collect_metrics_iteration(
         ));
     }
 
-    // Filter metrics
-    current_metrics.retain(|(curr_key, curr_val)| match cached_metrics.get(curr_key) {
-        Some(val) => val != curr_val,
-        None => true,
-    });
+    // Filter metrics, unless we want to send all metrics, including cached ones.
+    // See: https://github.com/neondatabase/neon/issues/3485
+    if !send_cached {
+        current_metrics.retain(|(curr_key, curr_val)| match cached_metrics.get(curr_key) {
+            Some(val) => val != curr_val,
+            None => true,
+        });
+    }
 
     if current_metrics.is_empty() {
         trace!("no new metrics to send");
-        return Ok(());
+        return;
     }
 
     // Send metrics.
@@ -244,6 +287,12 @@ pub async fn collect_metrics_iteration(
                     }
                 } else {
                     error!("metrics endpoint refused the sent metrics: {:?}", res);
+                    for metric in chunk_to_send.iter() {
+                        // Report if the metric value is suspiciously large
+                        if metric.value > (1u64 << 40) {
+                            error!("potentially abnormal metric value: {:?}", metric);
+                        }
+                    }
                 }
             }
             Err(err) => {
@@ -251,13 +300,12 @@ pub async fn collect_metrics_iteration(
             }
         }
     }
-
-    Ok(())
 }
 
 /// Caclculate synthetic size for each active tenant
 pub async fn calculate_synthetic_size_worker(
     synthetic_size_calculation_interval: Duration,
+    ctx: &RequestContext,
 ) -> anyhow::Result<()> {
     info!("starting calculate_synthetic_size_worker");
 
@@ -270,7 +318,13 @@ pub async fn calculate_synthetic_size_worker(
             },
         _ = ticker.tick() => {
 
-                let tenants = mgr::list_tenants().await;
+                let tenants = match mgr::list_tenants().await {
+                    Ok(tenants) => tenants,
+                    Err(e) => {
+                        warn!("cannot get tenant list: {e:#}");
+                        continue;
+                    }
+                };
                 // iterate through list of Active tenants and collect metrics
                 for (tenant_id, tenant_state) in tenants {
 
@@ -280,7 +334,7 @@ pub async fn calculate_synthetic_size_worker(
 
                     if let Ok(tenant) = mgr::get_tenant(tenant_id, true).await
                     {
-                        if let Err(e) = tenant.calculate_synthetic_size().await {
+                        if let Err(e) = tenant.calculate_synthetic_size(ctx).await {
                             error!("failed to calculate synthetic size for tenant {}: {}", tenant_id, e);
                         }
                     }

pageserver/src/context.rs

@@ -0,0 +1,199 @@
+//! This module defines `RequestContext`, a structure that we use throughout
+//! the pageserver to propagate high-level context from places
+//! that _originate_ activity down to the shared code paths at the
+//! heart of the pageserver. It's inspired by Golang's `context.Context`.
+//!
+//! For example, in `Timeline::get(page_nr, lsn)` we need to answer the following questions:
+//! 1. What high-level activity ([`TaskKind`]) needs this page?
+//!    We need that information as a categorical dimension for page access
+//!    statistics, which we, in turn, need to guide layer eviction policy design.
+//! 2. How should we behave if, to produce the page image, we need to
+//!    on-demand download a layer file ([`DownloadBehavior`]).
+//!
+//! [`RequestContext`] satisfies those needs.
+//! The current implementation is a small `struct` that is passed through
+//! the call chain by reference.
+//!
+//! ### Future Work
+//!
+//! However, we do not intend to stop here, since there are other needs that
+//! require carrying information from high to low levels of the app.
+//!
+//! Most importantly, **cancellation signaling** in response to
+//! 1. timeouts (page_service max response time) and
+//! 2. lifecycle requests (detach tenant, delete timeline).
+//!
+//! Related to that, there is sometimes a need to ensure that all tokio tasks spawned
+//! by the transitive callees of a request have finished. The keyword here
+//! is **Structured Concurrency**, and right now, we use `task_mgr` in most places,
+//! `TaskHandle` in some places, and careful code review around `FuturesUnordered`
+//! or `JoinSet` in other places.
+//!
+//! We do not yet have a systematic cancellation story in pageserver, and it is
+//! pretty clear that [`RequestContext`] will be responsible for that.
+//! So, the API already prepares for this role through the
+//! [`RequestContext::detached_child`] and [`RequestContext::attached_child`]  methods.
+//! See their doc comments for details on how we will use them in the future.
+//!
+//! It is not clear whether or how we will enforce Structured Concurrency, and
+//! what role [`RequestContext`] will play there.
+//! So, the API doesn't prepare us for this topic.
+//!
+//! Other future uses of `RequestContext`:
+//! - Communicate compute & IO priorities (user-initiated request vs. background-loop)
+//! - Request IDs for distributed tracing
+//! - Request/Timeline/Tenant-scoped log levels
+//!
+//! RequestContext might look quite different once it supports those features.
+//! Likely, it will have a shape similar to Golang's `context.Context`.
+//!
+//! ### Why A Struct Instead Of Method Parameters
+//!
+//! What's typical about such information is that it needs to be passed down
+//! along the call chain from high level to low level, but few of the functions
+//! in the middle need to understand it.
+//! Further, it is to be expected that we will need to propagate more data
+//! in the future (see the earlier section on future work).
+//! Hence, for functions in the middle of the call chain, we have the following
+//! requirements:
+//! 1. It should be easy to forward the context to callees.
+//! 2. To propagate more data from high-level to low-level code, the functions in
+//!    the middle should not need to be modified.
+//! The solution is to have a container structure ([`RequestContext`]) that
+//! carries the information. Functions that don't care about what's in it
+//! pass it along to callees.
+//!
+//! ### Why Not Task-Local Variables
+//!
+//! One could use task-local variables (the equivalent of thread-local variables)
+//! to address the immediate needs outlined above.
+//! However, we reject task-local variables because:
+//! 1. they are implicit, thereby making it harder to trace the data flow in code
+//!    reviews and during debugging,
+//! 2. they can be mutable, which enables implicit return data flow,
+//! 3. they are restrictive in that code which fans out into multiple tasks,
+//!    or even threads, needs to carefully propagate the state.
+//!
+//! In contrast, information flow with [`RequestContext`] is
+//! 1. always explicit,
+//! 2. strictly uni-directional because RequestContext is immutable,
+//! 3. tangible because a [`RequestContext`] is just a value.
+//!    When creating child activities, regardless of whether it's a task,
+//!    thread, or even an RPC to another service, the value can
+//!    be used like any other argument.
+//!
+//! The solution is that all code paths are infected with precisely one
+//! [`RequestContext`] argument. Functions in the middle of the call chain
+//! only need to pass it on.
+use crate::task_mgr::TaskKind;
+
+// The main structure of this module, see module-level comment.
+pub struct RequestContext {
+    task_kind: TaskKind,
+    download_behavior: DownloadBehavior,
+}
+
+/// Desired behavior if the operation requires an on-demand download
+/// to proceed.
+#[derive(Clone, Copy, PartialEq, Eq)]
+pub enum DownloadBehavior {
+    /// Download the layer file. It can take a while.
+    Download,
+
+    /// Download the layer file, but print a warning to the log. This should be used
+    /// in code where the layer file is expected to already exist locally.
+    Warn,
+
+    /// Return a PageReconstructError::NeedsDownload error
+    Error,
+}
+
+impl RequestContext {
+    /// Create a new RequestContext that has no parent.
+    ///
+    /// The function is called `new` because, once we add children
+    /// to it using `detached_child` or `attached_child`, the context
+    /// form a tree (not implemented yet since cancellation will be
+    /// the first feature that requires a tree).
+    ///
+    /// # Future: Cancellation
+    ///
+    /// The only reason why a context like this one can be canceled is
+    /// because someone explicitly canceled it.
+    /// It has no parent, so it cannot inherit cancellation from there.
+    pub fn new(task_kind: TaskKind, download_behavior: DownloadBehavior) -> Self {
+        RequestContext {
+            task_kind,
+            download_behavior,
+        }
+    }
+
+    /// Create a detached child context for a task that may outlive `self`.
+    ///
+    /// Use this when spawning new background activity that should complete
+    /// even if the current request is canceled.
+    ///
+    /// # Future: Cancellation
+    ///
+    /// Cancellation of `self` will not propagate to the child context returned
+    /// by this method.
+    ///
+    /// # Future: Structured Concurrency
+    ///
+    /// We could add the Future as a parameter to this function, spawn it as a task,
+    /// and pass to the new task the child context as an argument.
+    /// That would be an ergonomic improvement.
+    ///
+    /// We could make new calls to this function fail if `self` is already canceled.
+    pub fn detached_child(&self, task_kind: TaskKind, download_behavior: DownloadBehavior) -> Self {
+        self.child_impl(task_kind, download_behavior)
+    }
+
+    /// Create a child of context `self` for a task that shall not outlive `self`.
+    ///
+    /// Use this when fanning-out work to other async tasks.
+    ///
+    /// # Future: Cancellation
+    ///
+    /// Cancelling a context will propagate to its attached children.
+    ///
+    /// # Future: Structured Concurrency
+    ///
+    /// We could add the Future as a parameter to this function, spawn it as a task,
+    /// and track its `JoinHandle` inside the `RequestContext`.
+    ///
+    /// We could then provide another method to allow waiting for all child tasks
+    /// to finish.
+    ///
+    /// We could make new calls to this function fail if `self` is already canceled.
+    /// Alternatively, we could allow the creation but not spawn the task.
+    /// The method to wait for child tasks would return an error, indicating
+    /// that the child task was not started because the context was canceled.
+    pub fn attached_child(&self) -> Self {
+        self.child_impl(self.task_kind(), self.download_behavior())
+    }
+
+    /// Use this function when you should be creating a child context using
+    /// [`attached_child`] or [`detached_child`], but your caller doesn't provide
+    /// a context and you are unwilling to change all callers to provide one.
+    ///
+    /// Before we add cancellation, we should get rid of this method.
+    pub fn todo_child(task_kind: TaskKind, download_behavior: DownloadBehavior) -> Self {
+        Self::new(task_kind, download_behavior)
+    }
+
+    fn child_impl(&self, task_kind: TaskKind, download_behavior: DownloadBehavior) -> Self {
+        RequestContext {
+            task_kind,
+            download_behavior,
+        }
+    }
+
+    pub fn task_kind(&self) -> TaskKind {
+        self.task_kind
+    }
+
+    pub fn download_behavior(&self) -> DownloadBehavior {
+        self.download_behavior
+    }
+}

pageserver/src/http/openapi_spec.yml

@@ -437,6 +437,13 @@ paths:
           type: boolean
         description: |
           When true, skip calculation and only provide the model inputs (for debugging). Defaults to false.
+      - name: retention_period
+        in: query
+        required: false
+        schema:
+          type: integer
+        description: |
+          Override the default retention period (in bytes) used for size calculation.
     get:
       description: |
         Calculate tenant's size, which is a mixture of WAL (bytes) and logical_size (bytes).
@@ -664,6 +671,55 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/Error"
+  /v1/tenant/{tenant_id}/config/:
+    parameters:
+      - name: tenant_id
+        in: path
+        required: true
+        schema:
+          type: string
+          format: hex
+    get:
+      description: |
+        Returns tenant's config description: specific config overrides a tenant has
+        and the effective config.
+      responses:
+        "200":
+          description: Tenant config, specific and effective
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/TenantConfig"
+        "400":
+          description: Malformed get tenanant config request
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/Error"
+        "401":
+          description: Unauthorized Error
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/UnauthorizedError"
+        "403":
+          description: Forbidden Error
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/ForbiddenError"
+        "404":
+          description: Tenand or timeline were not found
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/NotFoundError"
+        "500":
+          description: Generic operation error
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/Error"
 components:
   securitySchemes:
     JWT:
@@ -724,10 +780,33 @@ components:
           type: integer
         checkpoint_timeout:
           type: string
+        compaction_target_size:
+          type: integer
         compaction_period:
           type: string
         compaction_threshold:
           type: string
+        image_creation_threshold:
+          type: integer
+        walreceiver_connect_timeout:
+          type: string
+        lagging_wal_timeout:
+          type: string
+        max_lsn_wal_lag:
+          type: integer
+        trace_read_requests:
+          type: boolean
+    TenantConfig:
+      type: object
+      properties:
+        tenant_specific_overrides:
+          type: object
+          schema:
+            $ref: "#/components/schemas/TenantConfigInfo"
+        effective_config:
+          type: object
+          schema:
+            $ref: "#/components/schemas/TenantConfigInfo"
     TimelineInfo:
       type: object
       required:

pageserver/src/http/routes.rs

@@ -1,19 +1,28 @@
+use std::collections::HashMap;
 use std::sync::Arc;
 
 use anyhow::{anyhow, Context, Result};
 use hyper::StatusCode;
 use hyper::{Body, Request, Response, Uri};
+use metrics::launch_timestamp::LaunchTimestamp;
 use pageserver_api::models::DownloadRemoteLayersTaskSpawnRequest;
 use remote_storage::GenericRemoteStorage;
+use tenant_size_model::{SizeResult, StorageModel};
 use tokio_util::sync::CancellationToken;
 use tracing::*;
+use utils::http::request::{get_request_param, must_get_query_param, parse_query_param};
 
 use super::models::{
     StatusResponse, TenantConfigRequest, TenantCreateRequest, TenantCreateResponse, TenantInfo,
-    TimelineCreateRequest, TimelineInfo,
+    TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
 };
+use crate::context::{DownloadBehavior, RequestContext};
 use crate::pgdatadir_mapping::LsnForTimestamp;
+use crate::task_mgr::TaskKind;
 use crate::tenant::config::TenantConfOpt;
+use crate::tenant::mgr::TenantMapInsertError;
+use crate::tenant::size::ModelInputs;
+use crate::tenant::storage_layer::LayerAccessStatsReset;
 use crate::tenant::{PageReconstructError, Timeline};
 use crate::{config::PageServerConf, tenant::mgr};
 use utils::{
@@ -31,7 +40,7 @@ use utils::{
 
 // Imports only used for testing APIs
 #[cfg(feature = "testing")]
-use super::models::{ConfigureFailpointsRequest, TimelineGcRequest};
+use super::models::ConfigureFailpointsRequest;
 
 struct State {
     conf: &'static PageServerConf,
@@ -81,18 +90,39 @@ fn check_permission(request: &Request<Body>, tenant_id: Option<TenantId>) -> Res
 fn apierror_from_prerror(err: PageReconstructError) -> ApiError {
     match err {
         PageReconstructError::Other(err) => ApiError::InternalServerError(err),
+        PageReconstructError::NeedsDownload(_, _) => {
+            // This shouldn't happen, because we use a RequestContext that requests to
+            // download any missing layer files on-demand.
+            ApiError::InternalServerError(anyhow::anyhow!("need to download remote layer file"))
+        }
+        PageReconstructError::Cancelled => {
+            ApiError::InternalServerError(anyhow::anyhow!("request was cancelled"))
+        }
         PageReconstructError::WalRedo(err) => {
             ApiError::InternalServerError(anyhow::Error::new(err))
         }
     }
 }
 
+fn apierror_from_tenant_map_insert_error(e: TenantMapInsertError) -> ApiError {
+    match e {
+        TenantMapInsertError::StillInitializing | TenantMapInsertError::ShuttingDown => {
+            ApiError::InternalServerError(anyhow::Error::new(e))
+        }
+        TenantMapInsertError::TenantAlreadyExists(id, state) => {
+            ApiError::Conflict(format!("tenant {id} already exists, state: {state:?}"))
+        }
+        TenantMapInsertError::Closure(e) => ApiError::InternalServerError(e),
+    }
+}
+
 // Helper function to construct a TimelineInfo struct for a timeline
 async fn build_timeline_info(
     timeline: &Arc<Timeline>,
     include_non_incremental_logical_size: bool,
+    ctx: &RequestContext,
 ) -> anyhow::Result<TimelineInfo> {
-    let mut info = build_timeline_info_common(timeline)?;
+    let mut info = build_timeline_info_common(timeline, ctx)?;
     if include_non_incremental_logical_size {
         // XXX we should be using spawn_ondemand_logical_size_calculation here.
         // Otherwise, if someone deletes the timeline / detaches the tenant while
@@ -102,6 +132,7 @@ async fn build_timeline_info(
                 .get_current_logical_size_non_incremental(
                     info.last_record_lsn,
                     CancellationToken::new(),
+                    ctx,
                 )
                 .await?,
         );
@@ -109,7 +140,10 @@ async fn build_timeline_info(
     Ok(info)
 }
 
-fn build_timeline_info_common(timeline: &Arc<Timeline>) -> anyhow::Result<TimelineInfo> {
+fn build_timeline_info_common(
+    timeline: &Arc<Timeline>,
+    ctx: &RequestContext,
+) -> anyhow::Result<TimelineInfo> {
     let last_record_lsn = timeline.get_last_record_lsn();
     let (wal_source_connstr, last_received_msg_lsn, last_received_msg_ts) = {
         let guard = timeline.last_received_wal.lock().unwrap();
@@ -129,7 +163,7 @@ fn build_timeline_info_common(timeline: &Arc<Timeline>) -> anyhow::Result<Timeli
         Lsn(0) => None,
         lsn @ Lsn(_) => Some(lsn),
     };
-    let current_logical_size = match timeline.get_current_logical_size() {
+    let current_logical_size = match timeline.get_current_logical_size(ctx) {
         Ok((size, _)) => Some(size),
         Err(err) => {
             error!("Timeline info creation failed to get current logical size: {err:?}");
@@ -180,6 +214,8 @@ async fn timeline_create_handler(mut request: Request<Body>) -> Result<Response<
         .new_timeline_id
         .unwrap_or_else(TimelineId::generate);
 
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
+
     let tenant = mgr::get_tenant(tenant_id, true)
         .await
         .map_err(ApiError::NotFound)?;
@@ -187,13 +223,14 @@ async fn timeline_create_handler(mut request: Request<Body>) -> Result<Response<
         new_timeline_id,
         request_data.ancestor_timeline_id.map(TimelineId::from),
         request_data.ancestor_start_lsn,
-        request_data.pg_version.unwrap_or(crate::DEFAULT_PG_VERSION)
+        request_data.pg_version.unwrap_or(crate::DEFAULT_PG_VERSION),
+        &ctx,
     )
     .instrument(info_span!("timeline_create", tenant = %tenant_id, new_timeline = ?request_data.new_timeline_id, timeline_id = %new_timeline_id, lsn=?request_data.ancestor_start_lsn, pg_version=?request_data.pg_version))
     .await {
         Ok(Some(new_timeline)) => {
             // Created. Construct a TimelineInfo for it.
-            let timeline_info = build_timeline_info_common(&new_timeline)
+            let timeline_info = build_timeline_info_common(&new_timeline, &ctx)
                 .map_err(ApiError::InternalServerError)?;
             json_response(StatusCode::CREATED, timeline_info)
         }
@@ -204,10 +241,12 @@ async fn timeline_create_handler(mut request: Request<Body>) -> Result<Response<
 
 async fn timeline_list_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
-    let include_non_incremental_logical_size =
-        query_param_present(&request, "include-non-incremental-logical-size");
+    let include_non_incremental_logical_size: Option<bool> =
+        parse_query_param(&request, "include-non-incremental-logical-size")?;
     check_permission(&request, Some(tenant_id))?;
 
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+
     let response_data = async {
         let tenant = mgr::get_tenant(tenant_id, true)
             .await
@@ -216,13 +255,14 @@ async fn timeline_list_handler(request: Request<Body>) -> Result<Response<Body>,
 
         let mut response_data = Vec::with_capacity(timelines.len());
         for timeline in timelines {
-            let timeline_info =
-                build_timeline_info(&timeline, include_non_incremental_logical_size)
-                    .await
-                    .context(
-                        "Failed to convert tenant timeline {timeline_id} into the local one: {e:?}",
-                    )
-                    .map_err(ApiError::InternalServerError)?;
+            let timeline_info = build_timeline_info(
+                &timeline,
+                include_non_incremental_logical_size.unwrap_or(false),
+                &ctx,
+            )
+            .await
+            .context("Failed to convert tenant timeline {timeline_id} into the local one: {e:?}")
+            .map_err(ApiError::InternalServerError)?;
 
             response_data.push(timeline_info);
         }
@@ -234,38 +274,16 @@ async fn timeline_list_handler(request: Request<Body>) -> Result<Response<Body>,
     json_response(StatusCode::OK, response_data)
 }
 
-/// Checks if a query param is present in the request's URL
-fn query_param_present(request: &Request<Body>, param: &str) -> bool {
-    request
-        .uri()
-        .query()
-        .map(|v| url::form_urlencoded::parse(v.as_bytes()).any(|(p, _)| p == param))
-        .unwrap_or(false)
-}
-
-fn get_query_param(request: &Request<Body>, param_name: &str) -> Result<String, ApiError> {
-    request.uri().query().map_or(
-        Err(ApiError::BadRequest(anyhow!("empty query in request"))),
-        |v| {
-            url::form_urlencoded::parse(v.as_bytes())
-                .find(|(k, _)| k == param_name)
-                .map_or(
-                    Err(ApiError::BadRequest(anyhow!(
-                        "no {param_name} specified in query parameters"
-                    ))),
-                    |(_, v)| Ok(v.into_owned()),
-                )
-        },
-    )
-}
-
 async fn timeline_detail_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
     let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
-    let include_non_incremental_logical_size =
-        query_param_present(&request, "include-non-incremental-logical-size");
+    let include_non_incremental_logical_size: Option<bool> =
+        parse_query_param(&request, "include-non-incremental-logical-size")?;
     check_permission(&request, Some(tenant_id))?;
 
+    // Logical size calculation needs downloading.
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+
     let timeline_info = async {
         let tenant = mgr::get_tenant(tenant_id, true)
             .await
@@ -275,10 +293,14 @@ async fn timeline_detail_handler(request: Request<Body>) -> Result<Response<Body
             .get_timeline(timeline_id, false)
             .map_err(ApiError::NotFound)?;
 
-        let timeline_info = build_timeline_info(&timeline, include_non_incremental_logical_size)
-            .await
-            .context("get local timeline info")
-            .map_err(ApiError::InternalServerError)?;
+        let timeline_info = build_timeline_info(
+            &timeline,
+            include_non_incremental_logical_size.unwrap_or(false),
+            &ctx,
+        )
+        .await
+        .context("get local timeline info")
+        .map_err(ApiError::InternalServerError)?;
 
         Ok::<_, ApiError>(timeline_info)
     }
@@ -293,18 +315,16 @@ async fn get_lsn_by_timestamp_handler(request: Request<Body>) -> Result<Response
     check_permission(&request, Some(tenant_id))?;
 
     let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
-    let timestamp_raw = get_query_param(&request, "timestamp")?;
-    let timestamp = humantime::parse_rfc3339(timestamp_raw.as_str())
+    let timestamp_raw = must_get_query_param(&request, "timestamp")?;
+    let timestamp = humantime::parse_rfc3339(&timestamp_raw)
         .with_context(|| format!("Invalid time: {:?}", timestamp_raw))
         .map_err(ApiError::BadRequest)?;
     let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);
 
-    let timeline = mgr::get_tenant(tenant_id, true)
-        .await
-        .and_then(|tenant| tenant.get_timeline(timeline_id, true))
-        .map_err(ApiError::NotFound)?;
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let timeline = active_timeline_of_active_tenant(tenant_id, timeline_id).await?;
     let result = timeline
-        .find_lsn_for_timestamp(timestamp_pg)
+        .find_lsn_for_timestamp(timestamp_pg, &ctx)
         .await
         .map_err(apierror_from_prerror)?;
 
@@ -322,16 +342,17 @@ async fn tenant_attach_handler(request: Request<Body>) -> Result<Response<Body>,
     let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
     check_permission(&request, Some(tenant_id))?;
 
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
+
     info!("Handling tenant attach {tenant_id}");
 
     let state = get_state(&request);
 
     if let Some(remote_storage) = &state.remote_storage {
-        // FIXME: distinguish between "Tenant already exists" and other errors
-        mgr::attach_tenant(state.conf, tenant_id, remote_storage.clone())
+        mgr::attach_tenant(state.conf, tenant_id, remote_storage.clone(), &ctx)
             .instrument(info_span!("tenant_attach", tenant = %tenant_id))
             .await
-            .map_err(ApiError::InternalServerError)?;
+            .map_err(apierror_from_tenant_map_insert_error)?;
     } else {
         return Err(ApiError::BadRequest(anyhow!(
             "attach_tenant is not possible because pageserver was configured without remote storage"
@@ -346,7 +367,9 @@ async fn timeline_delete_handler(request: Request<Body>) -> Result<Response<Body
     let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
     check_permission(&request, Some(tenant_id))?;
 
-    mgr::delete_timeline(tenant_id, timeline_id)
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
+
+    mgr::delete_timeline(tenant_id, timeline_id, &ctx)
         .instrument(info_span!("timeline_delete", tenant = %tenant_id, timeline = %timeline_id))
         .await
         // FIXME: Errors from `delete_timeline` can occur for a number of reasons, incuding both
@@ -377,11 +400,13 @@ async fn tenant_load_handler(request: Request<Body>) -> Result<Response<Body>, A
     let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
     check_permission(&request, Some(tenant_id))?;
 
