Dead-end websocket experiments

SELECT 1.0 is still not working

.config/hakari.toml

@@ -4,7 +4,7 @@
 hakari-package = "workspace_hack"
 
 # Format for `workspace-hack = ...` lines in other Cargo.tomls. Requires cargo-hakari 0.9.8 or above.
-dep-format-version = "3"
+dep-format-version = "4"
 
 # Setting workspace.resolver = "2" in the root Cargo.toml is HIGHLY recommended.
 # Hakari works much better with the new feature resolver.

.dockerignore

@@ -21,3 +21,4 @@
 !workspace_hack/
 !neon_local/
 !scripts/ninstall.sh
+!vm-cgconfig.conf

.github/PULL_REQUEST_TEMPLATE/release-pr.md

@@ -10,6 +10,7 @@
 <!-- List everything that should be done **before** release, any issues / setting changes / etc -->
 
 ### Checklist after release
+- [ ] Make sure instructions from PRs included in this release and labeled `manual_release_instructions` are executed (either by you or by people who wrote them).
 - [ ] Based on the merged commits write release notes and open a PR into `website` repo ([example](https://github.com/neondatabase/website/pull/219/files))
 - [ ] Check [#dev-production-stream](https://neondb.slack.com/archives/C03F5SM1N02) Slack channel
 - [ ] Check [stuck projects page](https://console.neon.tech/admin/projects?sort=last_active&order=desc&stuck=true)

.github/actions/allure-report/action.yml

@@ -15,20 +15,42 @@ outputs:
   report-url:
     description: 'Allure report URL'
     value: ${{ steps.generate-report.outputs.report-url }}
+  report-json-url:
+    description: 'Allure report JSON URL'
+    value: ${{ steps.generate-report.outputs.report-json-url }}
 
 runs:
   using: "composite"
+
   steps:
+    # We're using some of env variables quite offen, so let's set them once.
+    #
+    # It would be nice to have them set in common runs.env[0] section, but it doesn't work[1]
+    #
+    # - [0] https://docs.github.com/en/actions/creating-actions/metadata-syntax-for-github-actions#runsenv
+    # - [1] https://github.com/neondatabase/neon/pull/3907#discussion_r1154703456
+    #
+    - name: Set common environment variables
+      shell: bash -euxo pipefail {0}
+      run: |
+        echo "BUILD_TYPE=${BUILD_TYPE}"   >> $GITHUB_ENV
+        echo "BUCKET=${BUCKET}"           >> $GITHUB_ENV
+        echo "TEST_OUTPUT=${TEST_OUTPUT}" >> $GITHUB_ENV
+      env:
+        BUILD_TYPE: ${{ inputs.build_type }}
+        BUCKET: neon-github-public-dev
+        TEST_OUTPUT: /tmp/test_output
+
     - name: Validate input parameters
       shell: bash -euxo pipefail {0}
       run: |
         if [ "${{ inputs.action }}" != "store" ] && [ "${{ inputs.action }}" != "generate" ]; then
-          echo 2>&1 "Unknown inputs.action type '${{ inputs.action }}'; allowed 'generate' or 'store' only"
+          echo >&2 "Unknown inputs.action type '${{ inputs.action }}'; allowed 'generate' or 'store' only"
           exit 1
         fi
 
         if [ -z "${{ inputs.test_selection }}" ] && [ "${{ inputs.action }}" == "store" ]; then
-          echo 2>&1 "inputs.test_selection must be set for 'store' action"
+          echo >&2 "inputs.test_selection must be set for 'store' action"
           exit 2
         fi
 
@@ -76,16 +98,14 @@ runs:
           rm -f ${ALLURE_ZIP}
         fi
       env:
-        ALLURE_VERSION: 2.19.0
-        ALLURE_ZIP_MD5: ced21401a1a8b9dfb68cee9e4c210464
+        ALLURE_VERSION: 2.21.0
+        ALLURE_ZIP_MD5: c8db4dd8e2a7882583d569ed2c82879c
 
     - name: Upload Allure results
       if: ${{ inputs.action == 'store' }}
       env:
         REPORT_PREFIX: reports/${{ steps.calculate-vars.outputs.KEY }}/${{ inputs.build_type }}
         RAW_PREFIX: reports-raw/${{ steps.calculate-vars.outputs.KEY }}/${{ inputs.build_type }}
-        TEST_OUTPUT: /tmp/test_output
-        BUCKET: neon-github-public-dev
         TEST_SELECTION: ${{ steps.calculate-vars.outputs.TEST_SELECTION }}
       shell: bash -euxo pipefail {0}
       run: |
@@ -104,7 +124,7 @@ runs:
         EOF
         cat <<EOF > $TEST_OUTPUT/allure/results/environment.properties
           TEST_SELECTION=${{ inputs.test_selection }}
-          BUILD_TYPE=${{ inputs.build_type }}
+          BUILD_TYPE=${BUILD_TYPE}
         EOF
 
         ARCHIVE="${GITHUB_RUN_ID}-${TEST_SELECTION}-${GITHUB_RUN_ATTEMPT}-$(date +%s).tar.zst"
@@ -113,13 +133,12 @@ runs:
         tar -C ${TEST_OUTPUT}/allure/results -cf ${ARCHIVE} --zstd .
         aws s3 mv --only-show-errors ${ARCHIVE} "s3://${BUCKET}/${RAW_PREFIX}/${ARCHIVE}"
 
-    # Potentially we could have several running build for the same key (for example for the main branch),  so we use improvised lock for this
+    # Potentially we could have several running build for the same key (for example for the main branch), so we use improvised lock for this
     - name: Acquire Allure lock
       if: ${{ inputs.action == 'generate' }}
       shell: bash -euxo pipefail {0}
       env:
         LOCK_FILE: reports/${{ steps.calculate-vars.outputs.KEY }}/lock.txt
-        BUCKET: neon-github-public-dev
         TEST_SELECTION: ${{ steps.calculate-vars.outputs.TEST_SELECTION }}
       run: |
         LOCK_TIMEOUT=300 # seconds
@@ -149,8 +168,6 @@ runs:
       env:
         REPORT_PREFIX: reports/${{ steps.calculate-vars.outputs.KEY }}/${{ inputs.build_type }}
         RAW_PREFIX: reports-raw/${{ steps.calculate-vars.outputs.KEY }}/${{ inputs.build_type }}
-        TEST_OUTPUT: /tmp/test_output
-        BUCKET: neon-github-public-dev
       shell: bash -euxo pipefail {0}
       run: |
         # Get previously uploaded data for this run
@@ -186,24 +203,24 @@ runs:
         REPORT_URL=https://${BUCKET}.s3.amazonaws.com/${REPORT_PREFIX}/${GITHUB_RUN_ID}/index.html
 
         # Generate redirect
-        cat <<EOF > ./index.html
+        cat <<EOF > ${TEST_OUTPUT}/allure/index.html
           <!DOCTYPE html>
 
           <meta charset="utf-8">
           <title>Redirecting to ${REPORT_URL}</title>
           <meta http-equiv="refresh" content="0; URL=${REPORT_URL}">
         EOF
-        aws s3 cp --only-show-errors ./index.html "s3://${BUCKET}/${REPORT_PREFIX}/latest/index.html"
+        aws s3 cp --only-show-errors ${TEST_OUTPUT}/allure/index.html "s3://${BUCKET}/${REPORT_PREFIX}/latest/index.html"
 
         echo "[Allure Report](${REPORT_URL})" >> ${GITHUB_STEP_SUMMARY}
         echo "report-url=${REPORT_URL}" >> $GITHUB_OUTPUT
+        echo "report-json-url=${REPORT_URL%/index.html}/data/suites.json" >> $GITHUB_OUTPUT
 
     - name: Release Allure lock
       if: ${{ inputs.action == 'generate' && always() }}
       shell: bash -euxo pipefail {0}
       env:
         LOCK_FILE: reports/${{ steps.calculate-vars.outputs.KEY }}/lock.txt
-        BUCKET: neon-github-public-dev
         TEST_SELECTION: ${{ steps.calculate-vars.outputs.TEST_SELECTION }}
       run: |
         aws s3 cp --only-show-errors "s3://${BUCKET}/${LOCK_FILE}" ./lock.txt || exit 0
@@ -212,11 +229,16 @@ runs:
           aws s3 rm "s3://${BUCKET}/${LOCK_FILE}"
         fi
 
+    - name: Cleanup
+      if: always()
+      shell: bash -euxo pipefail {0}
+      run: |
+        rm -rf ${TEST_OUTPUT}/allure
+
     - uses: actions/github-script@v6
       if: ${{ inputs.action == 'generate' && always() }}
       env:
         REPORT_URL: ${{ steps.generate-report.outputs.report-url }}
-        BUILD_TYPE: ${{ inputs.build_type }}
         SHA: ${{ github.event.pull_request.head.sha || github.sha }}
       with:
         script: |

.github/actions/download/action.yml

@@ -37,7 +37,7 @@ runs:
             echo 'SKIPPED=true' >> $GITHUB_OUTPUT
             exit 0
           else
-            echo 2>&1 "Neither s3://${BUCKET}/${PREFIX}/${FILENAME} nor its version from previous attempts exist"
+            echo >&2 "Neither s3://${BUCKET}/${PREFIX}/${FILENAME} nor its version from previous attempts exist"
             exit 1
           fi
         fi

.github/actions/neon-branch-create/action.yml

@@ -58,7 +58,7 @@ runs:
         done
 
         if [ -z "${branch_id}" ] || [ "${branch_id}" == "null" ]; then
-          echo 2>&1 "Failed to create branch after 10 attempts, the latest response was: ${branch}"
+          echo >&2 "Failed to create branch after 10 attempts, the latest response was: ${branch}"
           exit 1
         fi
 
@@ -122,7 +122,7 @@ runs:
         done
 
         if [ -z "${password}" ] || [ "${password}" == "null" ]; then
-          echo 2>&1 "Failed to reset password after 10 attempts, the latest response was: ${reset_password}"
+          echo >&2 "Failed to reset password after 10 attempts, the latest response was: ${reset_password}"
           exit 1
         fi
 

.github/actions/neon-branch-delete/action.yml

@@ -48,7 +48,7 @@ runs:
         done
 
         if [ -z "${branch_id}" ] || [ "${branch_id}" == "null" ]; then
-          echo 2>&1 "Failed to delete branch after 10 attempts, the latest response was: ${deleted_branch}"
+          echo >&2 "Failed to delete branch after 10 attempts, the latest response was: ${deleted_branch}"
           exit 1
         fi
       env:

.github/actions/neon-project-create/action.yml

@@ -14,6 +14,12 @@ inputs:
   api_host:
     desctiption: 'Neon API host'
     default: console.stage.neon.tech
+  provisioner:
+    desctiption: 'k8s-pod or k8s-neonvm'
+    default: 'k8s-pod'
+  compute_units:
+    desctiption: '[Min, Max] compute units; Min and Max are used for k8s-neonvm with autoscaling, for k8s-pod values Min and Max should be equal'
+    default: '[1, 1]'
 
 outputs:
   dsn:
@@ -31,6 +37,10 @@ runs:
       # A shell without `set -x` to not to expose password/dsn in logs
       shell: bash -euo pipefail {0}
       run: |
+        if [ "${PROVISIONER}" == "k8s-pod" ] && [ "${MIN_CU}" != "${MAX_CU}" ]; then
+          echo >&2 "For k8s-pod provisioner MIN_CU should be equal to MAX_CU"
+        fi
+
         project=$(curl \
           "https://${API_HOST}/api/v2/projects" \
           --fail \
@@ -42,6 +52,9 @@ runs:
               \"name\": \"Created by actions/neon-project-create; GITHUB_RUN_ID=${GITHUB_RUN_ID}\",
               \"pg_version\": ${POSTGRES_VERSION},
               \"region_id\": \"${REGION_ID}\",
+              \"provisioner\": \"${PROVISIONER}\",
+              \"autoscaling_limit_min_cu\": ${MIN_CU},
+              \"autoscaling_limit_max_cu\": ${MAX_CU},
               \"settings\": { }
             }
           }")
@@ -62,3 +75,6 @@ runs:
         API_KEY: ${{ inputs.api_key }}
         REGION_ID: ${{ inputs.region_id }}
         POSTGRES_VERSION: ${{ inputs.postgres_version }}
+        PROVISIONER: ${{ inputs.provisioner }}
+        MIN_CU: ${{ fromJSON(inputs.compute_units)[0] }}
+        MAX_CU: ${{ fromJSON(inputs.compute_units)[1] }}

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

@@ -44,6 +44,10 @@ inputs:
     description: 'Secret access key'
     required: false
     default: ''
+  rerun_flaky:
+    description: 'Whether to rerun flaky tests'
+    required: false
+    default: 'false'
 
 runs:
   using: "composite"
@@ -101,6 +105,7 @@ runs:
         COMPATIBILITY_SNAPSHOT_DIR: /tmp/compatibility_snapshot_pg14
         ALLOW_BACKWARD_COMPATIBILITY_BREAKAGE: contains(github.event.pull_request.labels.*.name, 'backward compatibility breakage')
         ALLOW_FORWARD_COMPATIBILITY_BREAKAGE: contains(github.event.pull_request.labels.*.name, 'forward compatibility breakage')
+        RERUN_FLAKY: ${{ inputs.rerun_flaky }}
       shell: bash -euxo pipefail {0}
       run: |
         # PLATFORM will be embedded in the perf test report
@@ -143,6 +148,13 @@ runs:
           EXTRA_PARAMS="--out-dir $PERF_REPORT_DIR $EXTRA_PARAMS"
         fi
 
+        if [ "${RERUN_FLAKY}" == "true" ]; then
+          mkdir -p $TEST_OUTPUT
+          poetry run ./scripts/flaky_tests.py "${TEST_RESULT_CONNSTR}" --days 10 --output "$TEST_OUTPUT/flaky.json"
+
+          EXTRA_PARAMS="--flaky-tests-json $TEST_OUTPUT/flaky.json $EXTRA_PARAMS"
+        fi
+
         if [[ "${{ inputs.build_type }}" == "debug" ]]; then
           cov_prefix=(scripts/coverage "--profraw-prefix=$GITHUB_JOB" --dir=/tmp/coverage run)
         elif [[ "${{ inputs.build_type }}" == "release" ]]; then
@@ -190,7 +202,7 @@ runs:
         prefix: latest
 
     - name: Create Allure report
-      if: success() || failure()
+      if: ${{ !cancelled() }}
       uses: ./.github/actions/allure-report
       with:
         action: store

.github/actions/upload/action.yml

@@ -23,7 +23,7 @@ runs:
         mkdir -p $(dirname $ARCHIVE)
 
         if [ -f ${ARCHIVE} ]; then
-          echo 2>&1 "File ${ARCHIVE} already exist. Something went wrong before"
+          echo >&2 "File ${ARCHIVE} already exist. Something went wrong before"
           exit 1
         fi
 
@@ -33,10 +33,10 @@ runs:
         elif [ -f ${SOURCE} ]; then
           time tar -cf ${ARCHIVE} --zstd ${SOURCE}
         elif ! ls ${SOURCE} > /dev/null 2>&1; then
-          echo 2>&1 "${SOURCE} does not exist"
+          echo >&2 "${SOURCE} does not exist"
           exit 2
         else
-          echo 2>&1 "${SOURCE} is neither a directory nor a file, do not know how to handle it"
+          echo >&2 "${SOURCE} is neither a directory nor a file, do not know how to handle it"
           exit 3
         fi
 

.github/ansible/deploy.yaml

@@ -91,6 +91,15 @@
       tags:
       - pageserver
 
+    # used in `pageserver.service` template
+    - name: learn current availability_zone
+      shell:
+        cmd: "curl -s http://169.254.169.254/latest/meta-data/placement/availability-zone"
+      register: ec2_availability_zone
+
+    - set_fact: 
+        ec2_availability_zone={{ ec2_availability_zone.stdout }}
+
     - name: upload systemd service definition
       ansible.builtin.template:
         src: systemd/pageserver.service
@@ -118,7 +127,7 @@
         cmd: |
           INSTANCE_ID=$(curl -s http://169.254.169.254/latest/meta-data/instance-id)
           curl -sfS -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" {{ console_mgmt_base_url }}/management/api/v2/pageservers/$INSTANCE_ID | jq '.version = {{ current_version }}' > /tmp/new_version
-          curl -sfS -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" -X POST -d@/tmp/new_version {{ console_mgmt_base_url }}/management/api/v2/pageservers
+          curl -sfS -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" -H "Content-Type: application/json" -X POST -d@/tmp/new_version {{ console_mgmt_base_url }}/management/api/v2/pageservers
       tags:
       - pageserver
 
@@ -153,6 +162,15 @@
       tags:
       - safekeeper
 
+    # used in `safekeeper.service` template
+    - name: learn current availability_zone
+      shell:
+        cmd: "curl -s http://169.254.169.254/latest/meta-data/placement/availability-zone"
+      register: ec2_availability_zone
+
+    - set_fact: 
+        ec2_availability_zone={{ ec2_availability_zone.stdout }}
+
     # in the future safekeepers should discover pageservers byself
     # but currently use first pageserver that was discovered
     - name: set first pageserver var for safekeepers
@@ -188,6 +206,6 @@
         cmd: |
           INSTANCE_ID=$(curl -s http://169.254.169.254/latest/meta-data/instance-id)
           curl -sfS -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" {{ console_mgmt_base_url }}/management/api/v2/safekeepers/$INSTANCE_ID | jq '.version = {{ current_version }}' > /tmp/new_version
-          curl -sfS -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" -X POST -d@/tmp/new_version {{ console_mgmt_base_url }}/management/api/v2/safekeepers
+          curl -sfS -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" -H "Content-Type: application/json" -X POST -d@/tmp/new_version {{ console_mgmt_base_url }}/management/api/v2/safekeepers
       tags:
       - safekeeper

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

@@ -2,12 +2,22 @@ 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
+      disk_usage_based_eviction:
+        max_usage_pct: 85 # TODO: decrease to 80 after all pageservers are below 80
+        min_avail_bytes: 0
+        period: "10s"
+      tenant_config:
+        eviction_policy:
+          kind: "LayerAccessThreshold"
+          period: "10m"
+          threshold: &default_eviction_threshold "24h"
+        evictions_low_residence_duration_metric_threshold: *default_eviction_threshold
       remote_storage:
         bucket_name: "{{ bucket_name }}"
         bucket_region: "{{ bucket_region }}"
@@ -32,7 +42,7 @@ 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

@@ -2,12 +2,22 @@ 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
+      disk_usage_based_eviction:
+        max_usage_pct: 85 # TODO: decrease to 80 after all pageservers are below 80
+        min_avail_bytes: 0
+        period: "10s"
+      tenant_config:
+        eviction_policy:
+          kind: "LayerAccessThreshold"
+          period: "10m"
+          threshold: &default_eviction_threshold "24h"
+        evictions_low_residence_duration_metric_threshold: *default_eviction_threshold
       remote_storage:
         bucket_name: "{{ bucket_name }}"
         bucket_region: "{{ bucket_region }}"
@@ -27,6 +37,8 @@ storage:
           ansible_host:  i-0cd8d316ecbb715be
         pageserver-1.eu-central-1.aws.neon.tech:
           ansible_host:  i-090044ed3d383fef0
+        pageserver-2.eu-central-1.aws.neon.tech:
+          ansible_host:  i-033584edf3f4b6742
 
     safekeepers:
       hosts:

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

@@ -2,12 +2,22 @@ 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
+      disk_usage_based_eviction:
+        max_usage_pct: 85 # TODO: decrease to 80 after all pageservers are below 80
+        min_avail_bytes: 0
+        period: "10s"
+      tenant_config:
+        eviction_policy:
+          kind: "LayerAccessThreshold"
+          period: "10m"
+          threshold: &default_eviction_threshold "24h"
+        evictions_low_residence_duration_metric_threshold: *default_eviction_threshold
       remote_storage:
         bucket_name: "{{ bucket_name }}"
         bucket_region: "{{ bucket_region }}"
@@ -27,6 +37,8 @@ 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

@@ -2,12 +2,22 @@ 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
+      disk_usage_based_eviction:
+        max_usage_pct: 85 # TODO: decrease to 80 after all pageservers are below 80
+        min_avail_bytes: 0
+        period: "10s"
+      tenant_config:
+        eviction_policy:
+          kind: "LayerAccessThreshold"
+          period: "10m"
+          threshold: &default_eviction_threshold "24h"
+        evictions_low_residence_duration_metric_threshold: *default_eviction_threshold
       remote_storage:
         bucket_name: "{{ bucket_name }}"
         bucket_region: "{{ bucket_region }}"
@@ -24,11 +34,13 @@ storage:
     pageservers:
       hosts:
         pageserver-0.us-west-2.aws.neon.tech:
-          ansible_host: i-0d9f6dfae0e1c780d 
+          ansible_host: i-0d9f6dfae0e1c780d
         pageserver-1.us-west-2.aws.neon.tech:
           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:
@@ -37,5 +49,5 @@ storage:
         safekeeper-1.us-west-2.aws.neon.tech:
           ansible_host: i-074682f9d3c712e7c
         safekeeper-2.us-west-2.aws.neon.tech:
-          ansible_host: i-042b7efb1729d7966 
-          
+          ansible_host: i-042b7efb1729d7966
+

.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/scripts/init_pageserver.sh

@@ -3,6 +3,8 @@
 # fetch params from meta-data service
 INSTANCE_ID=$(curl -s http://169.254.169.254/latest/meta-data/instance-id)
 AZ_ID=$(curl -s http://169.254.169.254/latest/meta-data/placement/availability-zone)
+INSTANCE_TYPE=$(curl -s http://169.254.169.254/latest/meta-data/instance-type)
+DISK_SIZE=$(df -B1 /storage | tail -1 | awk '{print $2}')
 
 # store fqdn hostname in var
 HOST=$(hostname -f)
@@ -18,7 +20,9 @@ cat <<EOF | tee /tmp/payload
   "http_host": "${HOST}",
   "http_port": 9898,
   "active": false,
-  "availability_zone_id": "${AZ_ID}"
+  "availability_zone_id": "${AZ_ID}",
+  "disk_size": ${DISK_SIZE},
+  "instance_type": "${INSTANCE_TYPE}"
 }
 EOF
 
@@ -26,7 +30,7 @@ EOF
 if ! curl -sf -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" {{ console_mgmt_base_url }}/management/api/v2/pageservers/${INSTANCE_ID} -o /dev/null; then
 
     # not registered, so register it now
-    ID=$(curl -sf -X POST -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" {{ console_mgmt_base_url }}/management/api/v2/pageservers -d@/tmp/payload | jq -r '.id')
+    ID=$(curl -sf -X POST -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" -H "Content-Type: application/json" {{ console_mgmt_base_url }}/management/api/v2/pageservers -d@/tmp/payload | jq -r '.id')
 
     # init pageserver
     sudo -u pageserver /usr/local/bin/pageserver -c "id=${ID}" -c "pg_distrib_dir='/usr/local'" --init -D /storage/pageserver/data

.github/ansible/scripts/init_safekeeper.sh

@@ -25,7 +25,7 @@ EOF
 if ! curl -sf -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" {{ console_mgmt_base_url }}/management/api/v2/safekeepers/${INSTANCE_ID} -o /dev/null; then
 
     # not registered, so register it now
-    ID=$(curl -sf -X POST -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" {{ console_mgmt_base_url }}/management/api/v2/safekeepers -d@/tmp/payload | jq -r '.id')
+    ID=$(curl -sf -X POST -H "Authorization: Bearer {{ CONSOLE_API_TOKEN }}" -H "Content-Type: application/json" {{ console_mgmt_base_url }}/management/api/v2/safekeepers -d@/tmp/payload | jq -r '.id')
     # init safekeeper
     sudo -u safekeeper /usr/local/bin/safekeeper --id ${ID} --init -D /storage/safekeeper/data
 fi

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

@@ -2,12 +2,22 @@ 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
+      disk_usage_based_eviction:
+        max_usage_pct: 80
+        min_avail_bytes: 0
+        period: "10s"
+      tenant_config:
+        eviction_policy:
+          kind: "LayerAccessThreshold"
+          period: "20m"
+          threshold: &default_eviction_threshold "20m"
+        evictions_low_residence_duration_metric_threshold: *default_eviction_threshold
       remote_storage:
         bucket_name: "{{ bucket_name }}"
         bucket_region: "{{ bucket_region }}"

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

@@ -2,12 +2,22 @@ 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
+      disk_usage_based_eviction:
+        max_usage_pct: 80
+        min_avail_bytes: 0
+        period: "10s"
+      tenant_config:
+        eviction_policy:
+          kind: "LayerAccessThreshold"
+          period: "20m"
+          threshold: &default_eviction_threshold "20m"
+        evictions_low_residence_duration_metric_threshold: *default_eviction_threshold
       remote_storage:
         bucket_name: "{{ bucket_name }}"
         bucket_region: "{{ bucket_region }}"
@@ -31,6 +41,8 @@ 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 +52,5 @@ 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/ansible/systemd/pageserver.service

@@ -6,7 +6,7 @@ After=network.target auditd.service
 Type=simple
 User=pageserver
 Environment=RUST_BACKTRACE=1 NEON_REPO_DIR=/storage/pageserver LD_LIBRARY_PATH=/usr/local/v14/lib SENTRY_DSN={{ SENTRY_URL_PAGESERVER }} SENTRY_ENVIRONMENT={{ sentry_environment }}
-ExecStart=/usr/local/bin/pageserver -c "pg_distrib_dir='/usr/local'" -c "listen_pg_addr='0.0.0.0:6400'" -c "listen_http_addr='0.0.0.0:9898'" -c "broker_endpoint='{{ broker_endpoint }}'" -D /storage/pageserver/data
+ExecStart=/usr/local/bin/pageserver -c "pg_distrib_dir='/usr/local'" -c "listen_pg_addr='0.0.0.0:6400'" -c "listen_http_addr='0.0.0.0:9898'" -c "broker_endpoint='{{ broker_endpoint }}'" -c "availability_zone='{{ ec2_availability_zone }}'" -D /storage/pageserver/data
 ExecReload=/bin/kill -HUP $MAINPID
 KillMode=mixed
 KillSignal=SIGINT

.github/ansible/systemd/safekeeper.service

@@ -6,7 +6,7 @@ After=network.target auditd.service
 Type=simple
 User=safekeeper
 Environment=RUST_BACKTRACE=1 NEON_REPO_DIR=/storage/safekeeper/data LD_LIBRARY_PATH=/usr/local/v14/lib SENTRY_DSN={{ SENTRY_URL_SAFEKEEPER }} SENTRY_ENVIRONMENT={{ sentry_environment }}
-ExecStart=/usr/local/bin/safekeeper -l {{ inventory_hostname }}{{ hostname_suffix }}:6500 --listen-http {{ inventory_hostname }}{{ hostname_suffix }}:7676 -D /storage/safekeeper/data --broker-endpoint={{ broker_endpoint }} --remote-storage='{bucket_name="{{bucket_name}}", bucket_region="{{bucket_region}}", prefix_in_bucket="{{ safekeeper_s3_prefix }}"}'
+ExecStart=/usr/local/bin/safekeeper -l {{ inventory_hostname }}{{ hostname_suffix }}:6500 --listen-http {{ inventory_hostname }}{{ hostname_suffix }}:7676 -D /storage/safekeeper/data --broker-endpoint={{ broker_endpoint }} --remote-storage='{bucket_name="{{bucket_name}}", bucket_region="{{bucket_region}}", prefix_in_bucket="{{ safekeeper_s3_prefix }}"}' --availability-zone={{ ec2_availability_zone }}
 ExecReload=/bin/kill -HUP $MAINPID
 KillMode=mixed
 KillSignal=SIGINT

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

@@ -1,24 +1,38 @@
 # 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 5 minutes (5 * 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 300"]
+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
 podLabels:
-  zenith_service: proxy-scram
-  zenith_env: dev
-  zenith_region: eu-west-1
-  zenith_region_slug: eu-west-1
+  neon_service: proxy-scram
+  neon_env: dev
+  neon_region: eu-west-1
 
 exposedService:
   annotations:

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

@@ -10,15 +10,14 @@ settings:
   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
 podLabels:
-  zenith_service: proxy
-  zenith_env: dev
-  zenith_region: us-east-2
-  zenith_region_slug: us-east-2
+  neon_service: proxy
+  neon_env: dev
+  neon_region: us-east-2
 
 service:
   type: LoadBalancer

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

@@ -1,24 +1,39 @@
 # 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 5 minutes (5 * 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 300"]
+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: "*.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
 podLabels:
-  zenith_service: proxy-scram-legacy
-  zenith_env: dev
-  zenith_region: us-east-2
-  zenith_region_slug: us-east-2
+  neon_service: proxy-scram-legacy
+  neon_env: dev
+  neon_region: us-east-2
 
 exposedService:
   annotations:

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

@@ -1,24 +1,40 @@
 # 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 5 minutes (5 * 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 300"]
+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"
+  extraDomains: ["*.us-east-2.postgres.zenith.tech", "*.us-east-2.retooldb-staging.com"]
   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
 podLabels:
-  zenith_service: proxy-scram
-  zenith_env: dev
-  zenith_region: us-east-2
-  zenith_region_slug: us-east-2
+  neon_service: proxy-scram
+  neon_env: dev
+  neon_region: us-east-2
 
 exposedService:
   annotations:

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

@@ -1,24 +1,40 @@
 # 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 5 minutes (5 * 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 300"]
+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"
+  extraDomains: ["*.ap-southeast-1.retooldb.com", "*.ap-southeast-1.postgres.vercel-storage.com"]
   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: prod
-  zenith_region: ap-southeast-1
-  zenith_region_slug: ap-southeast-1
+  neon_service: proxy-scram
+  neon_env: prod
+  neon_region: ap-southeast-1
 
 exposedService:
   annotations:

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

@@ -1,24 +1,40 @@
 # 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 5 minutes (5 * 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 300"]
+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"
+  extraDomains: ["*.eu-central-1.retooldb.com", "*.eu-central-1.postgres.vercel-storage.com"]
   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: prod
-  zenith_region: eu-central-1
-  zenith_region_slug: eu-central-1
+  neon_service: proxy-scram
+  neon_env: prod
+  neon_region: eu-central-1
 
 exposedService:
   annotations:

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

@@ -13,10 +13,9 @@ settings:
 
 # -- Additional labels for zenith-proxy pods
 podLabels:
-  zenith_service: proxy
-  zenith_env: production
-  zenith_region: us-east-2
-  zenith_region_slug: us-east-2
+  neon_service: proxy
+  neon_env: production
+  neon_region: us-east-2
 
 service:
   type: LoadBalancer

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

@@ -1,24 +1,40 @@
 # 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 5 minutes (5 * 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 300"]
+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"
+  extraDomains: ["*.us-east-2.retooldb.com", "*.us-east-2.postgres.vercel-storage.com"]
   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: prod
-  zenith_region: us-east-2
-  zenith_region_slug: us-east-2
+  neon_service: proxy-scram
+  neon_env: prod
+  neon_region: us-east-2
 
 exposedService:
   annotations:

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

@@ -1,24 +1,39 @@
 # 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 5 minutes (5 * 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 300"]
+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: "*.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: prod
-  zenith_region: us-west-2
-  zenith_region_slug: us-west-2
+  neon_service: proxy-scram
+  neon_env: prod
+  neon_region: us-west-2
 
 exposedService:
   annotations:

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

@@ -1,24 +1,40 @@
 # 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 5 minutes (5 * 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 300"]
+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"
+  extraDomains: ["*.us-west-2.retooldb.com", "*.us-west-2.postgres.vercel-storage.com"]
   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: prod
-  zenith_region: us-west-2
-  zenith_region_slug: us-west-2
+  neon_service: proxy-scram
+  neon_env: prod
+  neon_region: us-west-2
 
 exposedService:
   annotations:

.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/pull_request_template.md

@@ -3,8 +3,12 @@
 ## Issue ticket number and link
 
 ## Checklist before requesting a review
+
 - [ ] I have performed a self-review of my code.
 - [ ] If it is a core feature, I have added thorough tests.
 - [ ] Do we need to implement analytics? if so did you add the relevant metrics to the dashboard?
 - [ ] If this PR requires public announcement, mark it with /release-notes label and add several sentences in this section.
 