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
+
     let state = get_state(&request);
-    mgr::load_tenant(state.conf, tenant_id, state.remote_storage.clone())
+    mgr::load_tenant(state.conf, tenant_id, state.remote_storage.clone(), &ctx)
         .instrument(info_span!("load", tenant = %tenant_id))
         .await
-        .map_err(ApiError::InternalServerError)?;
+        .map_err(apierror_from_tenant_map_insert_error)?;
 
     json_response(StatusCode::ACCEPTED, ())
 }
@@ -408,6 +433,8 @@ async fn tenant_list_handler(request: Request<Body>) -> Result<Response<Body>, A
     let response_data = mgr::list_tenants()
         .instrument(info_span!("tenant_list"))
         .await
+        .map_err(anyhow::Error::new)
+        .map_err(ApiError::InternalServerError)?
         .iter()
         .map(|(id, state)| TenantInfo {
             id: *id,
@@ -454,38 +481,50 @@ async fn tenant_status(request: Request<Body>) -> Result<Response<Body>, ApiErro
 /// to debug any of the calculations. Requires `tenant_id` request parameter, supports
 /// `inputs_only=true|false` (default false) which supports debugging failure to calculate model
 /// values.
+///
+/// 'retention_period' query parameter overrides the cutoff that is used to calculate the size
+/// (only if it is shorter than the real cutoff).
+///
+/// Note: we don't update the cached size and prometheus metric here.
+/// The retention period might be different, and it's nice to have a method to just calculate it
+/// without modifying anything anyway.
 async fn tenant_size_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
     check_permission(&request, Some(tenant_id))?;
+    let inputs_only: Option<bool> = parse_query_param(&request, "inputs_only")?;
+    let retention_period: Option<u64> = parse_query_param(&request, "retention_period")?;
+    let headers = request.headers();
 
-    let inputs_only = if query_param_present(&request, "inputs_only") {
-        get_query_param(&request, "inputs_only")?
-            .parse()
-            .map_err(|_| ApiError::BadRequest(anyhow!("failed to parse inputs_only")))?
-    } else {
-        false
-    };
-
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
     let tenant = mgr::get_tenant(tenant_id, true)
         .await
         .map_err(ApiError::InternalServerError)?;
 
     // this can be long operation
     let inputs = tenant
-        .gather_size_inputs()
+        .gather_size_inputs(retention_period, &ctx)
         .await
         .map_err(ApiError::InternalServerError)?;
 
-    let size = if !inputs_only {
-        Some(inputs.calculate().map_err(ApiError::InternalServerError)?)
-    } else {
-        None
-    };
+    let mut sizes = None;
+    if !inputs_only.unwrap_or(false) {
+        let storage_model = inputs
+            .calculate_model()
+            .map_err(ApiError::InternalServerError)?;
+        let size = storage_model.calculate();
 
-    /// Private response type with the additional "unstable" `inputs` field.
-    ///
-    /// The type is described with `id` and `size` in the openapi_spec file, but the `inputs` is
-    /// intentionally left out. The type resides in the pageserver not to expose `ModelInputs`.
+        // If request header expects html, return html
+        if headers["Accept"] == "text/html" {
+            return synthetic_size_html_response(inputs, storage_model, size);
+        }
+        sizes = Some(size);
+    } else if headers["Accept"] == "text/html" {
+        return Err(ApiError::BadRequest(anyhow!(
+            "inputs_only parameter is incompatible with html output request"
+        )));
+    }
+
+    /// The type resides in the pageserver not to expose `ModelInputs`.
     #[serde_with::serde_as]
     #[derive(serde::Serialize)]
     struct TenantHistorySize {
@@ -495,6 +534,9 @@ async fn tenant_size_handler(request: Request<Body>) -> Result<Response<Body>, A
         ///
         /// Will be none if `?inputs_only=true` was given.
         size: Option<u64>,
+        /// Size of each segment used in the model.
+        /// Will be null if `?inputs_only=true` was given.
+        segment_sizes: Option<Vec<tenant_size_model::SegmentSizeResult>>,
         inputs: crate::tenant::size::ModelInputs,
     }
 
@@ -502,12 +544,128 @@ async fn tenant_size_handler(request: Request<Body>) -> Result<Response<Body>, A
         StatusCode::OK,
         TenantHistorySize {
             id: tenant_id,
-            size,
+            size: sizes.as_ref().map(|x| x.total_size),
+            segment_sizes: sizes.map(|x| x.segments),
             inputs,
         },
     )
 }
 
+async fn layer_map_info_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
+    let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
+    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
+    let reset: LayerAccessStatsReset =
+        parse_query_param(&request, "reset")?.unwrap_or(LayerAccessStatsReset::NoReset);
+
+    check_permission(&request, Some(tenant_id))?;
+
+    let timeline = active_timeline_of_active_tenant(tenant_id, timeline_id).await?;
+    let layer_map_info = timeline.layer_map_info(reset);
+
+    json_response(StatusCode::OK, layer_map_info)
+}
+
+async fn layer_download_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
+    let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
+    check_permission(&request, Some(tenant_id))?;
+    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
+    let layer_file_name = get_request_param(&request, "layer_file_name")?;
+    check_permission(&request, Some(tenant_id))?;
+
+    let timeline = active_timeline_of_active_tenant(tenant_id, timeline_id).await?;
+    let downloaded = timeline
+        .download_layer(layer_file_name)
+        .await
+        .map_err(ApiError::InternalServerError)?;
+
+    match downloaded {
+        Some(true) => json_response(StatusCode::OK, ()),
+        Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
+        None => json_response(
+            StatusCode::BAD_REQUEST,
+            format!("Layer {tenant_id}/{timeline_id}/{layer_file_name} not found"),
+        ),
+    }
+}
+
+async fn evict_timeline_layer_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
+    let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
+    check_permission(&request, Some(tenant_id))?;
+    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
+    let layer_file_name = get_request_param(&request, "layer_file_name")?;
+
+    let timeline = active_timeline_of_active_tenant(tenant_id, timeline_id).await?;
+    let evicted = timeline
+        .evict_layer(layer_file_name)
+        .await
+        .map_err(ApiError::InternalServerError)?;
+
+    match evicted {
+        Some(true) => json_response(StatusCode::OK, ()),
+        Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
+        None => json_response(
+            StatusCode::BAD_REQUEST,
+            format!("Layer {tenant_id}/{timeline_id}/{layer_file_name} not found"),
+        ),
+    }
+}
+
+/// Get tenant_size SVG graph along with the JSON data.
+fn synthetic_size_html_response(
+    inputs: ModelInputs,
+    storage_model: StorageModel,
+    sizes: SizeResult,
+) -> Result<Response<Body>, ApiError> {
+    let mut timeline_ids: Vec<String> = Vec::new();
+    let mut timeline_map: HashMap<TimelineId, usize> = HashMap::new();
+    for (index, ti) in inputs.timeline_inputs.iter().enumerate() {
+        timeline_map.insert(ti.timeline_id, index);
+        timeline_ids.push(ti.timeline_id.to_string());
+    }
+    let seg_to_branch: Vec<usize> = inputs
+        .segments
+        .iter()
+        .map(|seg| *timeline_map.get(&seg.timeline_id).unwrap())
+        .collect();
+
+    let svg =
+        tenant_size_model::svg::draw_svg(&storage_model, &timeline_ids, &seg_to_branch, &sizes)
+            .map_err(ApiError::InternalServerError)?;
+
+    let mut response = String::new();
+
+    use std::fmt::Write;
+    write!(response, "<html>\n<body>\n").unwrap();
+    write!(response, "<div>\n{svg}\n</div>").unwrap();
+    writeln!(response, "Project size: {}", sizes.total_size).unwrap();
+    writeln!(response, "<pre>").unwrap();
+    writeln!(
+        response,
+        "{}",
+        serde_json::to_string_pretty(&inputs).unwrap()
+    )
+    .unwrap();
+    writeln!(
+        response,
+        "{}",
+        serde_json::to_string_pretty(&sizes.segments).unwrap()
+    )
+    .unwrap();
+    writeln!(response, "</pre>").unwrap();
+    write!(response, "</body>\n</html>\n").unwrap();
+
+    html_response(StatusCode::OK, response)
+}
+
+pub fn html_response(status: StatusCode, data: String) -> Result<Response<Body>, ApiError> {
+    let response = Response::builder()
+        .status(status)
+        .header(hyper::header::CONTENT_TYPE, "text/html")
+        .body(Body::from(data.as_bytes().to_vec()))
+        .map_err(|e| ApiError::InternalServerError(e.into()))?;
+    Ok(response)
+}
+
 // Helper function to standardize the error messages we produce on bad durations
 //
 // Intended to be used with anyhow's `with_context`, e.g.:
@@ -521,6 +679,8 @@ fn bad_duration<'a>(field_name: &'static str, value: &'a str) -> impl 'a + Fn()
 async fn tenant_create_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
     check_permission(&request, None)?;
 
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
+
     let request_data: TenantCreateRequest = json_request(&mut request).await?;
 
     let mut tenant_conf = TenantConfOpt::default();
@@ -598,42 +758,64 @@ async fn tenant_create_handler(mut request: Request<Body>) -> Result<Response<Bo
         tenant_conf,
         target_tenant_id,
         state.remote_storage.clone(),
+        &ctx,
     )
     .instrument(info_span!("tenant_create", tenant = ?target_tenant_id))
     .await
-    // FIXME: `create_tenant` can fail from both user and internal errors. Replace this
-    // with better error handling once the type permits it
-    .map_err(ApiError::InternalServerError)?;
+    .map_err(apierror_from_tenant_map_insert_error)?;
 
-    Ok(match new_tenant {
-        Some(tenant) => {
-            // We created the tenant. Existing API semantics are that the tenant
-            // is Active when this function returns.
-            if let res @ Err(_) = tenant.wait_to_become_active().await {
-                // This shouldn't happen because we just created the tenant directory
-                // in tenant::mgr::create_tenant, and there aren't any remote timelines
-                // to load, so, nothing can really fail during load.
-                // Don't do cleanup because we don't know how we got here.
-                // The tenant will likely be in `Broken` state and subsequent
-                // calls will fail.
-                res.context("created tenant failed to become active")
-                    .map_err(ApiError::InternalServerError)?;
-            }
-            json_response(
-                StatusCode::CREATED,
-                TenantCreateResponse(tenant.tenant_id()),
-            )?
-        }
-        None => json_response(StatusCode::CONFLICT, ())?,
-    })
+    // We created the tenant. Existing API semantics are that the tenant
+    // is Active when this function returns.
+    if let res @ Err(_) = new_tenant.wait_to_become_active().await {
+        // This shouldn't happen because we just created the tenant directory
+        // in tenant::mgr::create_tenant, and there aren't any remote timelines
+        // to load, so, nothing can really fail during load.
+        // Don't do cleanup because we don't know how we got here.
+        // The tenant will likely be in `Broken` state and subsequent
+        // calls will fail.
+        res.context("created tenant failed to become active")
+            .map_err(ApiError::InternalServerError)?;
+    }
+    json_response(
+        StatusCode::CREATED,
+        TenantCreateResponse(new_tenant.tenant_id()),
+    )
 }
 
-async fn tenant_config_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
+async fn get_tenant_config_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
+    let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
+    check_permission(&request, Some(tenant_id))?;
+
+    let tenant = mgr::get_tenant(tenant_id, false)
+        .await
+        .map_err(ApiError::NotFound)?;
+
+    let response = HashMap::from([
+        (
+            "tenant_specific_overrides",
+            serde_json::to_value(tenant.tenant_specific_overrides())
+                .context("serializing tenant specific overrides")
+                .map_err(ApiError::InternalServerError)?,
+        ),
+        (
+            "effective_config",
+            serde_json::to_value(tenant.effective_config())
+                .context("serializing effective config")
+                .map_err(ApiError::InternalServerError)?,
+        ),
+    ]);
+
+    json_response(StatusCode::OK, response)
+}
+
+async fn update_tenant_config_handler(
+    mut request: Request<Body>,
+) -> Result<Response<Body>, ApiError> {
     let request_data: TenantConfigRequest = json_request(&mut request).await?;
     let tenant_id = request_data.tenant_id;
     check_permission(&request, Some(tenant_id))?;
 
-    let mut tenant_conf: TenantConfOpt = Default::default();
+    let mut tenant_conf = TenantConfOpt::default();
     if let Some(gc_period) = request_data.gc_period {
         tenant_conf.gc_period = Some(
             humantime::parse_duration(&gc_period)
@@ -668,12 +850,8 @@ async fn tenant_config_handler(mut request: Request<Body>) -> Result<Response<Bo
                 .map_err(ApiError::BadRequest)?,
         );
     }
-    if let Some(max_lsn_wal_lag) = request_data.max_lsn_wal_lag {
-        tenant_conf.max_lsn_wal_lag = Some(max_lsn_wal_lag);
-    }
-    if let Some(trace_read_requests) = request_data.trace_read_requests {
-        tenant_conf.trace_read_requests = Some(trace_read_requests);
-    }
+    tenant_conf.max_lsn_wal_lag = request_data.max_lsn_wal_lag;
+    tenant_conf.trace_read_requests = request_data.trace_read_requests;
 
     tenant_conf.checkpoint_distance = request_data.checkpoint_distance;
     if let Some(checkpoint_timeout) = request_data.checkpoint_timeout {
@@ -694,8 +872,16 @@ async fn tenant_config_handler(mut request: Request<Body>) -> Result<Response<Bo
         );
     }
 
+    if let Some(eviction_policy) = request_data.eviction_policy {
+        tenant_conf.eviction_policy = Some(
+            serde_json::from_value(eviction_policy)
+                .context("parse field `eviction_policy`")
+                .map_err(ApiError::BadRequest)?,
+        );
+    }
+
     let state = get_state(&request);
-    mgr::update_tenant_config(state.conf, tenant_conf, tenant_id)
+    mgr::set_new_tenant_config(state.conf, tenant_conf, tenant_id)
         .instrument(info_span!("tenant_config", tenant = ?tenant_id))
         .await
         // FIXME: `update_tenant_config` can fail because of both user and internal errors.
@@ -739,7 +925,6 @@ async fn failpoints_handler(mut request: Request<Body>) -> Result<Response<Body>
 }
 
 // Run GC immediately on given timeline.
-#[cfg(feature = "testing")]
 async fn timeline_gc_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
     let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
@@ -747,7 +932,8 @@ async fn timeline_gc_handler(mut request: Request<Body>) -> Result<Response<Body
 
     let gc_req: TimelineGcRequest = json_request(&mut request).await?;
 
-    let wait_task_done = mgr::immediate_gc(tenant_id, timeline_id, gc_req).await?;
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let wait_task_done = mgr::immediate_gc(tenant_id, timeline_id, gc_req, &ctx).await?;
     let gc_result = wait_task_done
         .await
         .context("wait for gc task")
@@ -764,7 +950,8 @@ async fn timeline_compact_handler(request: Request<Body>) -> Result<Response<Bod
     let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
     check_permission(&request, Some(tenant_id))?;
 
-    let result_receiver = mgr::immediate_compact(tenant_id, timeline_id)
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let result_receiver = mgr::immediate_compact(tenant_id, timeline_id, &ctx)
         .await
         .context("spawn compaction task")
         .map_err(ApiError::InternalServerError)?;
@@ -785,18 +972,14 @@ async fn timeline_checkpoint_handler(request: Request<Body>) -> Result<Response<
     let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
     check_permission(&request, Some(tenant_id))?;
 
-    let tenant = mgr::get_tenant(tenant_id, true)
-        .await
-        .map_err(ApiError::NotFound)?;
-    let timeline = tenant
-        .get_timeline(timeline_id, true)
-        .map_err(ApiError::NotFound)?;
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let timeline = active_timeline_of_active_tenant(tenant_id, timeline_id).await?;
     timeline
         .freeze_and_flush()
         .await
         .map_err(ApiError::InternalServerError)?;
     timeline
-        .compact()
+        .compact(&ctx)
         .await
         .map_err(ApiError::InternalServerError)?;
 
@@ -811,12 +994,7 @@ async fn timeline_download_remote_layers_handler_post(
     let body: DownloadRemoteLayersTaskSpawnRequest = json_request(&mut request).await?;
     check_permission(&request, Some(tenant_id))?;
 
-    let tenant = mgr::get_tenant(tenant_id, true)
-        .await
-        .map_err(ApiError::NotFound)?;
-    let timeline = tenant
-        .get_timeline(timeline_id, true)
-        .map_err(ApiError::NotFound)?;
+    let timeline = active_timeline_of_active_tenant(tenant_id, timeline_id).await?;
     match timeline.spawn_download_all_remote_layers(body).await {
         Ok(st) => json_response(StatusCode::ACCEPTED, st),
         Err(st) => json_response(StatusCode::CONFLICT, st),
@@ -827,15 +1005,10 @@ async fn timeline_download_remote_layers_handler_get(
     request: Request<Body>,
 ) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
-    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
     check_permission(&request, Some(tenant_id))?;
+    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
 
-    let tenant = mgr::get_tenant(tenant_id, true)
-        .await
-        .map_err(ApiError::NotFound)?;
-    let timeline = tenant
-        .get_timeline(timeline_id, true)
-        .map_err(ApiError::NotFound)?;
+    let timeline = active_timeline_of_active_tenant(tenant_id, timeline_id).await?;
     let info = timeline
         .get_download_all_remote_layers_task_info()
         .context("task never started since last pageserver process start")
@@ -843,6 +1016,29 @@ async fn timeline_download_remote_layers_handler_get(
     json_response(StatusCode::OK, info)
 }
 
+async fn active_timeline_of_active_tenant(
+    tenant_id: TenantId,
+    timeline_id: TimelineId,
+) -> Result<Arc<Timeline>, ApiError> {
+    let tenant = mgr::get_tenant(tenant_id, true)
+        .await
+        .map_err(ApiError::NotFound)?;
+    tenant
+        .get_timeline(timeline_id, true)
+        .map_err(ApiError::NotFound)
+}
+
+async fn always_panic_handler(req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    // Deliberately cause a panic to exercise the panic hook registered via std::panic::set_hook().
+    // For pageserver, the relevant panic hook is `tracing_panic_hook` , and the `sentry` crate's wrapper around it.
+    // Use catch_unwind to ensure that tokio nor hyper are distracted by our panic.
+    let query = req.uri().query();
+    let _ = std::panic::catch_unwind(|| {
+        panic!("unconditional panic for testing panic hook integration; request query: {query:?}")
+    });
+    json_response(StatusCode::NO_CONTENT, ())
+}
+
 async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
     json_response(
         StatusCode::NOT_FOUND,
@@ -852,6 +1048,7 @@ async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
 
 pub fn make_router(
     conf: &'static PageServerConf,
+    launch_ts: &'static LaunchTimestamp,
     auth: Option<Arc<JwtAuth>>,
     remote_storage: Option<GenericRemoteStorage>,
 ) -> anyhow::Result<RouterBuilder<hyper::Body, ApiError>> {
@@ -868,6 +1065,14 @@ pub fn make_router(
         }))
     }
 
+    router = router.middleware(
+        endpoint::add_response_header_middleware(
+            "PAGESERVER_LAUNCH_TIMESTAMP",
+            &launch_ts.to_string(),
+        )
+        .expect("construct launch timestamp header middleware"),
+    );
+
     macro_rules! testing_api {
         ($handler_desc:literal, $handler:path $(,)?) => {{
             #[cfg(not(feature = "testing"))]
@@ -899,8 +1104,9 @@ pub fn make_router(
         .get("/v1/tenant", tenant_list_handler)
         .post("/v1/tenant", tenant_create_handler)
         .get("/v1/tenant/:tenant_id", tenant_status)
-        .get("/v1/tenant/:tenant_id/size", tenant_size_handler)
-        .put("/v1/tenant/config", tenant_config_handler)
+        .get("/v1/tenant/:tenant_id/synthetic_size", tenant_size_handler)
+        .put("/v1/tenant/config", update_tenant_config_handler)
+        .get("/v1/tenant/:tenant_id/config", get_tenant_config_handler)
         .get("/v1/tenant/:tenant_id/timeline", timeline_list_handler)
         .post("/v1/tenant/:tenant_id/timeline", timeline_create_handler)
         .post("/v1/tenant/:tenant_id/attach", tenant_attach_handler)
@@ -917,7 +1123,7 @@ pub fn make_router(
         )
         .put(
             "/v1/tenant/:tenant_id/timeline/:timeline_id/do_gc",
-            testing_api!("run timeline GC", timeline_gc_handler),
+            timeline_gc_handler,
         )
         .put(
             "/v1/tenant/:tenant_id/timeline/:timeline_id/compact",
@@ -939,5 +1145,18 @@ pub fn make_router(
             "/v1/tenant/:tenant_id/timeline/:timeline_id",
             timeline_delete_handler,
         )
+        .get(
+            "/v1/tenant/:tenant_id/timeline/:timeline_id/layer",
+            layer_map_info_handler,
+        )
+        .get(
+            "/v1/tenant/:tenant_id/timeline/:timeline_id/layer/:layer_file_name",
+            layer_download_handler,
+        )
+        .delete(
+            "/v1/tenant/:tenant_id/timeline/:timeline_id/layer/:layer_file_name",
+            evict_timeline_layer_handler,
+        )
+        .get("/v1/panic", always_panic_handler)
         .any(handler_404))
 }

pageserver/src/import_datadir.rs

@@ -12,6 +12,7 @@ use tokio_tar::Archive;
 use tracing::*;
 use walkdir::WalkDir;
 
+use crate::context::RequestContext;
 use crate::pgdatadir_mapping::*;
 use crate::tenant::Timeline;
 use crate::walingest::WalIngest;
@@ -47,6 +48,7 @@ pub async fn import_timeline_from_postgres_datadir(
     tline: &Timeline,
     pgdata_path: &Path,
     pgdata_lsn: Lsn,
+    ctx: &RequestContext,
 ) -> Result<()> {
     let mut pg_control: Option<ControlFileData> = None;
 
@@ -69,7 +71,7 @@ pub async fn import_timeline_from_postgres_datadir(
             let mut file = tokio::fs::File::open(absolute_path).await?;
             let len = metadata.len() as usize;
             if let Some(control_file) =
-                import_file(&mut modification, relative_path, &mut file, len).await?
+                import_file(&mut modification, relative_path, &mut file, len, ctx).await?
             {
                 pg_control = Some(control_file);
             }
@@ -99,6 +101,7 @@ pub async fn import_timeline_from_postgres_datadir(
         tline,
         Lsn(pg_control.checkPointCopy.redo),
         pgdata_lsn,
+        ctx,
     )
     .await?;
 
@@ -113,6 +116,7 @@ async fn import_rel(
     dboid: Oid,
     reader: &mut (impl AsyncRead + Send + Sync + Unpin),
     len: usize,
+    ctx: &RequestContext,
 ) -> anyhow::Result<()> {
     // Does it look like a relation file?
     trace!("importing rel file {}", path.display());
@@ -147,7 +151,10 @@ async fn import_rel(
     // FIXME: use proper error type for this, instead of parsing the error message.
     // Or better yet, keep track of which relations we've already created
     // https://github.com/neondatabase/neon/issues/3309
-    if let Err(e) = modification.put_rel_creation(rel, nblocks as u32).await {
+    if let Err(e) = modification
+        .put_rel_creation(rel, nblocks as u32, ctx)
+        .await
+    {
         if e.to_string().contains("already exists") {
             debug!("relation {} already exists. we must be extending it", rel);
         } else {
@@ -182,7 +189,7 @@ async fn import_rel(
     //
     // If we process rel segments out of order,
     // put_rel_extend will skip the update.
-    modification.put_rel_extend(rel, blknum).await?;
+    modification.put_rel_extend(rel, blknum, ctx).await?;
 
     Ok(())
 }
@@ -195,6 +202,7 @@ async fn import_slru(
     path: &Path,
     reader: &mut (impl AsyncRead + Send + Sync + Unpin),
     len: usize,
+    ctx: &RequestContext,
 ) -> anyhow::Result<()> {
     info!("importing slru file {path:?}");
 
@@ -211,7 +219,7 @@ async fn import_slru(
     ensure!(nblocks <= pg_constants::SLRU_PAGES_PER_SEGMENT as usize);
 
     modification
-        .put_slru_segment_creation(slru, segno, nblocks as u32)
+        .put_slru_segment_creation(slru, segno, nblocks as u32, ctx)
         .await?;
 
     let mut rpageno = 0;
@@ -252,15 +260,15 @@ async fn import_wal(
     tline: &Timeline,
     startpoint: Lsn,
     endpoint: Lsn,
+    ctx: &RequestContext,
 ) -> anyhow::Result<()> {
-    use std::io::Read;
     let mut waldecoder = WalStreamDecoder::new(startpoint, tline.pg_version);
 
     let mut segno = startpoint.segment_number(WAL_SEGMENT_SIZE);
     let mut offset = startpoint.segment_offset(WAL_SEGMENT_SIZE);
     let mut last_lsn = startpoint;
 
-    let mut walingest = WalIngest::new(tline, startpoint).await?;
+    let mut walingest = WalIngest::new(tline, startpoint, ctx).await?;
 
     while last_lsn <= endpoint {
         // FIXME: assume postgresql tli 1 for now
@@ -283,6 +291,7 @@ async fn import_wal(
             file.seek(std::io::SeekFrom::Start(offset as u64))?;
         }
 
+        use std::io::Read;
         let nread = file.read_to_end(&mut buf)?;
         if nread != WAL_SEGMENT_SIZE - offset {
             // Maybe allow this for .partial files?
@@ -297,7 +306,7 @@ async fn import_wal(
         while last_lsn <= endpoint {
             if let Some((lsn, recdata)) = waldecoder.poll_decode()? {
                 walingest
-                    .ingest_record(recdata, lsn, &mut modification, &mut decoded)
+                    .ingest_record(recdata, lsn, &mut modification, &mut decoded, ctx)
                     .await?;
                 last_lsn = lsn;
 
@@ -326,6 +335,7 @@ pub async fn import_basebackup_from_tar(
     tline: &Timeline,
     reader: &mut (impl AsyncRead + Send + Sync + Unpin),
     base_lsn: Lsn,
+    ctx: &RequestContext,
 ) -> Result<()> {
     info!("importing base at {base_lsn}");
     let mut modification = tline.begin_modification(base_lsn);
@@ -344,7 +354,7 @@ pub async fn import_basebackup_from_tar(
         match header.entry_type() {
             tokio_tar::EntryType::Regular => {
                 if let Some(res) =
-                    import_file(&mut modification, file_path.as_ref(), &mut entry, len).await?
+                    import_file(&mut modification, file_path.as_ref(), &mut entry, len, ctx).await?
                 {
                     // We found the pg_control file.
                     pg_control = Some(res);
@@ -376,13 +386,14 @@ pub async fn import_wal_from_tar(
     reader: &mut (impl AsyncRead + Send + Sync + Unpin),
     start_lsn: Lsn,
     end_lsn: Lsn,
+    ctx: &RequestContext,
 ) -> Result<()> {
     // Set up walingest mutable state
     let mut waldecoder = WalStreamDecoder::new(start_lsn, tline.pg_version);
     let mut segno = start_lsn.segment_number(WAL_SEGMENT_SIZE);
     let mut offset = start_lsn.segment_offset(WAL_SEGMENT_SIZE);
     let mut last_lsn = start_lsn;
-    let mut walingest = WalIngest::new(tline, start_lsn).await?;
+    let mut walingest = WalIngest::new(tline, start_lsn, ctx).await?;
 
     // Ingest wal until end_lsn
     info!("importing wal until {}", end_lsn);
@@ -431,7 +442,7 @@ pub async fn import_wal_from_tar(
         while last_lsn <= end_lsn {
             if let Some((lsn, recdata)) = waldecoder.poll_decode()? {
                 walingest
-                    .ingest_record(recdata, lsn, &mut modification, &mut decoded)
+                    .ingest_record(recdata, lsn, &mut modification, &mut decoded, ctx)
                     .await?;
                 last_lsn = lsn;
 
@@ -466,6 +477,7 @@ async fn import_file(
     file_path: &Path,
     reader: &mut (impl AsyncRead + Send + Sync + Unpin),
     len: usize,
+    ctx: &RequestContext,
 ) -> Result<Option<ControlFileData>> {
     let file_name = match file_path.file_name() {
         Some(name) => name.to_string_lossy(),
@@ -498,14 +510,16 @@ async fn import_file(
             }
             "pg_filenode.map" => {
                 let bytes = read_all_bytes(reader).await?;
-                modification.put_relmap_file(spcnode, dbnode, bytes).await?;
+                modification
+                    .put_relmap_file(spcnode, dbnode, bytes, ctx)
+                    .await?;
                 debug!("imported relmap file")
             }
             "PG_VERSION" => {
                 debug!("ignored PG_VERSION file");
             }
             _ => {
-                import_rel(modification, file_path, spcnode, dbnode, reader, len).await?;
+                import_rel(modification, file_path, spcnode, dbnode, reader, len, ctx).await?;
                 debug!("imported rel creation");
             }
         }
@@ -521,38 +535,40 @@ async fn import_file(
         match file_name.as_ref() {
             "pg_filenode.map" => {
                 let bytes = read_all_bytes(reader).await?;
-                modification.put_relmap_file(spcnode, dbnode, bytes).await?;
+                modification
+                    .put_relmap_file(spcnode, dbnode, bytes, ctx)
+                    .await?;
                 debug!("imported relmap file")
             }
             "PG_VERSION" => {
                 debug!("ignored PG_VERSION file");
             }
             _ => {
-                import_rel(modification, file_path, spcnode, dbnode, reader, len).await?;
+                import_rel(modification, file_path, spcnode, dbnode, reader, len, ctx).await?;
                 debug!("imported rel creation");
             }
         }
     } else if file_path.starts_with("pg_xact") {
         let slru = SlruKind::Clog;
 
-        import_slru(modification, slru, file_path, reader, len).await?;
+        import_slru(modification, slru, file_path, reader, len, ctx).await?;
         debug!("imported clog slru");
     } else if file_path.starts_with("pg_multixact/offsets") {
         let slru = SlruKind::MultiXactOffsets;
 
-        import_slru(modification, slru, file_path, reader, len).await?;
+        import_slru(modification, slru, file_path, reader, len, ctx).await?;
         debug!("imported multixact offsets slru");
     } else if file_path.starts_with("pg_multixact/members") {
         let slru = SlruKind::MultiXactMembers;
 
-        import_slru(modification, slru, file_path, reader, len).await?;
+        import_slru(modification, slru, file_path, reader, len, ctx).await?;
         debug!("imported multixact members slru");
     } else if file_path.starts_with("pg_twophase") {
         let xid = u32::from_str_radix(file_name.as_ref(), 16)?;
 
         let bytes = read_all_bytes(reader).await?;
         modification
-            .put_twophase_file(xid, Bytes::copy_from_slice(&bytes[..]))
+            .put_twophase_file(xid, Bytes::copy_from_slice(&bytes[..]), ctx)
             .await?;
         debug!("imported twophase file");
     } else if file_path.starts_with("pg_wal") {

pageserver/src/lib.rs

@@ -1,7 +1,9 @@
 mod auth;
 pub mod basebackup;
+pub mod broker_client;
 pub mod config;
 pub mod consumption_metrics;
+pub mod context;
 pub mod http;
 pub mod import_datadir;
 pub mod keyspace;
@@ -15,7 +17,6 @@ pub mod tenant;
 pub mod trace;
 pub mod virtual_file;
 pub mod walingest;
-pub mod walreceiver;
 pub mod walrecord;
 pub mod walredo;
 

pageserver/src/metrics.rs

@@ -1,10 +1,12 @@
 use metrics::core::{AtomicU64, GenericCounter};
 use metrics::{
-    register_histogram, register_histogram_vec, register_int_counter, register_int_counter_vec,
-    register_int_gauge, register_int_gauge_vec, register_uint_gauge_vec, Histogram, HistogramVec,
-    IntCounter, IntCounterVec, IntGauge, IntGaugeVec, UIntGauge, UIntGaugeVec,
+    register_counter_vec, register_histogram, register_histogram_vec, register_int_counter,
+    register_int_counter_vec, register_int_gauge, register_int_gauge_vec, register_uint_gauge_vec,
+    Counter, CounterVec, Histogram, HistogramVec, IntCounter, IntCounterVec, IntGauge, IntGaugeVec,
+    UIntGauge, UIntGaugeVec,
 };
 use once_cell::sync::Lazy;
+use pageserver_api::models::state;
 use utils::id::{TenantId, TimelineId};
 
 /// Prometheus histogram buckets (in seconds) that capture the majority of
@@ -35,11 +37,29 @@ const STORAGE_TIME_OPERATIONS: &[&str] = &[
     "gc",
 ];
 
-pub static STORAGE_TIME: Lazy<HistogramVec> = Lazy::new(|| {
-    register_histogram_vec!(
-        "pageserver_storage_operations_seconds",
-        "Time spent on storage operations",
+pub static STORAGE_TIME_SUM_PER_TIMELINE: Lazy<CounterVec> = Lazy::new(|| {
+    register_counter_vec!(
+        "pageserver_storage_operations_seconds_sum",
+        "Total time spent on storage operations with operation, tenant and timeline dimensions",
         &["operation", "tenant_id", "timeline_id"],
+    )
+    .expect("failed to define a metric")
+});
+
+pub static STORAGE_TIME_COUNT_PER_TIMELINE: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "pageserver_storage_operations_seconds_count",
+        "Count of storage operations with operation, tenant and timeline dimensions",
+        &["operation", "tenant_id", "timeline_id"],
+    )
+    .expect("failed to define a metric")
+});
+
+pub static STORAGE_TIME_GLOBAL: Lazy<HistogramVec> = Lazy::new(|| {
+    register_histogram_vec!(
+        "pageserver_storage_operations_seconds_global",
+        "Time spent on storage operations",
+        &["operation"],
         get_buckets_for_critical_operations(),
     )
     .expect("failed to define a metric")
@@ -112,6 +132,33 @@ static CURRENT_LOGICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
     .expect("failed to define current logical size metric")
 });
 
+// Metrics collected on tenant states.
+const TENANT_STATE_OPTIONS: &[&str] = &[
+    state::LOADING,
+    state::ATTACHING,
+    state::ACTIVE,
+    state::STOPPING,
+    state::BROKEN,
+];
+
+pub static TENANT_STATE_METRIC: Lazy<UIntGaugeVec> = Lazy::new(|| {
+    register_uint_gauge_vec!(
+        "pageserver_tenant_states_count",
+        "Count of tenants per state",
+        &["tenant_id", "state"]
+    )
+    .expect("Failed to register pageserver_tenant_states_count metric")
+});
+
+pub static TENANT_SYNTHETIC_SIZE_METRIC: Lazy<UIntGaugeVec> = Lazy::new(|| {
+    register_uint_gauge_vec!(
+        "pageserver_tenant_synthetic_size",
+        "Synthetic size of each tenant",
+        &["tenant_id"]
+    )
+    .expect("Failed to register pageserver_tenant_synthetic_size metric")
+});
+
 // Metrics for cloud upload. These metrics reflect data uploaded to cloud storage,
 // or in testing they estimate how much we would upload if we did.
 static NUM_PERSISTENT_FILES_CREATED: Lazy<IntCounterVec> = Lazy::new(|| {
@@ -375,18 +422,81 @@ pub static WAL_REDO_RECORD_COUNTER: Lazy<IntCounter> = Lazy::new(|| {
     .unwrap()
 });
 
+/// Similar to [`prometheus::HistogramTimer`] but does not record on drop.
+pub struct StorageTimeMetricsTimer {
+    metrics: StorageTimeMetrics,
+    start: Instant,
+}
+
+impl StorageTimeMetricsTimer {
+    fn new(metrics: StorageTimeMetrics) -> Self {
+        Self {
+            metrics,
+            start: Instant::now(),
+        }
+    }
+
+    /// Record the time from creation to now.
+    pub fn stop_and_record(self) {
+        let duration = self.start.elapsed().as_secs_f64();
+        self.metrics.timeline_sum.inc_by(duration);
+        self.metrics.timeline_count.inc();
+        self.metrics.global_histogram.observe(duration);
+    }
+}
+
+/// Timing facilities for an globally histogrammed metric, which is supported by per tenant and
+/// timeline total sum and count.
+#[derive(Clone, Debug)]
+pub struct StorageTimeMetrics {
+    /// Sum of f64 seconds, per operation, tenant_id and timeline_id
+    timeline_sum: Counter,
+    /// Number of oeprations, per operation, tenant_id and timeline_id
+    timeline_count: IntCounter,
+    /// Global histogram having only the "operation" label.
+    global_histogram: Histogram,
+}
+
+impl StorageTimeMetrics {
+    pub fn new(operation: &str, tenant_id: &str, timeline_id: &str) -> Self {
+        let timeline_sum = STORAGE_TIME_SUM_PER_TIMELINE
+            .get_metric_with_label_values(&[operation, tenant_id, timeline_id])
+            .unwrap();
+        let timeline_count = STORAGE_TIME_COUNT_PER_TIMELINE
+            .get_metric_with_label_values(&[operation, tenant_id, timeline_id])
+            .unwrap();
+        let global_histogram = STORAGE_TIME_GLOBAL
+            .get_metric_with_label_values(&[operation])
+            .unwrap();
+
+        StorageTimeMetrics {
+            timeline_sum,
+            timeline_count,
+            global_histogram,
+        }
+    }
+
+    /// Starts timing a new operation.
+    ///
+    /// Note: unlike [`prometheus::HistogramTimer`] the returned timer does not record on drop.
+    pub fn start_timer(&self) -> StorageTimeMetricsTimer {
+        StorageTimeMetricsTimer::new(self.clone())
+    }
+}
+
 #[derive(Debug)]
 pub struct TimelineMetrics {
     tenant_id: String,
     timeline_id: String,
     pub reconstruct_time_histo: Histogram,
     pub materialized_page_cache_hit_counter: GenericCounter<AtomicU64>,
-    pub flush_time_histo: Histogram,
-    pub compact_time_histo: Histogram,
-    pub create_images_time_histo: Histogram,
-    pub init_logical_size_histo: Histogram,
-    pub logical_size_histo: Histogram,
-    pub load_layer_map_histo: Histogram,
+    pub flush_time_histo: StorageTimeMetrics,
+    pub compact_time_histo: StorageTimeMetrics,
+    pub create_images_time_histo: StorageTimeMetrics,
+    pub init_logical_size_histo: StorageTimeMetrics,
+    pub logical_size_histo: StorageTimeMetrics,
+    pub load_layer_map_histo: StorageTimeMetrics,
+    pub garbage_collect_histo: StorageTimeMetrics,
     pub last_record_gauge: IntGauge,
     pub wait_lsn_time_histo: Histogram,
     pub resident_physical_size_gauge: UIntGauge,
@@ -406,24 +516,16 @@ impl TimelineMetrics {
         let materialized_page_cache_hit_counter = MATERIALIZED_PAGE_CACHE_HIT
             .get_metric_with_label_values(&[&tenant_id, &timeline_id])
             .unwrap();
-        let flush_time_histo = STORAGE_TIME
-            .get_metric_with_label_values(&["layer flush", &tenant_id, &timeline_id])
-            .unwrap();
-        let compact_time_histo = STORAGE_TIME
-            .get_metric_with_label_values(&["compact", &tenant_id, &timeline_id])
-            .unwrap();
-        let create_images_time_histo = STORAGE_TIME
-            .get_metric_with_label_values(&["create images", &tenant_id, &timeline_id])
-            .unwrap();
-        let init_logical_size_histo = STORAGE_TIME
-            .get_metric_with_label_values(&["init logical size", &tenant_id, &timeline_id])
-            .unwrap();
-        let logical_size_histo = STORAGE_TIME
-            .get_metric_with_label_values(&["logical size", &tenant_id, &timeline_id])
-            .unwrap();
-        let load_layer_map_histo = STORAGE_TIME
-            .get_metric_with_label_values(&["load layer map", &tenant_id, &timeline_id])
-            .unwrap();
+        let flush_time_histo = StorageTimeMetrics::new("layer flush", &tenant_id, &timeline_id);
+        let compact_time_histo = StorageTimeMetrics::new("compact", &tenant_id, &timeline_id);
+        let create_images_time_histo =
+            StorageTimeMetrics::new("create images", &tenant_id, &timeline_id);
+        let init_logical_size_histo =
+            StorageTimeMetrics::new("init logical size", &tenant_id, &timeline_id);
+        let logical_size_histo = StorageTimeMetrics::new("logical size", &tenant_id, &timeline_id);
+        let load_layer_map_histo =
+            StorageTimeMetrics::new("load layer map", &tenant_id, &timeline_id);
+        let garbage_collect_histo = StorageTimeMetrics::new("gc", &tenant_id, &timeline_id);
         let last_record_gauge = LAST_RECORD_LSN
             .get_metric_with_label_values(&[&tenant_id, &timeline_id])
             .unwrap();
@@ -453,6 +555,7 @@ impl TimelineMetrics {
             create_images_time_histo,
             init_logical_size_histo,
             logical_size_histo,
+            garbage_collect_histo,
             load_layer_map_histo,
             last_record_gauge,
             wait_lsn_time_histo,
@@ -478,7 +581,10 @@ impl Drop for TimelineMetrics {
         let _ = PERSISTENT_BYTES_WRITTEN.remove_label_values(&[tenant_id, timeline_id]);
 
         for op in STORAGE_TIME_OPERATIONS {
-            let _ = STORAGE_TIME.remove_label_values(&[op, tenant_id, timeline_id]);
+            let _ =
+                STORAGE_TIME_SUM_PER_TIMELINE.remove_label_values(&[op, tenant_id, timeline_id]);
+            let _ =
+                STORAGE_TIME_COUNT_PER_TIMELINE.remove_label_values(&[op, tenant_id, timeline_id]);
         }
         for op in STORAGE_IO_TIME_OPERATIONS {
             let _ = STORAGE_IO_TIME.remove_label_values(&[op, tenant_id, timeline_id]);
@@ -495,7 +601,11 @@ impl Drop for TimelineMetrics {
 }
 
 pub fn remove_tenant_metrics(tenant_id: &TenantId) {
-    let _ = STORAGE_TIME.remove_label_values(&["gc", &tenant_id.to_string(), "-"]);
+    let tid = tenant_id.to_string();
+    let _ = TENANT_SYNTHETIC_SIZE_METRIC.remove_label_values(&[&tid]);
+    for state in TENANT_STATE_OPTIONS {
+        let _ = TENANT_STATE_METRIC.remove_label_values(&[&tid, state]);
+    }
 }
 
 use futures::Future;

pageserver/src/page_service.rs

@@ -13,6 +13,7 @@ use anyhow::Context;
 use bytes::Buf;
 use bytes::Bytes;
 use futures::{Stream, StreamExt};
+use pageserver_api::models::TenantState;
 use pageserver_api::models::{
     PagestreamBeMessage, PagestreamDbSizeRequest, PagestreamDbSizeResponse,
     PagestreamErrorResponse, PagestreamExistsRequest, PagestreamExistsResponse,
@@ -30,19 +31,19 @@ use std::sync::Arc;
 use std::time::Duration;
 use tracing::*;
 use utils::id::ConnectionId;
-use utils::postgres_backend_async::QueryError;
 use utils::{
     auth::{Claims, JwtAuth, Scope},
     id::{TenantId, TimelineId},
     lsn::Lsn,
     postgres_backend::AuthType,
-    postgres_backend_async::{self, PostgresBackend},
+    postgres_backend_async::{self, is_expected_io_error, PostgresBackend, QueryError},
     simple_rcu::RcuReadGuard,
 };
 
 use crate::auth::check_permission;
 use crate::basebackup;
 use crate::config::PageServerConf;
+use crate::context::{DownloadBehavior, RequestContext};
 use crate::import_datadir::import_wal_from_tar;
 use crate::metrics::{LIVE_CONNECTIONS_COUNT, SMGR_QUERY_TIME};
 use crate::task_mgr;
@@ -123,6 +124,7 @@ pub async fn libpq_listener_main(
     auth: Option<Arc<JwtAuth>>,
     listener: TcpListener,
     auth_type: AuthType,
+    listener_ctx: RequestContext,
 ) -> anyhow::Result<()> {
     listener.set_nonblocking(true)?;
     let tokio_listener = tokio::net::TcpListener::from_std(listener)?;
@@ -146,6 +148,9 @@ pub async fn libpq_listener_main(
                 debug!("accepted connection from {}", peer_addr);
                 let local_auth = auth.clone();
 
+                let connection_ctx = listener_ctx
+                    .detached_child(TaskKind::PageRequestHandler, DownloadBehavior::Download);
+
                 // PageRequestHandler tasks are not associated with any particular
                 // timeline in the task manager. In practice most connections will
                 // only deal with a particular timeline, but we don't know which one
@@ -157,7 +162,7 @@ pub async fn libpq_listener_main(
                     None,
                     "serving compute connection task",
                     false,
-                    page_service_conn_main(conf, local_auth, socket, auth_type),
+                    page_service_conn_main(conf, local_auth, socket, auth_type, connection_ctx),
                 );
             }
             Err(err) => {
@@ -177,6 +182,7 @@ async fn page_service_conn_main(
     auth: Option<Arc<JwtAuth>>,
     socket: tokio::net::TcpStream,
     auth_type: AuthType,
+    connection_ctx: RequestContext,
 ) -> anyhow::Result<()> {
     // Immediately increment the gauge, then create a job to decrement it on task exit.
     // One of the pros of `defer!` is that this will *most probably*
@@ -191,24 +197,24 @@ async fn page_service_conn_main(
         .set_nodelay(true)
         .context("could not set TCP_NODELAY")?;
 
-    let mut conn_handler = PageServerHandler::new(conf, auth);
+    // XXX: pgbackend.run() should take the connection_ctx,
+    // and create a child per-query context when it invokes process_query.
+    // But it's in a shared crate, so, we store connection_ctx inside PageServerHandler
+    // and create the per-query context in process_query ourselves.
+    let mut conn_handler = PageServerHandler::new(conf, auth, connection_ctx);
     let pgbackend = PostgresBackend::new(socket, auth_type, None)?;
 
-    let result = pgbackend
+    match pgbackend
         .run(&mut conn_handler, task_mgr::shutdown_watcher)
-        .await;
-    match result {
+        .await
+    {
         Ok(()) => {
             // we've been requested to shut down
             Ok(())
         }
         Err(QueryError::Disconnected(ConnectionError::Socket(io_error))) => {
-            // `ConnectionReset` error happens when the Postgres client closes the connection.
-            // As this disconnection happens quite often and is expected,
-            // we decided to downgrade the logging level to `INFO`.
-            // See: https://github.com/neondatabase/neon/issues/1683.
-            if io_error.kind() == io::ErrorKind::ConnectionReset {
-                info!("Postgres client disconnected");
+            if is_expected_io_error(&io_error) {
+                info!("Postgres client disconnected ({io_error})");
                 Ok(())
             } else {
                 Err(io_error).context("Postgres connection error")
@@ -255,30 +261,42 @@ struct PageServerHandler {
     _conf: &'static PageServerConf,
     auth: Option<Arc<JwtAuth>>,
     claims: Option<Claims>,
+
+    /// The context created for the lifetime of the connection
+    /// services by this PageServerHandler.
+    /// For each query received over the connection,
+    /// `process_query` creates a child context from this one.
+    connection_ctx: RequestContext,
 }
 
 impl PageServerHandler {
-    pub fn new(conf: &'static PageServerConf, auth: Option<Arc<JwtAuth>>) -> Self {
+    pub fn new(
+        conf: &'static PageServerConf,
+        auth: Option<Arc<JwtAuth>>,
+        connection_ctx: RequestContext,
+    ) -> Self {
         PageServerHandler {
             _conf: conf,
             auth,
             claims: None,
+            connection_ctx,
         }
     }
 
-    #[instrument(skip(self, pgb))]
+    #[instrument(skip(self, pgb, ctx))]
     async fn handle_pagerequests(
         &self,
         pgb: &mut PostgresBackend,
         tenant_id: TenantId,
         timeline_id: TimelineId,
+        ctx: RequestContext,
     ) -> anyhow::Result<()> {
         // NOTE: pagerequests handler exits when connection is closed,
         //       so there is no need to reset the association
         task_mgr::associate_with(Some(tenant_id), Some(timeline_id));
 
         // Make request tracer if needed
-        let tenant = get_active_tenant_with_timeout(tenant_id).await?;
+        let tenant = get_active_tenant_with_timeout(tenant_id, &ctx).await?;
         let mut tracer = if tenant.get_trace_read_requests() {
             let connection_id = ConnectionId::generate();
             let path = tenant
@@ -329,22 +347,27 @@ impl PageServerHandler {
 
             let neon_fe_msg = PagestreamFeMessage::parse(&mut copy_data_bytes.reader())?;
 
+            // TODO: We could create a new per-request context here, with unique ID.
+            // Currently we use the same per-timeline context for all requests
+
             let response = match neon_fe_msg {
                 PagestreamFeMessage::Exists(req) => {
                     let _timer = metrics.get_rel_exists.start_timer();
-                    self.handle_get_rel_exists_request(&timeline, &req).await
+                    self.handle_get_rel_exists_request(&timeline, &req, &ctx)
+                        .await
                 }
                 PagestreamFeMessage::Nblocks(req) => {
                     let _timer = metrics.get_rel_size.start_timer();
-                    self.handle_get_nblocks_request(&timeline, &req).await
+                    self.handle_get_nblocks_request(&timeline, &req, &ctx).await
                 }
                 PagestreamFeMessage::GetPage(req) => {
                     let _timer = metrics.get_page_at_lsn.start_timer();
-                    self.handle_get_page_at_lsn_request(&timeline, &req).await
+                    self.handle_get_page_at_lsn_request(&timeline, &req, &ctx)
+                        .await
                 }
                 PagestreamFeMessage::DbSize(req) => {
                     let _timer = metrics.get_db_size.start_timer();
-                    self.handle_db_size_request(&timeline, &req).await
+                    self.handle_db_size_request(&timeline, &req, &ctx).await
                 }
             };
 
@@ -363,7 +386,8 @@ impl PageServerHandler {
         Ok(())
     }
 
-    #[instrument(skip(self, pgb))]
+    #[allow(clippy::too_many_arguments)]
+    #[instrument(skip(self, pgb, ctx))]
     async fn handle_import_basebackup(
         &self,
         pgb: &mut PostgresBackend,
@@ -372,12 +396,13 @@ impl PageServerHandler {
         base_lsn: Lsn,
         _end_lsn: Lsn,
         pg_version: u32,
+        ctx: RequestContext,
     ) -> Result<(), QueryError> {
         task_mgr::associate_with(Some(tenant_id), Some(timeline_id));
         // Create empty timeline
         info!("creating new timeline");
-        let tenant = get_active_tenant_with_timeout(tenant_id).await?;
-        let timeline = tenant.create_empty_timeline(timeline_id, base_lsn, pg_version)?;
+        let tenant = get_active_tenant_with_timeout(tenant_id, &ctx).await?;
+        let timeline = tenant.create_empty_timeline(timeline_id, base_lsn, pg_version, &ctx)?;
 
         // TODO mark timeline as not ready until it reaches end_lsn.
         // We might have some wal to import as well, and we should prevent compute
@@ -396,7 +421,7 @@ impl PageServerHandler {
 
         let mut copyin_stream = Box::pin(copyin_stream(pgb));
         timeline
-            .import_basebackup_from_tar(&mut copyin_stream, base_lsn)
+            .import_basebackup_from_tar(&mut copyin_stream, base_lsn, &ctx)
             .await?;
 