+## Checklist before merging
+
+- [ ] Do not forget to reformat commit message to not include the above checklist

.github/workflows/benchmarking.yml

@@ -30,7 +30,7 @@ defaults:
 
 concurrency:
   # Allow only one workflow per any non-`main` branch.
-  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.ref == 'refs/heads/main' && github.sha || 'anysha' }}
+  group: ${{ github.workflow }}-${{ github.ref_name }}-${{ github.ref_name == 'main' && github.sha || 'anysha' }}
   cancel-in-progress: true
 
 jobs:
@@ -42,7 +42,7 @@ jobs:
       DEFAULT_PG_VERSION: 14
       TEST_OUTPUT: /tmp/test_output
       BUILD_TYPE: remote
-      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref == 'refs/heads/main' ) }}
+      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
       PLATFORM: "neon-staging"
 
     runs-on: [ self-hosted, us-east-2, x64 ]
@@ -92,7 +92,7 @@ jobs:
         api_key: ${{ secrets.NEON_STAGING_API_KEY }}
 
     - name: Create Allure report
-      if: success() || failure()
+      if: ${{ !cancelled() }}
       uses: ./.github/actions/allure-report
       with:
         action: generate
@@ -107,25 +107,65 @@ jobs:
       env:
         SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
 
+  generate-matrices:
+    # Create matrices for the benchmarking jobs, so we run benchmarks on rds only once a week (on Saturday)
+    #
+    # Available platforms:
+    # - neon-captest-new: Freshly created project (1 CU)
+    # - neon-captest-freetier: Use freetier-sized compute (0.25 CU)
+    # - neon-captest-reuse: Reusing existing project
+    # - rds-aurora: Aurora Postgres Serverless v2 with autoscaling from 0.5 to 2 ACUs
+    # - rds-postgres: RDS Postgres db.m5.large instance (2 vCPU, 8 GiB) with gp3 EBS storage
+    runs-on: ubuntu-latest
+    outputs:
+      pgbench-compare-matrix: ${{ steps.pgbench-compare-matrix.outputs.matrix }}
+      olap-compare-matrix: ${{ steps.olap-compare-matrix.outputs.matrix }}
+
+    steps:
+    - name: Generate matrix for pgbench benchmark
+      id: pgbench-compare-matrix
+      run: |
+        matrix='{
+          "platform": [
+            "neon-captest-new",
+            "neon-captest-reuse"
+          ],
+          "db_size": [ "10gb" ],
+          "include": [
+            { "platform": "neon-captest-freetier", "db_size": "3gb"  },
+            { "platform": "neon-captest-new",      "db_size": "50gb" }
+          ]
+        }'
+
+        if [ "$(date +%A)" = "Saturday" ]; then
+          matrix=$(echo $matrix | jq '.include += [{ "platform": "rds-postgres", "db_size": "10gb"},
+                                                   { "platform": "rds-aurora",   "db_size": "50gb"}]')
+        fi
+
+        echo "matrix=$(echo $matrix | jq --compact-output '.')" >> $GITHUB_OUTPUT
+
+    - name: Generate matrix for OLAP benchmarks
+      id: olap-compare-matrix
+      run: |
+        matrix='{
+          "platform": [
+            "neon-captest-reuse"
+          ]
+        }'
+
+        if [ "$(date +%A)" = "Saturday" ]; then
+          matrix=$(echo $matrix | jq '.include += [{ "platform": "rds-postgres" },
+                                                   { "platform": "rds-aurora"   }]')
+        fi
+
+        echo "matrix=$(echo $matrix | jq --compact-output '.')" >> $GITHUB_OUTPUT
+
   pgbench-compare:
+    needs: [ generate-matrices ]
+
     strategy:
       fail-fast: false
-      matrix:
-        # neon-captest-new: Run pgbench in a freshly created project
-        # neon-captest-reuse: Same, but reusing existing project
-        # neon-captest-prefetch: Same, with prefetching enabled (new project)
-        # rds-aurora: Aurora Postgres Serverless v2 with autoscaling from 0.5 to 2 ACUs
-        # rds-postgres: RDS Postgres db.m5.large instance (2 vCPU, 8 GiB) with gp3 EBS storage
-        platform: [ neon-captest-reuse, neon-captest-prefetch, rds-postgres ]
-        db_size: [ 10gb ]
-        runner: [ us-east-2 ]
-        include:
-          - platform: neon-captest-prefetch
-            db_size: 50gb
-            runner: us-east-2
-          - platform: rds-aurora
-            db_size: 50gb
-            runner: us-east-2
+      matrix: ${{fromJson(needs.generate-matrices.outputs.pgbench-compare-matrix)}}
 
     env:
       TEST_PG_BENCH_DURATIONS_MATRIX: "60m"
@@ -134,10 +174,10 @@ jobs:
       DEFAULT_PG_VERSION: 14
       TEST_OUTPUT: /tmp/test_output
       BUILD_TYPE: remote
-      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref == 'refs/heads/main' ) }}
+      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
       PLATFORM: ${{ matrix.platform }}
 
-    runs-on: [ self-hosted, "${{ matrix.runner }}", x64 ]
+    runs-on: [ self-hosted, us-east-2, x64 ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
@@ -160,13 +200,14 @@ jobs:
         echo "${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin" >> $GITHUB_PATH
 
     - name: Create Neon Project
-      if: contains(fromJson('["neon-captest-new", "neon-captest-prefetch"]'), matrix.platform)
+      if: contains(fromJson('["neon-captest-new", "neon-captest-freetier"]'), matrix.platform)
       id: create-neon-project
       uses: ./.github/actions/neon-project-create
       with:
         region_id: ${{ github.event.inputs.region_id || 'aws-us-east-2' }}
         postgres_version: ${{ env.DEFAULT_PG_VERSION }}
         api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+        compute_units: ${{ (matrix.platform == 'neon-captest-freetier' && '[0.25, 0.25]') || '[1, 1]' }}
 
     - name: Set up Connection String
       id: set-up-connstr
@@ -175,7 +216,7 @@ jobs:
           neon-captest-reuse)
             CONNSTR=${{ secrets.BENCHMARK_CAPTEST_CONNSTR }}
             ;;
-          neon-captest-new | neon-captest-prefetch)
+          neon-captest-new | neon-captest-freetier)
             CONNSTR=${{ steps.create-neon-project.outputs.dsn }}
             ;;
           rds-aurora)
@@ -185,7 +226,7 @@ jobs:
             CONNSTR=${{ secrets.BENCHMARK_RDS_POSTGRES_CONNSTR }}
             ;;
           *)
-            echo 2>&1 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-reuse', 'neon-captest-new', 'neon-captest-prefetch', 'rds-aurora', or 'rds-postgres'"
+            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-reuse', 'neon-captest-new', 'neon-captest-freetier', 'rds-aurora', or 'rds-postgres'"
             exit 1
             ;;
         esac
@@ -194,17 +235,6 @@ jobs:
 
         psql ${CONNSTR} -c "SELECT version();"
 
-    - name: Set database options
-      if: matrix.platform == 'neon-captest-prefetch'
-      run: |
-        DB_NAME=$(psql ${BENCHMARK_CONNSTR} --no-align --quiet -t -c "SELECT current_database()")
-
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET enable_seqscan_prefetch=on"
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET effective_io_concurrency=32"
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET maintenance_io_concurrency=32"
-      env:
-        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
-
     - name: Benchmark init
       uses: ./.github/actions/run-python-test-set
       with:
@@ -252,7 +282,7 @@ jobs:
         api_key: ${{ secrets.NEON_STAGING_API_KEY }}
 
     - name: Create Allure report
-      if: success() || failure()
+      if: ${{ !cancelled() }}
       uses: ./.github/actions/allure-report
       with:
         action: generate
@@ -275,23 +305,19 @@ jobs:
     #
     # *_CLICKBENCH_CONNSTR: Genuine ClickBench DB with ~100M rows
     # *_CLICKBENCH_10M_CONNSTR: DB with the first 10M rows of ClickBench DB
-    if: success() || failure()
-    needs: [ pgbench-compare ]
+    if: ${{ !cancelled() }}
+    needs: [ generate-matrices, pgbench-compare ]
 
     strategy:
       fail-fast: false
-      matrix:
-        # neon-captest-prefetch: We have pre-created projects with prefetch enabled
-        # rds-aurora: Aurora Postgres Serverless v2 with autoscaling from 0.5 to 2 ACUs
-        # rds-postgres: RDS Postgres db.m5.large instance (2 vCPU, 8 GiB) with gp3 EBS storage
-        platform: [ neon-captest-prefetch, rds-postgres, rds-aurora ]
+      matrix: ${{ fromJson(needs.generate-matrices.outputs.olap-compare-matrix) }}
 
     env:
       POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
       DEFAULT_PG_VERSION: 14
       TEST_OUTPUT: /tmp/test_output
       BUILD_TYPE: remote
-      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref == 'refs/heads/main' ) }}
+      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
       PLATFORM: ${{ matrix.platform }}
 
     runs-on: [ self-hosted, us-east-2, x64 ]
@@ -320,7 +346,7 @@ jobs:
       id: set-up-connstr
       run: |
         case "${PLATFORM}" in
-          neon-captest-prefetch)
+          neon-captest-reuse)
             CONNSTR=${{ secrets.BENCHMARK_CAPTEST_CLICKBENCH_10M_CONNSTR }}
             ;;
           rds-aurora)
@@ -330,7 +356,7 @@ jobs:
             CONNSTR=${{ secrets.BENCHMARK_RDS_POSTGRES_CLICKBENCH_10M_CONNSTR }}
             ;;
           *)
-            echo 2>&1 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-prefetch', 'rds-aurora', or 'rds-postgres'"
+            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-reuse', 'rds-aurora', or 'rds-postgres'"
             exit 1
             ;;
         esac
@@ -339,17 +365,6 @@ jobs:
 
         psql ${CONNSTR} -c "SELECT version();"
 
-    - name: Set database options
-      if: matrix.platform == 'neon-captest-prefetch'
-      run: |
-        DB_NAME=$(psql ${BENCHMARK_CONNSTR} --no-align --quiet -t -c "SELECT current_database()")
-
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET enable_seqscan_prefetch=on"
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET effective_io_concurrency=32"
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET maintenance_io_concurrency=32"
-      env:
-        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
-
     - name: ClickBench benchmark
       uses: ./.github/actions/run-python-test-set
       with:
@@ -364,7 +379,7 @@ jobs:
         BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
 
     - name: Create Allure report
-      if: success() || failure()
+      if: ${{ !cancelled() }}
       uses: ./.github/actions/allure-report
       with:
         action: generate
@@ -386,23 +401,19 @@ jobs:
     # We might change it after https://github.com/neondatabase/neon/issues/2900.
     #
     # *_TPCH_S10_CONNSTR: DB generated with scale factor 10 (~10 GB)
-    if: success() || failure()
-    needs: [ clickbench-compare ]
+    if: ${{ !cancelled() }}
+    needs: [ generate-matrices, clickbench-compare ]
 
     strategy:
       fail-fast: false
-      matrix:
-        # neon-captest-prefetch: We have pre-created projects with prefetch enabled
-        # rds-aurora: Aurora Postgres Serverless v2 with autoscaling from 0.5 to 2 ACUs
-        # rds-postgres: RDS Postgres db.m5.large instance (2 vCPU, 8 GiB) with gp3 EBS storage
-        platform: [ neon-captest-prefetch, rds-postgres, rds-aurora ]
+      matrix: ${{ fromJson(needs.generate-matrices.outputs.olap-compare-matrix) }}
 
     env:
       POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
       DEFAULT_PG_VERSION: 14
       TEST_OUTPUT: /tmp/test_output
       BUILD_TYPE: remote
-      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref == 'refs/heads/main' ) }}
+      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
       PLATFORM: ${{ matrix.platform }}
 
     runs-on: [ self-hosted, us-east-2, x64 ]
@@ -431,7 +442,7 @@ jobs:
       id: set-up-connstr
       run: |
         case "${PLATFORM}" in
-          neon-captest-prefetch)
+          neon-captest-reuse)
             CONNSTR=${{ secrets.BENCHMARK_CAPTEST_TPCH_S10_CONNSTR }}
             ;;
           rds-aurora)
@@ -441,7 +452,7 @@ jobs:
             CONNSTR=${{ secrets.BENCHMARK_RDS_POSTGRES_TPCH_S10_CONNSTR }}
             ;;
           *)
-            echo 2>&1 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-prefetch', 'rds-aurora', or 'rds-postgres'"
+            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-reuse', 'rds-aurora', or 'rds-postgres'"
             exit 1
             ;;
         esac
@@ -450,17 +461,6 @@ jobs:
 
         psql ${CONNSTR} -c "SELECT version();"
 
-    - name: Set database options
-      if: matrix.platform == 'neon-captest-prefetch'
-      run: |
-        DB_NAME=$(psql ${BENCHMARK_CONNSTR} --no-align --quiet -t -c "SELECT current_database()")
-
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET enable_seqscan_prefetch=on"
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET effective_io_concurrency=32"
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET maintenance_io_concurrency=32"
-      env:
-        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
-
     - name: Run TPC-H benchmark
       uses: ./.github/actions/run-python-test-set
       with:
@@ -475,7 +475,7 @@ jobs:
         BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
 
     - name: Create Allure report
-      if: success() || failure()
+      if: ${{ !cancelled() }}
       uses: ./.github/actions/allure-report
       with:
         action: generate
@@ -491,23 +491,19 @@ jobs:
         SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
 
   user-examples-compare:
-    if: success() || failure()
-    needs: [ tpch-compare ]
+    if: ${{ !cancelled() }}
+    needs: [ generate-matrices, tpch-compare ]
 
     strategy:
       fail-fast: false
-      matrix:
-        # neon-captest-prefetch: We have pre-created projects with prefetch enabled
-        # rds-aurora: Aurora Postgres Serverless v2 with autoscaling from 0.5 to 2 ACUs
-        # rds-postgres: RDS Postgres db.m5.large instance (2 vCPU, 8 GiB) with gp3 EBS storage
-        platform: [ neon-captest-prefetch, rds-postgres, rds-aurora ]
+      matrix: ${{ fromJson(needs.generate-matrices.outputs.olap-compare-matrix) }}
 
     env:
       POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
       DEFAULT_PG_VERSION: 14
       TEST_OUTPUT: /tmp/test_output
       BUILD_TYPE: remote
-      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref == 'refs/heads/main' ) }}
+      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
       PLATFORM: ${{ matrix.platform }}
 
     runs-on: [ self-hosted, us-east-2, x64 ]
@@ -536,7 +532,7 @@ jobs:
       id: set-up-connstr
       run: |
         case "${PLATFORM}" in
-          neon-captest-prefetch)
+          neon-captest-reuse)
             CONNSTR=${{ secrets.BENCHMARK_USER_EXAMPLE_CAPTEST_CONNSTR }}
             ;;
           rds-aurora)
@@ -546,7 +542,7 @@ jobs:
             CONNSTR=${{ secrets.BENCHMARK_USER_EXAMPLE_RDS_POSTGRES_CONNSTR }}
             ;;
           *)
-            echo 2>&1 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-prefetch', 'rds-aurora', or 'rds-postgres'"
+            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-reuse', 'rds-aurora', or 'rds-postgres'"
             exit 1
             ;;
         esac
@@ -555,17 +551,6 @@ jobs:
 
         psql ${CONNSTR} -c "SELECT version();"
 
-    - name: Set database options
-      if: matrix.platform == 'neon-captest-prefetch'
-      run: |
-        DB_NAME=$(psql ${BENCHMARK_CONNSTR} --no-align --quiet -t -c "SELECT current_database()")
-
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET enable_seqscan_prefetch=on"
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET effective_io_concurrency=32"
-        psql ${BENCHMARK_CONNSTR} -c "ALTER DATABASE ${DB_NAME} SET maintenance_io_concurrency=32"
-      env:
-        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
-
     - name: Run user examples
       uses: ./.github/actions/run-python-test-set
       with:
@@ -580,7 +565,7 @@ jobs:
         BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
 
     - name: Create Allure report
-      if: success() || failure()
+      if: ${{ !cancelled() }}
       uses: ./.github/actions/allure-report
       with:
         action: generate

.github/workflows/build_and_test.yml

@@ -13,7 +13,7 @@ defaults:
 
 concurrency:
   # Allow only one workflow per any non-`main` branch.
-  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.ref == 'refs/heads/main' && github.sha || 'anysha' }}
+  group: ${{ github.workflow }}-${{ github.ref_name }}-${{ github.ref_name == 'main' && github.sha || 'anysha' }}
   cancel-in-progress: true
 
 env:
@@ -74,15 +74,12 @@ jobs:
       - name: Install Python deps
         run: ./scripts/pysync
 
-      - name: Run isort to ensure code format
-        run: poetry run isort --diff --check .
+      - name: Run ruff to ensure code format
+        run: poetry run ruff .
 
       - name: Run black to ensure code format
         run: poetry run black --diff --check .
 
-      - name: Run flake8 to ensure code format
-        run: poetry run flake8 .
-
       - name: Run mypy to check types
         run: poetry run mypy .
 
@@ -187,10 +184,10 @@ jobs:
           CARGO_FEATURES="--features testing"
           if [[ $BUILD_TYPE == "debug" ]]; then
             cov_prefix="scripts/coverage --profraw-prefix=$GITHUB_JOB --dir=/tmp/coverage run"
-            CARGO_FLAGS="--locked $CARGO_FEATURES"
+            CARGO_FLAGS="--locked"
           elif [[ $BUILD_TYPE == "release" ]]; then
             cov_prefix=""
-            CARGO_FLAGS="--locked --release $CARGO_FEATURES"
+            CARGO_FLAGS="--locked --release"
           fi
           echo "cov_prefix=${cov_prefix}" >> $GITHUB_ENV
           echo "CARGO_FEATURES=${CARGO_FEATURES}" >> $GITHUB_ENV
@@ -243,11 +240,18 @@ jobs:
 
       - name: Run cargo build
         run: |
-          ${cov_prefix} mold -run cargo build $CARGO_FLAGS --bins --tests
+          ${cov_prefix} mold -run cargo build $CARGO_FLAGS $CARGO_FEATURES --bins --tests
 
       - name: Run cargo test
         run: |
-          ${cov_prefix} cargo test $CARGO_FLAGS
+          ${cov_prefix} cargo test $CARGO_FLAGS $CARGO_FEATURES
+
+          # Run separate tests for real S3
+          export ENABLE_REAL_S3_REMOTE_STORAGE=nonempty
+          export REMOTE_STORAGE_S3_BUCKET=neon-github-public-dev
+          export REMOTE_STORAGE_S3_REGION=eu-central-1
+          # Avoid `$CARGO_FEATURES` since there's no `testing` feature in the e2e tests now
+          ${cov_prefix} cargo test $CARGO_FLAGS --package remote_storage --test pagination_tests -- s3_pagination_should_work --exact
 
       - name: Install rust binaries
         run: |
@@ -271,7 +275,7 @@ jobs:
             mkdir -p /tmp/neon/test_bin/
 
             test_exe_paths=$(
-              ${cov_prefix} cargo test $CARGO_FLAGS --message-format=json --no-run |
+              ${cov_prefix} cargo test $CARGO_FLAGS $CARGO_FEATURES --message-format=json --no-run |
               jq -r '.executable | select(. != null)'
             )
             for bin in $test_exe_paths; do
@@ -331,6 +335,10 @@ jobs:
           real_s3_region: us-west-2
           real_s3_access_key_id: "${{ secrets.AWS_ACCESS_KEY_ID_CI_TESTS_S3 }}"
           real_s3_secret_access_key: "${{ secrets.AWS_SECRET_ACCESS_KEY_CI_TESTS_S3 }}"
+          rerun_flaky: true
+        env:
+          TEST_RESULT_CONNSTR: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR }}
+          CHECK_ONDISK_DATA_COMPATIBILITY: nonempty
 
       - name: Merge and upload coverage data
         if: matrix.build_type == 'debug'
@@ -360,49 +368,97 @@ jobs:
           build_type: ${{ matrix.build_type }}
           test_selection: performance
           run_in_parallel: false
-          save_perf_report: ${{ github.ref == 'refs/heads/main' }}
+          save_perf_report: ${{ github.ref_name == 'main' }}
         env:
           VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
           PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
       # XXX: no coverage data handling here, since benchmarks are run on release builds,
       # while coverage is currently collected for the debug ones
 
-  merge-allure-report:
+  create-test-report:
     runs-on: [ self-hosted, gen3, small ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
     needs: [ regress-tests, benchmarks ]
     if: ${{ !cancelled() }}
-    strategy:
-      fail-fast: false
-      matrix:
-        build_type: [ debug, release ]
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-        with:
-          submodules: false
 
-      - name: Create Allure report
-        id: create-allure-report
+    steps:
+      - uses: actions/checkout@v3
+
+      - name: Create Allure report (debug)
+        if: ${{ !cancelled() }}
+        id: create-allure-report-debug
         uses: ./.github/actions/allure-report
         with:
           action: generate
-          build_type: ${{ matrix.build_type }}
+          build_type: debug
+
+      - name: Create Allure report (release)
+        if: ${{ !cancelled() }}
+        id: create-allure-report-release
+        uses: ./.github/actions/allure-report
+        with:
+          action: generate
+          build_type: release
+
+      - uses: actions/github-script@v6
+        if: >
+          !cancelled() &&
+          github.event_name == 'pull_request' && (
+            steps.create-allure-report-debug.outputs.report-url ||
+            steps.create-allure-report-release.outputs.report-url
+          )
+        with:
+          # Retry script for 5XX server errors: https://github.com/actions/github-script#retries
+          retries: 5
+          script: |
+            const reports = [{
+              buildType: "debug",
+              reportUrl: "${{ steps.create-allure-report-debug.outputs.report-url }}",
+              jsonUrl:   "${{ steps.create-allure-report-debug.outputs.report-json-url }}",
+            }, {
+              buildType: "release",
+              reportUrl: "${{ steps.create-allure-report-release.outputs.report-url }}",
+              jsonUrl:   "${{ steps.create-allure-report-release.outputs.report-json-url }}",
+            }]
+
+            const script = require("./scripts/pr-comment-test-report.js")
+            await script({
+              github,
+              context,
+              fetch,
+              reports,
+            })
 
       - name: Store Allure test stat in the DB
-        if: ${{ steps.create-allure-report.outputs.report-url }}
+        if: >
+          !cancelled() && (
+            steps.create-allure-report-debug.outputs.report-url ||
+            steps.create-allure-report-release.outputs.report-url
+          )
         env:
-          BUILD_TYPE: ${{ matrix.build_type }}
           SHA: ${{ github.event.pull_request.head.sha || github.sha }}
-          REPORT_URL: ${{ steps.create-allure-report.outputs.report-url }}
+          REPORT_JSON_URL_DEBUG: ${{ steps.create-allure-report-debug.outputs.report-json-url }}
+          REPORT_JSON_URL_RELEASE: ${{ steps.create-allure-report-release.outputs.report-json-url }}
           TEST_RESULT_CONNSTR: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR }}
         run: |
-          curl --fail --output suites.json ${REPORT_URL%/index.html}/data/suites.json
           ./scripts/pysync
 
-          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
+          for report_url in $REPORT_JSON_URL_DEBUG $REPORT_JSON_URL_RELEASE; do
+            if [ -z "$report_url" ]; then
+              continue
+            fi
+
+            if [[ "$report_url" == "$REPORT_JSON_URL_DEBUG" ]]; then
+              BUILD_TYPE=debug
+            else
+              BUILD_TYPE=release
+            fi
+
+            curl --fail --output suites.json "${report_url}"
+            DATABASE_URL="$TEST_RESULT_CONNSTR" poetry run python3 scripts/ingest_regress_test_result.py --revision ${SHA} --reference ${GITHUB_REF} --build-type ${BUILD_TYPE} --ingest suites.json
+          done
 
   coverage-report:
     runs-on: [ self-hosted, gen3, small ]
@@ -551,6 +607,48 @@ jobs:
       - name: Cleanup ECR folder
         run: rm -rf ~/.ecr
 
+
+  neon-image-depot:
+    # For testing this will run side-by-side for a few merges.
+    # This action is not really optimized yet, but gets the job done
+    runs-on: [ self-hosted, gen3, large ]
+    needs: [ tag ]
+    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/base:pinned
+    permissions:
+      contents: read
+      id-token: write
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+        with:
+          submodules: true
+          fetch-depth: 0
+
+      - name: Setup go
+        uses: actions/setup-go@v3
+        with:
+          go-version: '1.19'
+
+      - name: Set up Depot CLI
+        uses: depot/setup-action@v1
+
+      - name: Install Crane & ECR helper
+        run: 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: Build and push
+        uses: depot/build-push-action@v1
+        with:
+          # if no depot.json file is at the root of your repo, you must specify the project id
+          project: nrdv0s4kcs
+          push: true
+          tags: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:depot-${{needs.tag.outputs.build-tag}}
+
   compute-tools-image:
     runs-on: [ self-hosted, gen3, large ]
     needs: [ tag ]
@@ -611,34 +709,31 @@ 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: |
@@ -855,6 +950,16 @@ jobs:
     needs: [ push-docker-hub, tag, regress-tests ]
     if: ( github.ref_name == 'main' || github.ref_name == 'release' ) && github.event_name != 'workflow_dispatch'
     steps:
+      - name: Fix git ownership
+        run: |
+          # Workaround for `fatal: detected dubious ownership in repository at ...`
+          #
+          # Use both ${{ github.workspace }} and ${GITHUB_WORKSPACE} because they're different on host and in containers
+          #   Ref https://github.com/actions/checkout/issues/785
+          #
+          git config --global --add safe.directory ${{ github.workspace }}
+          git config --global --add safe.directory ${GITHUB_WORKSPACE}
+
       - name: Checkout
         uses: actions/checkout@v3
         with:
@@ -902,7 +1007,7 @@ jobs:
 
             S3_KEY=$(aws s3api list-objects-v2 --bucket ${BUCKET} --prefix ${OLD_PREFIX} | jq -r '.Contents[].Key' | grep ${FILENAME} | sort --version-sort | tail -1 || true)
             if [ -z "${S3_KEY}" ]; then
-              echo 2>&1 "Neither s3://${BUCKET}/${OLD_PREFIX}/${FILENAME} nor its version from previous attempts exist"
+              echo >&2 "Neither s3://${BUCKET}/${OLD_PREFIX}/${FILENAME} nor its version from previous attempts exist"
               exit 1
             fi
 