         // Drain the rest of the Copy data
@@ -418,7 +443,7 @@ impl PageServerHandler {
         Ok(())
     }
 
-    #[instrument(skip(self, pgb))]
+    #[instrument(skip(self, pgb, ctx))]
     async fn handle_import_wal(
         &self,
         pgb: &mut PostgresBackend,
@@ -426,10 +451,11 @@ impl PageServerHandler {
         timeline_id: TimelineId,
         start_lsn: Lsn,
         end_lsn: Lsn,
+        ctx: RequestContext,
     ) -> Result<(), QueryError> {
         task_mgr::associate_with(Some(tenant_id), Some(timeline_id));
 
-        let timeline = get_active_timeline_with_timeout(tenant_id, timeline_id).await?;
+        let timeline = get_active_tenant_timeline(tenant_id, timeline_id, &ctx).await?;
         let last_record_lsn = timeline.get_last_record_lsn();
         if last_record_lsn != start_lsn {
             return Err(QueryError::Other(
@@ -446,7 +472,7 @@ impl PageServerHandler {
         pgb.flush().await?;
         let mut copyin_stream = Box::pin(copyin_stream(pgb));
         let mut reader = tokio_util::io::StreamReader::new(&mut copyin_stream);
-        import_wal_from_tar(&timeline, &mut reader, start_lsn, end_lsn).await?;
+        import_wal_from_tar(&timeline, &mut reader, start_lsn, end_lsn, &ctx).await?;
         info!("wal import complete");
 
         // Drain the rest of the Copy data
@@ -492,6 +518,7 @@ impl PageServerHandler {
         mut lsn: Lsn,
         latest: bool,
         latest_gc_cutoff_lsn: &RcuReadGuard<Lsn>,
+        ctx: &RequestContext,
     ) -> anyhow::Result<Lsn> {
         if latest {
             // Latest page version was requested. If LSN is given, it is a hint
@@ -515,7 +542,7 @@ impl PageServerHandler {
             if lsn <= last_record_lsn {
                 lsn = last_record_lsn;
             } else {
-                timeline.wait_lsn(lsn).await?;
+                timeline.wait_lsn(lsn, ctx).await?;
                 // Since we waited for 'lsn' to arrive, that is now the last
                 // record LSN. (Or close enough for our purposes; the
                 // last-record LSN can advance immediately after we return
@@ -525,7 +552,7 @@ impl PageServerHandler {
             if lsn == Lsn(0) {
                 anyhow::bail!("invalid LSN(0) in request");
             }
-            timeline.wait_lsn(lsn).await?;
+            timeline.wait_lsn(lsn, ctx).await?;
         }
         anyhow::ensure!(
             lsn >= **latest_gc_cutoff_lsn,
@@ -535,52 +562,60 @@ impl PageServerHandler {
         Ok(lsn)
     }
 
-    #[instrument(skip(self, timeline, req), fields(rel = %req.rel, req_lsn = %req.lsn))]
+    #[instrument(skip(self, timeline, req, ctx), fields(rel = %req.rel, req_lsn = %req.lsn))]
     async fn handle_get_rel_exists_request(
         &self,
         timeline: &Timeline,
         req: &PagestreamExistsRequest,
+        ctx: &RequestContext,
     ) -> anyhow::Result<PagestreamBeMessage> {
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn)
-            .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
-        let exists = timeline.get_rel_exists(req.rel, lsn, req.latest).await?;
+        let exists = timeline
+            .get_rel_exists(req.rel, lsn, req.latest, ctx)
+            .await?;
 
         Ok(PagestreamBeMessage::Exists(PagestreamExistsResponse {
             exists,
         }))
     }
 
-    #[instrument(skip(self, timeline, req), fields(rel = %req.rel, req_lsn = %req.lsn))]
+    #[instrument(skip(self, timeline, req, ctx), fields(rel = %req.rel, req_lsn = %req.lsn))]
     async fn handle_get_nblocks_request(
         &self,
         timeline: &Timeline,
         req: &PagestreamNblocksRequest,
+        ctx: &RequestContext,
     ) -> anyhow::Result<PagestreamBeMessage> {
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn)
-            .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
-        let n_blocks = timeline.get_rel_size(req.rel, lsn, req.latest).await?;
+        let n_blocks = timeline.get_rel_size(req.rel, lsn, req.latest, ctx).await?;
 
         Ok(PagestreamBeMessage::Nblocks(PagestreamNblocksResponse {
             n_blocks,
         }))
     }
 
-    #[instrument(skip(self, timeline, req), fields(dbnode = %req.dbnode, req_lsn = %req.lsn))]
+    #[instrument(skip(self, timeline, req, ctx), fields(dbnode = %req.dbnode, req_lsn = %req.lsn))]
     async fn handle_db_size_request(
         &self,
         timeline: &Timeline,
         req: &PagestreamDbSizeRequest,
+        ctx: &RequestContext,
     ) -> anyhow::Result<PagestreamBeMessage> {
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn)
-            .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
 
         let total_blocks = timeline
-            .get_db_size(DEFAULTTABLESPACE_OID, req.dbnode, lsn, req.latest)
+            .get_db_size(DEFAULTTABLESPACE_OID, req.dbnode, lsn, req.latest, ctx)
             .await?;
         let db_size = total_blocks as i64 * BLCKSZ as i64;
 
@@ -589,15 +624,17 @@ impl PageServerHandler {
         }))
     }
 
-    #[instrument(skip(self, timeline, req), fields(rel = %req.rel, blkno = %req.blkno, req_lsn = %req.lsn))]
+    #[instrument(skip(self, timeline, req, ctx), fields(rel = %req.rel, blkno = %req.blkno, req_lsn = %req.lsn))]
     async fn handle_get_page_at_lsn_request(
         &self,
         timeline: &Timeline,
         req: &PagestreamGetPageRequest,
+        ctx: &RequestContext,
     ) -> anyhow::Result<PagestreamBeMessage> {
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
-        let lsn = Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn)
-            .await?;
+        let lsn =
+            Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
+                .await?;
         /*
         // Add a 1s delay to some requests. The delay helps the requests to
         // hit the race condition from github issue #1047 more easily.
@@ -608,7 +645,7 @@ impl PageServerHandler {
         */
 
         let page = timeline
-            .get_rel_page_at_lsn(req.rel, req.blkno, lsn, req.latest)
+            .get_rel_page_at_lsn(req.rel, req.blkno, lsn, req.latest, ctx)
             .await?;
 
         Ok(PagestreamBeMessage::GetPage(PagestreamGetPageResponse {
@@ -616,23 +653,25 @@ impl PageServerHandler {
         }))
     }
 
-    #[instrument(skip(self, pgb))]
+    #[allow(clippy::too_many_arguments)]
+    #[instrument(skip(self, pgb, ctx))]
     async fn handle_basebackup_request(
-        &self,
+        &mut self,
         pgb: &mut PostgresBackend,
         tenant_id: TenantId,
         timeline_id: TimelineId,
         lsn: Option<Lsn>,
         prev_lsn: Option<Lsn>,
         full_backup: bool,
+        ctx: RequestContext,
     ) -> anyhow::Result<()> {
         // check that the timeline exists
-        let timeline = get_active_timeline_with_timeout(tenant_id, timeline_id).await?;
+        let timeline = get_active_tenant_timeline(tenant_id, timeline_id, &ctx).await?;
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         if let Some(lsn) = lsn {
             // Backup was requested at a particular LSN. Wait for it to arrive.
             info!("waiting for {}", lsn);
-            timeline.wait_lsn(lsn).await?;
+            timeline.wait_lsn(lsn, &ctx).await?;
             timeline
                 .check_lsn_is_in_scope(lsn, &latest_gc_cutoff_lsn)
                 .context("invalid basebackup lsn")?;
@@ -645,8 +684,15 @@ impl PageServerHandler {
         // Send a tarball of the latest layer on the timeline
         {
             let mut writer = pgb.copyout_writer();
-            basebackup::send_basebackup_tarball(&mut writer, &timeline, lsn, prev_lsn, full_backup)
-                .await?;
+            basebackup::send_basebackup_tarball(
+                &mut writer,
+                &timeline,
+                lsn,
+                prev_lsn,
+                full_backup,
+                &ctx,
+            )
+            .await?;
         }
 
         pgb.write_message(&BeMessage::CopyDone)?;
@@ -717,6 +763,7 @@ impl postgres_backend_async::Handler for PageServerHandler {
         pgb: &mut PostgresBackend,
         query_string: &str,
     ) -> Result<(), QueryError> {
+        let ctx = self.connection_ctx.attached_child();
         debug!("process query {query_string:?}");
 
         if query_string.starts_with("pagestream ") {
@@ -734,7 +781,7 @@ impl postgres_backend_async::Handler for PageServerHandler {
 
             self.check_permission(Some(tenant_id))?;
 
-            self.handle_pagerequests(pgb, tenant_id, timeline_id)
+            self.handle_pagerequests(pgb, tenant_id, timeline_id, ctx)
                 .await?;
         } else if query_string.starts_with("basebackup ") {
             let (_, params_raw) = query_string.split_at("basebackup ".len());
@@ -763,7 +810,7 @@ impl postgres_backend_async::Handler for PageServerHandler {
             };
 
             // Check that the timeline exists
-            self.handle_basebackup_request(pgb, tenant_id, timeline_id, lsn, None, false)
+            self.handle_basebackup_request(pgb, tenant_id, timeline_id, lsn, None, false, ctx)
                 .await?;
             pgb.write_message(&BeMessage::CommandComplete(b"SELECT 1"))?;
         }
@@ -784,7 +831,7 @@ impl postgres_backend_async::Handler for PageServerHandler {
                 .with_context(|| format!("Failed to parse timeline id from {}", params[1]))?;
 
             self.check_permission(Some(tenant_id))?;
-            let timeline = get_active_timeline_with_timeout(tenant_id, timeline_id).await?;
+            let timeline = get_active_tenant_timeline(tenant_id, timeline_id, &ctx).await?;
 
             let end_of_timeline = timeline.get_last_record_rlsn();
 
@@ -835,7 +882,7 @@ impl postgres_backend_async::Handler for PageServerHandler {
             self.check_permission(Some(tenant_id))?;
 
             // Check that the timeline exists
-            self.handle_basebackup_request(pgb, tenant_id, timeline_id, lsn, prev_lsn, true)
+            self.handle_basebackup_request(pgb, tenant_id, timeline_id, lsn, prev_lsn, true, ctx)
                 .await?;
             pgb.write_message(&BeMessage::CommandComplete(b"SELECT 1"))?;
         } else if query_string.starts_with("import basebackup ") {
@@ -878,6 +925,7 @@ impl postgres_backend_async::Handler for PageServerHandler {
                     base_lsn,
                     end_lsn,
                     pg_version,
+                    ctx,
                 )
                 .await
             {
@@ -914,7 +962,7 @@ impl postgres_backend_async::Handler for PageServerHandler {
             self.check_permission(Some(tenant_id))?;
 
             match self
-                .handle_import_wal(pgb, tenant_id, timeline_id, start_lsn, end_lsn)
+                .handle_import_wal(pgb, tenant_id, timeline_id, start_lsn, end_lsn, ctx)
                 .await
             {
                 Ok(()) => pgb.write_message(&BeMessage::CommandComplete(b"SELECT 1"))?,
@@ -944,7 +992,7 @@ impl postgres_backend_async::Handler for PageServerHandler {
 
             self.check_permission(Some(tenant_id))?;
 
-            let tenant = get_active_tenant_with_timeout(tenant_id).await?;
+            let tenant = get_active_tenant_with_timeout(tenant_id, &ctx).await?;
             pgb.write_message(&BeMessage::RowDescription(&[
                 RowDescriptor::int8_col(b"checkpoint_distance"),
                 RowDescriptor::int8_col(b"checkpoint_timeout"),
@@ -990,27 +1038,66 @@ impl postgres_backend_async::Handler for PageServerHandler {
     }
 }
 
+#[derive(thiserror::Error, Debug)]
+enum GetActiveTenantError {
+    #[error(
+        "Timed out waiting {wait_time:?} for tenant active state. Latest state: {latest_state:?}"
+    )]
+    WaitForActiveTimeout {
+        latest_state: TenantState,
+        wait_time: Duration,
+    },
+    #[error(transparent)]
+    Other(#[from] anyhow::Error),
+}
+
+impl From<GetActiveTenantError> for QueryError {
+    fn from(e: GetActiveTenantError) -> Self {
+        match e {
+            GetActiveTenantError::WaitForActiveTimeout { .. } => QueryError::Disconnected(
+                ConnectionError::Socket(io::Error::new(io::ErrorKind::TimedOut, e.to_string())),
+            ),
+            GetActiveTenantError::Other(e) => QueryError::Other(e),
+        }
+    }
+}
+
 /// Get active tenant.
 ///
 /// If the tenant is Loading, waits for it to become Active, for up to 30 s. That
 /// ensures that queries don't fail immediately after pageserver startup, because
 /// all tenants are still loading.
-async fn get_active_tenant_with_timeout(tenant_id: TenantId) -> anyhow::Result<Arc<Tenant>> {
+async fn get_active_tenant_with_timeout(
+    tenant_id: TenantId,
+    _ctx: &RequestContext, /* require get a context to support cancellation in the future */
+) -> Result<Arc<Tenant>, GetActiveTenantError> {
     let tenant = mgr::get_tenant(tenant_id, false).await?;
-    match tokio::time::timeout(Duration::from_secs(30), tenant.wait_to_become_active()).await {
-        Ok(wait_result) => wait_result
-            // no .context(), the error message is good enough and some tests depend on it
-            .map(move |()| tenant),
-        Err(_) => anyhow::bail!("Timeout waiting for tenant {tenant_id} to become Active"),
+    let wait_time = Duration::from_secs(30);
+    match tokio::time::timeout(wait_time, tenant.wait_to_become_active()).await {
+        Ok(Ok(())) => Ok(tenant),
+        // no .context(), the error message is good enough and some tests depend on it
+        Ok(Err(wait_error)) => Err(GetActiveTenantError::Other(wait_error)),
+        Err(_) => {
+            let latest_state = tenant.current_state();
+            if latest_state == TenantState::Active {
+                Ok(tenant)
+            } else {
+                Err(GetActiveTenantError::WaitForActiveTimeout {
+                    latest_state,
+                    wait_time,
+                })
+            }
+        }
     }
 }
 
 /// Shorthand for getting a reference to a Timeline of an Active tenant.
-async fn get_active_timeline_with_timeout(
+async fn get_active_tenant_timeline(
     tenant_id: TenantId,
     timeline_id: TimelineId,
-) -> anyhow::Result<Arc<Timeline>> {
-    get_active_tenant_with_timeout(tenant_id)
-        .await
-        .and_then(|tenant| tenant.get_timeline(timeline_id, true))
+    ctx: &RequestContext,
+) -> Result<Arc<Timeline>, GetActiveTenantError> {
+    let tenant = get_active_tenant_with_timeout(tenant_id, ctx).await?;
+    let timeline = tenant.get_timeline(timeline_id, true)?;
+    Ok(timeline)
 }

pageserver/src/pgdatadir_mapping.rs

@@ -7,6 +7,7 @@
 //! Clarify that)
 //!
 use super::tenant::{PageReconstructError, Timeline};
+use crate::context::RequestContext;
 use crate::keyspace::{KeySpace, KeySpaceAccum};
 use crate::repository::*;
 use crate::walrecord::NeonWalRecord;
@@ -97,6 +98,7 @@ impl Timeline {
         blknum: BlockNumber,
         lsn: Lsn,
         latest: bool,
+        ctx: &RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         if tag.relnode == 0 {
             return Err(PageReconstructError::Other(anyhow::anyhow!(
@@ -104,7 +106,7 @@ impl Timeline {
             )));
         }
 
-        let nblocks = self.get_rel_size(tag, lsn, latest).await?;
+        let nblocks = self.get_rel_size(tag, lsn, latest, ctx).await?;
         if blknum >= nblocks {
             debug!(
                 "read beyond EOF at {} blk {} at {}, size is {}: returning all-zeros page",
@@ -114,7 +116,7 @@ impl Timeline {
         }
 
         let key = rel_block_to_key(tag, blknum);
-        self.get(key, lsn).await
+        self.get(key, lsn, ctx).await
     }
 
     // Get size of a database in blocks
@@ -124,13 +126,14 @@ impl Timeline {
         dbnode: Oid,
         lsn: Lsn,
         latest: bool,
+        ctx: &RequestContext,
     ) -> Result<usize, PageReconstructError> {
         let mut total_blocks = 0;
 
-        let rels = self.list_rels(spcnode, dbnode, lsn).await?;
+        let rels = self.list_rels(spcnode, dbnode, lsn, ctx).await?;
 
         for rel in rels {
-            let n_blocks = self.get_rel_size(rel, lsn, latest).await?;
+            let n_blocks = self.get_rel_size(rel, lsn, latest, ctx).await?;
             total_blocks += n_blocks as usize;
         }
         Ok(total_blocks)
@@ -142,6 +145,7 @@ impl Timeline {
         tag: RelTag,
         lsn: Lsn,
         latest: bool,
+        ctx: &RequestContext,
     ) -> Result<BlockNumber, PageReconstructError> {
         if tag.relnode == 0 {
             return Err(PageReconstructError::Other(anyhow::anyhow!(
@@ -154,7 +158,7 @@ impl Timeline {
         }
 
         if (tag.forknum == FSM_FORKNUM || tag.forknum == VISIBILITYMAP_FORKNUM)
-            && !self.get_rel_exists(tag, lsn, latest).await?
+            && !self.get_rel_exists(tag, lsn, latest, ctx).await?
         {
             // FIXME: Postgres sometimes calls smgrcreate() to create
             // FSM, and smgrnblocks() on it immediately afterwards,
@@ -164,7 +168,7 @@ impl Timeline {
         }
 
         let key = rel_size_to_key(tag);
-        let mut buf = self.get(key, lsn).await?;
+        let mut buf = self.get(key, lsn, ctx).await?;
         let nblocks = buf.get_u32_le();
 
         if latest {
@@ -186,6 +190,7 @@ impl Timeline {
         tag: RelTag,
         lsn: Lsn,
         _latest: bool,
+        ctx: &RequestContext,
     ) -> Result<bool, PageReconstructError> {
         if tag.relnode == 0 {
             return Err(PageReconstructError::Other(anyhow::anyhow!(
@@ -199,7 +204,7 @@ impl Timeline {
         }
         // fetch directory listing
         let key = rel_dir_to_key(tag.spcnode, tag.dbnode);
-        let buf = self.get(key, lsn).await?;
+        let buf = self.get(key, lsn, ctx).await?;
 
         match RelDirectory::des(&buf).context("deserialization failure") {
             Ok(dir) => {
@@ -216,10 +221,11 @@ impl Timeline {
         spcnode: Oid,
         dbnode: Oid,
         lsn: Lsn,
+        ctx: &RequestContext,
     ) -> Result<HashSet<RelTag>, PageReconstructError> {
         // fetch directory listing
         let key = rel_dir_to_key(spcnode, dbnode);
-        let buf = self.get(key, lsn).await?;
+        let buf = self.get(key, lsn, ctx).await?;
 
         match RelDirectory::des(&buf).context("deserialization failure") {
             Ok(dir) => {
@@ -244,9 +250,10 @@ impl Timeline {
         segno: u32,
         blknum: BlockNumber,
         lsn: Lsn,
+        ctx: &RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         let key = slru_block_to_key(kind, segno, blknum);
-        self.get(key, lsn).await
+        self.get(key, lsn, ctx).await
     }
 
     /// Get size of an SLRU segment
@@ -255,9 +262,10 @@ impl Timeline {
         kind: SlruKind,
         segno: u32,
         lsn: Lsn,
+        ctx: &RequestContext,
     ) -> Result<BlockNumber, PageReconstructError> {
         let key = slru_segment_size_to_key(kind, segno);
-        let mut buf = self.get(key, lsn).await?;
+        let mut buf = self.get(key, lsn, ctx).await?;
         Ok(buf.get_u32_le())
     }
 
@@ -267,10 +275,11 @@ impl Timeline {
         kind: SlruKind,
         segno: u32,
         lsn: Lsn,
+        ctx: &RequestContext,
     ) -> Result<bool, PageReconstructError> {
         // fetch directory listing
         let key = slru_dir_to_key(kind);
-        let buf = self.get(key, lsn).await?;
+        let buf = self.get(key, lsn, ctx).await?;
 
         match SlruSegmentDirectory::des(&buf).context("deserialization failure") {
             Ok(dir) => {
@@ -291,6 +300,7 @@ impl Timeline {
     pub async fn find_lsn_for_timestamp(
         &self,
         search_timestamp: TimestampTz,
+        ctx: &RequestContext,
     ) -> Result<LsnForTimestamp, PageReconstructError> {
         let gc_cutoff_lsn_guard = self.get_latest_gc_cutoff_lsn();
         let min_lsn = *gc_cutoff_lsn_guard;
@@ -313,6 +323,7 @@ impl Timeline {
                     Lsn(mid * 8),
                     &mut found_smaller,
                     &mut found_larger,
+                    ctx,
                 )
                 .await?;
 
@@ -362,14 +373,18 @@ impl Timeline {
         probe_lsn: Lsn,
         found_smaller: &mut bool,
         found_larger: &mut bool,
+        ctx: &RequestContext,
     ) -> Result<bool, PageReconstructError> {
-        for segno in self.list_slru_segments(SlruKind::Clog, probe_lsn).await? {
+        for segno in self
+            .list_slru_segments(SlruKind::Clog, probe_lsn, ctx)
+            .await?
+        {
             let nblocks = self
-                .get_slru_segment_size(SlruKind::Clog, segno, probe_lsn)
+                .get_slru_segment_size(SlruKind::Clog, segno, probe_lsn, ctx)
                 .await?;
             for blknum in (0..nblocks).rev() {
                 let clog_page = self
-                    .get_slru_page_at_lsn(SlruKind::Clog, segno, blknum, probe_lsn)
+                    .get_slru_page_at_lsn(SlruKind::Clog, segno, blknum, probe_lsn, ctx)
                     .await?;
 
                 if clog_page.len() == BLCKSZ as usize + 8 {
@@ -394,11 +409,12 @@ impl Timeline {
         &self,
         kind: SlruKind,
         lsn: Lsn,
+        ctx: &RequestContext,
     ) -> Result<HashSet<u32>, PageReconstructError> {
         // fetch directory entry
         let key = slru_dir_to_key(kind);
 
-        let buf = self.get(key, lsn).await?;
+        let buf = self.get(key, lsn, ctx).await?;
         match SlruSegmentDirectory::des(&buf).context("deserialization failure") {
             Ok(dir) => Ok(dir.segments),
             Err(e) => Err(PageReconstructError::from(e)),
@@ -410,18 +426,21 @@ impl Timeline {
         spcnode: Oid,
         dbnode: Oid,
         lsn: Lsn,
+        ctx: &RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         let key = relmap_file_key(spcnode, dbnode);
 
-        self.get(key, lsn).await
+        let buf = self.get(key, lsn, ctx).await?;
+        Ok(buf)
     }
 
     pub async fn list_dbdirs(
         &self,
         lsn: Lsn,
+        ctx: &RequestContext,
     ) -> Result<HashMap<(Oid, Oid), bool>, PageReconstructError> {
         // fetch directory entry
-        let buf = self.get(DBDIR_KEY, lsn).await?;
+        let buf = self.get(DBDIR_KEY, lsn, ctx).await?;
 
         match DbDirectory::des(&buf).context("deserialization failure") {
             Ok(dir) => Ok(dir.dbdirs),
@@ -433,18 +452,20 @@ impl Timeline {
         &self,
         xid: TransactionId,
         lsn: Lsn,
+        ctx: &RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         let key = twophase_file_key(xid);
-        let buf = self.get(key, lsn).await?;
+        let buf = self.get(key, lsn, ctx).await?;
         Ok(buf)
     }
 
     pub async fn list_twophase_files(
         &self,
         lsn: Lsn,
+        ctx: &RequestContext,
     ) -> Result<HashSet<TransactionId>, PageReconstructError> {
         // fetch directory entry
-        let buf = self.get(TWOPHASEDIR_KEY, lsn).await?;
+        let buf = self.get(TWOPHASEDIR_KEY, lsn, ctx).await?;
 
         match TwoPhaseDirectory::des(&buf).context("deserialization failure") {
             Ok(dir) => Ok(dir.xids),
@@ -452,12 +473,20 @@ impl Timeline {
         }
     }
 
-    pub async fn get_control_file(&self, lsn: Lsn) -> Result<Bytes, PageReconstructError> {
-        self.get(CONTROLFILE_KEY, lsn).await
+    pub async fn get_control_file(
+        &self,
+        lsn: Lsn,
+        ctx: &RequestContext,
+    ) -> Result<Bytes, PageReconstructError> {
+        self.get(CONTROLFILE_KEY, lsn, ctx).await
     }
 
-    pub async fn get_checkpoint(&self, lsn: Lsn) -> Result<Bytes, PageReconstructError> {
-        self.get(CHECKPOINT_KEY, lsn).await
+    pub async fn get_checkpoint(
+        &self,
+        lsn: Lsn,
+        ctx: &RequestContext,
+    ) -> Result<Bytes, PageReconstructError> {
+        self.get(CHECKPOINT_KEY, lsn, ctx).await
     }
 