.github/workflows/deploy-dev.yml

@@ -67,7 +67,7 @@ jobs:
           ./get_binaries.sh
 
           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 }}
+          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

.github/workflows/deploy-prod.yml

@@ -40,7 +40,9 @@ concurrency:
 jobs:
   deploy-prod-new:
     runs-on: prod
-    container: 093970136003.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
+    container:
+      image: 093970136003.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
+      options: --user root --privileged
     if: inputs.deployStorage && inputs.disclamerAcknowledged
     defaults:
       run:
@@ -66,7 +68,7 @@ jobs:
           ./get_binaries.sh
 
           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 }}
+          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:
@@ -163,78 +165,3 @@ jobs:
       - 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
-
-  # Deploy to old account below          
-
-  deploy:
-    runs-on: prod
-    container: 093970136003.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
-    if: inputs.deployStorage && inputs.disclamerAcknowledged
-    defaults:
-      run:
-        shell: bash
-    environment:
-      name: prod-old
-    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
-
-          eval $(ssh-agent)
-          echo "${{ secrets.TELEPORT_SSH_KEY }}"  | tr -d '\n'| base64 --decode >ssh-key
-          echo "${{ secrets.TELEPORT_SSH_CERT }}" | tr -d '\n'| base64 --decode >ssh-key-cert.pub
-          chmod 0600 ssh-key
-          ssh-add ssh-key
-          rm -f ssh-key ssh-key-cert.pub
-          ANSIBLE_CONFIG=./ansible.cfg ansible-galaxy collection install sivel.toiletwater
-          ANSIBLE_CONFIG=./ansible.cfg ansible-playbook deploy.yaml -i production.hosts.yaml -e CONSOLE_API_TOKEN=${{ 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
-
-      # Cleanup script fails otherwise - rm: cannot remove '/nvme/actions-runner/_work/_temp/_github_home/.ansible/collections': Permission denied
-      - name: Cleanup ansible folder
-        run: rm -rf ~/.ansible
-
-  deploy-storage-broker:
-    name: deploy storage broker on old staging and old prod
-    runs-on: [ self-hosted, gen3, small ]
-    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:pinned
-    if: inputs.deployStorageBroker && inputs.disclamerAcknowledged
-    defaults:
-      run:
-        shell: bash
-    environment:
-      name: prod-old
-    env:
-      KUBECONFIG: .kubeconfig
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-        with:
-          submodules: true
-          fetch-depth: 0
-          ref: ${{ inputs.branch }}
-
-      - name: Store kubeconfig file
-        run: |
-          echo "${{ secrets.PRODUCTION_KUBECONFIG_DATA }}" | base64 --decode > ${KUBECONFIG}
-          chmod 0600 ${KUBECONFIG}
-
-      - name: Add neon helm chart
-        run: helm repo add neondatabase https://neondatabase.github.io/helm-charts
-
-      - name: Deploy storage-broker
-        run:
-          helm upgrade neon-storage-broker neondatabase/neon-storage-broker --namespace neon-storage-broker --create-namespace --install --atomic -f .github/helm-values/production.neon-storage-broker.yaml --set image.tag=${{ inputs.dockerTag }} --set settings.sentryUrl=${{ secrets.SENTRY_URL_BROKER }} --wait --timeout 5m0s
-
-      - name: Cleanup helm folder
-        run: rm -rf ~/.cache

.github/workflows/neon_extra_builds.yml

@@ -12,7 +12,7 @@ defaults:
 
 concurrency:
   # Allow only one workflow per any non-`main` branch.
-  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.ref == 'refs/heads/main' && github.sha || 'anysha' }}
+  group: ${{ github.workflow }}-${{ github.ref_name }}-${{ github.ref_name == 'main' && github.sha || 'anysha' }}
   cancel-in-progress: true
 
 env:
@@ -53,14 +53,14 @@ jobs:
         uses: actions/cache@v3
         with:
           path: pg_install/v14
-          key: v1-${{ runner.os }}-${{ matrix.build_type }}-pg-${{ steps.pg_v14_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+          key: v1-${{ runner.os }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v14_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
 
       - name: Cache postgres v15 build
         id: cache_pg_15
         uses: actions/cache@v3
         with:
           path: pg_install/v15
-          key: v1-${{ runner.os }}-${{ matrix.build_type }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+          key: v1-${{ runner.os }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
 
       - name: Set extra env for macOS
         run: |

.github/workflows/pg_clients.yml

@@ -14,7 +14,7 @@ on:
 
 concurrency:
   # Allow only one workflow per any non-`main` branch.
-  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.ref == 'refs/heads/main' && github.sha || 'anysha' }}
+  group: ${{ github.workflow }}-${{ github.ref_name }}-${{ github.ref_name == 'main' && github.sha || 'anysha' }}
   cancel-in-progress: true
 
 jobs:

Cargo.toml

@@ -24,10 +24,10 @@ atty = "0.2.14"
 aws-config = { version = "0.51.0", default-features = false, features=["rustls"] }
 aws-sdk-s3 = "0.21.0"
 aws-smithy-http = "0.51.0"
-aws-types = "0.51.0"
+aws-types = "0.55"
 base64 = "0.13.0"
 bincode = "1.3"
-bindgen = "0.61"
+bindgen = "0.65"
 bstr = "1.0"
 byteorder = "1.4"
 bytes = "1.0"
@@ -38,6 +38,7 @@ 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"
@@ -49,12 +50,12 @@ git-version = "0.3"
 hashbrown = "0.13"
 hashlink = "0.8.1"
 hex = "0.4"
-hex-literal = "0.3"
+hex-literal = "0.4"
 hmac = "0.12.1"
 hostname = "0.3.1"
 humantime = "2.1"
 humantime-serde = "1.1.1"
-hyper = "0.14"
+hyper = { version = "0.14",  features = ["http2", "tcp", "runtime", "http1"]}
 hyper-tungstenite = "0.9"
 itertools = "0.10"
 jsonwebtoken = "8"
@@ -63,12 +64,12 @@ md5 = "0.7.0"
 memoffset = "0.8"
 nix = "0.26"
 notify = "5.0.0"
+num_cpus = "1.15"
 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"
-tracing-opentelemetry = "0.18.0"
 parking_lot = "0.12"
 pin-project-lite = "0.2"
 prometheus = {version = "0.13", default_features=false, features = ["process"]} # removes protobuf dependency
@@ -76,59 +77,68 @@ 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"
+rpds = "0.13"
 rustls = "0.20"
 rustls-pemfile = "1"
 rustls-split = "0.3"
 scopeguard = "1.1"
-sentry = { version = "0.29", default-features = false, features = ["backtrace", "contexts", "panic", "rustls", "reqwest" ] }
+sentry = { version = "0.30", default-features = false, features = ["backtrace", "contexts", "panic", "rustls", "reqwest" ] }
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1"
 serde_with = "2.0"
 sha2 = "0.10.2"
 signal-hook = "0.3"
-socket2 = "0.4.4"
+socket2 = "0.5"
 strum = "0.24"
 strum_macros = "0.24"
 svg_fmt = "0.4.1"
+sync_wrapper = "0.1.2"
 tar = "0.4"
+test-context = "0.1"
 thiserror = "1.0"
 tls-listener = { version = "0.6", features = ["rustls", "hyper-h1"] }
 tokio = { version = "1.17", features = ["macros"] }
+tokio-io-timeout = "1.2.0"
 tokio-postgres-rustls = "0.9.0"
 tokio-rustls = "0.23"
 tokio-stream = "0.1"
 tokio-util = { version = "0.7", features = ["io"] }
-toml = "0.5"
-toml_edit = { version = "0.17", features = ["easy"] }
-tonic = {version = "0.8", features = ["tls", "tls-roots"]}
+toml = "0.7"
+toml_edit = "0.19"
+tonic = {version = "0.9", 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"] }
 walkdir = "2.3.2"
-webpki-roots = "0.22.5"
-x509-parser = "0.14"
+webpki-roots = "0.23"
+x509-parser = "0.15"
+percent-encoding = "1.0"
 
 ## TODO replace this with tracing
 env_logger = "0.10"
 log = "0.4"
 
 ## Libraries from neondatabase/ git forks, ideally with changes to be upstreamed
-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="43e6db254a97fdecbce33d8bc0890accfd74495e" }
+postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="43e6db254a97fdecbce33d8bc0890accfd74495e", features = ["with-chrono-0_4", "array-impls"] }
 postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", rev="43e6db254a97fdecbce33d8bc0890accfd74495e" }
-postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", rev="43e6db254a97fdecbce33d8bc0890accfd74495e" }
-tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="43e6db254a97fdecbce33d8bc0890accfd74495e" }
+postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", rev="43e6db254a97fdecbce33d8bc0890accfd74495e", features = ["array-impls"] }
+tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev="43e6db254a97fdecbce33d8bc0890accfd74495e", features = ["array-impls"] }
 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
+compute_api = { version = "0.1", path = "./libs/compute_api/" }
 consumption_metrics = { version = "0.1", path = "./libs/consumption_metrics/" }
 metrics = { version = "0.1", path = "./libs/metrics/" }
 pageserver_api = { version = "0.1", path = "./libs/pageserver_api/" }
+postgres_backend = { version = "0.1", path = "./libs/postgres_backend/" }
 postgres_connection = { version = "0.1", path = "./libs/postgres_connection/" }
 postgres_ffi = { version = "0.1", path = "./libs/postgres_ffi/" }
 pq_proto = { version = "0.1", path = "./libs/pq_proto/" }
@@ -145,9 +155,9 @@ workspace_hack = { version = "0.1", path = "./workspace_hack/" }
 ## Build dependencies
 criterion = "0.4"
 rcgen = "0.10"
-rstest = "0.16"
-tempfile = "3.2"
-tonic-build = "0.8"
+rstest = "0.17"
+tempfile = "3.4"
+tonic-build = "0.9"
 
 # This is only needed for proxy's tests.
 # TODO: we should probably fork `tokio-postgres-rustls` instead.

Dockerfile

@@ -39,7 +39,7 @@ ARG CACHEPOT_BUCKET=neon-github-dev
 
 COPY --from=pg-build /home/nonroot/pg_install/v14/include/postgresql/server pg_install/v14/include/postgresql/server
 COPY --from=pg-build /home/nonroot/pg_install/v15/include/postgresql/server pg_install/v15/include/postgresql/server
-COPY . .
+COPY --chown=nonroot . .
 
 # Show build caching stats to check if it was used in the end.
 # Has to be the part of the same RUN since cachepot daemon is killed in the end of this RUN, losing the compilation stats.

Dockerfile.compute-node

@@ -1,3 +1,4 @@
+ARG PG_VERSION
 ARG REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com
 ARG IMAGE=rust
 ARG TAG=pinned
@@ -11,7 +12,7 @@ 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
+    libicu-dev libxslt1-dev liblz4-dev libzstd-dev
 
 #########################################################################################
 #
@@ -23,18 +24,30 @@ FROM build-deps AS pg-build
 ARG PG_VERSION
 COPY vendor/postgres-${PG_VERSION} postgres
 RUN cd postgres && \
-    ./configure CFLAGS='-O2 -g3' --enable-debug --with-openssl --with-uuid=ossp --with-icu && \
+    export CONFIGURE_CMD="./configure CFLAGS='-O2 -g3' --enable-debug --with-openssl --with-uuid=ossp \
+    --with-icu --with-libxml --with-libxslt --with-lz4" && \
+    if [ "${PG_VERSION}" != "v14" ]; then \
+        # zstd is available only from PG15
+        export CONFIGURE_CMD="${CONFIGURE_CMD} --with-zstd"; \
+    fi && \
+    eval $CONFIGURE_CMD && \
     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/pgrowlocks.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/pg_stat_statements.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/earthdistance.control
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/refint.control && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/xml2.control
 
 #########################################################################################
 #
@@ -50,17 +63,20 @@ RUN apt update && \
     libcgal-dev libgdal-dev libgmp-dev libmpfr-dev libopenscenegraph-dev libprotobuf-c-dev \
     protobuf-c-compiler xsltproc
 
-RUN wget https://gitlab.com/Oslandia/SFCGAL/-/archive/v1.3.10/SFCGAL-v1.3.10.tar.gz && \
-    tar zxvf SFCGAL-v1.3.10.tar.gz && \
-    cd SFCGAL-v1.3.10 && cmake . && make -j $(getconf _NPROCESSORS_ONLN) && \
+# 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 && \
+    echo "4e39b3b2adada6254a7bdba6d297bb28e1a9835a9f879b74f37e2dab70203232 SFCGAL.tar.gz" | sha256sum --check && \
+    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/* /
 
-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 && \
+ENV PATH "/usr/local/pgsql/bin:$PATH"
+
+RUN wget https://download.osgeo.org/postgis/source/postgis-3.3.2.tar.gz -O postgis.tar.gz && \
+    echo "9a2a219da005a1730a39d1959a1c7cec619b1efb009b65be80ffc25bad299068 postgis.tar.gz" | sha256sum --check && \
+    mkdir postgis-src && cd postgis-src && tar xvzf ../postgis.tar.gz --strip-components=1 -C . && \
     ./autogen.sh && \
-    export PATH="/usr/local/pgsql/bin:$PATH" && \
     ./configure --with-sfcgal=/usr/local/bin/sfcgal-config && \
     make -j $(getconf _NPROCESSORS_ONLN) install && \
     cd extensions/postgis && \
@@ -74,6 +90,16 @@ RUN wget https://download.osgeo.org/postgis/source/postgis-3.3.1.tar.gz && \
     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 && \
+    echo "cac297c07d34460887c4f3b522b35c470138760fe358e351ad1db4edb6ee306e pgrouting.tar.gz" | sha256sum --check && \
+    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"
@@ -83,30 +109,18 @@ RUN wget https://download.osgeo.org/postgis/source/postgis-3.3.1.tar.gz && \
 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
+    apt install -y ninja-build python3-dev libncurses5 binutils clang
 
-# 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 $(getconf _NPROCESSORS_ONLN) && make install && \
-    cp /usr/local/bin/ld.gold /usr/bin/gold
-
-# Sed is used to patch for https://github.com/plv8/plv8/issues/503
-RUN wget https://github.com/plv8/plv8/archive/refs/tags/v3.1.4.tar.gz && \
-    tar xvzf v3.1.4.tar.gz && \
-    cd plv8-3.1.4 && \
+RUN wget https://github.com/plv8/plv8/archive/refs/tags/v3.1.5.tar.gz -O plv8.tar.gz && \
+    echo "1e108d5df639e4c189e1c5bdfa2432a521c126ca89e7e5a969d46899ca7bf106 plv8.tar.gz" | sha256sum --check && \
+    mkdir plv8-src && cd plv8-src && tar xvzf ../plv8.tar.gz --strip-components=1 -C . && \
     export PATH="/usr/local/pgsql/bin:$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
+    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
 
 #########################################################################################
 #
@@ -120,24 +134,24 @@ 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 \
+      && echo "739d372726cb23129d57a539ce1432453448816e345e1545f6127296926b6754 /tmp/cmake-install.sh" | sha256sum --check \
       && 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 && \
+RUN wget https://github.com/uber/h3/archive/refs/tags/v4.1.0.tar.gz -O h3.tar.gz && \
+    echo "ec99f1f5974846bde64f4513cf8d2ea1b8d172d2218ab41803bf6a63532272bc h3.tar.gz" | sha256sum --check && \
+    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/v4.0.1.tar.gz -O h3-pg.tgz && \
-    tar xvzf h3-pg.tgz && \
-    cd h3-pg-4.0.1 && \
+RUN wget https://github.com/zachasme/h3-pg/archive/refs/tags/v4.1.2.tar.gz -O h3-pg.tar.gz && \
+    echo "c135aa45999b2ad1326d2537c1cadef96d52660838e4ca371706c08fdea1a956 h3-pg.tar.gz" | sha256sum --check && \
+    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 && \
@@ -153,9 +167,9 @@ RUN wget https://github.com/zachasme/h3-pg/archive/refs/tags/v4.0.1.tar.gz -O h3
 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 && \
-    tar xvzf 7.7.tar.gz && \
-    cd postgresql-unit-7.7 && \
+RUN wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.7.tar.gz -O postgresql-unit.tar.gz && \
+    echo "411d05beeb97e5a4abf17572bfcfbb5a68d98d1018918feff995f6ee3bb03e79 postgresql-unit.tar.gz" | sha256sum --check && \
+    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.
@@ -165,6 +179,305 @@ RUN wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.7.tar.gz &
     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 && \
+    echo "b76cf84ddad452cc880a6c8c661d137ddd8679c000a16332f4f03ecf6e10bcc8 pgvector.tar.gz" | sha256sum --check && \
+    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 && \
+    echo "cfdefb15007286f67d3d45510f04a6a7a495004be5b3aecb12cda667e774203f pgjwt.tar.gz" | sha256sum --check && \
+    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 && \
+    echo "e7f01ee0259dc1713f318a108f987663d60f3041948c2ada57a94b469565ca8e hypopg.tar.gz" | sha256sum --check && \
+    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 && \
+    echo "74576b992d9277c92196dd8d816baa2cc2d8046fe102f3dcd7f3c3febed6822a pg_hashids.tar.gz" | sha256sum --check && \
+    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 "rum-pg-build"
+# compile rum extension
+#
+#########################################################################################
+FROM build-deps AS rum-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN wget https://github.com/postgrespro/rum/archive/refs/tags/1.3.13.tar.gz -O rum.tar.gz && \
+    echo "6ab370532c965568df6210bd844ac6ba649f53055e48243525b0b7e5c4d69a7d rum.tar.gz" | sha256sum --check && \
+    mkdir rum-src && cd rum-src && tar xvzf ../rum.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/rum.control
+
+#########################################################################################
+#
+# Layer "pgtap-pg-build"
+# compile pgTAP extension
+#
+#########################################################################################
+FROM build-deps AS pgtap-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN wget https://github.com/theory/pgtap/archive/refs/tags/v1.2.0.tar.gz -O pgtap.tar.gz && \
+    echo "9c7c3de67ea41638e14f06da5da57bac6f5bd03fea05c165a0ec862205a5c052 pgtap.tar.gz" | sha256sum --check && \
+    mkdir pgtap-src && cd pgtap-src && tar xvzf ../pgtap.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/pgtap.control
+
+#########################################################################################
+#
+# Layer "ip4r-pg-build"
+# compile ip4r extension
+#
+#########################################################################################
+FROM build-deps AS ip4r-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN wget https://github.com/RhodiumToad/ip4r/archive/refs/tags/2.4.1.tar.gz -O ip4r.tar.gz && \
+    echo "78b9f0c1ae45c22182768fe892a32d533c82281035e10914111400bf6301c726 ip4r.tar.gz" | sha256sum --check && \
+    mkdir ip4r-src && cd ip4r-src && tar xvzf ../ip4r.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/ip4r.control
+
+#########################################################################################
+#
+# Layer "prefix-pg-build"
+# compile Prefix extension
+#
+#########################################################################################
+FROM build-deps AS prefix-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN wget https://github.com/dimitri/prefix/archive/refs/tags/v1.2.9.tar.gz -O prefix.tar.gz && \
+    echo "38d30a08d0241a8bbb8e1eb8f0152b385051665a8e621c8899e7c5068f8b511e prefix.tar.gz" | sha256sum --check && \
+    mkdir prefix-src && cd prefix-src && tar xvzf ../prefix.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/prefix.control
+
+#########################################################################################
+#
+# Layer "hll-pg-build"
+# compile hll extension
+#
+#########################################################################################
+FROM build-deps AS hll-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN wget https://github.com/citusdata/postgresql-hll/archive/refs/tags/v2.17.tar.gz -O hll.tar.gz && \
+    echo "9a18288e884f197196b0d29b9f178ba595b0dfc21fbf7a8699380e77fa04c1e9 hll.tar.gz" | sha256sum --check && \
+    mkdir hll-src && cd hll-src && tar xvzf ../hll.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/hll.control
+
+#########################################################################################
+#
+# Layer "plpgsql-check-pg-build"
+# compile plpgsql_check extension
+#
+#########################################################################################
+FROM build-deps AS plpgsql-check-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN wget https://github.com/okbob/plpgsql_check/archive/refs/tags/v2.3.2.tar.gz -O plpgsql_check.tar.gz && \
+    echo "9d81167c4bbeb74eebf7d60147b21961506161addc2aee537f95ad8efeae427b plpgsql_check.tar.gz" | sha256sum --check && \
+    mkdir plpgsql_check-src && cd plpgsql_check-src && tar xvzf ../plpgsql_check.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/plpgsql_check.control
+
+#########################################################################################
+#
+# Layer "timescaledb-pg-build"
+# compile timescaledb extension
+#
+#########################################################################################
+FROM build-deps AS timescaledb-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+ENV PATH "/usr/local/pgsql/bin:$PATH"
+
+RUN apt-get update && \
+    apt-get install -y cmake && \
+    wget https://github.com/timescale/timescaledb/archive/refs/tags/2.10.1.tar.gz -O timescaledb.tar.gz && \
+    echo "6fca72a6ed0f6d32d2b3523951ede73dc5f9b0077b38450a029a5f411fdb8c73 timescaledb.tar.gz" | sha256sum --check && \
+    mkdir timescaledb-src && cd timescaledb-src && tar xvzf ../timescaledb.tar.gz --strip-components=1 -C . && \
+    ./bootstrap -DSEND_TELEMETRY_DEFAULT:BOOL=OFF -DUSE_TELEMETRY:BOOL=OFF -DAPACHE_ONLY:BOOL=ON && \
+    cd build && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make install -j $(getconf _NPROCESSORS_ONLN) && \
+    echo "trusted = true" >> /usr/local/pgsql/share/extension/timescaledb.control
+
+#########################################################################################
+#
+# Layer "pg-hint-plan-pg-build"
+# compile pg_hint_plan extension
+#
+#########################################################################################
+FROM build-deps AS pg-hint-plan-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+ARG PG_VERSION
+ENV PATH "/usr/local/pgsql/bin:$PATH"
+
+RUN case "${PG_VERSION}" in \
+      "v14") \
+        export PG_HINT_PLAN_VERSION=14_1_4_1 \
+        export PG_HINT_PLAN_CHECKSUM=c3501becf70ead27f70626bce80ea401ceac6a77e2083ee5f3ff1f1444ec1ad1 \
+        ;; \
+      "v15") \
+        export PG_HINT_PLAN_VERSION=15_1_5_0 \
+        export PG_HINT_PLAN_CHECKSUM=564cbbf4820973ffece63fbf76e3c0af62c4ab23543142c7caaa682bc48918be \
+        ;; \
+      *) \
+        echo "Export the valid PG_HINT_PLAN_VERSION variable" && exit 1 \
+        ;; \
+    esac && \
+    wget https://github.com/ossc-db/pg_hint_plan/archive/refs/tags/REL${PG_HINT_PLAN_VERSION}.tar.gz -O pg_hint_plan.tar.gz && \
+    echo "${PG_HINT_PLAN_CHECKSUM} pg_hint_plan.tar.gz" | sha256sum --check && \
+    mkdir pg_hint_plan-src && cd pg_hint_plan-src && tar xvzf ../pg_hint_plan.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make install -j $(getconf _NPROCESSORS_ONLN) && \
+    echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_hint_plan.control
+
+#########################################################################################
+#
+# Layer "rust extensions"
+# This layer is used to build `pgx` deps
+#
+#########################################################################################
+FROM build-deps AS rust-extensions-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+RUN apt-get update && \
+    apt-get install -y curl libclang-dev cmake && \
+    useradd -ms /bin/bash nonroot -b /home
+
+ENV HOME=/home/nonroot
+ENV PATH="/home/nonroot/.cargo/bin:/usr/local/pgsql/bin/:$PATH"
+USER nonroot
+WORKDIR /home/nonroot
+ARG PG_VERSION
+
+RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && \
+    chmod +x rustup-init && \
+    ./rustup-init -y --no-modify-path --profile minimal --default-toolchain stable && \
+    rm rustup-init && \
+    cargo install --locked --version 0.7.3 cargo-pgx && \
+    /bin/bash -c 'cargo pgx init --pg${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
+
+# caeab60d70b2fd3ae421ec66466a3abbb37b7ee6 made on 06/03/2023
+# there is no release tag yet, but we need it due to the superuser fix in the control file, switch to git tag after release >= 0.1.5
+RUN wget https://github.com/supabase/pg_jsonschema/archive/caeab60d70b2fd3ae421ec66466a3abbb37b7ee6.tar.gz -O pg_jsonschema.tar.gz && \
+    echo "54129ce2e7ee7a585648dbb4cef6d73f795d94fe72f248ac01119992518469a4 pg_jsonschema.tar.gz" | sha256sum --check && \
+    mkdir pg_jsonschema-src && cd pg_jsonschema-src && tar xvzf ../pg_jsonschema.tar.gz --strip-components=1 -C . && \
+    sed -i 's/pgx = "0.7.1"/pgx = { version = "0.7.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    cargo pgx install --release && \
+    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
+
+# b4988843647450a153439be367168ed09971af85 made on 22/02/2023 (from remove-pgx-contrib-spiext branch)
+# Currently pgx version bump to >= 0.7.2  causes "call to unsafe function" compliation errors in
+# pgx-contrib-spiext. There is a branch that removes that dependency, so use it. It is on the
+# same 1.1 version we've used before.
+RUN wget https://github.com/yrashk/pg_graphql/archive/b4988843647450a153439be367168ed09971af85.tar.gz -O pg_graphql.tar.gz && \
+    echo "0c7b0e746441b2ec24187d0e03555faf935c2159e2839bddd14df6dafbc8c9bd pg_graphql.tar.gz" | sha256sum --check && \
+    mkdir pg_graphql-src && cd pg_graphql-src && tar xvzf ../pg_graphql.tar.gz --strip-components=1 -C . && \
+    sed -i 's/pgx = "~0.7.1"/pgx = { version = "0.7.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgx-tests = "~0.7.1"/pgx-tests = "0.7.3"/g' Cargo.toml && \
+    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 "pg-tiktoken-build"
+# Compile "pg_tiktoken" extension
+#
+#########################################################################################
+
+FROM rust-extensions-build AS pg-tiktoken-pg-build
+
+# 801f84f08c6881c8aa30f405fafbf00eec386a72 made on 10/03/2023
+RUN wget https://github.com/kelvich/pg_tiktoken/archive/801f84f08c6881c8aa30f405fafbf00eec386a72.tar.gz -O pg_tiktoken.tar.gz && \
+    echo "52f60ac800993a49aa8c609961842b611b6b1949717b69ce2ec9117117e16e4a pg_tiktoken.tar.gz" | sha256sum --check && \
+    mkdir pg_tiktoken-src && cd pg_tiktoken-src && tar xvzf ../pg_tiktoken.tar.gz --strip-components=1 -C . && \
+    cargo pgx install --release && \
+    echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_tiktoken.control
+
 #########################################################################################
 #
 # Layer "neon-pg-ext-build"
@@ -178,11 +491,30 @@ 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=pg-tiktoken-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 --from=rum-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=pgtap-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=ip4r-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=prefix-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=hll-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=plpgsql-check-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=timescaledb-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=pg-hint-plan-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 && \
+    make -j $(getconf _NPROCESSORS_ONLN) \
+        PG_CONFIG=/usr/local/pgsql/bin/pg_config \
+        -C pgxn/neon_utils \
         -s install
 
 #########################################################################################
@@ -228,20 +560,27 @@ RUN mkdir /var/db && useradd -m -d /var/db/postgres postgres && \
     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
+    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)
+# libicu67, locales for collations (including ICU and plpgsql_check)
+# liblz4-1 for lz4
 # libossp-uuid16 for extension ossp-uuid
 # libgeos, libgdal, libsfcgal1, libproj and libprotobuf-c1 for PostGIS
+# libxml2, libxslt1.1 for xml2
+# libzstd1 for zstd
 RUN apt update &&  \
     apt install --no-install-recommends -y \
+        gdb \
         locales \
         libicu67 \
+        liblz4-1 \
         libreadline8 \
         libossp-uuid16 \
         libgeos-c1v5 \
@@ -249,7 +588,10 @@ RUN apt update &&  \
         libproj19 \
         libprotobuf-c1 \
         libsfcgal1 \
-        gdb && \
+        libxml2 \
+        libxslt1.1 \
+        libzstd1 \
+        procps && \
     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
 

Dockerfile.vm-compute-node

@@ -0,0 +1,70 @@
+# Note: this file *mostly* just builds on Dockerfile.compute-node
+
+ARG SRC_IMAGE
+ARG VM_INFORMANT_VERSION=v0.1.14
+# on libcgroup update, make sure to check bootstrap.sh for changes
+ARG LIBCGROUP_VERSION=v2.0.3
+
+# Pull VM informant, to copy from later
+FROM neondatabase/vm-informant:$VM_INFORMANT_VERSION as informant
+
+# Build cgroup-tools
+#
+# At time of writing (2023-03-14), debian bullseye has a version of cgroup-tools (technically
+# libcgroup) that doesn't support cgroup v2 (version 0.41-11). Unfortunately, the vm-informant
+# requires cgroup v2, so we'll build cgroup-tools ourselves.
+FROM debian:bullseye-slim as libcgroup-builder
+ARG LIBCGROUP_VERSION
+
+RUN set -exu \
+	&& apt update \
+	&& apt install --no-install-recommends -y \
+		git \
+		ca-certificates \
+		automake \
+		cmake \
+		make \
+		gcc \
+		byacc \
+		flex \
+		libtool \
+		libpam0g-dev \
+	&& git clone --depth 1 -b $LIBCGROUP_VERSION https://github.com/libcgroup/libcgroup \
+	&& INSTALL_DIR="/libcgroup-install" \
+	&& mkdir -p "$INSTALL_DIR/bin" "$INSTALL_DIR/include" \
+	&& cd libcgroup \
+	# extracted from bootstrap.sh, with modified flags:
+	&& (test -d m4 || mkdir m4) \
+	&& autoreconf -fi \
+	&& rm -rf autom4te.cache \
+	&& CFLAGS="-O3" ./configure --prefix="$INSTALL_DIR" --sysconfdir=/etc --localstatedir=/var --enable-opaque-hierarchy="name=systemd" \
+	# actually build the thing...
+	&& make install
+
+# Combine, starting from non-VM compute node image.
+FROM $SRC_IMAGE as base
+
+# Temporarily set user back to root so we can run adduser, set inittab
+USER root
+RUN adduser vm-informant --disabled-password --no-create-home
+
+RUN set -e \
+	&& rm -f /etc/inittab \
+	&& touch /etc/inittab
+
+RUN set -e \
+	&& echo "::sysinit:cgconfigparser -l /etc/cgconfig.conf -s 1664" >> /etc/inittab \
+	&& CONNSTR="dbname=neondb user=cloud_admin sslmode=disable" \
+	&& ARGS="--auto-restart --cgroup=neon-postgres --pgconnstr=\"$CONNSTR\"" \
+	&& echo "::respawn:su vm-informant -c '/usr/local/bin/vm-informant $ARGS'" >> /etc/inittab
+
+USER postgres
+
+ADD vm-cgconfig.conf /etc/cgconfig.conf
+COPY --from=informant /usr/bin/vm-informant /usr/local/bin/vm-informant
+
+COPY --from=libcgroup-builder /libcgroup-install/bin/* /usr/bin/
+COPY --from=libcgroup-builder /libcgroup-install/lib/* /usr/lib/
+COPY --from=libcgroup-builder /libcgroup-install/sbin/* /usr/sbin/
+
+ENTRYPOINT ["/usr/sbin/cgexec", "-g", "*:neon-postgres", "/usr/local/bin/compute_ctl"]

Makefile

@@ -133,12 +133,26 @@ neon-pg-ext-%: postgres-%
 	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/$*/bin/pg_config CFLAGS='$(PG_CFLAGS) $(COPT)' \
 		-C $(POSTGRES_INSTALL_DIR)/build/neon-test-utils-$* \
 		-f $(ROOT_PROJECT_DIR)/pgxn/neon_test_utils/Makefile install
+	+@echo "Compiling neon_utils $*"
+	mkdir -p $(POSTGRES_INSTALL_DIR)/build/neon-utils-$*
+	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/$*/bin/pg_config CFLAGS='$(PG_CFLAGS) $(COPT)' \
+		-C $(POSTGRES_INSTALL_DIR)/build/neon-utils-$* \
+		-f $(ROOT_PROJECT_DIR)/pgxn/neon_utils/Makefile install
 
 .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
+	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/$*/bin/pg_config \
+	-C $(POSTGRES_INSTALL_DIR)/build/neon-utils-$* \
+	-f $(ROOT_PROJECT_DIR)/pgxn/neon_utils/Makefile clean
 
 .PHONY: neon-pg-ext
 neon-pg-ext: \

README.md

@@ -34,6 +34,13 @@ 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
+```
+
+Building Neon requires 3.15+ version of `protoc` (protobuf-compiler). If your distribution provides an older version, you can install a newer version from [here](https://github.com/protocolbuffers/protobuf/releases).
 