     /// Does the same as get_current_logical_size but counted on demand.
@@ -469,15 +498,16 @@ impl Timeline {
         &self,
         lsn: Lsn,
         cancel: CancellationToken,
+        ctx: &RequestContext,
     ) -> Result<u64, CalculateLogicalSizeError> {
         // Fetch list of database dirs and iterate them
-        let buf = self.get(DBDIR_KEY, lsn).await.context("read dbdir")?;
+        let buf = self.get(DBDIR_KEY, lsn, ctx).await.context("read dbdir")?;
         let dbdir = DbDirectory::des(&buf).context("deserialize db directory")?;
 
         let mut total_size: u64 = 0;
         for (spcnode, dbnode) in dbdir.dbdirs.keys() {
             for rel in self
-                .list_rels(*spcnode, *dbnode, lsn)
+                .list_rels(*spcnode, *dbnode, lsn, ctx)
                 .await
                 .context("list rels")?
             {
@@ -486,7 +516,7 @@ impl Timeline {
                 }
                 let relsize_key = rel_size_to_key(rel);
                 let mut buf = self
-                    .get(relsize_key, lsn)
+                    .get(relsize_key, lsn, ctx)
                     .await
                     .with_context(|| format!("read relation size of {rel:?}"))?;
                 let relsize = buf.get_u32_le();
@@ -501,7 +531,11 @@ impl Timeline {
     /// Get a KeySpace that covers all the Keys that are in use at the given LSN.
     /// Anything that's not listed maybe removed from the underlying storage (from
     /// that LSN forwards).
-    pub async fn collect_keyspace(&self, lsn: Lsn) -> anyhow::Result<KeySpace> {
+    pub async fn collect_keyspace(
+        &self,
+        lsn: Lsn,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<KeySpace> {
         // Iterate through key ranges, greedily packing them into partitions
         let mut result = KeySpaceAccum::new();
 
@@ -509,7 +543,7 @@ impl Timeline {
         result.add_key(DBDIR_KEY);
 
         // Fetch list of database dirs and iterate them
-        let buf = self.get(DBDIR_KEY, lsn).await?;
+        let buf = self.get(DBDIR_KEY, lsn, ctx).await?;
         let dbdir = DbDirectory::des(&buf).context("deserialization failure")?;
 
         let mut dbs: Vec<(Oid, Oid)> = dbdir.dbdirs.keys().cloned().collect();
@@ -519,14 +553,14 @@ impl Timeline {
             result.add_key(rel_dir_to_key(spcnode, dbnode));
 
             let mut rels: Vec<RelTag> = self
-                .list_rels(spcnode, dbnode, lsn)
+                .list_rels(spcnode, dbnode, lsn, ctx)
                 .await?
                 .into_iter()
                 .collect();
             rels.sort_unstable();
             for rel in rels {
                 let relsize_key = rel_size_to_key(rel);
-                let mut buf = self.get(relsize_key, lsn).await?;
+                let mut buf = self.get(relsize_key, lsn, ctx).await?;
                 let relsize = buf.get_u32_le();
 
                 result.add_range(rel_block_to_key(rel, 0)..rel_block_to_key(rel, relsize));
@@ -542,13 +576,13 @@ impl Timeline {
         ] {
             let slrudir_key = slru_dir_to_key(kind);
             result.add_key(slrudir_key);
-            let buf = self.get(slrudir_key, lsn).await?;
+            let buf = self.get(slrudir_key, lsn, ctx).await?;
             let dir = SlruSegmentDirectory::des(&buf).context("deserialization failure")?;
             let mut segments: Vec<u32> = dir.segments.iter().cloned().collect();
             segments.sort_unstable();
             for segno in segments {
                 let segsize_key = slru_segment_size_to_key(kind, segno);
-                let mut buf = self.get(segsize_key, lsn).await?;
+                let mut buf = self.get(segsize_key, lsn, ctx).await?;
                 let segsize = buf.get_u32_le();
 
                 result.add_range(
@@ -560,7 +594,7 @@ impl Timeline {
 
         // Then pg_twophase
         result.add_key(TWOPHASEDIR_KEY);
-        let buf = self.get(TWOPHASEDIR_KEY, lsn).await?;
+        let buf = self.get(TWOPHASEDIR_KEY, lsn, ctx).await?;
         let twophase_dir = TwoPhaseDirectory::des(&buf).context("deserialization failure")?;
         let mut xids: Vec<TransactionId> = twophase_dir.xids.iter().cloned().collect();
         xids.sort_unstable();
@@ -723,9 +757,10 @@ impl<'a> DatadirModification<'a> {
         spcnode: Oid,
         dbnode: Oid,
         img: Bytes,
+        ctx: &RequestContext,
     ) -> anyhow::Result<()> {
         // Add it to the directory (if it doesn't exist already)
-        let buf = self.get(DBDIR_KEY).await?;
+        let buf = self.get(DBDIR_KEY, ctx).await?;
         let mut dbdir = DbDirectory::des(&buf)?;
 
         let r = dbdir.dbdirs.insert((spcnode, dbnode), true);
@@ -755,9 +790,10 @@ impl<'a> DatadirModification<'a> {
         &mut self,
         xid: TransactionId,
         img: Bytes,
+        ctx: &RequestContext,
     ) -> anyhow::Result<()> {
         // Add it to the directory entry
-        let buf = self.get(TWOPHASEDIR_KEY).await?;
+        let buf = self.get(TWOPHASEDIR_KEY, ctx).await?;
         let mut dir = TwoPhaseDirectory::des(&buf)?;
         if !dir.xids.insert(xid) {
             anyhow::bail!("twophase file for xid {} already exists", xid);
@@ -781,16 +817,21 @@ impl<'a> DatadirModification<'a> {
         Ok(())
     }
 
-    pub async fn drop_dbdir(&mut self, spcnode: Oid, dbnode: Oid) -> anyhow::Result<()> {
+    pub async fn drop_dbdir(
+        &mut self,
+        spcnode: Oid,
+        dbnode: Oid,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<()> {
         let req_lsn = self.tline.get_last_record_lsn();
 
         let total_blocks = self
             .tline
-            .get_db_size(spcnode, dbnode, req_lsn, true)
+            .get_db_size(spcnode, dbnode, req_lsn, true, ctx)
             .await?;
 
         // Remove entry from dbdir
-        let buf = self.get(DBDIR_KEY).await?;
+        let buf = self.get(DBDIR_KEY, ctx).await?;
         let mut dir = DbDirectory::des(&buf)?;
         if dir.dbdirs.remove(&(spcnode, dbnode)).is_some() {
             let buf = DbDirectory::ser(&dir)?;
@@ -817,11 +858,12 @@ impl<'a> DatadirModification<'a> {
         &mut self,
         rel: RelTag,
         nblocks: BlockNumber,
+        ctx: &RequestContext,
     ) -> anyhow::Result<()> {
         anyhow::ensure!(rel.relnode != 0, "invalid relnode");
         // It's possible that this is the first rel for this db in this
         // tablespace.  Create the reldir entry for it if so.
-        let mut dbdir = DbDirectory::des(&self.get(DBDIR_KEY).await?)?;
+        let mut dbdir = DbDirectory::des(&self.get(DBDIR_KEY, ctx).await?)?;
         let rel_dir_key = rel_dir_to_key(rel.spcnode, rel.dbnode);
         let mut rel_dir = if dbdir.dbdirs.get(&(rel.spcnode, rel.dbnode)).is_none() {
             // Didn't exist. Update dbdir
@@ -833,7 +875,7 @@ impl<'a> DatadirModification<'a> {
             RelDirectory::default()
         } else {
             // reldir already exists, fetch it
-            RelDirectory::des(&self.get(rel_dir_key).await?)?
+            RelDirectory::des(&self.get(rel_dir_key, ctx).await?)?
         };
 
         // Add the new relation to the rel directory entry, and write it back
@@ -865,13 +907,14 @@ impl<'a> DatadirModification<'a> {
         &mut self,
         rel: RelTag,
         nblocks: BlockNumber,
+        ctx: &RequestContext,
     ) -> anyhow::Result<()> {
         anyhow::ensure!(rel.relnode != 0, "invalid relnode");
         let last_lsn = self.tline.get_last_record_lsn();
-        if self.tline.get_rel_exists(rel, last_lsn, true).await? {
+        if self.tline.get_rel_exists(rel, last_lsn, true, ctx).await? {
             let size_key = rel_size_to_key(rel);
             // Fetch the old size first
-            let old_size = self.get(size_key).await?.get_u32_le();
+            let old_size = self.get(size_key, ctx).await?.get_u32_le();
 
             // Update the entry with the new size.
             let buf = nblocks.to_le_bytes();
@@ -895,12 +938,13 @@ impl<'a> DatadirModification<'a> {
         &mut self,
         rel: RelTag,
         nblocks: BlockNumber,
+        ctx: &RequestContext,
     ) -> anyhow::Result<()> {
         anyhow::ensure!(rel.relnode != 0, "invalid relnode");
 
         // Put size
         let size_key = rel_size_to_key(rel);
-        let old_size = self.get(size_key).await?.get_u32_le();
+        let old_size = self.get(size_key, ctx).await?.get_u32_le();
 
         // only extend relation here. never decrease the size
         if nblocks > old_size {
@@ -916,12 +960,12 @@ impl<'a> DatadirModification<'a> {
     }
 
     /// Drop a relation.
-    pub async fn put_rel_drop(&mut self, rel: RelTag) -> anyhow::Result<()> {
+    pub async fn put_rel_drop(&mut self, rel: RelTag, ctx: &RequestContext) -> anyhow::Result<()> {
         anyhow::ensure!(rel.relnode != 0, "invalid relnode");
 
         // Remove it from the directory entry
         let dir_key = rel_dir_to_key(rel.spcnode, rel.dbnode);
-        let buf = self.get(dir_key).await?;
+        let buf = self.get(dir_key, ctx).await?;
         let mut dir = RelDirectory::des(&buf)?;
 
         if dir.rels.remove(&(rel.relnode, rel.forknum)) {
@@ -932,7 +976,7 @@ impl<'a> DatadirModification<'a> {
 
         // update logical size
         let size_key = rel_size_to_key(rel);
-        let old_size = self.get(size_key).await?.get_u32_le();
+        let old_size = self.get(size_key, ctx).await?.get_u32_le();
         self.pending_nblocks -= old_size as i64;
 
         // Remove enty from relation size cache
@@ -949,10 +993,11 @@ impl<'a> DatadirModification<'a> {
         kind: SlruKind,
         segno: u32,
         nblocks: BlockNumber,
+        ctx: &RequestContext,
     ) -> anyhow::Result<()> {
         // Add it to the directory entry
         let dir_key = slru_dir_to_key(kind);
-        let buf = self.get(dir_key).await?;
+        let buf = self.get(dir_key, ctx).await?;
         let mut dir = SlruSegmentDirectory::des(&buf)?;
 
         if !dir.segments.insert(segno) {
@@ -988,10 +1033,15 @@ impl<'a> DatadirModification<'a> {
     }
 
     /// This method is used for marking truncated SLRU files
-    pub async fn drop_slru_segment(&mut self, kind: SlruKind, segno: u32) -> anyhow::Result<()> {
+    pub async fn drop_slru_segment(
+        &mut self,
+        kind: SlruKind,
+        segno: u32,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<()> {
         // Remove it from the directory entry
         let dir_key = slru_dir_to_key(kind);
-        let buf = self.get(dir_key).await?;
+        let buf = self.get(dir_key, ctx).await?;
         let mut dir = SlruSegmentDirectory::des(&buf)?;
 
         if !dir.segments.remove(&segno) {
@@ -1015,9 +1065,13 @@ impl<'a> DatadirModification<'a> {
     }
 
     /// This method is used for marking truncated SLRU files
-    pub async fn drop_twophase_file(&mut self, xid: TransactionId) -> anyhow::Result<()> {
+    pub async fn drop_twophase_file(
+        &mut self,
+        xid: TransactionId,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<()> {
         // Remove it from the directory entry
-        let buf = self.get(TWOPHASEDIR_KEY).await?;
+        let buf = self.get(TWOPHASEDIR_KEY, ctx).await?;
         let mut dir = TwoPhaseDirectory::des(&buf)?;
 
         if !dir.xids.remove(&xid) {
@@ -1111,7 +1165,7 @@ impl<'a> DatadirModification<'a> {
 
     // Internal helper functions to batch the modifications
 
-    async fn get(&self, key: Key) -> Result<Bytes, PageReconstructError> {
+    async fn get(&self, key: Key, ctx: &RequestContext) -> Result<Bytes, PageReconstructError> {
         // Have we already updated the same key? Read the pending updated
         // version in that case.
         //
@@ -1132,7 +1186,7 @@ impl<'a> DatadirModification<'a> {
             }
         } else {
             let lsn = Lsn::max(self.tline.get_last_record_lsn(), self.lsn);
-            self.tline.get(key, lsn).await
+            self.tline.get(key, lsn, ctx).await
         }
     }
 
@@ -1542,10 +1596,11 @@ pub fn create_test_timeline(
     tenant: &crate::tenant::Tenant,
     timeline_id: utils::id::TimelineId,
     pg_version: u32,
+    ctx: &RequestContext,
 ) -> anyhow::Result<std::sync::Arc<Timeline>> {
     let tline = tenant
-        .create_empty_timeline(timeline_id, Lsn(8), pg_version)?
-        .initialize()?;
+        .create_empty_timeline(timeline_id, Lsn(8), pg_version, ctx)?
+        .initialize(ctx)?;
     let mut m = tline.begin_modification(Lsn(8));
     m.init_empty()?;
     m.commit()?;
@@ -1598,7 +1653,7 @@ mod tests {
         assert!(tline.list_rels(0, TESTDB, Lsn(0x30))?.contains(&TESTREL_A));
 
         // Create a branch, check that the relation is visible there
-        repo.branch_timeline(TIMELINE_ID, NEW_TIMELINE_ID, Lsn(0x30))?;
+        repo.branch_timeline(&tline, NEW_TIMELINE_ID, Lsn(0x30))?;
         let newtline = match repo.get_timeline(NEW_TIMELINE_ID)?.local_timeline() {
             Some(timeline) => timeline,
             None => panic!("Should have a local timeline"),

pageserver/src/repository.rs

@@ -7,11 +7,11 @@ use std::fmt;
 use std::ops::{AddAssign, Range};
 use std::time::Duration;
 
-#[derive(Debug, Clone, Copy, Hash, PartialEq, Eq, Ord, PartialOrd, Serialize, Deserialize)]
 /// Key used in the Repository kv-store.
 ///
 /// The Repository treats this as an opaque struct, but see the code in pgdatadir_mapping.rs
 /// for what we actually store in these fields.
+#[derive(Debug, Clone, Copy, Hash, PartialEq, Eq, Ord, PartialOrd, Serialize, Deserialize)]
 pub struct Key {
     pub field1: u8,
     pub field2: u32,

pageserver/src/task_mgr.rs

@@ -169,8 +169,18 @@ task_local! {
 /// Note that we don't try to limit how many task of a certain kind can be running
 /// at the same time.
 ///
-#[derive(Debug, PartialEq, Eq, Clone, Copy)]
+#[derive(
+    Debug,
+    // NB: enumset::EnumSetType derives PartialEq, Eq, Clone, Copy
+    enumset::EnumSetType,
+    serde::Serialize,
+    serde::Deserialize,
+    strum_macros::IntoStaticStr,
+)]
 pub enum TaskKind {
+    // Pageserver startup, i.e., `main`
+    Startup,
+
     // libpq listener task. It just accepts connection and spawns a
     // PageRequestHandler task for each connection.
     LibpqEndpointListener,
@@ -194,19 +204,26 @@ pub enum TaskKind {
     /// The `Connection` object is responsible for speaking the wire protocol.
     ///
     /// Walreceiver uses its own abstraction called `TaskHandle` to represent the activity of establishing and handling a connection.
-    /// That abstraction doesn't use `task_mgr` and hence, has no `TaskKind`.
+    /// That abstraction doesn't use `task_mgr`.
     /// The [`WalReceiverManager`] task ensures that this `TaskHandle` task does not outlive the [`WalReceiverManager`] task.
+    /// For the `RequestContext` that we hand to the TaskHandle, we use the [`WalReceiverConnectionHandler`] task kind.
     ///
     /// Once the connection is established, the `TaskHandle` task creates a
-    /// [`WalReceiverConnection`] task_mgr task that is responsible for polling
+    /// [`WalReceiverConnectionPoller`] task_mgr task that is responsible for polling
     /// the `Connection` object.
     /// A `CancellationToken` created by the `TaskHandle` task ensures
-    /// that the [`WalReceiverConnection`] task will cancel soon after as the `TaskHandle` is dropped.
+    /// that the [`WalReceiverConnectionPoller`] task will cancel soon after as the `TaskHandle` is dropped.
     WalReceiverManager,
 
-    /// The task that polls the `tokio-postgres::Connection` object.
+    /// The `TaskHandle` task that executes [`walreceiver_connection::handle_walreceiver_connection`].
+    /// Not a `task_mgr` task, but we use this `TaskKind` for its `RequestContext`.
     /// See the comment on [`WalReceiverManager`].
-    WalReceiverConnection,
+    WalReceiverConnectionHandler,
+
+    /// The task that polls the `tokio-postgres::Connection` object.
+    /// Spawned by task [`WalReceiverConnectionHandler`].
+    /// See the comment on [`WalReceiverManager`].
+    WalReceiverConnectionPoller,
 
     // Garbage collection worker. One per tenant
     GarbageCollector,
@@ -214,9 +231,14 @@ pub enum TaskKind {
     // Compaction. One per tenant.
     Compaction,
 
+    // Eviction. One per timeline.
+    Eviction,
+
     // Initial logical size calculation
     InitialLogicalSizeCalculation,
 
+    OndemandLogicalSizeCalculation,
+
     // Task that flushes frozen in-memory layers to disk
     LayerFlushTask,
 
@@ -239,6 +261,14 @@ pub enum TaskKind {
     DownloadAllRemoteLayers,
     // Task that calculates synthetis size for all active tenants
     CalculateSyntheticSize,
+
+    // A request that comes in via the pageserver HTTP API.
+    MgmtRequest,
+
+    DebugTool,
+
+    #[cfg(test)]
+    UnitTest,
 }
 
 #[derive(Default)]

pageserver/src/tenant/config.rs

@@ -28,7 +28,12 @@ pub mod defaults {
     pub const DEFAULT_COMPACTION_THRESHOLD: usize = 10;
 
     pub const DEFAULT_GC_HORIZON: u64 = 64 * 1024 * 1024;
-    pub const DEFAULT_GC_PERIOD: &str = "100 s";
+
+    // Large DEFAULT_GC_PERIOD is fine as long as PITR_INTERVAL is larger.
+    // If there's a need to decrease this value, first make sure that GC
+    // doesn't hold a layer map write lock for non-trivial operations.
+    // Relevant: https://github.com/neondatabase/neon/issues/3394
+    pub const DEFAULT_GC_PERIOD: &str = "1 hr";
     pub const DEFAULT_IMAGE_CREATION_THRESHOLD: usize = 3;
     pub const DEFAULT_PITR_INTERVAL: &str = "7 days";
     pub const DEFAULT_WALRECEIVER_CONNECT_TIMEOUT: &str = "2 seconds";
@@ -46,6 +51,7 @@ pub struct TenantConf {
     pub checkpoint_distance: u64,
     // Inmemory layer is also flushed at least once in checkpoint_timeout to
     // eventually upload WAL after activity is stopped.
+    #[serde(with = "humantime_serde")]
     pub checkpoint_timeout: Duration,
     // Target file size, when creating image and delta layers.
     // This parameter determines L1 layer file size.
@@ -85,30 +91,97 @@ pub struct TenantConf {
     /// to avoid eager reconnects.
     pub max_lsn_wal_lag: NonZeroU64,
     pub trace_read_requests: bool,
+    pub eviction_policy: EvictionPolicy,
 }
 
 /// Same as TenantConf, but this struct preserves the information about
 /// which parameters are set and which are not.
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, Default)]
 pub struct TenantConfOpt {
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
     pub checkpoint_distance: Option<u64>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
     pub checkpoint_timeout: Option<Duration>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
     pub compaction_target_size: Option<u64>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(with = "humantime_serde")]
+    #[serde(default)]
     pub compaction_period: Option<Duration>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
     pub compaction_threshold: Option<usize>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
     pub gc_horizon: Option<u64>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(with = "humantime_serde")]
+    #[serde(default)]
     pub gc_period: Option<Duration>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
     pub image_creation_threshold: Option<usize>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(with = "humantime_serde")]
+    #[serde(default)]
     pub pitr_interval: Option<Duration>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(with = "humantime_serde")]
+    #[serde(default)]
     pub walreceiver_connect_timeout: Option<Duration>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(with = "humantime_serde")]
+    #[serde(default)]
     pub lagging_wal_timeout: Option<Duration>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
     pub max_lsn_wal_lag: Option<NonZeroU64>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
     pub trace_read_requests: Option<bool>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
+    pub eviction_policy: Option<EvictionPolicy>,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(tag = "kind")]
+pub enum EvictionPolicy {
+    NoEviction,
+    LayerAccessThreshold(EvictionPolicyLayerAccessThreshold),
+}
+
+impl EvictionPolicy {
+    pub fn discriminant_str(&self) -> &'static str {
+        match self {
+            EvictionPolicy::NoEviction => "NoEviction",
+            EvictionPolicy::LayerAccessThreshold(_) => "LayerAccessThreshold",
+        }
+    }
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+pub struct EvictionPolicyLayerAccessThreshold {
+    #[serde(with = "humantime_serde")]
+    pub period: Duration,
+    #[serde(with = "humantime_serde")]
+    pub threshold: Duration,
 }
 
 impl TenantConfOpt {
@@ -145,6 +218,7 @@ impl TenantConfOpt {
             trace_read_requests: self
                 .trace_read_requests
                 .unwrap_or(global_conf.trace_read_requests),
+            eviction_policy: self.eviction_policy.unwrap_or(global_conf.eviction_policy),
         }
     }
 
@@ -217,6 +291,28 @@ impl Default for TenantConf {
             max_lsn_wal_lag: NonZeroU64::new(DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG)
                 .expect("cannot parse default max walreceiver Lsn wal lag"),
             trace_read_requests: false,
+            eviction_policy: EvictionPolicy::NoEviction,
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn de_serializing_pageserver_config_omits_empty_values() {
+        let small_conf = TenantConfOpt {
+            gc_horizon: Some(42),
+            ..TenantConfOpt::default()
+        };
+
+        let toml_form = toml_edit::easy::to_string(&small_conf).unwrap();
+        assert_eq!(toml_form, "gc_horizon = 42\n");
+        assert_eq!(small_conf, toml_edit::easy::from_str(&toml_form).unwrap());
+
+        let json_form = serde_json::to_string(&small_conf).unwrap();
+        assert_eq!(json_form, "{\"gc_horizon\":42}");
+        assert_eq!(small_conf, serde_json::from_str(&json_form).unwrap());
+    }
+}

pageserver/src/tenant/layer_map.rs

@@ -46,6 +46,7 @@
 mod historic_layer_coverage;
 mod layer_coverage;
 
+use crate::context::RequestContext;
 use crate::keyspace::KeyPartitioning;
 use crate::metrics::NUM_ONDISK_LAYERS;
 use crate::repository::Key;
@@ -58,6 +59,7 @@ use std::sync::Arc;
 use utils::lsn::Lsn;
 
 use historic_layer_coverage::BufferedHistoricLayerCoverage;
+pub use historic_layer_coverage::Replacement;
 
 use super::storage_layer::range_eq;
 
@@ -137,6 +139,30 @@ where
         self.layer_map.remove_historic_noflush(layer)
     }
 
+    /// Replaces existing layer iff it is the `expected`.
+    ///
+    /// If the expected layer has been removed it will not be inserted by this function.
+    ///
+    /// Returned `Replacement` describes succeeding in replacement or the reason why it could not
+    /// be done.
+    ///
+    /// TODO replacement can be done without buffering and rebuilding layer map updates.
+    ///      One way to do that is to add a layer of indirection for returned values, so
+    ///      that we can replace values only by updating a hashmap.
+    pub fn replace_historic(
+        &mut self,
+        expected: &Arc<L>,
+        new: Arc<L>,
+    ) -> anyhow::Result<Replacement<Arc<L>>> {
+        fail::fail_point!("layermap-replace-notfound", |_| Ok(
+            // this is not what happens if an L0 layer was not found a anyhow error but perhaps
+            // that should be changed. this is good enough to show a replacement failure.
+            Replacement::NotFound
+        ));
+
+        self.layer_map.replace_historic_noflush(expected, new)
+    }
+
     // We will flush on drop anyway, but this method makes it
     // more explicit that there is some work being done.
     /// Apply all updates
@@ -253,14 +279,8 @@ where
     /// Helper function for BatchedUpdates::insert_historic
     ///
     pub(self) fn insert_historic_noflush(&mut self, layer: Arc<L>) {
-        let kr = layer.get_key_range();
-        let lr = layer.get_lsn_range();
         self.historic.insert(
-            historic_layer_coverage::LayerKey {
-                key: kr.start.to_i128()..kr.end.to_i128(),
-                lsn: lr.start.0..lr.end.0,
-                is_image: !layer.is_incremental(),
-            },
+            historic_layer_coverage::LayerKey::from(&*layer),
             Arc::clone(&layer),
         );
 
@@ -277,30 +297,72 @@ where
     /// Helper function for BatchedUpdates::remove_historic
     ///
     pub fn remove_historic_noflush(&mut self, layer: Arc<L>) {
-        let kr = layer.get_key_range();
-        let lr = layer.get_lsn_range();
-        self.historic.remove(historic_layer_coverage::LayerKey {
-            key: kr.start.to_i128()..kr.end.to_i128(),
-            lsn: lr.start.0..lr.end.0,
-            is_image: !layer.is_incremental(),
-        });
+        self.historic
+            .remove(historic_layer_coverage::LayerKey::from(&*layer));
 
         if Self::is_l0(&layer) {
             let len_before = self.l0_delta_layers.len();
-
-            // FIXME: ptr_eq might fail to return true for 'dyn'
-            // references.  Clippy complains about this. In practice it
-            // seems to work, the assertion below would be triggered
-            // otherwise but this ought to be fixed.
-            #[allow(clippy::vtable_address_comparisons)]
             self.l0_delta_layers
-                .retain(|other| !Arc::ptr_eq(other, &layer));
-            assert_eq!(self.l0_delta_layers.len(), len_before - 1);
+                .retain(|other| !Self::compare_arced_layers(other, &layer));
+            // this assertion is related to use of Arc::ptr_eq in Self::compare_arced_layers,
+            // there's a chance that the comparison fails at runtime due to it comparing (pointer,
+            // vtable) pairs.
+            assert_eq!(
+                self.l0_delta_layers.len(),
+                len_before - 1,
+                "failed to locate removed historic layer from l0_delta_layers"
+            );
         }
 
         NUM_ONDISK_LAYERS.dec();
     }
 
+    pub(self) fn replace_historic_noflush(
+        &mut self,
+        expected: &Arc<L>,
+        new: Arc<L>,
+    ) -> anyhow::Result<Replacement<Arc<L>>> {
+        let key = historic_layer_coverage::LayerKey::from(&**expected);
+        let other = historic_layer_coverage::LayerKey::from(&*new);
+
+        let expected_l0 = Self::is_l0(expected);
+        let new_l0 = Self::is_l0(&new);
+
+        anyhow::ensure!(
+            key == other,
+            "expected and new must have equal LayerKeys: {key:?} != {other:?}"
+        );
+
+        anyhow::ensure!(
+            expected_l0 == new_l0,
+            "expected and new must both be l0 deltas or neither should be: {expected_l0} != {new_l0}"
+        );
+
+        let l0_index = if expected_l0 {
+            // find the index in case replace worked, we need to replace that as well
+            Some(
+                self.l0_delta_layers
+                    .iter()
+                    .position(|slot| Self::compare_arced_layers(slot, expected))
+                    .ok_or_else(|| anyhow::anyhow!("existing l0 delta layer was not found"))?,
+            )
+        } else {
+            None
+        };
+
+        let replaced = self.historic.replace(&key, new.clone(), |existing| {
+            Self::compare_arced_layers(existing, expected)
+        });
+
+        if let Replacement::Replaced { .. } = &replaced {
+            if let Some(index) = l0_index {
+                self.l0_delta_layers[index] = new;
+            }
+        }
+
+        Ok(replaced)
+    }
+
     /// Helper function for BatchedUpdates::drop.
     pub(self) fn flush_updates(&mut self) {
         self.historic.rebuild();
@@ -654,24 +716,139 @@ where
 
     /// debugging function to print out the contents of the layer map
     #[allow(unused)]
-    pub fn dump(&self, verbose: bool) -> Result<()> {
+    pub fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {
         println!("Begin dump LayerMap");
 
         println!("open_layer:");
         if let Some(open_layer) = &self.open_layer {
-            open_layer.dump(verbose)?;
+            open_layer.dump(verbose, ctx)?;
         }
 
         println!("frozen_layers:");
         for frozen_layer in self.frozen_layers.iter() {
-            frozen_layer.dump(verbose)?;
+            frozen_layer.dump(verbose, ctx)?;
         }
 
         println!("historic_layers:");
         for layer in self.iter_historic_layers() {
-            layer.dump(verbose)?;
+            layer.dump(verbose, ctx)?;
         }
         println!("End dump LayerMap");
         Ok(())
     }
+
+    /// Similar to `Arc::ptr_eq`, but only compares the object pointers, not vtables.
+    ///
+    /// Returns `true` if the two `Arc` point to the same layer, false otherwise.
+    #[inline(always)]
+    pub fn compare_arced_layers(left: &Arc<L>, right: &Arc<L>) -> bool {
+        // "dyn Trait" objects are "fat pointers" in that they have two components:
+        // - pointer to the object
+        // - pointer to the vtable
+        //
+        // rust does not provide a guarantee that these vtables are unique, but however
+        // `Arc::ptr_eq` as of writing (at least up to 1.67) uses a comparison where both the
+        // pointer and the vtable need to be equal.
+        //
+        // See: https://github.com/rust-lang/rust/issues/103763
+        //
+        // A future version of rust will most likely use this form below, where we cast each
+        // pointer into a pointer to unit, which drops the inaccessible vtable pointer, making it
+        // not affect the comparison.
+        //
+        // See: https://github.com/rust-lang/rust/pull/106450
+        let left = Arc::as_ptr(left) as *const ();
+        let right = Arc::as_ptr(right) as *const ();
+
+        left == right
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{LayerMap, Replacement};
+    use crate::tenant::storage_layer::{Layer, LayerDescriptor, LayerFileName};
+    use std::str::FromStr;
+    use std::sync::Arc;
+
+    mod l0_delta_layers_updated {
+
+        use super::*;
+
+        #[test]
+        fn for_full_range_delta() {
+            // l0_delta_layers are used by compaction, and should observe all buffered updates
+            l0_delta_layers_updated_scenario(
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000053423C21-0000000053424D69",
+                true
+            )
+        }
+
+        #[test]
+        fn for_non_full_range_delta() {
+            // has minimal uncovered areas compared to l0_delta_layers_updated_on_insert_replace_remove_for_full_range_delta
+            l0_delta_layers_updated_scenario(
+                "000000000000000000000000000000000001-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFE__0000000053423C21-0000000053424D69",
+                // because not full range
+                false
+            )
+        }
+
+        #[test]
+        fn for_image() {
+            l0_delta_layers_updated_scenario(
+                "000000000000000000000000000000000000-000000000000000000000000000000010000__0000000053424D69",
+                // code only checks if it is a full range layer, doesn't care about images, which must
+                // mean we should in practice never have full range images
+                false
+            )
+        }
+
+        fn l0_delta_layers_updated_scenario(layer_name: &str, expected_l0: bool) {
+            let name = LayerFileName::from_str(layer_name).unwrap();
+            let skeleton = LayerDescriptor::from(name);
+
+            let remote: Arc<dyn Layer> = Arc::new(skeleton.clone());
+            let downloaded: Arc<dyn Layer> = Arc::new(skeleton);
+
+            let mut map = LayerMap::default();
+
+            // two disjoint Arcs in different lifecycle phases.
+            assert!(!LayerMap::compare_arced_layers(&remote, &downloaded));
+
+            let expected_in_counts = (1, usize::from(expected_l0));
+
+            map.batch_update().insert_historic(remote.clone());
+            assert_eq!(count_layer_in(&map, &remote), expected_in_counts);
+
+            let replaced = map
+                .batch_update()
+                .replace_historic(&remote, downloaded.clone())
+                .expect("name derived attributes are the same");
+            assert!(
+                matches!(replaced, Replacement::Replaced { .. }),
+                "{replaced:?}"
+            );
+            assert_eq!(count_layer_in(&map, &downloaded), expected_in_counts);
+
+            map.batch_update().remove_historic(downloaded.clone());
+            assert_eq!(count_layer_in(&map, &downloaded), (0, 0));
+        }
+
+        fn count_layer_in(map: &LayerMap<dyn Layer>, layer: &Arc<dyn Layer>) -> (usize, usize) {
+            let historic = map
+                .iter_historic_layers()
+                .filter(|x| LayerMap::compare_arced_layers(x, layer))
+                .count();
+            let l0s = map
+                .get_level0_deltas()
+                .expect("why does this return a result");
+            let l0 = l0s
+                .iter()
+                .filter(|x| LayerMap::compare_arced_layers(x, layer))
+                .count();
+
+            (historic, l0)
+        }
+    }
 }

pageserver/src/tenant/layer_map/historic_layer_coverage.rs

@@ -41,6 +41,18 @@ impl Ord for LayerKey {
     }
 }
 
+impl<'a, L: crate::tenant::storage_layer::Layer + ?Sized> From<&'a L> for LayerKey {
+    fn from(layer: &'a L) -> Self {
+        let kr = layer.get_key_range();
+        let lr = layer.get_lsn_range();
+        LayerKey {
+            key: kr.start.to_i128()..kr.end.to_i128(),
+            lsn: lr.start.0..lr.end.0,
+            is_image: !layer.is_incremental(),
+        }
+    }
+}
+
 /// Efficiently queryable layer coverage for each LSN.
 ///
 /// Allows answering layer map queries very efficiently,
@@ -82,15 +94,13 @@ impl<Value: Clone> HistoricLayerCoverage<Value> {
         }
 
         // Insert into data structure
-        if layer_key.is_image {
-            self.head
-                .image_coverage
-                .insert(layer_key.key, layer_key.lsn.clone(), value);
+        let target = if layer_key.is_image {
+            &mut self.head.image_coverage
         } else {
-            self.head
-                .delta_coverage
-                .insert(layer_key.key, layer_key.lsn.clone(), value);
-        }
+            &mut self.head.delta_coverage
+        };
+
+        target.insert(layer_key.key, layer_key.lsn.clone(), value);
 
         // Remember history. Clone is O(1)
         self.historic.insert(layer_key.lsn.start, self.head.clone());
@@ -415,6 +425,59 @@ impl<Value: Clone> BufferedHistoricLayerCoverage<Value> {
         self.buffer.insert(layer_key, None);
     }
 
+    /// Replaces a previous layer with a new layer value.
+    ///
+    /// The replacement is conditional on:
+    /// - there is an existing `LayerKey` record
+    /// - there is no buffered removal for the given `LayerKey`
+    /// - the given closure returns true for the current `Value`
+    ///
+    /// The closure is used to compare the latest value (buffered insert, or existing layer)
+    /// against some expectation. This allows to use `Arc::ptr_eq` or similar which would be
+    /// inaccessible via `PartialEq` trait.
+    ///
+    /// Returns a `Replacement` value describing the outcome; only the case of
+    /// `Replacement::Replaced` modifies the map and requires a rebuild.
+    pub fn replace<F>(
+        &mut self,
+        layer_key: &LayerKey,
+        new: Value,
+        check_expected: F,
+    ) -> Replacement<Value>
+    where
+        F: FnOnce(&Value) -> bool,
+    {
+        let (slot, in_buffered) = match self.buffer.get(layer_key) {
+            Some(inner @ Some(_)) => {
+                // we compare against the buffered version, because there will be a later
+                // rebuild before querying
+                (inner.as_ref(), true)
+            }
+            Some(None) => {
+                // buffer has removal for this key; it will not be equivalent by any check_expected.
+                return Replacement::RemovalBuffered;
+            }
+            None => {
+                // no pending modification for the key, check layers
+                (self.layers.get(layer_key), false)
+            }
+        };
+
+        match slot {
+            Some(existing) if !check_expected(existing) => {
+                // unfortunate clone here, but otherwise the nll borrowck grows the region of
+                // 'a to cover the whole function, and we could not mutate in the other
+                // Some(existing) branch
+                Replacement::Unexpected(existing.clone())
+            }
+            None => Replacement::NotFound,
+            Some(_existing) => {
+                self.insert(layer_key.to_owned(), new);
+                Replacement::Replaced { in_buffered }
+            }
+        }
+    }
+
     pub fn rebuild(&mut self) {
         // Find the first LSN that needs to be rebuilt
         let rebuild_since: u64 = match self.buffer.iter().next() {
@@ -483,6 +546,22 @@ impl<Value: Clone> BufferedHistoricLayerCoverage<Value> {
     }
 }
 
+/// Outcome of the replace operation.
+#[derive(Debug)]
+pub enum Replacement<Value> {
+    /// Previous value was replaced with the new value.
+    Replaced {
+        /// Replacement happened for a scheduled insert.
+        in_buffered: bool,
+    },
+    /// Key was not found buffered updates or existing layers.
+    NotFound,
+    /// Key has been scheduled for removal, it was not replaced.
+    RemovalBuffered,
+    /// Previous value was rejected by the closure.
+    Unexpected(Value),
+}
+
 #[test]
 fn test_retroactive_regression_1() {
     let mut map = BufferedHistoricLayerCoverage::new();
@@ -548,7 +627,7 @@ fn test_retroactive_simple() {
         LayerKey {
             key: 2..5,
             lsn: 105..106,
-            is_image: true,
+            is_image: false,
         },
         "Delta 1".to_string(),
     );
@@ -556,17 +635,24 @@ fn test_retroactive_simple() {
     // Rebuild so we can start querying
     map.rebuild();
 
-    // Query key 4
-    let version = map.get().unwrap().get_version(90);
-    assert!(version.is_none());
-    let version = map.get().unwrap().get_version(102).unwrap();
-    assert_eq!(version.image_coverage.query(4), Some("Image 1".to_string()));
-    let version = map.get().unwrap().get_version(107).unwrap();
-    assert_eq!(version.image_coverage.query(4), Some("Delta 1".to_string()));
-    let version = map.get().unwrap().get_version(115).unwrap();
-    assert_eq!(version.image_coverage.query(4), Some("Image 2".to_string()));
-    let version = map.get().unwrap().get_version(125).unwrap();
-    assert_eq!(version.image_coverage.query(4), Some("Image 3".to_string()));
+    {
+        let map = map.get().expect("rebuilt");
+
+        let version = map.get_version(90);
+        assert!(version.is_none());
+        let version = map.get_version(102).unwrap();
+        assert_eq!(version.image_coverage.query(4), Some("Image 1".to_string()));
+
+        let version = map.get_version(107).unwrap();
+        assert_eq!(version.image_coverage.query(4), Some("Image 1".to_string()));
+        assert_eq!(version.delta_coverage.query(4), Some("Delta 1".to_string()));
+
+        let version = map.get_version(115).unwrap();
+        assert_eq!(version.image_coverage.query(4), Some("Image 2".to_string()));
+
+        let version = map.get_version(125).unwrap();
+        assert_eq!(version.image_coverage.query(4), Some("Image 3".to_string()));
+    }
 
     // Remove Image 3
     map.remove(LayerKey {
@@ -576,8 +662,147 @@ fn test_retroactive_simple() {
     });
     map.rebuild();
 
-    // Check deletion worked
-    let version = map.get().unwrap().get_version(125).unwrap();
-    assert_eq!(version.image_coverage.query(4), Some("Image 2".to_string()));
-    assert_eq!(version.image_coverage.query(8), Some("Image 4".to_string()));
+    {
+        // Check deletion worked
+        let map = map.get().expect("rebuilt");
+        let version = map.get_version(125).unwrap();
+        assert_eq!(version.image_coverage.query(4), Some("Image 2".to_string()));
+        assert_eq!(version.image_coverage.query(8), Some("Image 4".to_string()));
+    }
+}
+
+#[test]
+fn test_retroactive_replacement() {
+    let mut map = BufferedHistoricLayerCoverage::new();
+
+    let keys = [
+        LayerKey {
+            key: 0..5,
+            lsn: 100..101,
+            is_image: true,
+        },
+        LayerKey {
+            key: 3..9,
+            lsn: 110..111,
+            is_image: true,
+        },
+        LayerKey {
+            key: 4..6,
+            lsn: 120..121,
+            is_image: true,
+        },
+    ];
+
+    let layers = [
+        "Image 1".to_string(),
+        "Image 2".to_string(),
+        "Image 3".to_string(),
+    ];
+
+    for (key, layer) in keys.iter().zip(layers.iter()) {
+        map.insert(key.to_owned(), layer.to_owned());
+    }
+
+    // rebuild is not necessary here, because replace works for both buffered updates and existing
+    // layers.
+
+    for (key, orig_layer) in keys.iter().zip(layers.iter()) {
+        let replacement = format!("Remote {orig_layer}");
+
+        // evict
+        let ret = map.replace(key, replacement.clone(), |l| l == orig_layer);
+        assert!(
+            matches!(ret, Replacement::Replaced { .. }),
+            "replace {orig_layer}: {ret:?}"
+        );
+        map.rebuild();
+
+        let at = key.lsn.end + 1;
+
+        let version = map.get().expect("rebuilt").get_version(at).unwrap();
+        assert_eq!(
+            version.image_coverage.query(4).as_deref(),
+            Some(replacement.as_str()),
+            "query for 4 at version {at} after eviction",
+        );
+
+        // download
+        let ret = map.replace(key, orig_layer.clone(), |l| l == &replacement);
+        assert!(
+            matches!(ret, Replacement::Replaced { .. }),
+            "replace {orig_layer} back: {ret:?}"
+        );
+        map.rebuild();
+        let version = map.get().expect("rebuilt").get_version(at).unwrap();
+        assert_eq!(
+            version.image_coverage.query(4).as_deref(),
+            Some(orig_layer.as_str()),
+            "query for 4 at version {at} after download",
+        );
+    }
+}
+
+#[test]
+fn missing_key_is_not_inserted_with_replace() {
+    let mut map = BufferedHistoricLayerCoverage::new();
+    let key = LayerKey {
+        key: 0..5,
+        lsn: 100..101,
+        is_image: true,
+    };
+
+    let ret = map.replace(&key, "should not replace", |_| true);
+    assert!(matches!(ret, Replacement::NotFound), "{ret:?}");
+    map.rebuild();
+    assert!(map
+        .get()
+        .expect("no changes to rebuild")
+        .get_version(102)
+        .is_none());
+}
+
+#[test]
+fn replacing_buffered_insert_and_remove() {
+    let mut map = BufferedHistoricLayerCoverage::new();
+    let key = LayerKey {
+        key: 0..5,
+        lsn: 100..101,
+        is_image: true,
+    };
+
+    map.insert(key.clone(), "Image 1");
+    let ret = map.replace(&key, "Remote Image 1", |&l| l == "Image 1");
+    assert!(
+        matches!(ret, Replacement::Replaced { in_buffered: true }),
+        "{ret:?}"
+    );
+    map.rebuild();
+
+    assert_eq!(
+        map.get()
+            .expect("rebuilt")
+            .get_version(102)
+            .unwrap()
+            .image_coverage
+            .query(4),
+        Some("Remote Image 1")
+    );
+
+    map.remove(key.clone());
+    let ret = map.replace(&key, "should not replace", |_| true);
+    assert!(
+        matches!(ret, Replacement::RemovalBuffered),
+        "cannot replace after scheduled remove: {ret:?}"
+    );
+
+    map.rebuild();
+
+    let ret = map.replace(&key, "should not replace", |_| true);
+    assert!(
+        matches!(ret, Replacement::NotFound),
+        "cannot replace after remove + rebuild: {ret:?}"
+    );
+
+    let at_version = map.get().expect("rebuilt").get_version(102);
+    assert!(at_version.is_none());
 }

pageserver/src/tenant/mgr.rs

@@ -16,6 +16,7 @@ use remote_storage::GenericRemoteStorage;
 use utils::crashsafe;
 
 use crate::config::PageServerConf;
+use crate::context::{DownloadBehavior, RequestContext};
 use crate::task_mgr::{self, TaskKind};
 use crate::tenant::config::TenantConfOpt;
 use crate::tenant::{Tenant, TenantState};
@@ -24,8 +25,35 @@ use crate::IGNORED_TENANT_FILE_NAME;
 use utils::fs_ext::PathExt;
 use utils::id::{TenantId, TimelineId};
 
-static TENANTS: Lazy<RwLock<HashMap<TenantId, Arc<Tenant>>>> =
-    Lazy::new(|| RwLock::new(HashMap::new()));
+/// The tenants known to the pageserver.
+/// The enum variants are used to distinguish the different states that the pageserver can be in.
+enum TenantsMap {
+    /// [`init_tenant_mgr`] is not done yet.
+    Initializing,
+    /// [`init_tenant_mgr`] is done, all on-disk tenants have been loaded.
+    /// New tenants can be added using [`tenant_map_insert`].
+    Open(HashMap<TenantId, Arc<Tenant>>),
+    /// The pageserver has entered shutdown mode via [`shutdown_all_tenants`].
+    /// Existing tenants are still accessible, but no new tenants can be created.
+    ShuttingDown(HashMap<TenantId, Arc<Tenant>>),
+}
+
+impl TenantsMap {
+    fn get(&self, tenant_id: &TenantId) -> Option<&Arc<Tenant>> {
+        match self {
+            TenantsMap::Initializing => None,
+            TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => m.get(tenant_id),
+        }
+    }
+    fn remove(&mut self, tenant_id: &TenantId) -> Option<Arc<Tenant>> {
+        match self {
+            TenantsMap::Initializing => None,
+            TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => m.remove(tenant_id),
+        }
+    }
+}
+
+static TENANTS: Lazy<RwLock<TenantsMap>> = Lazy::new(|| RwLock::new(TenantsMap::Initializing));
 
 /// Initialize repositories with locally available timelines.
 /// Timelines that are only partially available locally (remote storage has more data than this pageserver)
@@ -36,13 +64,16 @@ pub async fn init_tenant_mgr(
     remote_storage: Option<GenericRemoteStorage>,
 ) -> anyhow::Result<()> {
     // Scan local filesystem for attached tenants
-    let mut number_of_tenants = 0;
     let tenants_dir = conf.tenants_path();
 
+    let mut tenants = HashMap::new();
+
     let mut dir_entries = fs::read_dir(&tenants_dir)
         .await
         .with_context(|| format!("Failed to list tenants dir {tenants_dir:?}"))?;
 
+    let ctx = RequestContext::todo_child(TaskKind::Startup, DownloadBehavior::Warn);
+
     loop {
         match dir_entries.next_entry().await {
             Ok(None) => break,
@@ -86,10 +117,10 @@ pub async fn init_tenant_mgr(
                         conf,
                         &tenant_dir_path,
                         remote_storage.clone(),
+                        &ctx,
                     ) {
                         Ok(tenant) => {
-                            TENANTS.write().await.insert(tenant.tenant_id(), tenant);
-                            number_of_tenants += 1;
+                            tenants.insert(tenant.tenant_id(), tenant);
                         }
                         Err(e) => {
                             error!("Failed to collect tenant files from dir {tenants_dir:?} for entry {dir_entry:?}, reason: {e:#}");
@@ -108,7 +139,11 @@ pub async fn init_tenant_mgr(
         }
     }
 
-    info!("Processed {number_of_tenants} local tenants at startup");
+    info!("Processed {} local tenants at startup", tenants.len());
+
+    let mut tenants_map = TENANTS.write().await;
+    assert!(matches!(&*tenants_map, &TenantsMap::Initializing));
+    *tenants_map = TenantsMap::Open(tenants);
     Ok(())
 }
 
@@ -116,6 +151,7 @@ pub fn schedule_local_tenant_processing(
     conf: &'static PageServerConf,
     tenant_path: &Path,
     remote_storage: Option<GenericRemoteStorage>,
+    ctx: &RequestContext,
 ) -> anyhow::Result<Arc<Tenant>> {
     anyhow::ensure!(
         tenant_path.is_dir(),
@@ -150,7 +186,7 @@ pub fn schedule_local_tenant_processing(
     let tenant = if conf.tenant_attaching_mark_file_path(&tenant_id).exists() {
         info!("tenant {tenant_id} has attaching mark file, resuming its attach operation");
         if let Some(remote_storage) = remote_storage {
-            Tenant::spawn_attach(conf, tenant_id, remote_storage)
+            Tenant::spawn_attach(conf, tenant_id, remote_storage, ctx)
         } else {
             warn!("tenant {tenant_id} has attaching mark file, but pageserver has no remote storage configured");
             Tenant::create_broken_tenant(conf, tenant_id)
@@ -158,7 +194,7 @@ pub fn schedule_local_tenant_processing(
     } else {
         info!("tenant {tenant_id} is assumed to be loadable, starting load operation");
         // Start loading the tenant into memory. It will initially be in Loading state.
-        Tenant::spawn_load(conf, tenant_id, remote_storage)
+        Tenant::spawn_load(conf, tenant_id, remote_storage, ctx)
     };
     Ok(tenant)
 }
@@ -166,21 +202,44 @@ pub fn schedule_local_tenant_processing(
 ///
 /// Shut down all tenants. This runs as part of pageserver shutdown.
 ///
+/// NB: We leave the tenants in the map, so that they remain accessible through
+/// the management API until we shut it down. If we removed the shut-down tenants
+/// from the tenants map, the management API would return 404 for these tenants,
+/// because TenantsMap::get() now returns `None`.
+/// That could be easily misinterpreted by control plane, the consumer of the
+/// management API. For example, it could attach the tenant on a different pageserver.
+/// We would then be in split-brain once this pageserver restarts.
 pub async fn shutdown_all_tenants() {
+    // Prevent new tenants from being created.
     let tenants_to_shut_down = {
         let mut m = TENANTS.write().await;
-        let mut tenants_to_shut_down = Vec::with_capacity(m.len());
-        for (_, tenant) in m.drain() {
-            if tenant.is_active() {
-                // updates tenant state, forbidding new GC and compaction iterations from starting
-                tenant.set_stopping();
-                tenants_to_shut_down.push(tenant)
+        match &mut *m {
+            TenantsMap::Initializing => {
+                *m = TenantsMap::ShuttingDown(HashMap::default());
+                info!("tenants map is empty");
+                return;
+            }
+            TenantsMap::Open(tenants) => {
+                let tenants_clone = tenants.clone();
+                *m = TenantsMap::ShuttingDown(std::mem::take(tenants));
+                tenants_clone
+            }
+            TenantsMap::ShuttingDown(_) => {
+                error!("already shutting down, this function isn't supposed to be called more than once");
+                return;
             }
         }
-        drop(m);
-        tenants_to_shut_down
     };
 