 2. [Install Rust](https://www.rust-lang.org/tools/install)
 ```
@@ -41,11 +48,14 @@ dnf install flex bison readline-devel zlib-devel openssl-devel \
 curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
 ```
 
-#### Installing dependencies on OSX (12.3.1)
+#### Installing dependencies on macOS (12.3.1)
 1. Install XCode and dependencies
 ```
 xcode-select --install
 brew install protobuf openssl flex bison
+
+# add openssl to PATH, required for ed25519 keys generation in neon_local
+echo 'export PATH="$(brew --prefix openssl)/bin:$PATH"' >> ~/.zshrc
 ```
 
 2. [Install Rust](https://www.rust-lang.org/tools/install)
@@ -83,9 +93,10 @@ 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
@@ -99,9 +110,10 @@ 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`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
@@ -135,15 +147,15 @@ Created an initial timeline 'de200bd42b49cc1814412c7e592dd6e9' at Lsn 0/16B5A50
 Setting tenant 9ef87a5bf0d92544f6fafeeb3239695c as a default one
 
 # start postgres compute node
-> ./target/debug/neon_local pg start main
-Starting new postgres (v14) main on timeline de200bd42b49cc1814412c7e592dd6e9 ...
+> ./target/debug/neon_local endpoint start main
+Starting new endpoint main (PostgreSQL v14) on timeline de200bd42b49cc1814412c7e592dd6e9 ...
 Extracting base backup to create postgres instance: path=.neon/pgdatadirs/tenants/9ef87a5bf0d92544f6fafeeb3239695c/main port=55432
-Starting postgres node at 'host=127.0.0.1 port=55432 user=cloud_admin dbname=postgres'
+Starting postgres at 'host=127.0.0.1 port=55432 user=cloud_admin dbname=postgres'
 