+    let mut tenants_to_freeze_and_flush = Vec::with_capacity(tenants_to_shut_down.len());
+    for (_, tenant) in tenants_to_shut_down {
+        if tenant.is_active() {
+            // updates tenant state, forbidding new GC and compaction iterations from starting
+            tenant.set_stopping();
+            tenants_to_freeze_and_flush.push(tenant);
+        }
+    }
+
     // Shut down all existing walreceiver connections and stop accepting the new ones.
     task_mgr::shutdown_tasks(Some(TaskKind::WalReceiverManager), None, None).await;
 
@@ -192,7 +251,7 @@ pub async fn shutdown_all_tenants() {
     // should be no more activity in any of the repositories.
     //
     // On error, log it but continue with the shutdown for other tenants.
-    for tenant in tenants_to_shut_down {
+    for tenant in tenants_to_freeze_and_flush {
         let tenant_id = tenant.tenant_id();
         debug!("shutdown tenant {tenant_id}");
 
@@ -207,39 +266,41 @@ pub async fn create_tenant(
     tenant_conf: TenantConfOpt,
     tenant_id: TenantId,
     remote_storage: Option<GenericRemoteStorage>,
-) -> anyhow::Result<Option<Arc<Tenant>>> {
-    match TENANTS.write().await.entry(tenant_id) {
-        hash_map::Entry::Occupied(_) => {
-            debug!("tenant {tenant_id} already exists");
-            Ok(None)
-        }
-        hash_map::Entry::Vacant(v) => {
-            // Hold the write_tenants() lock, since all of this is local IO.
-            // If this section ever becomes contentious, introduce a new `TenantState::Creating`.
-            let tenant_directory = super::create_tenant_files(conf, tenant_conf, tenant_id)?;
-            let created_tenant =
-                schedule_local_tenant_processing(conf, &tenant_directory, remote_storage)?;
-            let crated_tenant_id = created_tenant.tenant_id();
-            anyhow::ensure!(
+    ctx: &RequestContext,
+) -> Result<Arc<Tenant>, TenantMapInsertError> {
+    tenant_map_insert(tenant_id, |vacant_entry| {
+        // We're holding the tenants lock in write mode while doing local IO.
+        // If this section ever becomes contentious, introduce a new `TenantState::Creating`
+        // and do the work in that state.
+        let tenant_directory = super::create_tenant_files(conf, tenant_conf, tenant_id)?;
+        let created_tenant =
+            schedule_local_tenant_processing(conf, &tenant_directory, remote_storage, ctx)?;
+        let crated_tenant_id = created_tenant.tenant_id();
+        anyhow::ensure!(
                 tenant_id == crated_tenant_id,
                 "loaded created tenant has unexpected tenant id (expect {tenant_id} != actual {crated_tenant_id})",
             );
-            v.insert(Arc::clone(&created_tenant));
-            Ok(Some(created_tenant))
-        }
-    }
+        vacant_entry.insert(Arc::clone(&created_tenant));
+        Ok(created_tenant)
+    }).await
 }
 
-pub async fn update_tenant_config(
+pub async fn set_new_tenant_config(
     conf: &'static PageServerConf,
-    tenant_conf: TenantConfOpt,
+    new_tenant_conf: TenantConfOpt,
     tenant_id: TenantId,
 ) -> anyhow::Result<()> {
     info!("configuring tenant {tenant_id}");
-    get_tenant(tenant_id, true)
-        .await?
-        .update_tenant_config(tenant_conf);
-    Tenant::persist_tenant_config(&conf.tenant_config_path(tenant_id), tenant_conf, false)?;
+    let tenant = get_tenant(tenant_id, true).await?;
+
+    let tenant_config_path = conf.tenant_config_path(tenant_id);
+    Tenant::persist_tenant_config(
+        &tenant.tenant_id(),
+        &tenant_config_path,
+        new_tenant_conf,
+        false,
+    )?;
+    tenant.set_new_tenant_config(new_tenant_conf);
     Ok(())
 }
 
@@ -260,10 +321,14 @@ pub async fn get_tenant(tenant_id: TenantId, active_only: bool) -> anyhow::Resul
     }
 }
 
-pub async fn delete_timeline(tenant_id: TenantId, timeline_id: TimelineId) -> anyhow::Result<()> {
+pub async fn delete_timeline(
+    tenant_id: TenantId,
+    timeline_id: TimelineId,
+    ctx: &RequestContext,
+) -> anyhow::Result<()> {
     match get_tenant(tenant_id, true).await {
         Ok(tenant) => {
-            tenant.delete_timeline(timeline_id).await?;
+            tenant.delete_timeline(timeline_id, ctx).await?;
         }
         Err(e) => anyhow::bail!("Cannot access tenant {tenant_id} in local tenant state: {e:?}"),
     }
@@ -291,8 +356,9 @@ pub async fn load_tenant(
     conf: &'static PageServerConf,
     tenant_id: TenantId,
     remote_storage: Option<GenericRemoteStorage>,
-) -> anyhow::Result<()> {
-    run_if_no_tenant_in_memory(tenant_id, |vacant_entry| {
+    ctx: &RequestContext,
+) -> Result<(), TenantMapInsertError> {
+    tenant_map_insert(tenant_id, |vacant_entry| {
         let tenant_path = conf.tenant_path(&tenant_id);
         let tenant_ignore_mark = conf.tenant_ignore_mark_file_path(tenant_id);
         if tenant_ignore_mark.exists() {
@@ -300,7 +366,7 @@ pub async fn load_tenant(
                 .with_context(|| format!("Failed to remove tenant ignore mark {tenant_ignore_mark:?} during tenant loading"))?;
         }
 
-        let new_tenant = schedule_local_tenant_processing(conf, &tenant_path, remote_storage)
+        let new_tenant = schedule_local_tenant_processing(conf, &tenant_path, remote_storage, ctx)
             .with_context(|| {
                 format!("Failed to schedule tenant processing in path {tenant_path:?}")
             })?;
@@ -329,16 +395,24 @@ pub async fn ignore_tenant(
     .await
 }
 
+#[derive(Debug, thiserror::Error)]
+pub enum TenantMapListError {
+    #[error("tenant map is still initiailizing")]
+    Initializing,
+}
+
 ///
 /// Get list of tenants, for the mgmt API
 ///
-pub async fn list_tenants() -> Vec<(TenantId, TenantState)> {
-    TENANTS
-        .read()
-        .await
-        .iter()
+pub async fn list_tenants() -> Result<Vec<(TenantId, TenantState)>, TenantMapListError> {
+    let tenants = TENANTS.read().await;
+    let m = match &*tenants {
+        TenantsMap::Initializing => return Err(TenantMapListError::Initializing),
+        TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => m,
+    };
+    Ok(m.iter()
         .map(|(id, tenant)| (*id, tenant.current_state()))
-        .collect()
+        .collect())
 }
 
 /// Execute Attach mgmt API command.
@@ -349,34 +423,62 @@ pub async fn attach_tenant(
     conf: &'static PageServerConf,
     tenant_id: TenantId,
     remote_storage: GenericRemoteStorage,
-) -> anyhow::Result<()> {
-    run_if_no_tenant_in_memory(tenant_id, |vacant_entry| {
+    ctx: &RequestContext,
+) -> Result<(), TenantMapInsertError> {
+    tenant_map_insert(tenant_id, |vacant_entry| {
         let tenant_path = conf.tenant_path(&tenant_id);
         anyhow::ensure!(
             !tenant_path.exists(),
             "Cannot attach tenant {tenant_id}, local tenant directory already exists"
         );
 
-        let tenant = Tenant::spawn_attach(conf, tenant_id, remote_storage);
+        let tenant = Tenant::spawn_attach(conf, tenant_id, remote_storage, ctx);
         vacant_entry.insert(tenant);
-
         Ok(())
     })
     .await
 }
 
-async fn run_if_no_tenant_in_memory<F, V>(tenant_id: TenantId, run: F) -> anyhow::Result<V>
+#[derive(Debug, thiserror::Error)]
+pub enum TenantMapInsertError {
+    #[error("tenant map is still initializing")]
+    StillInitializing,
+    #[error("tenant map is shutting down")]
+    ShuttingDown,
+    #[error("tenant {0} already exists, state: {1:?}")]
+    TenantAlreadyExists(TenantId, TenantState),
+    #[error(transparent)]
+    Closure(#[from] anyhow::Error),
+}
+
+/// Give the given closure access to the tenants map entry for the given `tenant_id`, iff that
+/// entry is vacant. The closure is responsible for creating the tenant object and inserting
+/// it into the tenants map through the vacnt entry that it receives as argument.
+///
+/// NB: the closure should return quickly because the current implementation of tenants map
+/// serializes access through an `RwLock`.
+async fn tenant_map_insert<F, V>(
+    tenant_id: TenantId,
+    insert_fn: F,
+) -> Result<V, TenantMapInsertError>
 where
     F: FnOnce(hash_map::VacantEntry<TenantId, Arc<Tenant>>) -> anyhow::Result<V>,
 {
-    match TENANTS.write().await.entry(tenant_id) {
-        hash_map::Entry::Occupied(e) => {
-            anyhow::bail!(
-                "tenant {tenant_id} already exists, state: {:?}",
-                e.get().current_state()
-            )
-        }
-        hash_map::Entry::Vacant(v) => run(v),
+    let mut guard = TENANTS.write().await;
+    let m = match &mut *guard {
+        TenantsMap::Initializing => return Err(TenantMapInsertError::StillInitializing),
+        TenantsMap::ShuttingDown(_) => return Err(TenantMapInsertError::ShuttingDown),
+        TenantsMap::Open(m) => m,
+    };
+    match m.entry(tenant_id) {
+        hash_map::Entry::Occupied(e) => Err(TenantMapInsertError::TenantAlreadyExists(
+            tenant_id,
+            e.get().current_state(),
+        )),
+        hash_map::Entry::Vacant(v) => match insert_fn(v) {
+            Ok(v) => Ok(v),
+            Err(e) => Err(TenantMapInsertError::Closure(e)),
+        },
     }
 }
 
@@ -438,20 +540,18 @@ where
     }
 }
 
-#[cfg(feature = "testing")]
 use {
     crate::repository::GcResult, pageserver_api::models::TimelineGcRequest,
     utils::http::error::ApiError,
 };
 
-#[cfg(feature = "testing")]
 pub async fn immediate_gc(
     tenant_id: TenantId,
     timeline_id: TimelineId,
     gc_req: TimelineGcRequest,
+    ctx: &RequestContext,
 ) -> Result<tokio::sync::oneshot::Receiver<Result<GcResult, anyhow::Error>>, ApiError> {
     let guard = TENANTS.read().await;
-
     let tenant = guard
         .get(&tenant_id)
         .map(Arc::clone)
@@ -462,7 +562,8 @@ pub async fn immediate_gc(
     // Use tenant's pitr setting
     let pitr = tenant.get_pitr_interval();
 
-    // Run in task_mgr to avoid race with detach operation
+    // Run in task_mgr to avoid race with tenant_detach operation
+    let ctx = ctx.detached_child(TaskKind::GarbageCollector, DownloadBehavior::Download);
     let (task_done, wait_task_done) = tokio::sync::oneshot::channel();
     task_mgr::spawn(
         &tokio::runtime::Handle::current(),
@@ -474,7 +575,7 @@ pub async fn immediate_gc(
         async move {
             fail::fail_point!("immediate_gc_task_pre");
             let result = tenant
-                .gc_iteration(Some(timeline_id), gc_horizon, pitr)
+                .gc_iteration(Some(timeline_id), gc_horizon, pitr, &ctx)
                 .instrument(info_span!("manual_gc", tenant = %tenant_id, timeline = %timeline_id))
                 .await;
                 // FIXME: `gc_iteration` can return an error for multiple reasons; we should handle it
@@ -497,6 +598,7 @@ pub async fn immediate_gc(
 pub async fn immediate_compact(
     tenant_id: TenantId,
     timeline_id: TimelineId,
+    ctx: &RequestContext,
 ) -> Result<tokio::sync::oneshot::Receiver<anyhow::Result<()>>, ApiError> {
     let guard = TENANTS.read().await;
 
@@ -510,7 +612,8 @@ pub async fn immediate_compact(
         .get_timeline(timeline_id, true)
         .map_err(ApiError::NotFound)?;
 
-    // Run in task_mgr to avoid race with detach operation
+    // Run in task_mgr to avoid race with tenant_detach operation
+    let ctx = ctx.detached_child(TaskKind::Compaction, DownloadBehavior::Download);
     let (task_done, wait_task_done) = tokio::sync::oneshot::channel();
     task_mgr::spawn(
         &tokio::runtime::Handle::current(),
@@ -523,7 +626,7 @@ pub async fn immediate_compact(
         false,
         async move {
             let result = timeline
-                .compact()
+                .compact(&ctx)
                 .instrument(
                     info_span!("manual_compact", tenant = %tenant_id, timeline = %timeline_id),
                 )

pageserver/src/tenant/remote_timeline_client.rs

@@ -571,14 +571,15 @@ impl RemoteTimelineClient {
         Ok(())
     }
 
-    ///
     /// Launch a delete operation in the background.
     ///
+    /// The operation does not modify local state but assumes the local files have already been
+    /// deleted, and is used to mirror those changes to remote.
+    ///
     /// Note: This schedules an index file upload before the deletions.  The
     /// deletion won't actually be performed, until any previously scheduled
     /// upload operations, and the index file upload, have completed
     /// succesfully.
-    ///
     pub fn schedule_layer_file_deletion(
         self: &Arc<Self>,
         names: &[LayerFileName],
@@ -1010,7 +1011,10 @@ impl RemoteTimelineClient {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::tenant::harness::{TenantHarness, TIMELINE_ID};
+    use crate::{
+        tenant::harness::{TenantHarness, TIMELINE_ID},
+        DEFAULT_PG_VERSION,
+    };
     use remote_storage::{RemoteStorageConfig, RemoteStorageKind};
     use std::{collections::HashSet, path::Path};
     use utils::lsn::Lsn;
@@ -1064,9 +1068,19 @@ mod tests {
     // Test scheduling
     #[test]
     fn upload_scheduling() -> anyhow::Result<()> {
+        // Use a current-thread runtime in the test
+        let runtime = Box::leak(Box::new(
+            tokio::runtime::Builder::new_current_thread()
+                .enable_all()
+                .build()?,
+        ));
+        let _entered = runtime.enter();
+
         let harness = TenantHarness::create("upload_scheduling")?;
+        let (tenant, ctx) = runtime.block_on(harness.load());
+        let _timeline =
+            tenant.create_empty_timeline(TIMELINE_ID, Lsn(0), DEFAULT_PG_VERSION, &ctx)?;
         let timeline_path = harness.timeline_path(&TIMELINE_ID);
-        std::fs::create_dir_all(&timeline_path)?;
 
         let remote_fs_dir = harness.conf.workdir.join("remote_fs");
         std::fs::create_dir_all(remote_fs_dir)?;
@@ -1084,14 +1098,6 @@ mod tests {
             storage: RemoteStorageKind::LocalFs(remote_fs_dir.clone()),
         };
 
-        // Use a current-thread runtime in the test
-        let runtime = Box::leak(Box::new(
-            tokio::runtime::Builder::new_current_thread()
-                .enable_all()
-                .build()?,
-        ));
-        let _entered = runtime.enter();
-
         // Test outline:
         //
         // Schedule upload of a bunch of layers. Check that they are started immediately, not queued
@@ -1130,18 +1136,29 @@ mod tests {
         client.init_upload_queue_for_empty_remote(&metadata)?;
 
         // Create a couple of dummy files,  schedule upload for them
-        let content_foo = dummy_contents("foo");
-        let content_bar = dummy_contents("bar");
-        std::fs::write(timeline_path.join("foo"), &content_foo)?;
-        std::fs::write(timeline_path.join("bar"), &content_bar)?;
+        let layer_file_name_1: LayerFileName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
+        let layer_file_name_2: LayerFileName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D9-00000000016B5A52".parse().unwrap();
+        let layer_file_name_3: LayerFileName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59DA-00000000016B5A53".parse().unwrap();
+        let content_1 = dummy_contents("foo");
+        let content_2 = dummy_contents("bar");
+        let content_3 = dummy_contents("baz");
+        std::fs::write(
+            timeline_path.join(layer_file_name_1.file_name()),
+            &content_1,
+        )?;
+        std::fs::write(
+            timeline_path.join(layer_file_name_2.file_name()),
+            &content_2,
+        )?;
+        std::fs::write(timeline_path.join(layer_file_name_3.file_name()), content_3)?;
 
         client.schedule_layer_file_upload(
-            &LayerFileName::Test("foo".to_owned()),
-            &LayerFileMetadata::new(content_foo.len() as u64),
+            &layer_file_name_1,
+            &LayerFileMetadata::new(content_1.len() as u64),
         )?;
         client.schedule_layer_file_upload(
-            &LayerFileName::Test("bar".to_owned()),
-            &LayerFileMetadata::new(content_bar.len() as u64),
+            &layer_file_name_2,
+            &LayerFileMetadata::new(content_2.len() as u64),
         )?;
 
         // Check that they are started immediately, not queued
@@ -1178,7 +1195,13 @@ mod tests {
 
         // Download back the index.json, and check that the list of files is correct
         let index_part = runtime.block_on(client.download_index_file())?;
-        assert_file_list(&index_part.timeline_layers, &["foo", "bar"]);
+        assert_file_list(
+            &index_part.timeline_layers,
+            &[
+                &layer_file_name_1.file_name(),
+                &layer_file_name_2.file_name(),
+            ],
+        );
         let downloaded_metadata = index_part.parse_metadata()?;
         assert_eq!(downloaded_metadata, metadata);
 
@@ -1186,10 +1209,10 @@ mod tests {
         let content_baz = dummy_contents("baz");
         std::fs::write(timeline_path.join("baz"), &content_baz)?;
         client.schedule_layer_file_upload(
-            &LayerFileName::Test("baz".to_owned()),
+            &layer_file_name_3,
             &LayerFileMetadata::new(content_baz.len() as u64),
         )?;
-        client.schedule_layer_file_deletion(&[LayerFileName::Test("foo".to_owned())])?;
+        client.schedule_layer_file_deletion(&[layer_file_name_1.clone()])?;
         {
             let mut guard = client.upload_queue.lock().unwrap();
             let upload_queue = guard.initialized_mut().unwrap();
@@ -1201,12 +1224,26 @@ mod tests {
             assert!(upload_queue.num_inprogress_deletions == 0);
             assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 0);
         }
-        assert_remote_files(&["foo", "bar", "index_part.json"], &remote_timeline_dir);
+        assert_remote_files(
+            &[
+                &layer_file_name_1.file_name(),
+                &layer_file_name_2.file_name(),
+                "index_part.json",
+            ],
+            &remote_timeline_dir,
+        );
 
         // Finish them
         runtime.block_on(client.wait_completion())?;
 
-        assert_remote_files(&["bar", "baz", "index_part.json"], &remote_timeline_dir);
+        assert_remote_files(
+            &[
+                &layer_file_name_2.file_name(),
+                &layer_file_name_3.file_name(),
+                "index_part.json",
+            ],
+            &remote_timeline_dir,
+        );
 
         Ok(())
     }

pageserver/src/tenant/remote_timeline_client/index.rs

@@ -8,7 +8,8 @@ use serde::{Deserialize, Serialize};
 use serde_with::{serde_as, DisplayFromStr};
 use tracing::warn;
 
-use crate::tenant::{metadata::TimelineMetadata, storage_layer::LayerFileName};
+use crate::tenant::metadata::TimelineMetadata;
+use crate::tenant::storage_layer::LayerFileName;
 
 use utils::lsn::Lsn;
 
@@ -274,7 +275,7 @@ mod tests {
             "timeline_layers":["000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9"],
             "layer_metadata":{
                 "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9": { "file_size": 25600000 },
-                "LAYER_FILE_NAME::test/not_a_real_layer_but_adding_coverage": { "file_size": 9007199254741001 }
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51": { "file_size": 9007199254741001 }
             },
             "disk_consistent_lsn":"0/16960E8",
             "metadata_bytes":[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
@@ -288,7 +289,7 @@ mod tests {
                 ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9".parse().unwrap(), IndexLayerMetadata {
                     file_size: Some(25600000),
                 }),
-                (LayerFileName::new_test("not_a_real_layer_but_adding_coverage"), IndexLayerMetadata {
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), IndexLayerMetadata {
                     // serde_json should always parse this but this might be a double with jq for
                     // example.
                     file_size: Some(9007199254741001),
@@ -312,7 +313,7 @@ mod tests {
             "missing_layers":["This shouldn't fail deserialization"],
             "layer_metadata":{
                 "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9": { "file_size": 25600000 },
-                "LAYER_FILE_NAME::test/not_a_real_layer_but_adding_coverage": { "file_size": 9007199254741001 }
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51": { "file_size": 9007199254741001 }
             },
             "disk_consistent_lsn":"0/16960E8",
             "metadata_bytes":[112,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
@@ -326,7 +327,7 @@ mod tests {
                 ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9".parse().unwrap(), IndexLayerMetadata {
                     file_size: Some(25600000),
                 }),
-                (LayerFileName::new_test("not_a_real_layer_but_adding_coverage"), IndexLayerMetadata {
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), IndexLayerMetadata {
                     // serde_json should always parse this but this might be a double with jq for
                     // example.
                     file_size: Some(9007199254741001),

pageserver/src/tenant/storage_layer.rs

@@ -1,18 +1,30 @@
 //! Common traits and structs for layers
 
-mod delta_layer;
+pub mod delta_layer;
 mod filename;
 mod image_layer;
 mod inmemory_layer;
 mod remote_layer;
 
+use crate::config::PageServerConf;
+use crate::context::RequestContext;
 use crate::repository::{Key, Value};
+use crate::task_mgr::TaskKind;
 use crate::walrecord::NeonWalRecord;
 use anyhow::Result;
 use bytes::Bytes;
+use either::Either;
+use enum_map::EnumMap;
+use enumset::EnumSet;
+use pageserver_api::models::LayerAccessKind;
+use pageserver_api::models::{
+    HistoricLayerInfo, LayerResidenceEvent, LayerResidenceEventReason, LayerResidenceStatus,
+};
 use std::ops::Range;
 use std::path::PathBuf;
-use std::sync::Arc;
+use std::sync::{Arc, Mutex};
+use std::time::{SystemTime, UNIX_EPOCH};
+use utils::history_buffer::HistoryBufferWithDropCounter;
 
 use utils::{
     id::{TenantId, TimelineId},
@@ -20,7 +32,7 @@ use utils::{
 };
 
 pub use delta_layer::{DeltaLayer, DeltaLayerWriter};
-pub use filename::{DeltaFileName, ImageFileName, LayerFileName, PathOrConf};
+pub use filename::{DeltaFileName, ImageFileName, LayerFileName};
 pub use image_layer::{ImageLayer, ImageLayerWriter};
 pub use inmemory_layer::InMemoryLayer;
 pub use remote_layer::RemoteLayer;
@@ -80,9 +92,191 @@ pub enum ValueReconstructResult {
     Missing,
 }
 