 # check list of running postgres instances
-> ./target/debug/neon_local pg list
- NODE  ADDRESS          TIMELINE                          BRANCH NAME  LSN        STATUS
- main  127.0.0.1:55432  de200bd42b49cc1814412c7e592dd6e9  main         0/16B5BA8  running
+> ./target/debug/neon_local endpoint list
+ ENDPOINT  ADDRESS          TIMELINE                          BRANCH NAME  LSN        STATUS
+ main      127.0.0.1:55432  de200bd42b49cc1814412c7e592dd6e9  main         0/16B5BA8  running
 ```
 
 2. Now, it is possible to connect to postgres and run some queries:
@@ -172,14 +184,14 @@ Created timeline 'b3b863fa45fa9e57e615f9f2d944e601' at Lsn 0/16F9A00 for tenant:
 (L) ┗━ @0/16F9A00: migration_check [b3b863fa45fa9e57e615f9f2d944e601]
 
 # start postgres on that branch
-> ./target/debug/neon_local pg start migration_check --branch-name migration_check
-Starting new postgres migration_check on timeline b3b863fa45fa9e57e615f9f2d944e601 ...
+> ./target/debug/neon_local endpoint start migration_check --branch-name migration_check
+Starting new endpoint migration_check (PostgreSQL v14) on timeline b3b863fa45fa9e57e615f9f2d944e601 ...
 Extracting base backup to create postgres instance: path=.neon/pgdatadirs/tenants/9ef87a5bf0d92544f6fafeeb3239695c/migration_check port=55433
-Starting postgres node at 'host=127.0.0.1 port=55433 user=cloud_admin dbname=postgres'
+Starting postgres at 'host=127.0.0.1 port=55433 user=cloud_admin dbname=postgres'
 
 # check the new list of running postgres instances
-> ./target/debug/neon_local pg list
- NODE             ADDRESS          TIMELINE                          BRANCH NAME      LSN        STATUS
+> ./target/debug/neon_local endpoint list
+ ENDPOINT         ADDRESS          TIMELINE                          BRANCH NAME      LSN        STATUS
  main             127.0.0.1:55432  de200bd42b49cc1814412c7e592dd6e9  main             0/16F9A38  running
  migration_check  127.0.0.1:55433  b3b863fa45fa9e57e615f9f2d944e601  migration_check  0/16F9A70  running
 

compute_tools/Cargo.toml

@@ -11,12 +11,14 @@ clap.workspace = true
 futures.workspace = true
 hyper = { workspace = true, features = ["full"] }
 notify.workspace = true
+num_cpus.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
@@ -25,4 +27,6 @@ tracing-subscriber.workspace = true
 tracing-utils.workspace = true
 url.workspace = true
 
+compute_api.workspace = true
+utils.workspace = true
 workspace_hack.workspace = true

compute_tools/src/bin/compute_ctl.rs

@@ -34,23 +34,24 @@ use std::fs::File;
 use std::panic;
 use std::path::Path;
 use std::process::exit;
-use std::sync::{Arc, RwLock};
+use std::sync::{mpsc, Arc, Condvar, Mutex};
 use std::{thread, time::Duration};
 
 use anyhow::{Context, Result};
 use chrono::Utc;
 use clap::Arg;
 use tracing::{error, info};
+use url::Url;
 
-use compute_tools::compute::{ComputeMetrics, ComputeNode, ComputeState, ComputeStatus};
+use compute_api::responses::ComputeStatus;
+
+use compute_tools::compute::{ComputeNode, ComputeState, ParsedSpec};
+use compute_tools::configurator::launch_configurator;
 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::*;
-use compute_tools::pg_helpers::*;
 use compute_tools::spec::*;
-use url::Url;
 
 fn main() -> Result<()> {
     init_tracing_and_logging(DEFAULT_LOG_LEVEL)?;
@@ -63,27 +64,93 @@ fn main() -> Result<()> {
     let connstr = matches
         .get_one::<String>("connstr")
         .expect("Postgres connection string is required");
-    let spec = matches.get_one::<String>("spec");
+    let spec_json = 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();
 
-    let spec: ComputeSpec = match spec {
+    let mut spec = None;
+    let mut live_config_allowed = false;
+    match spec_json {
         // First, try to get cluster spec from the cli argument
-        Some(json) => serde_json::from_str(json)?,
+        Some(json) => {
+            spec = Some(serde_json::from_str(json)?);
+        }
         None => {
             // Second, try to read it from the file if path is provided
             if let Some(sp) = spec_path {
                 let path = Path::new(sp);
                 let file = File::open(path)?;
-                serde_json::from_reader(file)?
+                spec = Some(serde_json::from_reader(file)?);
+            } else if let Some(id) = compute_id {
+                if let Some(cp_base) = control_plane_uri {
+                    live_config_allowed = true;
+                    if let Ok(s) = get_spec_from_control_plane(cp_base, id) {
+                        spec = Some(s);
+                    }
+                } else {
+                    panic!("must specify both --control-plane-uri and --compute-id or none");
+                }
             } else {
-                panic!("cluster spec should be provided via --spec or --spec-path argument");
+                panic!(
+                    "compute spec should be provided by one of the following ways: \
+                    --spec OR --spec-path OR --control-plane-uri and --compute-id"
+                );
             }
         }
     };
 
+    let mut new_state = ComputeState::new();
+    let spec_set;
+    if let Some(spec) = spec {
+        let pspec = ParsedSpec::try_from(spec).map_err(|msg| anyhow::anyhow!(msg))?;
+        new_state.pspec = Some(pspec);
+        spec_set = true;
+    } else {
+        spec_set = false;
+    }
+    let compute_node = ComputeNode {
+        start_time: Utc::now(),
+        connstr: Url::parse(connstr).context("cannot parse connstr as a URL")?,
+        pgdata: pgdata.to_string(),
+        pgbin: pgbin.to_string(),
+        live_config_allowed,
+        state: Mutex::new(new_state),
+        state_changed: Condvar::new(),
+    };
+    let compute = Arc::new(compute_node);
+
+    // Launch http service first, so we were able to serve control-plane
+    // requests, while configuration is still in progress.
+    let _http_handle = launch_http_server(&compute).expect("cannot launch http endpoint thread");
+
+    if !spec_set {
+        // No spec provided, hang waiting for it.
+        info!("no compute spec provided, waiting");
+        let mut state = compute.state.lock().unwrap();
+        while state.status != ComputeStatus::ConfigurationPending {
+            state = compute.state_changed.wait(state).unwrap();
+
+            if state.status == ComputeStatus::ConfigurationPending {
+                info!("got spec, continue configuration");
+                // Spec is already set by the http server handler.
+                break;
+            }
+        }
+    }
+
+    // We got all we need, update the state.
+    let mut state = compute.state.lock().unwrap();
+    let pspec = state.pspec.as_ref().expect("spec must be set");
+    let startup_tracing_context = pspec.spec.startup_tracing_context.clone();
+    state.status = ComputeStatus::Init;
+    compute.state_changed.notify_all();
+    drop(state);
+
     // Extract OpenTelemetry context for the startup actions from the spec, and
     // attach it to the current tracing context.
     //
@@ -99,7 +166,7 @@ fn main() -> Result<()> {
     // 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 {
+    let startup_context_guard = if let Some(ref carrier) = startup_tracing_context {
         use opentelemetry::propagation::TextMapPropagator;
         use opentelemetry::sdk::propagation::TraceContextPropagator;
         Some(TraceContextPropagator::new().extract(carrier).attach())
@@ -107,42 +174,10 @@ fn main() -> Result<()> {
         None
     };
 
-    let pageserver_connstr = spec
-        .cluster
-        .settings
-        .find("neon.pageserver_connstring")
-        .expect("pageserver connstr should be provided");
-    let tenant = spec
-        .cluster
-        .settings
-        .find("neon.tenant_id")
-        .expect("tenant id should be provided");
-    let timeline = spec
-        .cluster
-        .settings
-        .find("neon.timeline_id")
-        .expect("tenant id should be provided");
-
-    let compute_state = ComputeNode {
-        start_time: Utc::now(),
-        connstr: Url::parse(connstr).context("cannot parse connstr as a URL")?,
-        pgdata: pgdata.to_string(),
-        pgbin: pgbin.to_string(),
-        spec,
-        tenant,
-        timeline,
-        pageserver_connstr,
-        metrics: ComputeMetrics::default(),
-        state: RwLock::new(ComputeState::new()),
-    };
-    let compute = Arc::new(compute_state);
-
-    // Launch service threads first, so we were able to serve availability
-    // requests, while configuration is still in progress.
-    let _http_handle = launch_http_server(&compute).expect("cannot launch http endpoint thread");
+    // Launch remaining service threads
     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");
+    let _configurator_handle =
+        launch_configurator(&compute).expect("cannot launch configurator thread");
 
     // Start Postgres
     let mut delay_exit = false;
@@ -151,7 +186,7 @@ fn main() -> Result<()> {
         Ok(pg) => Some(pg),
         Err(err) => {
             error!("could not start the compute node: {:?}", err);
-            let mut state = compute.state.write().unwrap();
+            let mut state = compute.state.lock().unwrap();
             state.error = Some(format!("{:?}", err));
             state.status = ComputeStatus::Failed;
             drop(state);
@@ -182,13 +217,29 @@ fn main() -> Result<()> {
     if delay_exit {
         info!("giving control plane 30s to collect the error before shutdown");
         thread::sleep(Duration::from_secs(30));
-        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();
+    // pending traces before we exit. Shutting down OTEL tracing provider may
+    // hang for quite some time, see, for example:
+    // - https://github.com/open-telemetry/opentelemetry-rust/issues/868
+    // - and our problems with staging https://github.com/neondatabase/cloud/issues/3707#issuecomment-1493983636
+    //
+    // Yet, we want computes to shut down fast enough, as we may need a new one
+    // for the same timeline ASAP. So wait no longer than 2s for the shutdown to
+    // complete, then just error out and exit the main thread.
+    info!("shutting down tracing");
+    let (sender, receiver) = mpsc::channel();
+    let _ = thread::spawn(move || {
+        tracing_utils::shutdown_tracing();
+        sender.send(()).ok()
+    });
+    let shutdown_res = receiver.recv_timeout(Duration::from_millis(2000));
+    if shutdown_res.is_err() {
+        error!("timed out while shutting down tracing, exiting anyway");
+    }
 
+    info!("shutting down");
     exit(exit_code.unwrap_or(1))
 }
 
@@ -230,6 +281,18 @@ 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_API_BASE_URI"),
+        )
 }
 
 #[test]

compute_tools/src/checker.rs

@@ -1,12 +1,28 @@
 use anyhow::{anyhow, Result};
-use postgres::Client;
 use tokio_postgres::NoTls;
 use tracing::{error, instrument};
 
 use crate::compute::ComputeNode;
 
+/// Update timestamp in a row in a special service table to check
+/// that we can actually write some data in this particular timeline.
+/// Create table if it's missing.
 #[instrument(skip_all)]
-pub fn create_writability_check_data(client: &mut Client) -> Result<()> {
+pub async fn check_writability(compute: &ComputeNode) -> Result<()> {
+    // Connect to the database.
+    let (client, connection) = tokio_postgres::connect(compute.connstr.as_str(), NoTls).await?;
+    if client.is_closed() {
+        return Err(anyhow!("connection to postgres closed"));
+    }
+
+    // The connection object performs the actual communication with the database,
+    // so spawn it off to run on its own.
+    tokio::spawn(async move {
+        if let Err(e) = connection.await {
+            error!("connection error: {}", e);
+        }
+    });
+
     let query = "
     CREATE TABLE IF NOT EXISTS health_check (
         id serial primary key,
@@ -15,31 +31,15 @@ pub fn create_writability_check_data(client: &mut Client) -> Result<()> {
     INSERT INTO health_check VALUES (1, now())
         ON CONFLICT (id) DO UPDATE
          SET updated_at = now();";
-    let result = client.simple_query(query)?;
-    if result.len() < 2 {
-        return Err(anyhow::format_err!("executed  {} queries", result.len()));
-    }
-    Ok(())
-}
-
-#[instrument(skip_all)]
-pub async fn check_writability(compute: &ComputeNode) -> Result<()> {
-    let (client, connection) = tokio_postgres::connect(compute.connstr.as_str(), NoTls).await?;
-    if client.is_closed() {
-        return Err(anyhow!("connection to postgres closed"));
-    }
-    tokio::spawn(async move {
-        if let Err(e) = connection.await {
-            error!("connection error: {}", e);
-        }
-    });
-
-    let result = client
-        .simple_query("UPDATE health_check SET updated_at = now() WHERE id = 1;")
-        .await?;
-
-    if result.len() != 1 {
-        return Err(anyhow!("statement can't be executed"));
+
+    let result = client.simple_query(query).await?;
+
+    if result.len() != 2 {
+        return Err(anyhow::format_err!(
+            "expected 2 query results, but got {}",
+            result.len()
+        ));
     }
+
     Ok(())
 }

compute_tools/src/compute.rs

@@ -18,16 +18,20 @@ use std::fs;
 use std::os::unix::fs::PermissionsExt;
 use std::path::Path;
 use std::process::{Command, Stdio};
-use std::sync::atomic::{AtomicU64, Ordering};
-use std::sync::RwLock;
+use std::str::FromStr;
+use std::sync::{Condvar, Mutex};
 
 use anyhow::{Context, Result};
 use chrono::{DateTime, Utc};
 use postgres::{Client, NoTls};
-use serde::{Serialize, Serializer};
+use tokio_postgres;
 use tracing::{info, instrument, warn};
+use utils::id::{TenantId, TimelineId};
+use utils::lsn::Lsn;
+
+use compute_api::responses::{ComputeMetrics, ComputeStatus};
+use compute_api::spec::ComputeSpec;
 
-use crate::checker::create_writability_check_data;
 use crate::config;
 use crate::pg_helpers::*;
 use crate::spec::*;
@@ -39,40 +43,45 @@ pub struct ComputeNode {
     pub connstr: url::Url,
     pub pgdata: String,
     pub pgbin: String,
-    pub spec: ComputeSpec,
-    pub tenant: String,
-    pub timeline: String,
-    pub pageserver_connstr: String,
-    pub metrics: ComputeMetrics,
-    /// Volatile part of the `ComputeNode` so should be used under `RwLock`
-    /// to allow HTTP API server to serve status requests, while configuration
-    /// is in progress.
-    pub state: RwLock<ComputeState>,
+    /// We should only allow live re- / configuration of the compute node if
+    /// it uses 'pull model', i.e. it can go to control-plane and fetch
+    /// the latest configuration. Otherwise, there could be a case:
+    /// - we start compute with some spec provided as argument
+    /// - we push new spec and it does reconfiguration
+    /// - but then something happens and compute pod / VM is destroyed,
+    ///   so k8s controller starts it again with the **old** spec
+    /// and the same for empty computes:
+    /// - we started compute without any spec
+    /// - we push spec and it does configuration
+    /// - but then it is restarted without any spec again
+    pub live_config_allowed: bool,
+    /// Volatile part of the `ComputeNode`, which should be used under `Mutex`.
+    /// To allow HTTP API server to serving status requests, while configuration
+    /// is in progress, lock should be held only for short periods of time to do
+    /// read/write, not the whole configuration process.
+    pub state: Mutex<ComputeState>,
+    /// `Condvar` to allow notifying waiters about state changes.
+    pub state_changed: Condvar,
 }
 
-fn rfc3339_serialize<S>(x: &DateTime<Utc>, s: S) -> Result<S::Ok, S::Error>
-where
-    S: Serializer,
-{
-    x.to_rfc3339().serialize(s)
-}
-
-#[derive(Serialize)]
-#[serde(rename_all = "snake_case")]
+#[derive(Clone, Debug)]
 pub struct ComputeState {
     pub status: ComputeStatus,
     /// Timestamp of the last Postgres activity
-    #[serde(serialize_with = "rfc3339_serialize")]
     pub last_active: DateTime<Utc>,
     pub error: Option<String>,
+    pub pspec: Option<ParsedSpec>,
+    pub metrics: ComputeMetrics,
 }
 
 impl ComputeState {
     pub fn new() -> Self {
         Self {
-            status: ComputeStatus::Init,
+            status: ComputeStatus::Empty,
             last_active: Utc::now(),
             error: None,
+            pspec: None,
+            metrics: ComputeMetrics::default(),
         }
     }
 }
@@ -83,29 +92,58 @@ impl Default for ComputeState {
     }
 }
 
-#[derive(Serialize, Clone, Copy, PartialEq, Eq)]
-#[serde(rename_all = "snake_case")]
-pub enum ComputeStatus {
-    Init,
-    Running,
-    Failed,
+#[derive(Clone, Debug)]
+pub struct ParsedSpec {
+    pub spec: ComputeSpec,
+    pub tenant_id: TenantId,
+    pub timeline_id: TimelineId,
+    pub pageserver_connstr: String,
+    pub storage_auth_token: Option<String>,
 }
 
-#[derive(Default, Serialize)]
-pub struct ComputeMetrics {
-    pub sync_safekeepers_ms: AtomicU64,
-    pub basebackup_ms: AtomicU64,
-    pub config_ms: AtomicU64,
-    pub total_startup_ms: AtomicU64,
+impl TryFrom<ComputeSpec> for ParsedSpec {
+    type Error = String;
+    fn try_from(spec: ComputeSpec) -> Result<Self, String> {
+        let pageserver_connstr = spec
+            .cluster
+            .settings
+            .find("neon.pageserver_connstring")
+            .ok_or("pageserver connstr should be provided")?;
+        let storage_auth_token = spec.storage_auth_token.clone();
+        let tenant_id: TenantId = spec
+            .cluster
+            .settings
+            .find("neon.tenant_id")
+            .ok_or("tenant id should be provided")
+            .map(|s| TenantId::from_str(&s))?
+            .or(Err("invalid tenant id"))?;
+        let timeline_id: TimelineId = spec
+            .cluster
+            .settings
+            .find("neon.timeline_id")
+            .ok_or("timeline id should be provided")
+            .map(|s| TimelineId::from_str(&s))?
+            .or(Err("invalid timeline id"))?;
+
+        Ok(ParsedSpec {
+            spec,
+            pageserver_connstr,
+            storage_auth_token,
+            tenant_id,
+            timeline_id,
+        })
+    }
 }
 
 impl ComputeNode {
     pub fn set_status(&self, status: ComputeStatus) {
-        self.state.write().unwrap().status = status;
+        let mut state = self.state.lock().unwrap();
+        state.status = status;
+        self.state_changed.notify_all();
     }
 
     pub fn get_status(&self) -> ComputeStatus {
-        self.state.read().unwrap().status
+        self.state.lock().unwrap().status
     }
 
     // Remove `pgdata` directory and create it again with right permissions.
@@ -121,14 +159,26 @@ impl ComputeNode {
 
     // Get basebackup from the libpq connection to pageserver using `connstr` and
     // unarchive it to `pgdata` directory overriding all its previous content.
-    #[instrument(skip(self))]
-    fn get_basebackup(&self, lsn: &str) -> Result<()> {
+    #[instrument(skip(self, compute_state))]
+    fn get_basebackup(&self, compute_state: &ComputeState, lsn: Lsn) -> Result<()> {
+        let spec = compute_state.pspec.as_ref().expect("spec must be set");
         let start_time = Utc::now();
 
-        let mut client = Client::connect(&self.pageserver_connstr, NoTls)?;
+        let mut config = postgres::Config::from_str(&spec.pageserver_connstr)?;
+
+        // Use the storage auth token from the config file, if given.
+        // Note: this overrides any password set in the connection string.
+        if let Some(storage_auth_token) = &spec.storage_auth_token {
+            info!("Got storage auth token from spec file");
+            config.password(storage_auth_token);
+        } else {
+            info!("Storage auth token not set");
+        }
+
+        let mut client = config.connect(NoTls)?;
         let basebackup_cmd = match lsn {
-            "0/0" => format!("basebackup {} {}", &self.tenant, &self.timeline), // First start of the compute
-            _ => format!("basebackup {} {} {}", &self.tenant, &self.timeline, lsn),
+            Lsn(0) => format!("basebackup {} {}", spec.tenant_id, spec.timeline_id), // First start of the compute
+            _ => format!("basebackup {} {} {}", spec.tenant_id, spec.timeline_id, lsn),
         };
         let copyreader = client.copy_out(basebackup_cmd.as_str())?;
 
@@ -141,27 +191,28 @@ impl ComputeNode {
         ar.set_ignore_zeros(true);
         ar.unpack(&self.pgdata)?;
 
-        self.metrics.basebackup_ms.store(
-            Utc::now()
-                .signed_duration_since(start_time)
-                .to_std()
-                .unwrap()
-                .as_millis() as u64,
-            Ordering::Relaxed,
-        );
-
+        self.state.lock().unwrap().metrics.basebackup_ms = Utc::now()
+            .signed_duration_since(start_time)
+            .to_std()
+            .unwrap()
+            .as_millis() as u64;
         Ok(())
     }
 
     // Run `postgres` in a special mode with `--sync-safekeepers` argument
     // and return the reported LSN back to the caller.
-    #[instrument(skip(self))]
-    fn sync_safekeepers(&self) -> Result<String> {
+    #[instrument(skip(self, storage_auth_token))]
+    fn sync_safekeepers(&self, storage_auth_token: Option<String>) -> Result<Lsn> {
         let start_time = Utc::now();
 
         let sync_handle = Command::new(&self.pgbin)
             .args(["--sync-safekeepers"])
             .env("PGDATA", &self.pgdata) // we cannot use -D in this mode
+            .envs(if let Some(storage_auth_token) = &storage_auth_token {
+                vec![("NEON_AUTH_TOKEN", storage_auth_token)]
+            } else {
+                vec![]
+            })
             .stdout(Stdio::piped())
             .spawn()
             .expect("postgres --sync-safekeepers failed to start");
@@ -182,45 +233,42 @@ impl ComputeNode {
             );
         }
 
-        self.metrics.sync_safekeepers_ms.store(
-            Utc::now()
-                .signed_duration_since(start_time)
-                .to_std()
-                .unwrap()
-                .as_millis() as u64,
-            Ordering::Relaxed,
-        );
+        self.state.lock().unwrap().metrics.sync_safekeepers_ms = Utc::now()
+            .signed_duration_since(start_time)
+            .to_std()
+            .unwrap()
+            .as_millis() as u64;
 
-        let lsn = String::from(String::from_utf8(sync_output.stdout)?.trim());
+        let lsn = Lsn::from_str(String::from_utf8(sync_output.stdout)?.trim())?;
 
         Ok(lsn)
     }
 
     /// Do all the preparations like PGDATA directory creation, configuration,
     /// safekeepers sync, basebackup, etc.
-    #[instrument(skip(self))]
-    pub fn prepare_pgdata(&self) -> Result<()> {
-        let spec = &self.spec;
+    #[instrument(skip(self, compute_state))]
+    pub fn prepare_pgdata(&self, compute_state: &ComputeState) -> Result<()> {
+        let pspec = compute_state.pspec.as_ref().expect("spec must be set");
         let pgdata_path = Path::new(&self.pgdata);
 
         // Remove/create an empty pgdata directory and put configuration there.
         self.create_pgdata()?;
-        config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), spec)?;
+        config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), &pspec.spec)?;
 
         info!("starting safekeepers syncing");
         let lsn = self
-            .sync_safekeepers()
+            .sync_safekeepers(pspec.storage_auth_token.clone())
             .with_context(|| "failed to sync safekeepers")?;
         info!("safekeepers synced at LSN {}", lsn);
 
         info!(
             "getting basebackup@{} from pageserver {}",
-            lsn, &self.pageserver_connstr
+            lsn, &pspec.pageserver_connstr
         );
-        self.get_basebackup(&lsn).with_context(|| {
+        self.get_basebackup(compute_state, lsn).with_context(|| {
             format!(
                 "failed to get basebackup@{} from pageserver {}",
-                lsn, &self.pageserver_connstr
+                lsn, &pspec.pageserver_connstr
             )
         })?;
 
@@ -233,12 +281,20 @@ impl ComputeNode {
     /// Start Postgres as a child process and manage DBs/roles.
     /// After that this will hang waiting on the postmaster process to exit.
     #[instrument(skip(self))]
-    pub fn start_postgres(&self) -> Result<std::process::Child> {
+    pub fn start_postgres(
+        &self,
+        storage_auth_token: Option<String>,
+    ) -> Result<std::process::Child> {
         let pgdata_path = Path::new(&self.pgdata);
 
         // Run postgres as a child process.
         let mut pg = Command::new(&self.pgbin)
             .args(["-D", &self.pgdata])
+            .envs(if let Some(storage_auth_token) = &storage_auth_token {
+                vec![("NEON_AUTH_TOKEN", storage_auth_token)]
+            } else {
+                vec![]
+            })
             .spawn()
             .expect("cannot start postgres process");
 
@@ -247,8 +303,9 @@ impl ComputeNode {
         Ok(pg)
     }
 
-    #[instrument(skip(self))]
-    pub fn apply_config(&self) -> Result<()> {
+    /// Do initial configuration of the already started Postgres.
+    #[instrument(skip(self, compute_state))]
+    pub fn apply_config(&self, compute_state: &ComputeState) -> Result<()> {
         // If connection fails,
         // it may be the old node with `zenith_admin` superuser.
         //
@@ -279,18 +336,61 @@ impl ComputeNode {
         };
 
         // Proceed with post-startup configuration. Note, that order of operations is important.
-        handle_roles(&self.spec, &mut client)?;
-        handle_databases(&self.spec, &mut client)?;
-        handle_role_deletions(self, &mut client)?;
-        handle_grants(self, &mut client)?;
-        create_writability_check_data(&mut client)?;
+        let spec = &compute_state.pspec.as_ref().expect("spec must be set").spec;
+        handle_roles(spec, &mut client)?;
+        handle_databases(spec, &mut client)?;
+        handle_role_deletions(spec, self.connstr.as_str(), &mut client)?;
+        handle_grants(spec, self.connstr.as_str(), &mut client)?;
+        handle_extensions(spec, &mut client)?;
 
         // 'Close' connection
         drop(client);
 
         info!(
             "finished configuration of compute for project {}",
-            self.spec.cluster.cluster_id
+            spec.cluster.cluster_id
+        );
+
+        Ok(())
+    }
+
+    // We could've wrapped this around `pg_ctl reload`, but right now we don't use
+    // `pg_ctl` for start / stop, so this just seems much easier to do as we already
+    // have opened connection to Postgres and superuser access.
+    #[instrument(skip(self, client))]
+    fn pg_reload_conf(&self, client: &mut Client) -> Result<()> {
+        client.simple_query("SELECT pg_reload_conf()")?;
+        Ok(())
+    }
+
+    /// Similar to `apply_config()`, but does a bit different sequence of operations,
+    /// as it's used to reconfigure a previously started and configured Postgres node.
+    #[instrument(skip(self))]
+    pub fn reconfigure(&self) -> Result<()> {
+        let spec = self.state.lock().unwrap().pspec.clone().unwrap().spec;
+
+        // Write new config
+        let pgdata_path = Path::new(&self.pgdata);
+        config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), &spec)?;
+
+        let mut client = Client::connect(self.connstr.as_str(), NoTls)?;
+        self.pg_reload_conf(&mut client)?;
+
+        // Proceed with post-startup configuration. Note, that order of operations is important.
+        handle_roles(&spec, &mut client)?;
+        handle_databases(&spec, &mut client)?;
+        handle_role_deletions(&spec, self.connstr.as_str(), &mut client)?;
+        handle_grants(&spec, self.connstr.as_str(), &mut client)?;
+        handle_extensions(&spec, &mut client)?;
+
+        // 'Close' connection
+        drop(client);
+
+        let unknown_op = "unknown".to_string();
+        let op_id = spec.operation_uuid.as_ref().unwrap_or(&unknown_op);
+        info!(
+            "finished reconfiguration of compute node for operation {}",
+            op_id
         );
 
         Ok(())
@@ -298,40 +398,38 @@ impl ComputeNode {
 
     #[instrument(skip(self))]
     pub fn start_compute(&self) -> Result<std::process::Child> {
+        let compute_state = self.state.lock().unwrap().clone();
+        let spec = compute_state.pspec.as_ref().expect("spec must be set");
         info!(
             "starting compute for project {}, operation {}, tenant {}, timeline {}",
-            self.spec.cluster.cluster_id,
-            self.spec.operation_uuid.as_ref().unwrap(),
-            self.tenant,
-            self.timeline,
+            spec.spec.cluster.cluster_id,
+            spec.spec.operation_uuid.as_deref().unwrap_or("None"),
+            spec.tenant_id,
+            spec.timeline_id,
         );
 
-        self.prepare_pgdata()?;
+        self.prepare_pgdata(&compute_state)?;
 
         let start_time = Utc::now();
 
-        let pg = self.start_postgres()?;
+        let pg = self.start_postgres(spec.storage_auth_token.clone())?;
 
-        self.apply_config()?;
+        self.apply_config(&compute_state)?;
 
         let startup_end_time = Utc::now();
-        self.metrics.config_ms.store(
-            startup_end_time
+        {
+            let mut state = self.state.lock().unwrap();
+            state.metrics.config_ms = startup_end_time
                 .signed_duration_since(start_time)
                 .to_std()
                 .unwrap()
-                .as_millis() as u64,
-            Ordering::Relaxed,
-        );
-        self.metrics.total_startup_ms.store(
-            startup_end_time
+                .as_millis() as u64;
+            state.metrics.total_startup_ms = startup_end_time
                 .signed_duration_since(self.start_time)
                 .to_std()
                 .unwrap()
-                .as_millis() as u64,
-            Ordering::Relaxed,
-        );
-
+                .as_millis() as u64;
+        }
         self.set_status(ComputeStatus::Running);
 
         Ok(pg)
@@ -400,4 +498,43 @@ impl ComputeNode {
 
         Ok(())
     }
+
+    /// Select `pg_stat_statements` data and return it as a stringified JSON
+    pub async fn collect_insights(&self) -> String {
+        let mut result_rows: Vec<String> = Vec::new();
+        let connect_result = tokio_postgres::connect(self.connstr.as_str(), NoTls).await;
+        let (client, connection) = connect_result.unwrap();
+        tokio::spawn(async move {
+            if let Err(e) = connection.await {
+                eprintln!("connection error: {}", e);
+            }
+        });
+        let result = client
+            .simple_query(
+                "SELECT
+    row_to_json(pg_stat_statements)
+FROM
+    pg_stat_statements
+WHERE
+    userid != 'cloud_admin'::regrole::oid
+ORDER BY
+    (mean_exec_time + mean_plan_time) DESC
+LIMIT 100",
+            )
+            .await;
+
+        if let Ok(raw_rows) = result {
+            for message in raw_rows.iter() {
+                if let postgres::SimpleQueryMessage::Row(row) = message {
+                    if let Some(json) = row.get(0) {
+                        result_rows.push(json.to_string());
+                    }
+                }
+            }
+
+            format!("{{\"pg_stat_statements\": [{}]}}", result_rows.join(","))
+        } else {
+            "{{\"pg_stat_statements\": []}}".to_string()
+        }
+    }
 }

compute_tools/src/config.rs

@@ -6,7 +6,7 @@ use std::path::Path;
 use anyhow::Result;
 
 use crate::pg_helpers::PgOptionsSerialize;
-use crate::spec::ComputeSpec;
+use compute_api::spec::ComputeSpec;
 
 /// Check that `line` is inside a text file and put it there if it is not.
 /// Create file if it doesn't exist.

compute_tools/src/configurator.rs

@@ -0,0 +1,54 @@
+use std::sync::Arc;
+use std::thread;
+
+use anyhow::Result;
+use tracing::{error, info, instrument};
+
+use compute_api::responses::ComputeStatus;
+
+use crate::compute::ComputeNode;
+
+#[instrument(skip(compute))]
+fn configurator_main_loop(compute: &Arc<ComputeNode>) {
+    info!("waiting for reconfiguration requests");
+    loop {
+        let state = compute.state.lock().unwrap();
+        let mut state = compute.state_changed.wait(state).unwrap();
+
+        if state.status == ComputeStatus::ConfigurationPending {
+            info!("got configuration request");
+            state.status = ComputeStatus::Configuration;
+            compute.state_changed.notify_all();
+            drop(state);
+
+            let mut new_status = ComputeStatus::Failed;
+            if let Err(e) = compute.reconfigure() {
+                error!("could not configure compute node: {}", e);
+            } else {
+                new_status = ComputeStatus::Running;
+                info!("compute node configured");
+            }
+
+            // XXX: used to test that API is blocking
+            // std::thread::sleep(std::time::Duration::from_millis(10000));
+
+            compute.set_status(new_status);
+        } else if state.status == ComputeStatus::Failed {
+            info!("compute node is now in Failed state, exiting");
+            break;
+        } else {
+            info!("woken up for compute status: {:?}, sleeping", state.status);
+        }
+    }
+}
+
+pub fn launch_configurator(compute: &Arc<ComputeNode>) -> Result<thread::JoinHandle<()>> {
+    let compute = Arc::clone(compute);
+
+    Ok(thread::Builder::new()
+        .name("compute-configurator".into())
+        .spawn(move || {
+            configurator_main_loop(&compute);
+            info!("configurator thread is exited");
+        })?)
+}

compute_tools/src/http/api.rs

@@ -3,14 +3,34 @@ use std::net::SocketAddr;
 use std::sync::Arc;
 use std::thread;
 
+use crate::compute::{ComputeNode, ComputeState, ParsedSpec};
+use compute_api::requests::ConfigurationRequest;
+use compute_api::responses::{ComputeStatus, ComputeStatusResponse, GenericAPIError};
+
 use anyhow::Result;
 use hyper::service::{make_service_fn, service_fn};
 use hyper::{Body, Method, Request, Response, Server, StatusCode};
+use num_cpus;
 use serde_json;
+use tokio::task;
 use tracing::{error, info};
 use tracing_utils::http::OtelName;
 
-use crate::compute::ComputeNode;
+fn status_response_from_state(state: &ComputeState) -> ComputeStatusResponse {
+    ComputeStatusResponse {
+        tenant: state
+            .pspec
+            .as_ref()
+            .map(|pspec| pspec.tenant_id.to_string()),
+        timeline: state
+            .pspec
+            .as_ref()
+            .map(|pspec| pspec.timeline_id.to_string()),
+        status: state.status,
+        last_active: state.last_active,
+        error: state.error.clone(),
+    }
+}
 
 // Service function to handle all available routes.
 async fn routes(req: Request<Body>, compute: &Arc<ComputeNode>) -> Response<Body> {
@@ -23,23 +43,80 @@ async fn routes(req: Request<Body>, compute: &Arc<ComputeNode>) -> Response<Body
         // Serialized compute state.
         (&Method::GET, "/status") => {
             info!("serving /status GET request");
-            let state = compute.state.read().unwrap();
-            Response::new(Body::from(serde_json::to_string(&*state).unwrap()))
+            let state = compute.state.lock().unwrap();
+            let status_response = status_response_from_state(&state);
+            Response::new(Body::from(serde_json::to_string(&status_response).unwrap()))
         }
 
         // Startup metrics in JSON format. Keep /metrics reserved for a possible
         // future use for Prometheus metrics format.
         (&Method::GET, "/metrics.json") => {
             info!("serving /metrics.json GET request");
-            Response::new(Body::from(serde_json::to_string(&compute.metrics).unwrap()))
+            let metrics = compute.state.lock().unwrap().metrics.clone();
+            Response::new(Body::from(serde_json::to_string(&metrics).unwrap()))
+        }
+
+        // Collect Postgres current usage insights
+        (&Method::GET, "/insights") => {
+            info!("serving /insights GET request");
+            let status = compute.get_status();
+            if status != ComputeStatus::Running {
+                let msg = format!("compute is not running, current status: {:?}", status);
+                error!(msg);
+                return Response::new(Body::from(msg));
+            }
+
+            let insights = compute.collect_insights().await;
+            Response::new(Body::from(insights))
         }
 
         (&Method::POST, "/check_writability") => {
             info!("serving /check_writability POST request");
+            let status = compute.get_status();
+            if status != ComputeStatus::Running {
+                let msg = format!(
+                    "invalid compute status for check_writability request: {:?}",
+                    status
+                );
+                error!(msg);
+                return Response::new(Body::from(msg));
+            }
+
             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())),
+                Err(e) => {
+                    error!("check_writability failed: {}", e);
+                    Response::new(Body::from(e.to_string()))
+                }
+            }
+        }
+
+        (&Method::GET, "/info") => {
+            let num_cpus = num_cpus::get_physical();
+            info!("serving /info GET request. num_cpus: {}", num_cpus);
+            Response::new(Body::from(
+                serde_json::json!({
+                    "num_cpus": num_cpus,
+                })
+                .to_string(),
+            ))
+        }
+
+        // Accept spec in JSON format and request compute configuration. If
+        // anything goes wrong after we set the compute status to `ConfigurationPending`
+        // and update compute state with new spec, we basically leave compute
+        // in the potentially wrong state. That said, it's control-plane's
+        // responsibility to watch compute state after reconfiguration request
+        // and to clean restart in case of errors.
+        (&Method::POST, "/configure") => {
+            info!("serving /configure POST request");
+            match handle_configure_request(req, compute).await {
+                Ok(msg) => Response::new(Body::from(msg)),
+                Err((msg, code)) => {
+                    error!("error handling /configure request: {msg}");
+                    render_json_error(&msg, code)
+                }
             }
         }
 
@@ -52,6 +129,94 @@ async fn routes(req: Request<Body>, compute: &Arc<ComputeNode>) -> Response<Body
     }
 }
 
+async fn handle_configure_request(
+    req: Request<Body>,
+    compute: &Arc<ComputeNode>,
+) -> Result<String, (String, StatusCode)> {
+    if !compute.live_config_allowed {
+        return Err((
+            "live configuration is not allowed for this compute node".to_string(),
+            StatusCode::PRECONDITION_FAILED,
+        ));
+    }
+
+    let body_bytes = hyper::body::to_bytes(req.into_body()).await.unwrap();
+    let spec_raw = String::from_utf8(body_bytes.to_vec()).unwrap();
+    if let Ok(request) = serde_json::from_str::<ConfigurationRequest>(&spec_raw) {
+        let spec = request.spec;
+
+        let parsed_spec = match ParsedSpec::try_from(spec) {
+            Ok(ps) => ps,
+            Err(msg) => return Err((msg, StatusCode::PRECONDITION_FAILED)),
+        };
+
+        // XXX: wrap state update under lock in code blocks. Otherwise,
+        // we will try to `Send` `mut state` into the spawned thread
+        // bellow, which will cause error:
+        // ```
+        // error: future cannot be sent between threads safely
+        // ```
+        {
+            let mut state = compute.state.lock().unwrap();
+            if state.status != ComputeStatus::Empty && state.status != ComputeStatus::Running {
+                let msg = format!(
+                    "invalid compute status for configuration request: {:?}",
+                    state.status.clone()
+                );
+                return Err((msg, StatusCode::PRECONDITION_FAILED));
+            }
+            state.pspec = Some(parsed_spec);
+            state.status = ComputeStatus::ConfigurationPending;
+            compute.state_changed.notify_all();
+            drop(state);
+            info!("set new spec and notified waiters");
+        }
+
+        // Spawn a blocking thread to wait for compute to become Running.
+        // This is needed to do not block the main pool of workers and
+        // be able to serve other requests while some particular request
+        // is waiting for compute to finish configuration.
+        let c = compute.clone();
+        task::spawn_blocking(move || {
+            let mut state = c.state.lock().unwrap();
+            while state.status != ComputeStatus::Running {
+                state = c.state_changed.wait(state).unwrap();
+                info!(
+                    "waiting for compute to become Running, current status: {:?}",
+                    state.status
+                );
+
+                if state.status == ComputeStatus::Failed {
+                    let err = state.error.as_ref().map_or("unknown error", |x| x);
+                    let msg = format!("compute configuration failed: {:?}", err);
+                    return Err((msg, StatusCode::INTERNAL_SERVER_ERROR));
+                }
+            }
+
+            Ok(())
+        })
+        .await
+        .unwrap()?;
+
+        // Return current compute state if everything went well.
+        let state = compute.state.lock().unwrap().clone();
+        let status_response = status_response_from_state(&state);
+        Ok(serde_json::to_string(&status_response).unwrap())
+    } else {
+        Err(("invalid spec".to_string(), StatusCode::BAD_REQUEST))
+    }
+}
+
+fn render_json_error(e: &str, status: StatusCode) -> Response<Body> {
+    let error = GenericAPIError {
+        error: e.to_string(),
+    };
+    Response::builder()
+        .status(status)
+        .body(Body::from(serde_json::to_string(&error).unwrap()))
+        .unwrap()
+}
+
 // Main Hyper HTTP server function that runs it and blocks waiting on it forever.
 #[tokio::main]
 async fn serve(state: Arc<ComputeNode>) {

compute_tools/src/http/openapi_spec.yaml

@@ -10,12 +10,12 @@ paths:
   /status:
     get:
       tags:
-      - "info"
-      summary: Get compute node internal status
+      - Info
+      summary: Get compute node internal status.
       description: ""
       operationId: getComputeStatus
       responses:
-        "200":
+        200:
           description: ComputeState
           content:
             application/json:
@@ -25,35 +25,121 @@ paths:
   /metrics.json:
     get:
       tags:
-      - "info"
-      summary: Get compute node startup metrics in JSON format
+      - Info
+      summary: Get compute node startup metrics in JSON format.
       description: ""
       operationId: getComputeMetricsJSON
       responses:
-        "200":
+        200:
           description: ComputeMetrics
           content:
             application/json:
               schema:
                 $ref: "#/components/schemas/ComputeMetrics"
 
+  /insights:
+    get:
+      tags:
+      - Info
+      summary: Get current compute insights in JSON format.
+      description: |
+        Note, that this doesn't include any historical data.
+      operationId: getComputeInsights
+      responses:
+        200:
+          description: Compute insights
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/ComputeInsights"
+
+  /info:
+    get:
+      tags:
+      - Info
+      summary: Get info about the compute pod / VM.
+      description: ""
+      operationId: getInfo
+      responses:
+        200:
+          description: Info
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/Info"
+
   /check_writability:
     post:
       tags:
-      - "check"
-      summary: Check that we can write new data on this compute
+      - Check
+      summary: Check that we can write new data on this compute.
       description: ""
       operationId: checkComputeWritability
       responses:
-        "200":
+        200:
           description: Check result
           content:
             text/plain:
               schema:
                 type: string
-                description: Error text or 'true' if check passed
+                description: Error text or 'true' if check passed.
                 example: "true"
 
+  /configure:
+    post:
+      tags:
+      - Configure
+      summary: Perform compute node configuration.
+      description: |
+        This is a blocking API endpoint, i.e. it blocks waiting until
+        compute is finished configuration and is in `Running` state.
+        Optional non-blocking mode could be added later.
+      operationId: configureCompute
+      requestBody:
+        description: Configuration request.
+        required: true
+        content:
+          application/json:
+            schema:
+              type: object
+              required:
+                - spec
+              properties:
+                spec:
+                  # XXX: I don't want to explain current spec in the OpenAPI format,
+                  # as it could be changed really soon. Consider doing it later.
+                  type: object
+      responses:
+        200:
+          description: Compute configuration finished.
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/ComputeState"
+        400:
+          description: Provided spec is invalid.
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/GenericError"
+        412:
+          description: |
+            It's not possible to do live-configuration of the compute.
+            It's either in the wrong state, or compute doesn't use pull
+            mode of configuration.
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/GenericError"
+        500:
+          description: |
+            Compute configuration request was processed, but error
+            occurred. Compute will likely shutdown soon.
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/GenericError"
+
 components:
   securitySchemes:
     JWT:
@@ -64,7 +150,7 @@ components:
   schemas:
     ComputeMetrics:
       type: object
-      description: Compute startup metrics
+      description: Compute startup metrics.
       required:
         - sync_safekeepers_ms
         - basebackup_ms
@@ -80,6 +166,15 @@ components:
         total_startup_ms:
           type: integer
 
+    Info:
+      type: object
+      description: Information about VM/Pod.
+      required:
+        - num_cpus
+      properties:
+        num_cpus:
+          type: integer
+
     ComputeState:
       type: object
       required:
@@ -90,11 +185,29 @@ components:
           $ref: '#/components/schemas/ComputeStatus'
         last_active:
           type: string
-          description: The last detected compute activity timestamp in UTC and RFC3339 format
+          description: The last detected compute activity timestamp in UTC and RFC3339 format.
           example: "2022-10-12T07:20:50.52Z"
         error:
           type: string
-          description: Text of the error during compute startup, if any
+          description: Text of the error during compute startup, if any.
+          example: ""
+        tenant:
+          type: string
+          description: Identifier of the current tenant served by compute node, if any.
+          example: c9269c359e9a199fad1ea0981246a78f
+        timeline:
+          type: string
+          description: Identifier of the current timeline served by compute node, if any.
+          example: ece7de74d4b8cbe5433a68ce4d1b97b4
+
+    ComputeInsights:
+      type: object
+      properties:
+        pg_stat_statements:
+          description: Contains raw output from pg_stat_statements in JSON format.
+          type: array
+          items:
+            type: object
 
     ComputeStatus:
       type: string
@@ -102,6 +215,19 @@ components:
         - init
         - failed
         - running
+      example: running
+
+    #
+    # Errors
+    #
+
+    GenericError:
+      type: object
+      required:
+        - error
+      properties:
+        error:
+          type: string
 
 security:
   - JWT: []

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

@@ -4,11 +4,11 @@
 //!
 pub mod checker;
 pub mod config;
+pub mod configurator;
 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/monitor.rs

@@ -46,7 +46,7 @@ fn watch_compute_activity(compute: &ComputeNode) {
                             AND usename != 'cloud_admin';", // XXX: find a better way to filter other monitors?
                         &[],
                     );
-                let mut last_active = compute.state.read().unwrap().last_active;
+                let mut last_active = compute.state.lock().unwrap().last_active;
 
                 if let Ok(backs) = backends {
                     let mut idle_backs: Vec<DateTime<Utc>> = vec![];
@@ -87,7 +87,7 @@ fn watch_compute_activity(compute: &ComputeNode) {
                 }
 
                 // Update the last activity in the shared state if we got a more recent one.
-                let mut state = compute.state.write().unwrap();
+                let mut state = compute.state.lock().unwrap();
                 if last_active > state.last_active {
                     state.last_active = last_active;
                     debug!("set the last compute activity time to: {}", last_active);

compute_tools/src/pg_helpers.rs

@@ -10,49 +10,34 @@ use std::time::{Duration, Instant};
 use anyhow::{bail, Result};
 use notify::{RecursiveMode, Watcher};
 use postgres::{Client, Transaction};
-use serde::Deserialize;
 use tracing::{debug, instrument};
 
+use compute_api::spec::{Database, GenericOption, GenericOptions, PgIdent, Role};
+
 const POSTGRES_WAIT_TIMEOUT: Duration = Duration::from_millis(60 * 1000); // milliseconds
 
-/// Rust representation of Postgres role info with only those fields
-/// that matter for us.
-#[derive(Clone, Deserialize)]
-pub struct Role {
-    pub name: PgIdent,
-    pub encrypted_password: Option<String>,
-    pub options: GenericOptions,
+/// Escape a string for including it in a SQL literal
+fn escape_literal(s: &str) -> String {
+    s.replace('\'', "''").replace('\\', "\\\\")
 }
 
-/// Rust representation of Postgres database info with only those fields
-/// that matter for us.
-#[derive(Clone, Deserialize)]
-pub struct Database {
-    pub name: PgIdent,
-    pub owner: PgIdent,
-    pub options: GenericOptions,
+/// Escape a string so that it can be used in postgresql.conf.
+/// Same as escape_literal, currently.
+fn escape_conf_value(s: &str) -> String {
+    s.replace('\'', "''").replace('\\', "\\\\")
 }
 
-/// Common type representing both SQL statement params with or without value,
-/// like `LOGIN` or `OWNER username` in the `CREATE/ALTER ROLE`, and config
-/// options like `wal_level = logical`.
-#[derive(Clone, Deserialize)]
-pub struct GenericOption {
-    pub name: String,
-    pub value: Option<String>,
-    pub vartype: String,
+trait GenericOptionExt {
+    fn to_pg_option(&self) -> String;
+    fn to_pg_setting(&self) -> String;
 }
 
-/// Optional collection of `GenericOption`'s. Type alias allows us to
-/// declare a `trait` on it.
-pub type GenericOptions = Option<Vec<GenericOption>>;
-
-impl GenericOption {
+impl GenericOptionExt for GenericOption {
     /// Represent `GenericOption` as SQL statement parameter.
-    pub fn to_pg_option(&self) -> String {
+    fn to_pg_option(&self) -> String {
         if let Some(val) = &self.value {
             match self.vartype.as_ref() {
-                "string" => format!("{} '{}'", self.name, val),
+                "string" => format!("{} '{}'", self.name, escape_literal(val)),
                 _ => format!("{} {}", self.name, val),
             }
         } else {
@@ -61,18 +46,11 @@ impl GenericOption {
     }
 
     /// Represent `GenericOption` as configuration option.
-    pub fn to_pg_setting(&self) -> String {
+    fn to_pg_setting(&self) -> String {
         if let Some(val) = &self.value {
-            let name = match self.name.as_str() {
-                "safekeepers" => "neon.safekeepers",
-                "wal_acceptor_reconnect" => "neon.safekeeper_reconnect_timeout",
-                "wal_acceptor_connection_timeout" => "neon.safekeeper_connection_timeout",
-                it => it,
-            };
-
             match self.vartype.as_ref() {
-                "string" => format!("{} = '{}'", name, val),
-                _ => format!("{} = {}", name, val),
+                "string" => format!("{} = '{}'", self.name, escape_conf_value(val)),
+                _ => format!("{} = {}", self.name, val),
             }
         } else {
             self.name.to_owned()
@@ -107,6 +85,7 @@ impl PgOptionsSerialize for GenericOptions {
                 .map(|op| op.to_pg_setting())
                 .collect::<Vec<String>>()
                 .join("\n")
+                + "\n" // newline after last setting
         } else {
             "".to_string()
         }
@@ -126,10 +105,14 @@ impl GenericOptionsSearch for GenericOptions {
     }
 }
 
-impl Role {
+pub trait RoleExt {
+    fn to_pg_options(&self) -> String;
+}
+
+impl RoleExt for Role {
     /// Serialize a list of role parameters into a Postgres-acceptable
     /// string of arguments.
-    pub fn to_pg_options(&self) -> String {
+    fn to_pg_options(&self) -> String {
         // XXX: consider putting LOGIN as a default option somewhere higher, e.g. in control-plane.
         // For now, we do not use generic `options` for roles. Once used, add
         // `self.options.as_pg_options()` somewhere here.
@@ -154,21 +137,17 @@ impl Role {
     }
 }
 
-impl Database {
-    pub fn new(name: PgIdent, owner: PgIdent) -> Self {
-        Self {
-            name,
-            owner,
-            options: None,
-        }
-    }
+pub trait DatabaseExt {
+    fn to_pg_options(&self) -> String;
+}
 
+impl DatabaseExt for Database {
     /// Serialize a list of database parameters into a Postgres-acceptable
     /// string of arguments.
     /// NB: `TEMPLATE` is actually also an identifier, but so far we only need
     /// to use `template0` and `template1`, so it is not a problem. Yet in the future
     /// it may require a proper quoting too.
-    pub fn to_pg_options(&self) -> String {
+    fn to_pg_options(&self) -> String {
         let mut params: String = self.options.as_pg_options();
         write!(params, " OWNER {}", &self.owner.pg_quote())
             .expect("String is documented to not to error during write operations");
@@ -177,10 +156,6 @@ impl Database {
     }
 }
 
-/// String type alias representing Postgres identifier and
-/// intended to be used for DB / role names.
-pub type PgIdent = String;
-
 /// Generic trait used to provide quoting / encoding for strings used in the
 /// Postgres SQL queries and DATABASE_URL.
 pub trait Escaping {
@@ -221,7 +196,11 @@ pub fn get_existing_dbs(client: &mut Client) -> Result<Vec<Database>> {
             &[],
         )?
         .iter()
-        .map(|row| Database::new(row.get("datname"), row.get("owner")))
+        .map(|row| Database {
+            name: row.get("datname"),
+            owner: row.get("owner"),
+            options: None,
+        })
         .collect();
 
     Ok(postgres_dbs)

compute_tools/src/spec.rs

@@ -1,55 +1,45 @@
-use std::collections::HashMap;
 use std::path::Path;
 use std::str::FromStr;
 
-use anyhow::Result;
+use anyhow::{anyhow, bail, Result};
 use postgres::config::Config;
 use postgres::{Client, NoTls};
-use serde::Deserialize;
 use tracing::{info, info_span, instrument, span_enabled, warn, Level};
 
-use crate::compute::ComputeNode;
 use crate::config;
 use crate::params::PG_HBA_ALL_MD5;
 use crate::pg_helpers::*;
 
-/// Cluster spec or configuration represented as an optional number of
-/// delta operations + final cluster state description.
-#[derive(Clone, Deserialize)]
-pub struct ComputeSpec {
-    pub format_version: f32,
-    pub timestamp: String,
-    pub operation_uuid: Option<String>,
-    /// Expected cluster state at the end of transition process.
-    pub cluster: Cluster,
-    pub delta_operations: Option<Vec<DeltaOp>>,
+use compute_api::responses::ControlPlaneSpecResponse;
+use compute_api::spec::{ComputeSpec, Database, PgIdent, Role};
 
-    pub startup_tracing_context: Option<HashMap<String, String>>,
-}
+/// Request spec from the control-plane by compute_id. If `NEON_CONSOLE_JWT`
+/// env variable is set, it will be used for authorization.
+pub fn get_spec_from_control_plane(base_uri: &str, compute_id: &str) -> Result<ComputeSpec> {
+    let cp_uri = format!("{base_uri}/management/api/v2/computes/{compute_id}/spec");
+    let jwt: String = match std::env::var("NEON_CONSOLE_JWT") {
+        Ok(v) => v,
+        Err(_) => "".to_string(),
+    };
+    info!("getting spec from control plane: {}", cp_uri);
 
-/// Cluster state seen from the perspective of the external tools
-/// like Rails web console.
-#[derive(Clone, Deserialize)]
-pub struct Cluster {
-    pub cluster_id: String,
-    pub name: String,
-    pub state: Option<String>,
-    pub roles: Vec<Role>,
-    pub databases: Vec<Database>,
-    pub settings: GenericOptions,
-}
+    // TODO: check the response. We should distinguish cases when it's
+    // - network error, then retry
+    // - no spec for compute yet, then wait
+    // - compute id is unknown or any other error, then bail out
+    let resp: ControlPlaneSpecResponse = reqwest::blocking::Client::new()
+        .get(cp_uri)
+        .header("Authorization", jwt)
+        .send()
+        .map_err(|e| anyhow!("could not send spec request to control plane: {}", e))?
+        .json()
+        .map_err(|e| anyhow!("could not get compute spec from control plane: {}", e))?;
 
-/// Single cluster state changing operation that could not be represented as
-/// a static `Cluster` structure. For example:
-/// - DROP DATABASE
-/// - DROP ROLE
-/// - ALTER ROLE name RENAME TO new_name
-/// - ALTER DATABASE name RENAME TO new_name
-#[derive(Clone, Deserialize)]
-pub struct DeltaOp {
-    pub action: String,
-    pub name: PgIdent,
-    pub new_name: Option<PgIdent>,
+    if let Some(spec) = resp.spec {
+        Ok(spec)
+    } else {
+        bail!("could not get compute spec from control plane")
+    }
 }
 
 /// It takes cluster specification and does the following:
@@ -224,8 +214,8 @@ pub fn handle_roles(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
 
 /// Reassign all dependent objects and delete requested roles.
 #[instrument(skip_all)]
-pub fn handle_role_deletions(node: &ComputeNode, client: &mut Client) -> Result<()> {
-    if let Some(ops) = &node.spec.delta_operations {
+pub fn handle_role_deletions(spec: &ComputeSpec, connstr: &str, client: &mut Client) -> Result<()> {
+    if let Some(ops) = &spec.delta_operations {
         // First, reassign all dependent objects to db owners.
         info!("reassigning dependent objects of to-be-deleted roles");
 
@@ -242,7 +232,7 @@ pub fn handle_role_deletions(node: &ComputeNode, client: &mut Client) -> Result<
             // Check that role is still present in Postgres, as this could be a
             // restart with the same spec after role deletion.
             if op.action == "delete_role" && existing_roles.iter().any(|r| r.name == op.name) {
-                reassign_owned_objects(node, &op.name)?;
+                reassign_owned_objects(spec, connstr, &op.name)?;
             }
         }
 
@@ -266,10 +256,10 @@ pub fn handle_role_deletions(node: &ComputeNode, client: &mut Client) -> Result<
 }
 
 // Reassign all owned objects in all databases to the owner of the database.
-fn reassign_owned_objects(node: &ComputeNode, role_name: &PgIdent) -> Result<()> {
-    for db in &node.spec.cluster.databases {
+fn reassign_owned_objects(spec: &ComputeSpec, connstr: &str, role_name: &PgIdent) -> Result<()> {
+    for db in &spec.cluster.databases {
         if db.owner != *role_name {
-            let mut conf = Config::from_str(node.connstr.as_str())?;
+            let mut conf = Config::from_str(connstr)?;
             conf.dbname(&db.name);
 
             let mut client = conf.connect(NoTls)?;
@@ -414,9 +404,7 @@ pub fn handle_databases(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
 /// Grant CREATE ON DATABASE to the database owner and do some other alters and grants
 /// to allow users creating trusted extensions and re-creating `public` schema, for example.
 #[instrument(skip_all)]
-pub fn handle_grants(node: &ComputeNode, client: &mut Client) -> Result<()> {
-    let spec = &node.spec;
-
+pub fn handle_grants(spec: &ComputeSpec, connstr: &str, client: &mut Client) -> Result<()> {
     info!("cluster spec grants:");
 
     // We now have a separate `web_access` role to connect to the database
@@ -448,8 +436,8 @@ pub fn handle_grants(node: &ComputeNode, client: &mut Client) -> Result<()> {
     // Do some per-database access adjustments. We'd better do this at db creation time,
     // but CREATE DATABASE isn't transactional. So we cannot create db + do some grants
     // atomically.
-    for db in &node.spec.cluster.databases {
-        let mut conf = Config::from_str(node.connstr.as_str())?;
+    for db in &spec.cluster.databases {
+        let mut conf = Config::from_str(connstr)?;
         conf.dbname(&db.name);
 
         let mut db_client = conf.connect(NoTls)?;
@@ -515,3 +503,18 @@ pub fn handle_grants(node: &ComputeNode, client: &mut Client) -> Result<()> {
 
     Ok(())
 }
+
+/// Create required system extensions
+#[instrument(skip_all)]
+pub fn handle_extensions(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
+    if let Some(libs) = spec.cluster.settings.find("shared_preload_libraries") {
+        if libs.contains("pg_stat_statements") {
+            // Create extension only if this compute really needs it
+            let query = "CREATE EXTENSION IF NOT EXISTS pg_stat_statements";
+            info!("creating system extensions with query: {}", query);
+            client.simple_query(query)?;
+        }
+    }
+
+    Ok(())
+}

compute_tools/tests/pg_helpers_tests.rs

@@ -1,14 +1,13 @@
 #[cfg(test)]
 mod pg_helpers_tests {
-
     use std::fs::File;
 
+    use compute_api::spec::{ComputeSpec, GenericOption, GenericOptions, PgIdent};
     use compute_tools::pg_helpers::*;
-    use compute_tools::spec::ComputeSpec;
 
     #[test]
     fn params_serialize() {
-        let file = File::open("tests/cluster_spec.json").unwrap();
+        let file = File::open("../libs/compute_api/tests/cluster_spec.json").unwrap();
         let spec: ComputeSpec = serde_json::from_reader(file).unwrap();
 
         assert_eq!(
@@ -23,12 +22,35 @@ mod pg_helpers_tests {
 
     #[test]
     fn settings_serialize() {
-        let file = File::open("tests/cluster_spec.json").unwrap();
+        let file = File::open("../libs/compute_api/tests/cluster_spec.json").unwrap();
         let spec: ComputeSpec = serde_json::from_reader(file).unwrap();
 
         assert_eq!(
             spec.cluster.settings.as_pg_settings(),
-            "fsync = off\nwal_level = replica\nhot_standby = on\nneon.safekeepers = '127.0.0.1:6502,127.0.0.1:6503,127.0.0.1:6501'\nwal_log_hints = on\nlog_connections = on\nshared_buffers = 32768\nport = 55432\nmax_connections = 100\nmax_wal_senders = 10\nlisten_addresses = '0.0.0.0'\nwal_sender_timeout = 0\npassword_encryption = md5\nmaintenance_work_mem = 65536\nmax_parallel_workers = 8\nmax_worker_processes = 8\nneon.tenant_id = 'b0554b632bd4d547a63b86c3630317e8'\nmax_replication_slots = 10\nneon.timeline_id = '2414a61ffc94e428f14b5758fe308e13'\nshared_preload_libraries = 'neon'\nsynchronous_standby_names = 'walproposer'\nneon.pageserver_connstring = 'host=127.0.0.1 port=6400'"
+            r#"fsync = off
+wal_level = replica
+hot_standby = on
+neon.safekeepers = '127.0.0.1:6502,127.0.0.1:6503,127.0.0.1:6501'
+wal_log_hints = on
+log_connections = on
+shared_buffers = 32768
+port = 55432
+max_connections = 100
+max_wal_senders = 10
+listen_addresses = '0.0.0.0'
+wal_sender_timeout = 0
+password_encryption = md5
+maintenance_work_mem = 65536
+max_parallel_workers = 8
+max_worker_processes = 8
+neon.tenant_id = 'b0554b632bd4d547a63b86c3630317e8'
+max_replication_slots = 10
+neon.timeline_id = '2414a61ffc94e428f14b5758fe308e13'
+shared_preload_libraries = 'neon'
+synchronous_standby_names = 'walproposer'
+neon.pageserver_connstring = 'host=127.0.0.1 port=6400'
+test.escaping = 'here''s a backslash \\ and a quote '' and a double-quote " hooray'
+"#
         );
     }
 

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
@@ -23,6 +24,7 @@ url.workspace = true
 # Note: Do not directly depend on pageserver or safekeeper; use pageserver_api or safekeeper_api
 # instead, so that recompile times are better.
 pageserver_api.workspace = true
+postgres_backend.workspace = true
 safekeeper_api.workspace = true
 postgres_connection.workspace = true
 storage_broker.workspace = true

control_plane/safekeepers.conf

@@ -2,7 +2,8 @@
 [pageserver]
 listen_pg_addr = '127.0.0.1:64000'
 listen_http_addr = '127.0.0.1:9898'
-auth_type = 'Trust'
+pg_auth_type = 'Trust'
+http_auth_type = 'Trust'
 
 [[safekeepers]]
 id = 1

control_plane/simple.conf

@@ -3,7 +3,8 @@
 [pageserver]
 listen_pg_addr = '127.0.0.1:64000'
 listen_http_addr = '127.0.0.1:9898'
-auth_type = 'Trust'
+pg_auth_type = 'Trust'
+http_auth_type = 'Trust'
 
 [[safekeepers]]
 id = 1

control_plane/src/bin/neon_local.rs

@@ -7,7 +7,7 @@
 //!
 use anyhow::{anyhow, bail, Context, Result};
 use clap::{value_parser, Arg, ArgAction, ArgMatches, Command};
-use control_plane::compute::ComputeControlPlane;
+use control_plane::endpoint::ComputeControlPlane;
 use control_plane::local_env::LocalEnv;
 use control_plane::pageserver::PageServerNode;
 use control_plane::safekeeper::SafekeeperNode;
@@ -17,6 +17,7 @@ use pageserver_api::{
     DEFAULT_HTTP_LISTEN_ADDR as DEFAULT_PAGESERVER_HTTP_ADDR,
     DEFAULT_PG_LISTEN_ADDR as DEFAULT_PAGESERVER_PG_ADDR,
 };
+use postgres_backend::AuthType;
 use safekeeper_api::{
     DEFAULT_HTTP_LISTEN_PORT as DEFAULT_SAFEKEEPER_HTTP_PORT,
     DEFAULT_PG_LISTEN_PORT as DEFAULT_SAFEKEEPER_PG_PORT,
@@ -30,7 +31,6 @@ use utils::{
     auth::{Claims, Scope},
     id::{NodeId, TenantId, TenantTimelineId, TimelineId},
     lsn::Lsn,
-    postgres_backend::AuthType,
     project_git_version,
 };
 
@@ -53,14 +53,15 @@ listen_addr = '{DEFAULT_BROKER_ADDR}'
 id = {DEFAULT_PAGESERVER_ID}
 listen_pg_addr = '{DEFAULT_PAGESERVER_PG_ADDR}'
 listen_http_addr = '{DEFAULT_PAGESERVER_HTTP_ADDR}'
-auth_type = '{pageserver_auth_type}'
+pg_auth_type = '{trust_auth}'
+http_auth_type = '{trust_auth}'
 
 [[safekeepers]]
 id = {DEFAULT_SAFEKEEPER_ID}
 pg_port = {DEFAULT_SAFEKEEPER_PG_PORT}
 http_port = {DEFAULT_SAFEKEEPER_HTTP_PORT}
 "#,
-        pageserver_auth_type = AuthType::Trust,
+        trust_auth = AuthType::Trust,
     )
 }
 
@@ -105,8 +106,9 @@ fn main() -> Result<()> {
             "start" => handle_start_all(sub_args, &env),
             "stop" => handle_stop_all(sub_args, &env),
             "pageserver" => handle_pageserver(sub_args, &env),
-            "pg" => handle_pg(sub_args, &env),
             "safekeeper" => handle_safekeeper(sub_args, &env),
+            "endpoint" => handle_endpoint(sub_args, &env),
+            "pg" => bail!("'pg' subcommand has been renamed to 'endpoint'"),
             _ => bail!("unexpected subcommand {sub_name}"),
         };
 
@@ -469,10 +471,10 @@ fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::LocalEnv) -
             let mut cplane = ComputeControlPlane::load(env.clone())?;
             println!("Importing timeline into pageserver ...");
             pageserver.timeline_import(tenant_id, timeline_id, base, pg_wal, pg_version)?;
-            println!("Creating node for imported timeline ...");
             env.register_branch_mapping(name.to_string(), tenant_id, timeline_id)?;
 
-            cplane.new_node(tenant_id, name, timeline_id, None, None, pg_version)?;
+            println!("Creating endpoint for imported timeline ...");
+            cplane.new_endpoint(tenant_id, name, timeline_id, None, None, pg_version)?;
             println!("Done");
         }
         Some(("branch", branch_match)) => {
@@ -520,10 +522,10 @@ fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::LocalEnv) -
     Ok(())
 }
 
-fn handle_pg(pg_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
-    let (sub_name, sub_args) = match pg_match.subcommand() {
-        Some(pg_subcommand_data) => pg_subcommand_data,
-        None => bail!("no pg subcommand provided"),
+fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
+    let (sub_name, sub_args) = match ep_match.subcommand() {
+        Some(ep_subcommand_data) => ep_subcommand_data,
+        None => bail!("no endpoint subcommand provided"),
     };
 
     let mut cplane = ComputeControlPlane::load(env.clone())?;
@@ -545,7 +547,7 @@ fn handle_pg(pg_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
             table.load_preset(comfy_table::presets::NOTHING);
 
             table.set_header([
-                "NODE",
+                "ENDPOINT",
                 "ADDRESS",
                 "TIMELINE",
                 "BRANCH NAME",
@@ -553,39 +555,39 @@ fn handle_pg(pg_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
                 "STATUS",
             ]);
 
-            for ((_, node_name), node) in cplane
-                .nodes
+            for (endpoint_id, endpoint) in cplane
+                .endpoints
                 .iter()
-                .filter(|((node_tenant_id, _), _)| node_tenant_id == &tenant_id)
+                .filter(|(_, endpoint)| endpoint.tenant_id == tenant_id)
             {
-                let lsn_str = match node.lsn {
+                let lsn_str = match endpoint.lsn {
                     None => {
-                        // -> primary node
+                        // -> primary endpoint
                         // Use the LSN at the end of the timeline.
                         timeline_infos
-                            .get(&node.timeline_id)
+                            .get(&endpoint.timeline_id)
                             .map(|bi| bi.last_record_lsn.to_string())
                             .unwrap_or_else(|| "?".to_string())
                     }
                     Some(lsn) => {
-                        // -> read-only node
-                        // Use the node's LSN.
+                        // -> read-only endpoint
+                        // Use the endpoint's LSN.
                         lsn.to_string()
                     }
                 };
 
                 let branch_name = timeline_name_mappings
-                    .get(&TenantTimelineId::new(tenant_id, node.timeline_id))
+                    .get(&TenantTimelineId::new(tenant_id, endpoint.timeline_id))
                     .map(|name| name.as_str())
                     .unwrap_or("?");
 
                 table.add_row([
-                    node_name.as_str(),
-                    &node.address.to_string(),
-                    &node.timeline_id.to_string(),
+                    endpoint_id.as_str(),
+                    &endpoint.address.to_string(),
+                    &endpoint.timeline_id.to_string(),
                     branch_name,
                     lsn_str.as_str(),
-                    node.status(),
+                    endpoint.status(),
                 ]);
             }
 
@@ -596,10 +598,10 @@ fn handle_pg(pg_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
                 .get_one::<String>("branch-name")
                 .map(|s| s.as_str())
                 .unwrap_or(DEFAULT_BRANCH_NAME);
-            let node_name = sub_args
-                .get_one::<String>("node")
-                .map(|node_name| node_name.to_string())
-                .unwrap_or_else(|| format!("{branch_name}_node"));
+            let endpoint_id = sub_args
+                .get_one::<String>("endpoint_id")
+                .map(String::to_string)
+                .unwrap_or_else(|| format!("ep-{branch_name}"));
 
             let lsn = sub_args
                 .get_one::<String>("lsn")
@@ -617,17 +619,17 @@ fn handle_pg(pg_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
                 .copied()
                 .context("Failed to parse postgres version from the argument string")?;
 
-            cplane.new_node(tenant_id, &node_name, timeline_id, lsn, port, pg_version)?;
+            cplane.new_endpoint(tenant_id, &endpoint_id, timeline_id, lsn, port, pg_version)?;
         }
         "start" => {
             let port: Option<u16> = sub_args.get_one::<u16>("port").copied();
-            let node_name = sub_args
-                .get_one::<String>("node")
-                .ok_or_else(|| anyhow!("No node name was provided to start"))?;
+            let endpoint_id = sub_args
+                .get_one::<String>("endpoint_id")
+                .ok_or_else(|| anyhow!("No endpoint ID was provided to start"))?;
 
-            let node = cplane.nodes.get(&(tenant_id, node_name.to_string()));
+            let endpoint = cplane.endpoints.get(endpoint_id.as_str());
 
-            let auth_token = if matches!(env.pageserver.auth_type, AuthType::NeonJWT) {
+            let auth_token = if matches!(env.pageserver.pg_auth_type, AuthType::NeonJWT) {
                 let claims = Claims::new(Some(tenant_id), Scope::Tenant);
 
                 Some(env.generate_auth_token(&claims)?)
@@ -635,9 +637,9 @@ fn handle_pg(pg_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
                 None
             };
 
-            if let Some(node) = node {
-                println!("Starting existing postgres {node_name}...");
-                node.start(&auth_token)?;
+            if let Some(endpoint) = endpoint {
+                println!("Starting existing endpoint {endpoint_id}...");
+                endpoint.start(&auth_token)?;
             } else {
                 let branch_name = sub_args
                     .get_one::<String>("branch-name")
@@ -662,27 +664,33 @@ fn handle_pg(pg_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
                 // start --port X
                 // stop
                 // start <-- will also use port X even without explicit port argument
-                println!("Starting new postgres (v{pg_version}) {node_name} on timeline {timeline_id} ...");
+                println!("Starting new endpoint {endpoint_id} (PostgreSQL v{pg_version}) on timeline {timeline_id} ...");
 
-                let node =
-                    cplane.new_node(tenant_id, node_name, timeline_id, lsn, port, pg_version)?;
-                node.start(&auth_token)?;
+                let ep = cplane.new_endpoint(
+                    tenant_id,
+                    endpoint_id,
+                    timeline_id,
+                    lsn,
+                    port,
+                    pg_version,
+                )?;
+                ep.start(&auth_token)?;
             }
         }
         "stop" => {
-            let node_name = sub_args
-                .get_one::<String>("node")
-                .ok_or_else(|| anyhow!("No node name was provided to stop"))?;
+            let endpoint_id = sub_args
+                .get_one::<String>("endpoint_id")
+                .ok_or_else(|| anyhow!("No endpoint ID was provided to stop"))?;
             let destroy = sub_args.get_flag("destroy");
 
-            let node = cplane
-                .nodes
-                .get(&(tenant_id, node_name.to_string()))
-                .with_context(|| format!("postgres {node_name} is not found"))?;
-            node.stop(destroy)?;
+            let endpoint = cplane
+                .endpoints
+                .get(endpoint_id.as_str())
+                .with_context(|| format!("postgres endpoint {endpoint_id} is not found"))?;
+            endpoint.stop(destroy)?;
         }
 
-        _ => bail!("Unexpected pg subcommand '{sub_name}'"),
+        _ => bail!("Unexpected endpoint subcommand '{sub_name}'"),
     }
 
     Ok(())
@@ -801,7 +809,7 @@ fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Resul
 }
 
 fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> anyhow::Result<()> {
-    // Postgres nodes are not started automatically
+    // Endpoints are not started automatically
 
     broker::start_broker_process(env)?;
 
@@ -835,10 +843,10 @@ fn handle_stop_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<
 fn try_stop_all(env: &local_env::LocalEnv, immediate: bool) {
     let pageserver = PageServerNode::from_env(env);
 
-    // Stop all compute nodes
+    // Stop all endpoints
     match ComputeControlPlane::load(env.clone()) {
         Ok(cplane) => {
-            for (_k, node) in cplane.nodes {
+            for (_k, node) in cplane.endpoints {
                 if let Err(e) = node.stop(false) {
                     eprintln!("postgres stop failed: {e:#}");
                 }
@@ -871,7 +879,9 @@ fn cli() -> Command {
         .help("Name of the branch to be created or used as an alias for other services")
         .required(false);
 
-    let pg_node_arg = Arg::new("node").help("Postgres node name").required(false);
+    let endpoint_id_arg = Arg::new("endpoint_id")
+        .help("Postgres endpoint id")
+        .required(false);
 
     let safekeeper_id_arg = Arg::new("id").help("safekeeper id").required(false);
 
@@ -1025,27 +1035,27 @@ fn cli() -> Command {
                 )
         )
         .subcommand(
-            Command::new("pg")
+            Command::new("endpoint")
                 .arg_required_else_help(true)
                 .about("Manage postgres instances")
                 .subcommand(Command::new("list").arg(tenant_id_arg.clone()))
                 .subcommand(Command::new("create")
-                    .about("Create a postgres compute node")
-                    .arg(pg_node_arg.clone())
+                    .about("Create a compute endpoint")
+                    .arg(endpoint_id_arg.clone())
                     .arg(branch_name_arg.clone())
                     .arg(tenant_id_arg.clone())
                     .arg(lsn_arg.clone())
                     .arg(port_arg.clone())
                     .arg(
                         Arg::new("config-only")
-                            .help("Don't do basebackup, create compute node with only config files")
+                            .help("Don't do basebackup, create endpoint directory with only config files")
                             .long("config-only")
                             .required(false))
                     .arg(pg_version_arg.clone())
                 )
                 .subcommand(Command::new("start")
-                    .about("Start a postgres compute node.\n This command actually creates new node from scratch, but preserves existing config files")
-                    .arg(pg_node_arg.clone())
+                    .about("Start postgres.\n If the endpoint doesn't exist yet, it is created.")
+                    .arg(endpoint_id_arg.clone())
                     .arg(tenant_id_arg.clone())
                     .arg(branch_name_arg)
                     .arg(timeline_id_arg)
@@ -1055,7 +1065,7 @@ fn cli() -> Command {
                 )
                 .subcommand(
                     Command::new("stop")
-                    .arg(pg_node_arg)
+                    .arg(endpoint_id_arg)
                     .arg(tenant_id_arg)
                     .arg(
                         Arg::new("destroy")
@@ -1067,6 +1077,13 @@ fn cli() -> Command {
                 )
 
         )
+        // Obsolete old name for 'endpoint'. We now just print an error if it's used.
+        .subcommand(
+            Command::new("pg")
+                .hide(true)
+                .arg(Arg::new("ignore-rest").allow_hyphen_values(true).num_args(0..).required(false))
+                .trailing_var_arg(true)
+        )
         .subcommand(
             Command::new("start")
                 .about("Start page server and safekeepers")

control_plane/src/endpoint.rs

@@ -14,7 +14,6 @@ use anyhow::{Context, Result};
 use utils::{
     id::{TenantId, TimelineId},
     lsn::Lsn,
-    postgres_backend::AuthType,
 };
 
 use crate::local_env::{LocalEnv, DEFAULT_PG_VERSION};
@@ -26,54 +25,45 @@ use crate::postgresql_conf::PostgresConf;
 //
 pub struct ComputeControlPlane {
     base_port: u16,
-    pageserver: Arc<PageServerNode>,
-    pub nodes: BTreeMap<(TenantId, String), Arc<PostgresNode>>,
+
+    // endpoint ID is the key
+    pub endpoints: BTreeMap<String, Arc<Endpoint>>,
+
     env: LocalEnv,
+    pageserver: Arc<PageServerNode>,
 }
 
 impl ComputeControlPlane {
-    // Load current nodes with ports from data directories on disk
-    // Directory structure has the following layout:
-    // pgdatadirs
-    // |- tenants
-    // |  |- <tenant_id>
-    // |  |   |- <node name>
+    // Load current endpoints from the endpoints/ subdirectories
     pub fn load(env: LocalEnv) -> Result<ComputeControlPlane> {
         let pageserver = Arc::new(PageServerNode::from_env(&env));
 
-        let mut nodes = BTreeMap::default();
-        let pgdatadirspath = &env.pg_data_dirs_path();
-
-        for tenant_dir in fs::read_dir(pgdatadirspath)
-            .with_context(|| format!("failed to list {}", pgdatadirspath.display()))?
+        let mut endpoints = BTreeMap::default();
+        for endpoint_dir in fs::read_dir(env.endpoints_path())
+            .with_context(|| format!("failed to list {}", env.endpoints_path().display()))?
         {
-            let tenant_dir = tenant_dir?;
-            for timeline_dir in fs::read_dir(tenant_dir.path())
-                .with_context(|| format!("failed to list {}", tenant_dir.path().display()))?
-            {
-                let node = PostgresNode::from_dir_entry(timeline_dir?, &env, &pageserver)?;
-                nodes.insert((node.tenant_id, node.name.clone()), Arc::new(node));
-            }
+            let ep = Endpoint::from_dir_entry(endpoint_dir?, &env, &pageserver)?;
+            endpoints.insert(ep.name.clone(), Arc::new(ep));
         }
 
         Ok(ComputeControlPlane {
             base_port: 55431,
-            pageserver,
-            nodes,
+            endpoints,
             env,
+            pageserver,
         })
     }
 
     fn get_port(&mut self) -> u16 {
         1 + self
-            .nodes
+            .endpoints
             .values()
-            .map(|node| node.address.port())
+            .map(|ep| ep.address.port())
             .max()
             .unwrap_or(self.base_port)
     }
 
-    pub fn new_node(
+    pub fn new_endpoint(
         &mut self,
         tenant_id: TenantId,
         name: &str,
@@ -81,56 +71,58 @@ impl ComputeControlPlane {
         lsn: Option<Lsn>,
         port: Option<u16>,
         pg_version: u32,
-    ) -> Result<Arc<PostgresNode>> {
+    ) -> Result<Arc<Endpoint>> {
         let port = port.unwrap_or_else(|| self.get_port());
-        let node = Arc::new(PostgresNode {
+        let ep = Arc::new(Endpoint {
             name: name.to_owned(),
             address: SocketAddr::new("127.0.0.1".parse().unwrap(), port),
             env: self.env.clone(),
             pageserver: Arc::clone(&self.pageserver),
-            is_test: false,
             timeline_id,
             lsn,
             tenant_id,
-            uses_wal_proposer: false,
             pg_version,
         });
 
-        node.create_pgdata()?;
-        node.setup_pg_conf(self.env.pageserver.auth_type)?;
+        ep.create_pgdata()?;
+        ep.setup_pg_conf()?;
 
-        self.nodes
-            .insert((tenant_id, node.name.clone()), Arc::clone(&node));
+        self.endpoints.insert(ep.name.clone(), Arc::clone(&ep));
 
-        Ok(node)
+        Ok(ep)
     }
 }
 
 ///////////////////////////////////////////////////////////////////////////////
 
 #[derive(Debug)]
-pub struct PostgresNode {
-    pub address: SocketAddr,
+pub struct Endpoint {
+    /// used as the directory name
     name: String,
+    pub tenant_id: TenantId,
+    pub timeline_id: TimelineId,
+    // Some(lsn) if this is a read-only endpoint anchored at 'lsn'. None for the primary.
+    pub lsn: Option<Lsn>,
+
+    // port and address of the Postgres server
+    pub address: SocketAddr,
+    pg_version: u32,
+
+    // These are not part of the endpoint as such, but the environment
+    // the endpoint runs in.
     pub env: LocalEnv,
     pageserver: Arc<PageServerNode>,
-    is_test: bool,
-    pub timeline_id: TimelineId,
-    pub lsn: Option<Lsn>, // if it's a read-only node. None for primary
-    pub tenant_id: TenantId,
-    uses_wal_proposer: bool,
-    pg_version: u32,
 }
 
-impl PostgresNode {
+impl Endpoint {
     fn from_dir_entry(
         entry: std::fs::DirEntry,
         env: &LocalEnv,
         pageserver: &Arc<PageServerNode>,
-    ) -> Result<PostgresNode> {
+    ) -> Result<Endpoint> {
         if !entry.file_type()?.is_dir() {
             anyhow::bail!(
-                "PostgresNode::from_dir_entry failed: '{}' is not a directory",
+                "Endpoint::from_dir_entry failed: '{}' is not a directory",
                 entry.path().display()
             );
         }
@@ -140,7 +132,7 @@ impl PostgresNode {
         let name = fname.to_str().unwrap().to_string();
 
         // Read config file into memory
-        let cfg_path = entry.path().join("postgresql.conf");
+        let cfg_path = entry.path().join("pgdata").join("postgresql.conf");
         let cfg_path_str = cfg_path.to_string_lossy();
         let mut conf_file = File::open(&cfg_path)
             .with_context(|| format!("failed to open config file in {}", cfg_path_str))?;
@@ -152,7 +144,6 @@ impl PostgresNode {
         let port: u16 = conf.parse_field("port", &context)?;
         let timeline_id: TimelineId = conf.parse_field("neon.timeline_id", &context)?;
         let tenant_id: TenantId = conf.parse_field("neon.tenant_id", &context)?;
-        let uses_wal_proposer = conf.get("neon.safekeepers").is_some();
 
         // Read postgres version from PG_VERSION file to determine which postgres version binary to use.
         // If it doesn't exist, assume broken data directory and use default pg version.
@@ -167,16 +158,14 @@ impl PostgresNode {
             conf.parse_field_optional("recovery_target_lsn", &context)?;
 
         // ok now
-        Ok(PostgresNode {
+        Ok(Endpoint {
             address: SocketAddr::new("127.0.0.1".parse().unwrap(), port),
             name,
             env: env.clone(),
             pageserver: Arc::clone(pageserver),
-            is_test: false,
             timeline_id,
             lsn: recovery_target_lsn,
             tenant_id,
-            uses_wal_proposer,
             pg_version,
         })
     }
@@ -277,8 +266,8 @@ impl PostgresNode {
     }
 
     // Write postgresql.conf with default configuration
-    // and PG_VERSION file to the data directory of a new node.
-    fn setup_pg_conf(&self, auth_type: AuthType) -> Result<()> {
+    // and PG_VERSION file to the data directory of a new endpoint.
+    fn setup_pg_conf(&self) -> Result<()> {
         let mut conf = PostgresConf::new();
         conf.append("max_wal_senders", "10");
         conf.append("wal_log_hints", "off");
@@ -297,34 +286,17 @@ impl PostgresNode {
         // walproposer panics when basebackup is invalid, it is pointless to restart in this case.
         conf.append("restart_after_crash", "off");
 
-        // Configure the node to fetch pages from pageserver
+        // Configure the Neon Postgres extension to fetch pages from pageserver
         let pageserver_connstr = {
             let config = &self.pageserver.pg_connection_config;
             let (host, port) = (config.host(), config.port());
 
-            // Set up authentication
-            //
-            // $NEON_AUTH_TOKEN will be replaced with value from environment
-            // variable during compute pg startup. It is done this way because
-            // otherwise user will be able to retrieve the value using SHOW
-            // command or pg_settings
-            let password = if let AuthType::NeonJWT = auth_type {
-                "$NEON_AUTH_TOKEN"
-            } else {
-                ""
-            };
-            // NOTE avoiding spaces in connection string, because it is less error prone if we forward it somewhere.
-            // Also note that not all parameters are supported here. Because in compute we substitute $NEON_AUTH_TOKEN
-            // We parse this string and build it back with token from env var, and for simplicity rebuild
-            // uses only needed variables namely host, port, user, password.
-            format!("postgresql://no_user:{password}@{host}:{port}")
+            // NOTE: avoid spaces in connection string, because it is less error prone if we forward it somewhere.
+            format!("postgresql://no_user@{host}:{port}")
         };
         conf.append("shared_preload_libraries", "neon");
         conf.append_line("");
         conf.append("neon.pageserver_connstring", &pageserver_connstr);
-        if let AuthType::NeonJWT = auth_type {
-            conf.append("neon.safekeeper_token_env", "$NEON_AUTH_TOKEN");
-        }
         conf.append("neon.tenant_id", &self.tenant_id.to_string());
         conf.append("neon.timeline_id", &self.timeline_id.to_string());
         if let Some(lsn) = self.lsn {
@@ -350,7 +322,7 @@ impl PostgresNode {
         conf.append("max_replication_flush_lag", "10GB");
 
         if !self.env.safekeepers.is_empty() {
-            // Configure the node to connect to the safekeepers
+            // Configure Postgres to connect to the safekeepers
             conf.append("synchronous_standby_names", "walproposer");
 
             let safekeepers = self
@@ -385,7 +357,7 @@ impl PostgresNode {
     fn load_basebackup(&self, auth_token: &Option<String>) -> Result<()> {
         let backup_lsn = if let Some(lsn) = self.lsn {
             Some(lsn)
-        } else if self.uses_wal_proposer {
+        } else if !self.env.safekeepers.is_empty() {
             // LSN 0 means that it is bootstrap and we need to download just
             // latest data from the pageserver. That is a bit clumsy but whole bootstrap
             // procedure evolves quite actively right now, so let's think about it again
@@ -405,8 +377,12 @@ impl PostgresNode {
         Ok(())
     }
 
+    pub fn endpoint_path(&self) -> PathBuf {
+        self.env.endpoints_path().join(&self.name)
+    }
+
     pub fn pgdata(&self) -> PathBuf {
-        self.env.pg_data_dir(&self.tenant_id, &self.name)
+        self.endpoint_path().join("pgdata")
     }
 
     pub fn status(&self) -> &str {
@@ -424,7 +400,7 @@ impl PostgresNode {
 
     fn pg_ctl(&self, args: &[&str], auth_token: &Option<String>) -> Result<()> {
         let pg_ctl_path = self.env.pg_bin_dir(self.pg_version)?.join("pg_ctl");
-        let mut cmd = Command::new(pg_ctl_path);
+        let mut cmd = Command::new(&pg_ctl_path);
         cmd.args(
             [
                 &[
@@ -447,11 +423,15 @@ impl PostgresNode {
             "DYLD_LIBRARY_PATH",
             self.env.pg_lib_dir(self.pg_version)?.to_str().unwrap(),
         );
+
+        // Pass authentication token used for the connections to pageserver and safekeepers
         if let Some(token) = auth_token {
             cmd.env("NEON_AUTH_TOKEN", token);
         }
 
-        let pg_ctl = cmd.output().context("pg_ctl failed")?;
+        let pg_ctl = cmd
+            .output()
+            .context(format!("{} failed", pg_ctl_path.display()))?;
         if !pg_ctl.status.success() {
             anyhow::bail!(
                 "pg_ctl failed, exit code: {}, stdout: {}, stderr: {}",
@@ -464,12 +444,11 @@ impl PostgresNode {
     }
 
     pub fn start(&self, auth_token: &Option<String>) -> Result<()> {
-        // Bail if the node already running.
         if self.status() == "running" {
-            anyhow::bail!("The node is already running");
+            anyhow::bail!("The endpoint is already running");
         }
 
-        // 1. We always start compute node from scratch, so
+        // 1. We always start Postgres from scratch, so
         // if old dir exists, preserve 'postgresql.conf' and drop the directory
         let postgresql_conf_path = self.pgdata().join("postgresql.conf");
         let postgresql_conf = fs::read(&postgresql_conf_path).with_context(|| {
@@ -491,21 +470,17 @@ impl PostgresNode {
             File::create(self.pgdata().join("standby.signal"))?;
         }
 
-        // 4. Finally start the compute node postgres
-        println!("Starting postgres node at '{}'", self.connstr());
+        // 4. Finally start postgres
+        println!("Starting postgres at '{}'", self.connstr());
         self.pg_ctl(&["start"], auth_token)
     }
 
-    pub fn restart(&self, auth_token: &Option<String>) -> Result<()> {
-        self.pg_ctl(&["restart"], auth_token)
-    }
-
     pub fn stop(&self, destroy: bool) -> Result<()> {
         // If we are going to destroy data directory,
         // use immediate shutdown mode, otherwise,
         // shutdown gracefully to leave the data directory sane.
         //
-        // Compute node always starts from scratch, so stop
+        // Postgres is always started from scratch, so stop
         // without destroy only used for testing and debugging.
         //
         if destroy {
@@ -514,7 +489,7 @@ impl PostgresNode {
                 "Destroying postgres data directory '{}'",
                 self.pgdata().to_str().unwrap()
             );
-            fs::remove_dir_all(self.pgdata())?;
+            fs::remove_dir_all(self.endpoint_path())?;
         } else {
             self.pg_ctl(&["stop"], &None)?;
         }
@@ -530,26 +505,4 @@ impl PostgresNode {
             "postgres"
         )
     }
-
-    // XXX: cache that in control plane
-    pub fn whoami(&self) -> String {
-        let output = Command::new("whoami")
-            .output()
-            .expect("failed to execute whoami");
-
-        assert!(output.status.success(), "whoami failed");
-
-        String::from_utf8(output.stdout).unwrap().trim().to_string()
-    }
-}
-
-impl Drop for PostgresNode {
-    // destructor to clean up state after test is done
-    // XXX: we may detect failed test by setting some flag in catch_unwind()
-    // and checking it here. But let just clean datadirs on start.
-    fn drop(&mut self) {
-        if self.is_test {
-            let _ = self.stop(true);
-        }
-    }
 }

control_plane/src/lib.rs

@@ -9,7 +9,7 @@
 
 mod background_process;
 pub mod broker;
-pub mod compute;
+pub mod endpoint;
 pub mod local_env;
 pub mod pageserver;
 pub mod postgresql_conf;

control_plane/src/local_env.rs

@@ -5,6 +5,7 @@
 
 use anyhow::{bail, ensure, Context};
 
+use postgres_backend::AuthType;
 use reqwest::Url;
 use serde::{Deserialize, Serialize};
 use serde_with::{serde_as, DisplayFromStr};
@@ -17,9 +18,8 @@ use std::net::SocketAddr;
 use std::path::{Path, PathBuf};
 use std::process::{Command, Stdio};
 use utils::{
-    auth::{encode_from_key_file, Claims, Scope},
+    auth::{encode_from_key_file, Claims},
     id::{NodeId, TenantId, TenantTimelineId, TimelineId},
-    postgres_backend::AuthType,
 };
 
 use crate::safekeeper::SafekeeperNode;
@@ -110,15 +110,14 @@ impl NeonBroker {
 pub struct PageServerConf {
     // node id
     pub id: NodeId,
+
     // Pageserver connection settings
     pub listen_pg_addr: String,
     pub listen_http_addr: String,
 
-    // used to determine which auth type is used
-    pub auth_type: AuthType,
-
-    // jwt auth token used for communication with pageserver
-    pub auth_token: String,
+    // auth type used for the PG and HTTP ports
+    pub pg_auth_type: AuthType,
+    pub http_auth_type: AuthType,
 }
 
 impl Default for PageServerConf {
@@ -127,8 +126,8 @@ impl Default for PageServerConf {
             id: NodeId(0),
             listen_pg_addr: String::new(),
             listen_http_addr: String::new(),
-            auth_type: AuthType::Trust,
-            auth_token: String::new(),
+            pg_auth_type: AuthType::Trust,
+            http_auth_type: AuthType::Trust,
         }
     }
 }
@@ -201,14 +200,8 @@ impl LocalEnv {
         self.neon_distrib_dir.join("storage_broker")
     }
 
-    pub fn pg_data_dirs_path(&self) -> PathBuf {
-        self.base_data_dir.join("pgdatadirs").join("tenants")
-    }
-
-    pub fn pg_data_dir(&self, tenant_id: &TenantId, branch_name: &str) -> PathBuf {
-        self.pg_data_dirs_path()
-            .join(tenant_id.to_string())
-            .join(branch_name)
+    pub fn endpoints_path(&self) -> PathBuf {
+        self.base_data_dir.join("endpoints")
     }
 
     // TODO: move pageserver files into ./pageserver
@@ -401,49 +394,34 @@ impl LocalEnv {
 
         fs::create_dir(base_path)?;
 
-        // generate keys for jwt
-        // openssl genrsa -out private_key.pem 2048
-        let private_key_path;
+        // Generate keypair for JWT.
+        //
+        // The keypair is only needed if authentication is enabled in any of the
+        // components. For convenience, we generate the keypair even if authentication
+        // is not enabled, so that you can easily enable it after the initialization
+        // step. However, if the key generation fails, we treat it as non-fatal if
+        // authentication was not enabled.
         if self.private_key_path == PathBuf::new() {
-            private_key_path = base_path.join("auth_private_key.pem");
-            let keygen_output = Command::new("openssl")
-                .arg("genrsa")
-                .args(["-out", private_key_path.to_str().unwrap()])
-                .arg("2048")
-                .stdout(Stdio::null())
-                .output()
-                .context("failed to generate auth private key")?;
-            if !keygen_output.status.success() {
-                bail!(
-                    "openssl failed: '{}'",
-                    String::from_utf8_lossy(&keygen_output.stderr)
-                );
-            }
-            self.private_key_path = PathBuf::from("auth_private_key.pem");
-
-            let public_key_path = base_path.join("auth_public_key.pem");
-            // openssl rsa -in private_key.pem -pubout -outform PEM -out public_key.pem
-            let keygen_output = Command::new("openssl")
-                .arg("rsa")
-                .args(["-in", private_key_path.to_str().unwrap()])
-                .arg("-pubout")
-                .args(["-outform", "PEM"])
-                .args(["-out", public_key_path.to_str().unwrap()])
-                .stdout(Stdio::null())
-                .output()
-                .context("failed to generate auth private key")?;
-            if !keygen_output.status.success() {
-                bail!(
-                    "openssl failed: '{}'",
-                    String::from_utf8_lossy(&keygen_output.stderr)
-                );
+            match generate_auth_keys(
+                base_path.join("auth_private_key.pem").as_path(),
+                base_path.join("auth_public_key.pem").as_path(),
+            ) {
+                Ok(()) => {
+                    self.private_key_path = PathBuf::from("auth_private_key.pem");
+                }
+                Err(e) => {
+                    if !self.auth_keys_needed() {
+                        eprintln!("Could not generate keypair for JWT authentication: {e}");
+                        eprintln!("Continuing anyway because authentication was not enabled");
+                        self.private_key_path = PathBuf::from("auth_private_key.pem");
+                    } else {
+                        return Err(e);
+                    }
+                }
             }
         }
 
-        self.pageserver.auth_token =
-            self.generate_auth_token(&Claims::new(None, Scope::PageServerApi))?;
-
-        fs::create_dir_all(self.pg_data_dirs_path())?;
+        fs::create_dir_all(self.endpoints_path())?;
 
         for safekeeper in &self.safekeepers {
             fs::create_dir_all(SafekeeperNode::datadir_path_by_id(self, safekeeper.id))?;
@@ -451,6 +429,12 @@ impl LocalEnv {
 
         self.persist_config(base_path)
     }
+
+    fn auth_keys_needed(&self) -> bool {
+        self.pageserver.pg_auth_type == AuthType::NeonJWT
+            || self.pageserver.http_auth_type == AuthType::NeonJWT
+            || self.safekeepers.iter().any(|sk| sk.auth_enabled)
+    }
 }
 
 fn base_path() -> PathBuf {
@@ -460,6 +444,43 @@ fn base_path() -> PathBuf {
     }
 }
 
+/// Generate a public/private key pair for JWT authentication
+fn generate_auth_keys(private_key_path: &Path, public_key_path: &Path) -> anyhow::Result<()> {
+    // Generate the key pair
+    //
+    // openssl genpkey -algorithm ed25519 -out auth_private_key.pem
+    let keygen_output = Command::new("openssl")
+        .arg("genpkey")
+        .args(["-algorithm", "ed25519"])
+        .args(["-out", private_key_path.to_str().unwrap()])
+        .stdout(Stdio::null())
+        .output()
+        .context("failed to generate auth private key")?;
+    if !keygen_output.status.success() {
+        bail!(
+            "openssl failed: '{}'",
+            String::from_utf8_lossy(&keygen_output.stderr)
+        );
+    }
+    // Extract the public key from the private key file
+    //
+    // openssl pkey -in auth_private_key.pem -pubout -out auth_public_key.pem
+    let keygen_output = Command::new("openssl")
+        .arg("pkey")
+        .args(["-in", private_key_path.to_str().unwrap()])
+        .arg("-pubout")
+        .args(["-out", public_key_path.to_str().unwrap()])
+        .output()
+        .context("failed to extract public key from private key")?;
+    if !keygen_output.status.success() {
+        bail!(
+            "openssl failed: '{}'",
+            String::from_utf8_lossy(&keygen_output.stderr)
+        );
+    }
+    Ok(())
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;

control_plane/src/pageserver.rs

@@ -11,6 +11,7 @@ use anyhow::{bail, Context};
 use pageserver_api::models::{
     TenantConfigRequest, TenantCreateRequest, TenantInfo, TimelineCreateRequest, TimelineInfo,
 };
+use postgres_backend::AuthType;
 use postgres_connection::{parse_host_port, PgConnectionConfig};
 use reqwest::blocking::{Client, RequestBuilder, Response};
 use reqwest::{IntoUrl, Method};
@@ -20,7 +21,6 @@ use utils::{
     http::error::HttpErrorBody,
     id::{TenantId, TimelineId},
     lsn::Lsn,
-    postgres_backend::AuthType,
 };
 
 use crate::{background_process, local_env::LocalEnv};
@@ -82,15 +82,8 @@ impl PageServerNode {
         let (host, port) = parse_host_port(&env.pageserver.listen_pg_addr)
             .expect("Unable to parse listen_pg_addr");
         let port = port.unwrap_or(5432);
-        let password = if env.pageserver.auth_type == AuthType::NeonJWT {
-            Some(env.pageserver.auth_token.clone())
-        } else {
-            None
-        };
-
         Self {
-            pg_connection_config: PgConnectionConfig::new_host_port(host, port)
-                .set_password(password),
+            pg_connection_config: PgConnectionConfig::new_host_port(host, port),
             env: env.clone(),
             http_client: Client::new(),
             http_base_url: format!("http://{}/v1", env.pageserver.listen_http_addr),
@@ -106,25 +99,32 @@ impl PageServerNode {
             self.env.pg_distrib_dir_raw().display()
         );
 
-        let authg_type_param = format!("auth_type='{}'", self.env.pageserver.auth_type);
+        let http_auth_type_param =
+            format!("http_auth_type='{}'", self.env.pageserver.http_auth_type);
         let listen_http_addr_param = format!(
             "listen_http_addr='{}'",
             self.env.pageserver.listen_http_addr
         );
+
+        let pg_auth_type_param = format!("pg_auth_type='{}'", self.env.pageserver.pg_auth_type);
         let listen_pg_addr_param =
             format!("listen_pg_addr='{}'", self.env.pageserver.listen_pg_addr);
+
         let broker_endpoint_param = format!("broker_endpoint='{}'", self.env.broker.client_url());
 
         let mut overrides = vec![
             id,
             pg_distrib_dir_param,
-            authg_type_param,
+            http_auth_type_param,
+            pg_auth_type_param,
             listen_http_addr_param,
             listen_pg_addr_param,
             broker_endpoint_param,
         ];
 
-        if self.env.pageserver.auth_type != AuthType::Trust {
+        if self.env.pageserver.http_auth_type != AuthType::Trust
+            || self.env.pageserver.pg_auth_type != AuthType::Trust
+        {
             overrides.push("auth_validation_public_key_path='auth_public_key.pem'".to_owned());
         }
         overrides
@@ -247,7 +247,10 @@ impl PageServerNode {
     }
 
     fn pageserver_env_variables(&self) -> anyhow::Result<Vec<(String, String)>> {
-        Ok(if self.env.pageserver.auth_type != AuthType::Trust {
+        // FIXME: why is this tied to pageserver's auth type? Whether or not the safekeeper
+        // needs a token, and how to generate that token, seems independent to whether
+        // the pageserver requires a token in incoming requests.
+        Ok(if self.env.pageserver.http_auth_type != AuthType::Trust {
             // Generate a token to connect from the pageserver to a safekeeper
             let token = self
                 .env
@@ -270,27 +273,30 @@ impl PageServerNode {
         background_process::stop_process(immediate, "pageserver", &self.pid_file())
     }
 
-    pub fn page_server_psql(&self, sql: &str) -> Vec<postgres::SimpleQueryMessage> {
-        let mut client = self.pg_connection_config.connect_no_tls().unwrap();
-
-        println!("Pageserver query: '{sql}'");
-        client.simple_query(sql).unwrap()
-    }
-
-    pub fn page_server_psql_client(&self) -> result::Result<postgres::Client, postgres::Error> {
-        self.pg_connection_config.connect_no_tls()
-    }
-
-    fn http_request<U: IntoUrl>(&self, method: Method, url: U) -> RequestBuilder {
-        let mut builder = self.http_client.request(method, url);
-        if self.env.pageserver.auth_type == AuthType::NeonJWT {
-            builder = builder.bearer_auth(&self.env.pageserver.auth_token)
+    pub fn page_server_psql_client(&self) -> anyhow::Result<postgres::Client> {
+        let mut config = self.pg_connection_config.clone();
+        if self.env.pageserver.pg_auth_type == AuthType::NeonJWT {
+            let token = self
+                .env
+                .generate_auth_token(&Claims::new(None, Scope::PageServerApi))?;
+            config = config.set_password(Some(token));
         }
-        builder
+        Ok(config.connect_no_tls()?)
+    }
+
+    fn http_request<U: IntoUrl>(&self, method: Method, url: U) -> anyhow::Result<RequestBuilder> {
+        let mut builder = self.http_client.request(method, url);
+        if self.env.pageserver.http_auth_type == AuthType::NeonJWT {
+            let token = self
+                .env
+                .generate_auth_token(&Claims::new(None, Scope::PageServerApi))?;
+            builder = builder.bearer_auth(token)
+        }
+        Ok(builder)
     }
 
     pub fn check_status(&self) -> Result<()> {
-        self.http_request(Method::GET, format!("{}/status", self.http_base_url))
+        self.http_request(Method::GET, format!("{}/status", self.http_base_url))?
             .send()?
             .error_from_body()?;
         Ok(())
@@ -298,7 +304,7 @@ impl PageServerNode {
 
     pub fn tenant_list(&self) -> Result<Vec<TenantInfo>> {
         Ok(self
-            .http_request(Method::GET, format!("{}/tenant", self.http_base_url))
+            .http_request(Method::GET, format!("{}/tenant", self.http_base_url))?
             .send()?
             .error_from_body()?
             .json()?)
@@ -352,11 +358,24 @@ 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")?,
+            min_resident_size_override: settings
+                .remove("min_resident_size_override")
+                .map(|x| x.parse::<u64>())
+                .transpose()
+                .context("Failed to parse 'min_resident_size_override' as integer")?,
+            evictions_low_residence_duration_metric_threshold: settings
+                .remove("evictions_low_residence_duration_metric_threshold")
+                .map(|x| x.to_string()),
         };
         if !settings.is_empty() {
             bail!("Unrecognized tenant settings: {settings:?}")
         }
-        self.http_request(Method::POST, format!("{}/tenant", self.http_base_url))
+        self.http_request(Method::POST, format!("{}/tenant", self.http_base_url))?
             .json(&request)
             .send()?
             .error_from_body()?
@@ -373,7 +392,7 @@ impl PageServerNode {
     }
 
     pub fn tenant_config(&self, tenant_id: TenantId, settings: HashMap<&str, &str>) -> Result<()> {
-        self.http_request(Method::PUT, format!("{}/tenant/config", self.http_base_url))
+        self.http_request(Method::PUT, format!("{}/tenant/config", self.http_base_url))?
             .json(&TenantConfigRequest {
                 tenant_id,
                 checkpoint_distance: settings
@@ -419,6 +438,19 @@ 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")?,
+                min_resident_size_override: settings
+                    .get("min_resident_size_override")
+                    .map(|x| x.parse::<u64>())
+                    .transpose()
+                    .context("Failed to parse 'min_resident_size_override' as an integer")?,
+                evictions_low_residence_duration_metric_threshold: settings
+                    .get("evictions_low_residence_duration_metric_threshold")
+                    .map(|x| x.to_string()),
             })
             .send()?
             .error_from_body()?;
@@ -431,7 +463,7 @@ impl PageServerNode {
             .http_request(
                 Method::GET,
                 format!("{}/tenant/{}/timeline", self.http_base_url, tenant_id),
-            )
+            )?
             .send()?
             .error_from_body()?
             .json()?;
@@ -450,7 +482,7 @@ impl PageServerNode {
         self.http_request(
             Method::POST,
             format!("{}/tenant/{}/timeline", self.http_base_url, tenant_id),
-        )
+        )?
         .json(&TimelineCreateRequest {
             new_timeline_id,
             ancestor_start_lsn,
@@ -487,7 +519,7 @@ impl PageServerNode {
         pg_wal: Option<(Lsn, PathBuf)>,
         pg_version: u32,
     ) -> anyhow::Result<()> {
-        let mut client = self.pg_connection_config.connect_no_tls().unwrap();
+        let mut client = self.page_server_psql_client()?;
 
         // Init base reader
         let (start_lsn, base_tarfile_path) = base;

control_plane/src/safekeeper.rs

@@ -1,7 +1,6 @@
 use std::io::Write;
 use std::path::PathBuf;
 use std::process::Child;
-use std::sync::Arc;
 use std::{io, result};
 
 use anyhow::Context;
@@ -11,7 +10,6 @@ use reqwest::{IntoUrl, Method};
 use thiserror::Error;
 use utils::{http::error::HttpErrorBody, id::NodeId};
 
-use crate::pageserver::PageServerNode;
 use crate::{
     background_process,
     local_env::{LocalEnv, SafekeeperConf},
@@ -65,14 +63,10 @@ pub struct SafekeeperNode {
     pub env: LocalEnv,
     pub http_client: Client,
     pub http_base_url: String,
-
-    pub pageserver: Arc<PageServerNode>,
 }
 
 impl SafekeeperNode {
     pub fn from_env(env: &LocalEnv, conf: &SafekeeperConf) -> SafekeeperNode {
-        let pageserver = Arc::new(PageServerNode::from_env(env));
-
         SafekeeperNode {
             id: conf.id,
             conf: conf.clone(),
@@ -80,7 +74,6 @@ impl SafekeeperNode {
             env: env.clone(),
             http_client: Client::new(),
             http_base_url: format!("http://127.0.0.1:{}/v1", conf.http_port),
-            pageserver,
         }
     }
 
@@ -115,6 +108,10 @@ impl SafekeeperNode {
         let datadir = self.datadir_path();
 
         let id_string = id.to_string();
+        // TODO: add availability_zone to the config.
+        // Right now we just specify any value here and use it to check metrics in tests.
+        let availability_zone = format!("sk-{}", id_string);
+
         let mut args = vec![
             "-D",
             datadir.to_str().with_context(|| {
@@ -126,6 +123,8 @@ impl SafekeeperNode {
             &listen_pg,
             "--listen-http",
             &listen_http,
+            "--availability-zone",
+            &availability_zone,
         ];
         if !self.conf.sync {
             args.push("--no-sync");
@@ -157,7 +156,7 @@ impl SafekeeperNode {
         }
 
         background_process::start_process(
-            &format!("safekeeper {id}"),
+            &format!("safekeeper-{id}"),
             &datadir,
             &self.env.safekeeper_bin(),
             &args,

docker-compose/docker-compose.yml

@@ -160,6 +160,7 @@ services:
     build:
       context: ./compute_wrapper/
       args:
+        - REPOSITORY=${REPOSITORY:-neondatabase}
         - COMPUTE_IMAGE=compute-node-v${PG_VERSION:-14}
         - TAG=${TAG:-latest}
         - http_proxy=$http_proxy

docs/authentication.md

@@ -29,12 +29,54 @@ These components should not have access to the private key and may only get toke
 The key pair is generated once for an installation of compute/pageserver/safekeeper, e.g. by `neon_local init`.
 There is currently no way to rotate the key without bringing down all components.
 
+### Best practices
+
+See [RFC 8725: JSON Web Token Best Current Practices](https://www.rfc-editor.org/rfc/rfc8725)
+
+
+### Token format
+
+The JWT tokens in Neon use "EdDSA" as the algorithm (defined in [RFC8037](https://www.rfc-editor.org/rfc/rfc8037)).
+
+Example:
+
+Header:
+
+```
+{
+  "alg": "EdDSA",
+  "typ": "JWT"
+}
+```
+
+Payload:
+
+```
+{
+  "scope": "tenant",  # "tenant", "pageserverapi", or "safekeeperdata"
+  "tenant_id": "5204921ff44f09de8094a1390a6a50f6",
+}
+```
+
+
+Meanings of scope:
+
+"tenant": Provides access to all data for a specific tenant
+
+"pageserverapi": Provides blanket access to all tenants on the pageserver plus pageserver-wide APIs.
+Should only be used e.g. for status check/tenant creation/list.
+
+"safekeeperdata": Provides blanket access to all data on the safekeeper plus safekeeper-wide APIs.
+Should only be used e.g. for status check.
+Currently also used for connection from any pageserver to any safekeeper.
+
+
 ### CLI
 CLI generates a key pair during call to `neon_local init` with the following commands:
 