+#[derive(Debug)]
+pub struct LayerAccessStats(Mutex<LayerAccessStatsLocked>);
+
+/// This struct holds two instances of [`LayerAccessStatsInner`].
+/// Accesses are recorded to both instances.
+/// The `for_scraping_api`instance can be reset from the management API via [`LayerAccessStatsReset`].
+/// The `for_eviction_policy` is never reset.
+#[derive(Debug, Default, Clone)]
+struct LayerAccessStatsLocked {
+    for_scraping_api: LayerAccessStatsInner,
+    for_eviction_policy: LayerAccessStatsInner,
+}
+
+impl LayerAccessStatsLocked {
+    fn iter_mut(&mut self) -> impl Iterator<Item = &mut LayerAccessStatsInner> {
+        [&mut self.for_scraping_api, &mut self.for_eviction_policy].into_iter()
+    }
+}
+
+#[derive(Debug, Default, Clone)]
+struct LayerAccessStatsInner {
+    first_access: Option<LayerAccessStatFullDetails>,
+    count_by_access_kind: EnumMap<LayerAccessKind, u64>,
+    task_kind_flag: EnumSet<TaskKind>,
+    last_accesses: HistoryBufferWithDropCounter<LayerAccessStatFullDetails, 16>,
+    last_residence_changes: HistoryBufferWithDropCounter<LayerResidenceEvent, 16>,
+}
+
+#[derive(Debug, Clone, Copy)]
+pub(super) struct LayerAccessStatFullDetails {
+    pub(super) when: SystemTime,
+    pub(super) task_kind: TaskKind,
+    pub(super) access_kind: LayerAccessKind,
+}
+
+#[derive(Clone, Copy, strum_macros::EnumString)]
+pub enum LayerAccessStatsReset {
+    NoReset,
+    JustTaskKindFlags,
+    AllStats,
+}
+
+fn system_time_to_millis_since_epoch(ts: &SystemTime) -> u64 {
+    ts.duration_since(UNIX_EPOCH)
+        .expect("better to die in this unlikely case than report false stats")
+        .as_millis()
+        .try_into()
+        .expect("64 bits is enough for few more years")
+}
+
+impl LayerAccessStatFullDetails {
+    fn as_api_model(&self) -> pageserver_api::models::LayerAccessStatFullDetails {
+        let Self {
+            when,
+            task_kind,
+            access_kind,
+        } = self;
+        pageserver_api::models::LayerAccessStatFullDetails {
+            when_millis_since_epoch: system_time_to_millis_since_epoch(when),
+            task_kind: task_kind.into(), // into static str, powered by strum_macros
+            access_kind: *access_kind,
+        }
+    }
+}
+
+impl LayerAccessStats {
+    pub(crate) fn for_loading_layer(status: LayerResidenceStatus) -> Self {
+        let new = LayerAccessStats(Mutex::new(LayerAccessStatsLocked::default()));
+        new.record_residence_event(status, LayerResidenceEventReason::LayerLoad);
+        new
+    }
+
+    pub(crate) fn for_new_layer_file() -> Self {
+        let new = LayerAccessStats(Mutex::new(LayerAccessStatsLocked::default()));
+        new.record_residence_event(
+            LayerResidenceStatus::Resident,
+            LayerResidenceEventReason::LayerCreate,
+        );
+        new
+    }
+
+    /// Creates a clone of `self` and records `new_status` in the clone.
+    /// The `new_status` is not recorded in `self`
+    pub(crate) fn clone_for_residence_change(
+        &self,
+        new_status: LayerResidenceStatus,
+    ) -> LayerAccessStats {
+        let clone = {
+            let inner = self.0.lock().unwrap();
+            inner.clone()
+        };
+        let new = LayerAccessStats(Mutex::new(clone));
+        new.record_residence_event(new_status, LayerResidenceEventReason::ResidenceChange);
+        new
+    }
+
+    fn record_residence_event(
+        &self,
+        status: LayerResidenceStatus,
+        reason: LayerResidenceEventReason,
+    ) {
+        let mut locked = self.0.lock().unwrap();
+        locked.iter_mut().for_each(|inner| {
+            inner
+                .last_residence_changes
+                .write(LayerResidenceEvent::new(status, reason))
+        });
+    }
+
+    fn record_access(&self, access_kind: LayerAccessKind, task_kind: TaskKind) {
+        let this_access = LayerAccessStatFullDetails {
+            when: SystemTime::now(),
+            task_kind,
+            access_kind,
+        };
+
+        let mut locked = self.0.lock().unwrap();
+        locked.iter_mut().for_each(|inner| {
+            inner.first_access.get_or_insert(this_access);
+            inner.count_by_access_kind[access_kind] += 1;
+            inner.task_kind_flag |= task_kind;
+            inner.last_accesses.write(this_access);
+        })
+    }
+
+    fn as_api_model(
+        &self,
+        reset: LayerAccessStatsReset,
+    ) -> pageserver_api::models::LayerAccessStats {
+        let mut locked = self.0.lock().unwrap();
+        let inner = &mut locked.for_scraping_api;
+        let LayerAccessStatsInner {
+            first_access,
+            count_by_access_kind,
+            task_kind_flag,
+            last_accesses,
+            last_residence_changes,
+        } = inner;
+        let ret = pageserver_api::models::LayerAccessStats {
+            access_count_by_access_kind: count_by_access_kind
+                .iter()
+                .map(|(kind, count)| (kind, *count))
+                .collect(),
+            task_kind_access_flag: task_kind_flag
+                .iter()
+                .map(|task_kind| task_kind.into()) // into static str, powered by strum_macros
+                .collect(),
+            first: first_access.as_ref().map(|a| a.as_api_model()),
+            accesses_history: last_accesses.map(|m| m.as_api_model()),
+            residence_events_history: last_residence_changes.clone(),
+        };
+        match reset {
+            LayerAccessStatsReset::NoReset => (),
+            LayerAccessStatsReset::JustTaskKindFlags => {
+                inner.task_kind_flag.clear();
+            }
+            LayerAccessStatsReset::AllStats => {
+                *inner = LayerAccessStatsInner::default();
+            }
+        }
+        ret
+    }
+
+    pub(super) fn most_recent_access_or_residence_event(
+        &self,
+    ) -> Either<LayerAccessStatFullDetails, LayerResidenceEvent> {
+        let locked = self.0.lock().unwrap();
+        let inner = &locked.for_eviction_policy;
+        match inner.last_accesses.recent() {
+            Some(a) => Either::Left(*a),
+            None => match inner.last_residence_changes.recent() {
+                Some(e) => Either::Right(e.clone()),
+                None => unreachable!("constructors for LayerAccessStats ensure that there's always a residence change event"),
+            }
+        }
+    }
+}
+
 /// Supertrait of the [`Layer`] trait that captures the bare minimum interface
 /// required by [`LayerMap`].
-pub trait Layer: Send + Sync {
+///
+/// All layers should implement a minimal `std::fmt::Debug` without tenant or
+/// timeline names, because those are known in the context of which the layers
+/// are used in (timeline).
+pub trait Layer: std::fmt::Debug + Send + Sync {
     /// Range of keys that this layer covers
     fn get_key_range(&self) -> Range<Key>;
 
@@ -117,13 +311,14 @@ pub trait Layer: Send + Sync {
         key: Key,
         lsn_range: Range<Lsn>,
         reconstruct_data: &mut ValueReconstructState,
+        ctx: &RequestContext,
     ) -> Result<ValueReconstructResult>;
 
     /// A short ID string that uniquely identifies the given layer within a [`LayerMap`].
     fn short_id(&self) -> String;
 
     /// Dump summary of the contents of the layer to stdout
-    fn dump(&self, verbose: bool) -> Result<()>;
+    fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()>;
 }
 
 /// Returned by [`Layer::iter`]
@@ -144,8 +339,7 @@ pub type LayerKeyIter<'i> = Box<dyn Iterator<Item = (Key, Lsn, u64)> + 'i>;
 /// Furthermore, there are two kinds of on-disk layers: delta and image layers.
 /// A delta layer contains all modifications within a range of LSNs and keys.
 /// An image layer is a snapshot of all the data in a key-range, at a single
-/// LSN
-///
+/// LSN.
 pub trait PersistentLayer: Layer {
     fn get_tenant_id(&self) -> TenantId;
 
@@ -161,11 +355,11 @@ pub trait PersistentLayer: Layer {
     fn local_path(&self) -> Option<PathBuf>;
 
     /// Iterate through all keys and values stored in the layer
-    fn iter(&self) -> Result<LayerIter<'_>>;
+    fn iter(&self, ctx: &RequestContext) -> Result<LayerIter<'_>>;
 
     /// Iterate through all keys stored in the layer. Returns key, lsn and value size
     /// It is used only for compaction and so is currently implemented only for DeltaLayer
-    fn key_iter(&self) -> Result<LayerKeyIter<'_>> {
+    fn key_iter(&self, _ctx: &RequestContext) -> Result<LayerKeyIter<'_>> {
         panic!("Not implemented")
     }
 
@@ -185,6 +379,10 @@ pub trait PersistentLayer: Layer {
     /// Should not change over the lifetime of the layer object because
     /// current_physical_size is computed as the som of this value.
     fn file_size(&self) -> Option<u64>;
+
+    fn info(&self, reset: LayerAccessStatsReset) -> HistoricLayerInfo;
+
+    fn access_stats(&self) -> &LayerAccessStats;
 }
 
 pub fn downcast_remote_layer(
@@ -197,15 +395,11 @@ pub fn downcast_remote_layer(
     }
 }
 
-impl std::fmt::Debug for dyn Layer {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        f.debug_struct("Layer")
-            .field("short_id", &self.short_id())
-            .finish()
-    }
-}
-
 /// Holds metadata about a layer without any content. Used mostly for testing.
+///
+/// To use filenames as fixtures, parse them as [`LayerFileName`] then convert from that to a
+/// LayerDescriptor.
+#[derive(Clone, Debug)]
 pub struct LayerDescriptor {
     pub key: Range<Key>,
     pub lsn: Range<Lsn>,
@@ -231,6 +425,7 @@ impl Layer for LayerDescriptor {
         _key: Key,
         _lsn_range: Range<Lsn>,
         _reconstruct_data: &mut ValueReconstructState,
+        _ctx: &RequestContext,
     ) -> Result<ValueReconstructResult> {
         todo!("This method shouldn't be part of the Layer trait")
     }
@@ -239,7 +434,65 @@ impl Layer for LayerDescriptor {
         self.short_id.clone()
     }
 
-    fn dump(&self, _verbose: bool) -> Result<()> {
+    fn dump(&self, _verbose: bool, _ctx: &RequestContext) -> Result<()> {
         todo!()
     }
 }
+
+impl From<DeltaFileName> for LayerDescriptor {
+    fn from(value: DeltaFileName) -> Self {
+        let short_id = value.to_string();
+        LayerDescriptor {
+            key: value.key_range,
+            lsn: value.lsn_range,
+            is_incremental: true,
+            short_id,
+        }
+    }
+}
+
+impl From<ImageFileName> for LayerDescriptor {
+    fn from(value: ImageFileName) -> Self {
+        let short_id = value.to_string();
+        let lsn = value.lsn_as_range();
+        LayerDescriptor {
+            key: value.key_range,
+            lsn,
+            is_incremental: false,
+            short_id,
+        }
+    }
+}
+
+impl From<LayerFileName> for LayerDescriptor {
+    fn from(value: LayerFileName) -> Self {
+        match value {
+            LayerFileName::Delta(d) => Self::from(d),
+            LayerFileName::Image(i) => Self::from(i),
+        }
+    }
+}
+
+/// Helper enum to hold a PageServerConf, or a path
+///
+/// This is used by DeltaLayer and ImageLayer. Normally, this holds a reference to the
+/// global config, and paths to layer files are constructed using the tenant/timeline
+/// path from the config. But in the 'pageserver_binutils' binary, we need to construct a Layer
+/// struct for a file on disk, without having a page server running, so that we have no
+/// config. In that case, we use the Path variant to hold the full path to the file on
+/// disk.
+enum PathOrConf {
+    Path(PathBuf),
+    Conf(&'static PageServerConf),
+}
+
+/// Range wrapping newtype, which uses display to render Debug.
+///
+/// Useful with `Key`, which has too verbose `{:?}` for printing multiple layers.
+struct RangeDisplayDebug<'a, T: std::fmt::Display>(&'a Range<T>);
+
+impl<'a, T: std::fmt::Display> std::fmt::Debug for RangeDisplayDebug<'a, T> {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}..{}", self.0.start, self.0.end)
+    }
+}

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -24,6 +24,7 @@
 //! "values" part.
 //!
 use crate::config::PageServerConf;
+use crate::context::RequestContext;
 use crate::page_cache::{PageReadGuard, PAGE_SZ};
 use crate::repository::{Key, Value, KEY_SIZE};
 use crate::tenant::blob_io::{BlobCursor, BlobWriter, WriteBlobWriter};
@@ -36,6 +37,7 @@ use crate::virtual_file::VirtualFile;
 use crate::{walrecord, TEMP_FILE_SUFFIX};
 use crate::{DELTA_FILE_MAGIC, STORAGE_FORMAT_VERSION};
 use anyhow::{bail, ensure, Context, Result};
+use pageserver_api::models::{HistoricLayerInfo, LayerAccessKind};
 use rand::{distributions::Alphanumeric, Rng};
 use serde::{Deserialize, Serialize};
 use std::fs::{self, File};
@@ -53,7 +55,10 @@ use utils::{
     lsn::Lsn,
 };
 
-use super::{DeltaFileName, Layer, LayerFileName, LayerIter, LayerKeyIter, PathOrConf};
+use super::{
+    DeltaFileName, Layer, LayerAccessStats, LayerAccessStatsReset, LayerFileName, LayerIter,
+    LayerKeyIter, LayerResidenceStatus, PathOrConf,
+};
 
 ///
 /// Header stored in the beginning of the file
@@ -62,7 +67,7 @@ use super::{DeltaFileName, Layer, LayerFileName, LayerIter, LayerKeyIter, PathOr
 /// the 'index' starts at the block indicated by 'index_start_blk'
 ///
 #[derive(Debug, Serialize, Deserialize, PartialEq, Eq)]
-struct Summary {
+pub struct Summary {
     /// Magic value to identify this as a neon delta file. Always DELTA_FILE_MAGIC.
     magic: u16,
     format_version: u16,
@@ -73,9 +78,9 @@ struct Summary {
     lsn_range: Range<Lsn>,
 
     /// Block number where the 'index' part of the file begins.
-    index_start_blk: u32,
+    pub index_start_blk: u32,
     /// Block within the 'index', where the B-tree root page is stored
-    index_root_blk: u32,
+    pub index_root_blk: u32,
 }
 
 impl From<&DeltaLayer> for Summary {
@@ -125,7 +130,7 @@ impl BlobRef {
     }
 }
 
-const DELTA_KEY_SIZE: usize = KEY_SIZE + 8;
+pub const DELTA_KEY_SIZE: usize = KEY_SIZE + 8;
 struct DeltaKey([u8; DELTA_KEY_SIZE]);
 
 ///
@@ -165,14 +170,13 @@ impl DeltaKey {
     }
 }
 
+/// DeltaLayer is the in-memory data structure associated with an on-disk delta
+/// file.
 ///
-/// DeltaLayer is the in-memory data structure associated with an
-/// on-disk delta file.  We keep a DeltaLayer in memory for each
-/// file, in the LayerMap. If a layer is in "loaded" state, we have a
-/// copy of the index in memory, in 'inner'. Otherwise the struct is
-/// just a placeholder for a file that exists on disk, and it needs to
-/// be loaded before using it in queries.
-///
+/// We keep a DeltaLayer in memory for each file, in the LayerMap. If a layer
+/// is in "loaded" state, we have a copy of the index in memory, in 'inner'.
+/// Otherwise the struct is just a placeholder for a file that exists on disk,
+/// and it needs to be loaded before using it in queries.
 pub struct DeltaLayer {
     path_or_conf: PathOrConf,
 
@@ -183,9 +187,24 @@ pub struct DeltaLayer {
 
     pub file_size: u64,
 
+    access_stats: LayerAccessStats,
+
     inner: RwLock<DeltaLayerInner>,
 }
 
+impl std::fmt::Debug for DeltaLayer {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        use super::RangeDisplayDebug;
+
+        f.debug_struct("DeltaLayer")
+            .field("key_range", &RangeDisplayDebug(&self.key_range))
+            .field("lsn_range", &self.lsn_range)
+            .field("file_size", &self.file_size)
+            .field("inner", &self.inner)
+            .finish()
+    }
+}
+
 pub struct DeltaLayerInner {
     /// If false, the fields below have not been loaded into memory yet.
     loaded: bool,
@@ -198,6 +217,16 @@ pub struct DeltaLayerInner {
     file: Option<FileBlockReader<VirtualFile>>,
 }
 
+impl std::fmt::Debug for DeltaLayerInner {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("DeltaLayerInner")
+            .field("loaded", &self.loaded)
+            .field("index_start_blk", &self.index_start_blk)
+            .field("index_root_blk", &self.index_root_blk)
+            .finish()
+    }
+}
+
 impl Layer for DeltaLayer {
     fn get_key_range(&self) -> Range<Key> {
         self.key_range.clone()
@@ -214,7 +243,7 @@ impl Layer for DeltaLayer {
         self.filename().file_name()
     }
     /// debugging function to print out the contents of the layer
-    fn dump(&self, verbose: bool) -> Result<()> {
+    fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {
         println!(
             "----- delta layer for ten {} tli {} keys {}-{} lsn {}-{} ----",
             self.tenant_id,
@@ -229,7 +258,7 @@ impl Layer for DeltaLayer {
             return Ok(());
         }
 
-        let inner = self.load()?;
+        let inner = self.load(LayerAccessKind::Dump, ctx)?;
 
         println!(
             "index_start_blk: {}, root {}",
@@ -293,6 +322,7 @@ impl Layer for DeltaLayer {
         key: Key,
         lsn_range: Range<Lsn>,
         reconstruct_state: &mut ValueReconstructState,
+        ctx: &RequestContext,
     ) -> anyhow::Result<ValueReconstructResult> {
         ensure!(lsn_range.start >= self.lsn_range.start);
         let mut need_image = true;
@@ -301,7 +331,7 @@ impl Layer for DeltaLayer {
 
         {
             // Open the file and lock the metadata in memory
-            let inner = self.load()?;
+            let inner = self.load(LayerAccessKind::GetValueReconstructData, ctx)?;
 
             // Scan the page versions backwards, starting from `lsn`.
             let file = inner.file.as_ref().unwrap();
@@ -391,16 +421,18 @@ impl PersistentLayer for DeltaLayer {
         Some(self.path())
     }
 
-    fn iter(&self) -> Result<LayerIter<'_>> {
-        let inner = self.load().context("load delta layer")?;
+    fn iter(&self, ctx: &RequestContext) -> Result<LayerIter<'_>> {
+        let inner = self
+            .load(LayerAccessKind::KeyIter, ctx)
+            .context("load delta layer")?;
         Ok(match DeltaValueIter::new(inner) {
             Ok(iter) => Box::new(iter),
             Err(err) => Box::new(std::iter::once(Err(err))),
         })
     }
 
-    fn key_iter(&self) -> Result<LayerKeyIter<'_>> {
-        let inner = self.load()?;
+    fn key_iter(&self, ctx: &RequestContext) -> Result<LayerKeyIter<'_>> {
+        let inner = self.load(LayerAccessKind::KeyIter, ctx)?;
         Ok(Box::new(
             DeltaKeyIter::new(inner).context("Layer index is corrupted")?,
         ))
@@ -415,6 +447,26 @@ impl PersistentLayer for DeltaLayer {
     fn file_size(&self) -> Option<u64> {
         Some(self.file_size)
     }
+
+    fn info(&self, reset: LayerAccessStatsReset) -> HistoricLayerInfo {
+        let layer_file_name = self.filename().file_name();
+        let lsn_range = self.get_lsn_range();
+
+        let access_stats = self.access_stats.as_api_model(reset);
+
+        HistoricLayerInfo::Delta {
+            layer_file_name,
+            layer_file_size: Some(self.file_size),
+            lsn_start: lsn_range.start,
+            lsn_end: lsn_range.end,
+            remote: false,
+            access_stats,
+        }
+    }
+
+    fn access_stats(&self) -> &LayerAccessStats {
+        &self.access_stats
+    }
 }
 
 impl DeltaLayer {
@@ -459,7 +511,13 @@ impl DeltaLayer {
     /// Open the underlying file and read the metadata into memory, if it's
     /// not loaded already.
     ///
-    fn load(&self) -> Result<RwLockReadGuard<DeltaLayerInner>> {
+    fn load(
+        &self,
+        access_kind: LayerAccessKind,
+        ctx: &RequestContext,
+    ) -> Result<RwLockReadGuard<DeltaLayerInner>> {
+        self.access_stats
+            .record_access(access_kind, ctx.task_kind());
         loop {
             // Quick exit if already loaded
             let inner = self.inner.read().unwrap();
@@ -540,6 +598,7 @@ impl DeltaLayer {
         tenant_id: TenantId,
         filename: &DeltaFileName,
         file_size: u64,
+        access_stats: LayerAccessStats,
     ) -> DeltaLayer {
         DeltaLayer {
             path_or_conf: PathOrConf::Conf(conf),
@@ -548,6 +607,7 @@ impl DeltaLayer {
             key_range: filename.key_range.clone(),
             lsn_range: filename.lsn_range.clone(),
             file_size,
+            access_stats,
             inner: RwLock::new(DeltaLayerInner {
                 loaded: false,
                 file: None,
@@ -577,6 +637,7 @@ impl DeltaLayer {
             key_range: summary.key_range,
             lsn_range: summary.lsn_range,
             file_size: metadata.len(),
+            access_stats: LayerAccessStats::for_loading_layer(LayerResidenceStatus::Resident),
             inner: RwLock::new(DeltaLayerInner {
                 loaded: false,
                 file: None,
@@ -747,6 +808,7 @@ impl DeltaLayerWriterInner {
             key_range: self.key_start..key_end,
             lsn_range: self.lsn_range.clone(),
             file_size: metadata.len(),
+            access_stats: LayerAccessStats::for_new_layer_file(),
             inner: RwLock::new(DeltaLayerInner {
                 loaded: false,
                 file: None,

pageserver/src/tenant/storage_layer/filename.rs

@@ -1,23 +1,32 @@
 //!
 //! Helper functions for dealing with filenames of the image and delta layer files.
 //!
-use crate::config::PageServerConf;
 use crate::repository::Key;
 use std::cmp::Ordering;
 use std::fmt;
 use std::ops::Range;
-use std::path::PathBuf;
 use std::str::FromStr;
 
 use utils::lsn::Lsn;
 
 // Note: Timeline::load_layer_map() relies on this sort order
-#[derive(Debug, PartialEq, Eq, Clone, Hash)]
+#[derive(PartialEq, Eq, Clone, Hash)]
 pub struct DeltaFileName {
     pub key_range: Range<Key>,
     pub lsn_range: Range<Lsn>,
 }
 
+impl std::fmt::Debug for DeltaFileName {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        use super::RangeDisplayDebug;
+
+        f.debug_struct("DeltaFileName")
+            .field("key_range", &RangeDisplayDebug(&self.key_range))
+            .field("lsn_range", &self.lsn_range)
+            .finish()
+    }
+}
+
 impl PartialOrd for DeltaFileName {
     fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
         Some(self.cmp(other))
@@ -102,12 +111,23 @@ impl fmt::Display for DeltaFileName {
     }
 }
 
-#[derive(Debug, PartialEq, Eq, Clone, Hash)]
+#[derive(PartialEq, Eq, Clone, Hash)]
 pub struct ImageFileName {
     pub key_range: Range<Key>,
     pub lsn: Lsn,
 }
 
+impl std::fmt::Debug for ImageFileName {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        use super::RangeDisplayDebug;
+
+        f.debug_struct("ImageFileName")
+            .field("key_range", &RangeDisplayDebug(&self.key_range))
+            .field("lsn", &self.lsn)
+            .finish()
+    }
+}
+
 impl PartialOrd for ImageFileName {
     fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
         Some(self.cmp(other))
@@ -130,6 +150,13 @@ impl Ord for ImageFileName {
     }
 }
 
+impl ImageFileName {
+    pub fn lsn_as_range(&self) -> Range<Lsn> {
+        // Saves from having to copypaste this all over
+        self.lsn..(self.lsn + 1)
+    }
+}
+
 ///
 /// Represents the filename of an ImageLayer
 ///
@@ -177,49 +204,32 @@ impl fmt::Display for ImageFileName {
 pub enum LayerFileName {
     Image(ImageFileName),
     Delta(DeltaFileName),
-    #[cfg(test)]
-    Test(String),
 }
 
 impl LayerFileName {
     pub fn file_name(&self) -> String {
         match self {
-            LayerFileName::Image(fname) => format!("{fname}"),
-            LayerFileName::Delta(fname) => format!("{fname}"),
-            #[cfg(test)]
-            LayerFileName::Test(fname) => fname.to_string(),
+            Self::Image(fname) => fname.to_string(),
+            Self::Delta(fname) => fname.to_string(),
         }
     }
-    #[cfg(test)]
-    pub(crate) fn new_test(name: &str) -> LayerFileName {
-        LayerFileName::Test(name.to_owned())
-    }
 }
 
 impl From<ImageFileName> for LayerFileName {
     fn from(fname: ImageFileName) -> Self {
-        LayerFileName::Image(fname)
+        Self::Image(fname)
     }
 }
 impl From<DeltaFileName> for LayerFileName {
     fn from(fname: DeltaFileName) -> Self {
-        LayerFileName::Delta(fname)
+        Self::Delta(fname)
     }
 }
 
-// include a `/` in the name as an additional layer of robustness
-// because `/` chars are not allowed in UNIX paths
-#[cfg(test)]
-const LAYER_FILE_NAME_TEST_PREFIX: &str = "LAYER_FILE_NAME::test/";
-
 impl FromStr for LayerFileName {
     type Err = String;
 
     fn from_str(value: &str) -> Result<Self, Self::Err> {
-        #[cfg(test)]
-        if let Some(value) = value.strip_prefix(LAYER_FILE_NAME_TEST_PREFIX) {
-            return Ok(LayerFileName::Test(value.to_owned()));
-        }
         let delta = DeltaFileName::parse_str(value);
         let image = ImageFileName::parse_str(value);
         let ok = match (delta, image) {
@@ -228,8 +238,8 @@ impl FromStr for LayerFileName {
                     "neither delta nor image layer file name: {value:?}"
                 ))
             }
-            (Some(delta), None) => LayerFileName::Delta(delta),
-            (None, Some(image)) => LayerFileName::Image(image),
+            (Some(delta), None) => Self::Delta(delta),
+            (None, Some(image)) => Self::Image(image),
             (Some(_), Some(_)) => unreachable!(),
         };
         Ok(ok)
@@ -242,12 +252,8 @@ impl serde::Serialize for LayerFileName {
         S: serde::Serializer,
     {
         match self {
-            LayerFileName::Image(fname) => serializer.serialize_str(&format!("{}", fname)),
-            LayerFileName::Delta(fname) => serializer.serialize_str(&format!("{}", fname)),
-            #[cfg(test)]
-            LayerFileName::Test(t) => {
-                serializer.serialize_str(&format!("{LAYER_FILE_NAME_TEST_PREFIX}{t}"))
-            }
+            Self::Image(fname) => serializer.serialize_str(&fname.to_string()),
+            Self::Delta(fname) => serializer.serialize_str(&fname.to_string()),
         }
     }
 }
@@ -270,16 +276,3 @@ impl<'de> serde::de::Visitor<'de> for LayerFileNameVisitor {
         v.parse().map_err(|e| E::custom(e))
     }
 }
-
-/// Helper enum to hold a PageServerConf, or a path
-///
-/// This is used by DeltaLayer and ImageLayer. Normally, this holds a reference to the
-/// global config, and paths to layer files are constructed using the tenant/timeline
-/// path from the config. But in the 'pageserver_binutils' binary, we need to construct a Layer
-/// struct for a file on disk, without having a page server running, so that we have no
-/// config. In that case, we use the Path variant to hold the full path to the file on
-/// disk.
-pub enum PathOrConf {
-    Path(PathBuf),
-    Conf(&'static PageServerConf),
-}