 ```bash
-openssl genrsa -out auth_private_key.pem 2048
-openssl rsa -in auth_private_key.pem -pubout -outform PEM -out auth_public_key.pem
+openssl genpkey -algorithm ed25519 -out auth_private_key.pem
+openssl pkey -in auth_private_key.pem -pubout -out auth_public_key.pem
 ```
 
 Configuration files for all components point to `public_key.pem` for JWT validation.
@@ -64,20 +106,22 @@ Their authentication is just plain PostgreSQL authentication and out of scope fo
 There is no administrative API except those provided by PostgreSQL.
 
 #### Outgoing connections
-Compute connects to Pageserver for getting pages.
-The connection string is configured by the `neon.pageserver_connstring` PostgreSQL GUC, e.g. `postgresql://no_user:$NEON_AUTH_TOKEN@localhost:15028`.
-The environment variable inside the connection string is substituted with
-the JWT token.
+Compute connects to Pageserver for getting pages. The connection string is
+configured by the `neon.pageserver_connstring` PostgreSQL GUC,
+e.g. `postgresql://no_user@localhost:15028`. If the `$NEON_AUTH_TOKEN`
+environment variable is set, it is used as the password for the connection. (The
+pageserver uses JWT tokens for authentication, so the password is really a
+token.)
 
-Compute connects to Safekeepers to write and commit data.
-The token is the same for all safekeepers.
-It's stored in an environment variable, whose name is configured
-by the `neon.safekeeper_token_env` PostgreSQL GUC.
-If the GUC is unset, no token is passed.
+Compute connects to Safekeepers to write and commit data. The list of safekeeper
+addresses is given in the `neon.safekeepers` GUC. The connections to the
+safekeepers take the password from the `$NEON_AUTH_TOKEN` environment
+variable, if set.
 
-Note that both tokens can be (and typically are) the same;
-the scope is the tenant and the token is usually passed through the
-`$NEON_AUTH_TOKEN` environment variable.
+The `compute_ctl` binary that runs before the PostgreSQL server, and launches
+PostgreSQL, also makes a connection to the pageserver. It uses it to fetch the
+initial "base backup" dump, to initialize the PostgreSQL data directory. It also
+uses `$NEON_AUTH_TOKEN` as the password for the connection.
 
 ### Pageserver
 #### Overview
@@ -102,10 +146,12 @@ Each compute should present a token valid for the timeline's tenant.
 Pageserver also has HTTP API: some parts are per-tenant,
 some parts are server-wide, these are different scopes.
 
-The `auth_type` configuration variable in Pageserver's config may have
-either of three values:
+Authentication can be enabled separately for the HTTP mgmt API, and
+for the libpq connections from compute. The `http_auth_type` and
+`pg_auth_type` configuration variables in Pageserver's config may
+have one of these values:
 
-* `Trust` removes all authentication. The outdated `MD5` value does likewise
+* `Trust` removes all authentication.
 * `NeonJWT` enables JWT validation.
    Tokens are validated using the public key which lies in a PEM file
    specified in the `auth_validation_public_key_path` config.

docs/docker.md

@@ -37,9 +37,9 @@ You can specify version of neon cluster using following environment values.
 - PG_VERSION: postgres version for compute (default is 14)
 - TAG: the tag version of [docker image](https://registry.hub.docker.com/r/neondatabase/neon/tags) (default is latest), which is tagged in [CI test](/.github/workflows/build_and_test.yml)
 ```
-$ cd docker-compose/docker-compose.yml
+$ cd docker-compose/
 $ docker-compose down   # remove the conainers if exists
-$ PG_VERSION=15 TAG=2221 docker-compose up --build -d  # You can specify the postgres and image version
+$ PG_VERSION=15 TAG=2937 docker-compose up --build -d  # You can specify the postgres and image version
 Creating network "dockercompose_default" with the default driver
 Creating docker-compose_storage_broker_1       ... done
 (...omit...)

docs/rfcs/022-pageserver-delete-from-s3.md

@@ -0,0 +1,269 @@
+# Deleting pageserver part of tenants data from s3
+
+Created on 08.03.23
+
+## Motivation
+
+Currently we dont delete pageserver part of the data from s3 when project is deleted. (The same is true for safekeepers, but this outside of the scope of this RFC).
+
+This RFC aims to spin a discussion to come to a robust deletion solution that wont put us in into a corner for features like postponed deletion (when we keep data for user to be able to restore a project if it was deleted by accident)
+
+## Summary
+
+TLDR; There are two options, one based on control plane issuing actual delete requests to s3 and the other one that keeps s3 stuff bound to pageserver. Each one has its pros and cons.
+
+The decision is to stick with pageserver centric approach. For motivation see [Decision](#decision).
+
+## Components
+
+pageserver, control-plane
+
+## Requirements
+
+Deletion should successfully finish (eventually) without leaving dangling files in presense of:
+
+- component restarts
+- component outage
+- pageserver loss
+
+## Proposed implementation
+
+Before the options are discussed, note that deletion can be quite long process. For deletion from s3 the obvious choice is [DeleteObjects](https://docs.aws.amazon.com/AmazonS3/latest/API/API_DeleteObjects.html) API call. It allows to batch deletion of up to 1k objects in one API call. So deletion operation linearly depends on number of layer files.
+
+Another design limitation is that there is no cheap `mv` operation available for s3. `mv` from `aws s3 mv` uses `copy(src, dst) + delete(src)`. So `mv`-like operation is not feasible as a building block because it actually amplifies the problem with both duration and resulting cost of the operation.
+
+The case when there are multiple pageservers handling the same tenants is largely out of scope of the RFC. We still consider case with migration from one PS to another, but do not consider case when tenant exists on multiple pageservers for extended period of time. The case with multiple pageservers can be reduced to case with one pageservers by calling detach on all pageservers except the last one, for it actual delete needs to be called.
+
+For simplicity lets look into deleting tenants. Differences in deletion process between tenants and timelines are mentioned in paragraph ["Differences between tenants and timelines"](#differences-between-tenants-and-timelines)
+
+### 1. Pageserver owns deletion machinery
+
+#### The sequence
+
+TLDR; With this approach control plane needs to call delete on a tenant and poll for progress. As much as possible is handled on pageserver. Lets see the sequence.
+
+Happy path:
+
+```mermaid
+sequenceDiagram
+    autonumber
+    participant CP as Control Plane
+    participant PS as Pageserver
+    participant S3
+
+    CP->>PS: Delete tenant
+    PS->>S3: Create deleted mark file at <br> /tenant/meta/deleted
+    PS->>PS: Create deleted mark file locally
+    PS->>CP: Accepted
+    PS->>PS: delete local files other than deleted mark
+    loop Delete layers for each timeline
+        PS->>S3: delete(..)
+        CP->>PS: Finished?
+        PS->>CP: False
+    end
+    PS->>S3: Delete mark file
+    PS->>PS: Delete local mark file
+
+    loop Poll for status
+        CP->>PS: Finished?
+        PS->>CP: True or False
+    end
+```
+
+Why two mark files?
+Remote one is needed for cases when pageserver is lost during deletion so other pageserver can learn the deletion from s3 during attach.
+
+Why local mark file is needed?
+
+If we dont have one, we have two choices, delete local data before deleting the remote part or do that after.
+
+If we delete local data before remote then during restart pageserver wont pick up remote tenant at all because nothing is available locally (pageserver looks for remote conuterparts of locally available tenants).
+
+If we delete local data after remote then at the end of the sequence when remote mark file is deleted if pageserver restart happens then the state is the same to situation when pageserver just missing data on remote without knowing the fact that this data is intended to be deleted. In this case the current behavior is upload everything local-only to remote.
+
+Thus we need local record of tenant being deleted as well.
+
+##### Handle pageserver crashes
+
+Lets explore sequences with various crash points.
+
+Pageserver crashes before `deleted` mark file is persisted in s3:
+
+```mermaid
+sequenceDiagram
+    autonumber
+    participant CP as Control Plane
+    participant PS as Pageserver
+    participant S3
+
+    CP->>PS: Delete tenant
+    note over PS: Crash point 1.
+    CP->>PS: Retry delete request
+
+    PS->>S3: Create deleted mark file at <br> /tenant/meta/deleted
+    PS->>PS: Create deleted mark file locally
+
+    PS->>CP: Accepted
+
+    PS->>PS: delete local files other than deleted mark
+
+    loop Delete layers for each timeline
+        PS->>S3: delete(..)
+        CP->>PS: Finished?
+        PS->>CP: False
+    end
+    PS->>S3: Delete mark file
+    PS->>PS: Delete local mark file
+
+    CP->>PS: Finished?
+    PS->>CP: True
+```
+
+Pageserver crashed when deleted mark was about to be persisted in s3, before Control Plane gets a response:
+
+```mermaid
+sequenceDiagram
+    autonumber
+    participant CP as Control Plane
+    participant PS as Pageserver
+    participant S3
+
+    CP->>PS: Delete tenant
+    PS->>S3: Create deleted mark file at <br> /tenant/meta/deleted
+
+    note over PS: Crash point 2.
+    note over PS: During startup we reconcile <br> with remote and see <br> whether the remote mark exists
+    alt Remote mark exists
+        PS->>PS: create local mark if its missing
+        PS->>PS: delete local files other than deleted mark
+        loop Delete layers for each timeline
+            PS->>S3: delete(..)
+        end
+
+        note over CP: Eventually console should <br> retry delete request
+
+        CP->>PS: Retry delete tenant
+        PS->>CP: Not modified
+    else Mark is missing
+        note over PS: Continue to operate the tenant as if deletion didnt happen
+
+        note over CP: Eventually console should <br> retry delete request
+
+        CP->>PS: Retry delete tenant
+        PS->>S3: Create deleted mark file at <br> /tenant/meta/deleted
+        PS->>CP: Delete tenant
+    end
+
+    PS->>PS: Continue with layer file deletions
+    loop Delete layers for each timeline
+        PS->>S3: delete(..)
+        CP->>PS: Finished?
+        PS->>CP: False
+    end
+
+    PS->>S3: Delete mark file
+    PS->>PS: Delete local mark file
+
+    CP->>PS: Finished?
+    PS->>CP: True
+```
+
+Similar sequence applies when both local and remote marks were persisted but Control Plane still didnt receive a response.
+
+If pageserver crashes after both mark files were deleted then it will reply to control plane status poll request with 404 which should be treated by control plane as success.
+
+The same applies if pageserver crashes in the end, when remote mark is deleted but before local one gets deleted. In this case on restart pageserver moves forward with deletion of local mark and Control Plane will receive 404.
+
+##### Differences between tenants and timelines
+
+For timeline the sequence is the same with the following differences:
+
+- remote delete mark file can be replaced with a boolean "deleted" flag in index_part.json
+- local deletion mark is not needed, because whole tenant is kept locally so situation described in motivation for local mark is impossible
+
+##### Handle pageserver loss
+
+If pageseserver is lost then the deleted tenant should be attached to different pageserver and delete request needs to be retried against new pageserver. Then attach logic is shared with one described for pageserver restarts (local deletion mark wont be available so needs to be created).
+
+##### Restrictions for tenant that is in progress of being deleted
+
+I propose to add another state to tenant/timeline - PendingDelete. This state shouldnt allow executing any operations aside from polling the deletion status.
+
+#### Summary
+
+Pros:
+
+- Storage is not dependent on control plane. Storage can be restarted even if control plane is not working.
+- Allows for easier dogfooding, console can use Neon backed database as primary operational data store. If storage depends on control plane and control plane depends on storage we're stuck.
+- No need to share inner s3 workings with control plane. Pageserver presents api contract and S3 paths are not part of this contract.
+- No need to pass list of alive timelines to attach call. This will be solved by pageserver observing deleted flag. See
+
+Cons:
+
+- Logic is a tricky, needs good testing
+- Anything else?
+
+### 2. Control plane owns deletion machinery
+
+In this case the only action performed on pageserver is removal of local files.
+
+Everything else is done by control plane. The steps are as follows:
+
+1. Control plane marks tenant as "delete pending" in its database
+2. It lists the s3 for all the files and repeatedly calls delete until nothing is left behind
+3. When no files are left marks deletion as completed
+
+In case of restart it selects all tenants marked as "delete pending" and continues the deletion.
+
+For tenants it is simple. For timelines there are caveats.
+
+Assume that the same workflow is used for timelines.
+
+If a tenant gets relocated during timeline deletion the attach call with its current logic will pick up deleted timeline in its half deleted state.
+
+Available options:
+
+- require list of alive timelines to be passed to attach call
+- use the same schema with flag in index_part.json (again part of the caveats around pageserver restart applies). In this case nothing stops pageserver from implementing deletion inside if we already have these deletion marks.
+
+With first option the following problem becomes apparent:
+
+Who is the source of truth regarding timeline liveness?
+
+Imagine:
+PS1 fails.
+PS2 gets assigned the tenant.
+New branch gets created
+PS1 starts up (is it possible or we just recycle it?)
+PS1 is unaware of the new branch. It can either fall back to s3 ls, or ask control plane.
+
+So here comes the dependency of storage on control plane. During restart storage needs to know which timelines are valid for operation. If there is nothing on s3 that can answer that question storage neeeds to ask control plane.
+
+### Summary
+
+Cons:
+
+- Potential thundering herd-like problem during storage restart (requests to control plane)
+- Potential increase in storage startup time (additional request to control plane)
+- Storage startup starts to depend on console
+- Erroneous attach call can attach tenant in half deleted state
+
+Pros:
+
+- Easier to reason about if you dont have to account for pageserver restarts
+
+### Extra notes
+
+There was a concern that having deletion code in pageserver is a littlebit scary, but we need to have this code somewhere. So to me it is equally scary to have that in whatever place it ends up at.
+
+Delayed deletion can be done with both approaches. As discussed with Anna (@stepashka) this is only relevant for tenants (projects) not for timelines. For first approach detach can be called immediately and deletion can be done later with attach + delete. With second approach control plane needs to start the deletion whenever necessary.
+
+## Decision
+
+After discussion in comments I see that we settled on two options (though a bit different from ones described in rfc). First one is the same - pageserver owns as much as possible. The second option is that pageserver owns markers thing, but actual deletion happens in control plane by repeatedly calling ls + delete.
+
+To my mind the only benefit of the latter approach is possible code reuse between safekeepers and pageservers. Otherwise poking around integrating s3 library into control plane, configuring shared knowledge abouth paths in s3 - are the downsides. Another downside of relying on control plane is the testing process. Control plane resides in different repository so it is quite hard to test pageserver related changes there. e2e test suite there doesnt support shutting down pageservers, which are separate docker containers there instead of just processes.
+
+With pageserver owning everything we still give the retry logic to control plane but its easier to duplicate if needed compared to sharing inner s3 workings. We will have needed tests for retry logic in neon repo.
+
+So the decision is to proceed with pageserver centric approach.

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/sourcetree.md

@@ -129,13 +129,12 @@ Run `poetry shell` to activate the virtual environment.
 Alternatively, use `poetry run` to run a single command in the venv, e.g. `poetry run pytest`.
 
 ### Obligatory checks
-We force code formatting via `black`, `isort` and type hints via `mypy`.
+We force code formatting via `black`, `ruff`, and type hints via `mypy`.
 Run the following commands in the repository's root (next to `pyproject.toml`):
 
 ```bash
-poetry run isort .  # Imports are reformatted
 poetry run black .  # All code is reformatted
-poetry run flake8 .  # Python linter
+poetry run ruff .  # Python linter
 poetry run mypy .  # Ensure there are no typing errors
 ```
 

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/compute_api/Cargo.toml

@@ -0,0 +1,14 @@
+[package]
+name = "compute_api"
+version = "0.1.0"
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+anyhow.workspace = true
+chrono.workspace = true
+serde.workspace = true
+serde_with.workspace = true
+serde_json.workspace = true
+
+workspace_hack.workspace = true

libs/compute_api/src/lib.rs

@@ -0,0 +1,3 @@
+pub mod requests;
+pub mod responses;
+pub mod spec;

libs/compute_api/src/requests.rs

@@ -0,0 +1,14 @@
+//! Structs representing the JSON formats used in the compute_ctl's HTTP API.
+
+use crate::spec::ComputeSpec;
+use serde::Deserialize;
+
+/// Request of the /configure API
+///
+/// We now pass only `spec` in the configuration request, but later we can
+/// extend it and something like `restart: bool` or something else. So put
+/// `spec` into a struct initially to be more flexible in the future.
+#[derive(Deserialize, Debug)]
+pub struct ConfigurationRequest {
+    pub spec: ComputeSpec,
+}

libs/compute_api/src/responses.rs

@@ -0,0 +1,78 @@
+//! Structs representing the JSON formats used in the compute_ctl's HTTP API.
+
+use chrono::{DateTime, Utc};
+use serde::{Deserialize, Serialize, Serializer};
+
+use crate::spec::ComputeSpec;
+
+#[derive(Serialize, Debug)]
+pub struct GenericAPIError {
+    pub error: String,
+}
+
+/// Response of the /status API
+#[derive(Serialize, Debug)]
+#[serde(rename_all = "snake_case")]
+pub struct ComputeStatusResponse {
+    pub tenant: Option<String>,
+    pub timeline: Option<String>,
+    pub status: ComputeStatus,
+    #[serde(serialize_with = "rfc3339_serialize")]
+    pub last_active: DateTime<Utc>,
+    pub error: Option<String>,
+}
+
+#[derive(Serialize)]
+#[serde(rename_all = "snake_case")]
+pub struct ComputeState {
+    pub status: ComputeStatus,
+    /// Timestamp of the last Postgres activity
+    #[serde(serialize_with = "rfc3339_serialize")]
+    pub last_active: DateTime<Utc>,
+    pub error: Option<String>,
+}
+
+#[derive(Serialize, Clone, Copy, Debug, PartialEq, Eq)]
+#[serde(rename_all = "snake_case")]
+pub enum ComputeStatus {
+    // Spec wasn't provided at start, waiting for it to be
+    // provided by control-plane.
+    Empty,
+    // Compute configuration was requested.
+    ConfigurationPending,
+    // Compute node has spec and initial startup and
+    // configuration is in progress.
+    Init,
+    // Compute is configured and running.
+    Running,
+    // New spec is being applied.
+    Configuration,
+    // Either startup or configuration failed,
+    // compute will exit soon or is waiting for
+    // control-plane to terminate it.
+    Failed,
+}
+
+fn rfc3339_serialize<S>(x: &DateTime<Utc>, s: S) -> Result<S::Ok, S::Error>
+where
+    S: Serializer,
+{
+    x.to_rfc3339().serialize(s)
+}
+
+/// Response of the /metrics.json API
+#[derive(Clone, Debug, Default, Serialize)]
+pub struct ComputeMetrics {
+    pub sync_safekeepers_ms: u64,
+    pub basebackup_ms: u64,
+    pub config_ms: u64,
+    pub total_startup_ms: u64,
+}
+
+/// Response of the `/computes/{compute_id}/spec` control-plane API.
+/// This is not actually a compute API response, so consider moving
+/// to a different place.
+#[derive(Deserialize, Debug)]
+pub struct ControlPlaneSpecResponse {
+    pub spec: Option<ComputeSpec>,
+}

libs/compute_api/src/spec.rs

@@ -0,0 +1,97 @@
+//! `ComputeSpec` represents the contents of the spec.json file.
+//!
+//! The spec.json file is used to pass information to 'compute_ctl'. It contains
+//! all the information needed to start up the right version of PostgreSQL,
+//! and connect it to the storage nodes.
+use serde::Deserialize;
+use std::collections::HashMap;
+
+/// String type alias representing Postgres identifier and
+/// intended to be used for DB / role names.
+pub type PgIdent = String;
+
+/// Cluster spec or configuration represented as an optional number of
+/// delta operations + final cluster state description.
+#[derive(Clone, Debug, Default, Deserialize)]
+pub struct ComputeSpec {
+    pub format_version: f32,
+
+    // The control plane also includes a 'timestamp' field in the JSON document,
+    // but we don't use it for anything. Serde will ignore missing fields when
+    // deserializing it.
+    pub operation_uuid: Option<String>,
+    /// Expected cluster state at the end of transition process.
+    pub cluster: Cluster,
+    pub delta_operations: Option<Vec<DeltaOp>>,
+
+    pub storage_auth_token: Option<String>,
+
+    pub startup_tracing_context: Option<HashMap<String, String>>,
+}
+
+#[derive(Clone, Debug, Default, Deserialize)]
+pub struct Cluster {
+    pub cluster_id: String,
+    pub name: String,
+    pub state: Option<String>,
+    pub roles: Vec<Role>,
+    pub databases: Vec<Database>,
+    pub settings: GenericOptions,
+}
+
+/// Single cluster state changing operation that could not be represented as
+/// a static `Cluster` structure. For example:
+/// - DROP DATABASE
+/// - DROP ROLE
+/// - ALTER ROLE name RENAME TO new_name
+/// - ALTER DATABASE name RENAME TO new_name
+#[derive(Clone, Debug, Deserialize)]
+pub struct DeltaOp {
+    pub action: String,
+    pub name: PgIdent,
+    pub new_name: Option<PgIdent>,
+}
+
+/// Rust representation of Postgres role info with only those fields
+/// that matter for us.
+#[derive(Clone, Debug, Deserialize)]
+pub struct Role {
+    pub name: PgIdent,
+    pub encrypted_password: Option<String>,
+    pub options: GenericOptions,
+}
+
+/// Rust representation of Postgres database info with only those fields
+/// that matter for us.
+#[derive(Clone, Debug, Deserialize)]
+pub struct Database {
+    pub name: PgIdent,
+    pub owner: PgIdent,
+    pub options: GenericOptions,
+}
+
+/// Common type representing both SQL statement params with or without value,
+/// like `LOGIN` or `OWNER username` in the `CREATE/ALTER ROLE`, and config
+/// options like `wal_level = logical`.
+#[derive(Clone, Debug, Deserialize)]
+pub struct GenericOption {
+    pub name: String,
+    pub value: Option<String>,
+    pub vartype: String,
+}
+
+/// Optional collection of `GenericOption`'s. Type alias allows us to
+/// declare a `trait` on it.
+pub type GenericOptions = Option<Vec<GenericOption>>;
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::fs::File;
+
+    #[test]
+    fn parse_spec_file() {
+        let file = File::open("tests/cluster_spec.json").unwrap();
+        let _spec: ComputeSpec = serde_json::from_reader(file).unwrap();
+    }
+}

libs/compute_api/tests/cluster_spec.json

@@ -178,6 +178,11 @@
                 "name": "neon.pageserver_connstring",
                 "value": "host=127.0.0.1 port=6400",
                 "vartype": "string"
+            },
+            {
+                "name": "test.escaping",
+                "value": "here's a backslash \\ and a quote ' and a double-quote \" hooray",
+                "vartype": "string"
             }
         ]
     },

libs/consumption_metrics/Cargo.toml

@@ -4,13 +4,12 @@ version = "0.1.0"
 edition = "2021"
 license = "Apache-2.0"
 
-# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
-
 [dependencies]
-anyhow = "1.0.68"
-chrono = { version = "0.4", default-features = false, features = ["clock", "serde"] }
-rand = "0.8.3"
-serde = "1.0.152"
-serde_with = "2.1.0"
-utils = { version = "0.1.0", path = "../utils" }
-workspace_hack = { version = "0.1.0", path = "../../workspace_hack" }
+anyhow.workspace = true
+chrono.workspace = true
+rand.workspace = true
+serde.workspace = true
+serde_with.workspace = true
+utils.workspace = true
+
+workspace_hack.workspace = true

libs/pageserver_api/Cargo.toml

@@ -7,6 +7,7 @@ license.workspace = true
 [dependencies]
 serde.workspace = true
 serde_with.workspace = true
+serde_json.workspace = true
 const_format.workspace = true
 anyhow.workspace = true
 bytes.workspace = true
@@ -14,5 +15,7 @@ byteorder.workspace = true
 utils.workspace = true
 postgres_ffi.workspace = true
 enum-map.workspace = true
+strum.workspace = true
+strum_macros.workspace = true
 
 workspace_hack.workspace = true

libs/pageserver_api/src/models.rs

@@ -7,6 +7,7 @@ use std::{
 use byteorder::{BigEndian, ReadBytesExt};
 use serde::{Deserialize, Serialize};
 use serde_with::{serde_as, DisplayFromStr};
+use strum_macros;
 use utils::{
     history_buffer::HistoryBufferWithDropCounter,
     id::{NodeId, TenantId, TimelineId},
@@ -18,11 +19,23 @@ use anyhow::bail;
 use bytes::{BufMut, Bytes, BytesMut};
 
 /// A state of a tenant in pageserver's memory.
-#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
+#[derive(
+    Clone,
+    PartialEq,
+    Eq,
+    serde::Serialize,
+    serde::Deserialize,
+    strum_macros::Display,
+    strum_macros::EnumString,
+    strum_macros::EnumVariantNames,
+    strum_macros::AsRefStr,
+    strum_macros::IntoStaticStr,
+)]
+#[serde(tag = "slug", content = "data")]
 pub enum TenantState {
-    // This tenant is being loaded from local disk
+    /// This tenant is being loaded from local disk
     Loading,
-    // This tenant is being downloaded from cloud storage.
+    /// This tenant is being downloaded from cloud storage.
     Attaching,
     /// Tenant is fully operational
     Active,
@@ -31,15 +44,7 @@ pub enum TenantState {
     Stopping,
     /// A tenant is recognized by the pageserver, but can no longer be used for
     /// any operations, because it failed to be activated.
-    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";
+    Broken { reason: String, backtrace: String },
 }
 
 impl TenantState {
@@ -49,17 +54,26 @@ impl TenantState {
             Self::Attaching => true,
             Self::Active => false,
             Self::Stopping => false,
-            Self::Broken => false,
+            Self::Broken { .. } => false,
         }
     }
 
-    pub fn as_str(&self) -> &'static str {
+    pub fn broken_from_reason(reason: String) -> Self {
+        let backtrace_str: String = format!("{}", std::backtrace::Backtrace::force_capture());
+        Self::Broken {
+            reason,
+            backtrace: backtrace_str,
+        }
+    }
+}
+
+impl std::fmt::Debug for TenantState {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         match self {
-            TenantState::Loading => state::LOADING,
-            TenantState::Attaching => state::ATTACHING,
-            TenantState::Active => state::ACTIVE,
-            TenantState::Stopping => state::STOPPING,
-            TenantState::Broken => state::BROKEN,
+            Self::Broken { reason, backtrace } if !reason.is_empty() => {
+                write!(f, "Broken due to: {reason}. Backtrace:\n{backtrace}")
+            }
+            _ => write!(f, "{self}"),
         }
     }
 }
@@ -115,6 +129,13 @@ pub struct TenantCreateRequest {
     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>,
+    pub min_resident_size_override: Option<u64>,
+    pub evictions_low_residence_duration_metric_threshold: Option<String>,
 }
 
 #[serde_as]
@@ -155,6 +176,13 @@ 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>,
+    pub min_resident_size_override: Option<u64>,
+    pub evictions_low_residence_duration_metric_threshold: Option<String>,
 }
 
 impl TenantConfigRequest {
@@ -174,6 +202,9 @@ impl TenantConfigRequest {
             lagging_wal_timeout: None,
             max_lsn_wal_lag: None,
             trace_read_requests: None,
+            eviction_policy: None,
+            min_resident_size_override: None,
+            evictions_low_residence_duration_metric_threshold: None,
         }
     }
 }
@@ -263,11 +294,11 @@ pub struct LayerResidenceEvent {
     ///
     #[serde(rename = "timestamp_millis_since_epoch")]
     #[serde_as(as = "serde_with::TimestampMilliSeconds")]
-    timestamp: SystemTime,
+    pub timestamp: SystemTime,
     /// The new residence status of the layer.
-    status: LayerResidenceStatus,
+    pub status: LayerResidenceStatus,
     /// The reason why we had to record this event.
-    reason: LayerResidenceEventReason,
+    pub reason: LayerResidenceEventReason,
 }
 
 /// The reason for recording a given [`ResidenceEvent`].
@@ -335,7 +366,7 @@ pub enum InMemoryLayerInfo {
 pub enum HistoricLayerInfo {
     Delta {
         layer_file_name: String,
-        layer_file_size: Option<u64>,
+        layer_file_size: u64,
 
         #[serde_as(as = "DisplayFromStr")]
         lsn_start: Lsn,
@@ -346,7 +377,7 @@ pub enum HistoricLayerInfo {
     },
     Image {
         layer_file_name: String,
-        layer_file_size: Option<u64>,
+        layer_file_size: u64,
 
         #[serde_as(as = "DisplayFromStr")]
         lsn_start: Lsn,
@@ -601,6 +632,7 @@ impl PagestreamBeMessage {
 #[cfg(test)]
 mod tests {
     use bytes::Buf;
+    use serde_json::json;
 
     use super::*;
 
@@ -651,4 +683,57 @@ mod tests {
             assert!(msg == reconstructed);
         }
     }
+
+    #[test]
+    fn test_tenantinfo_serde() {
+        // Test serialization/deserialization of TenantInfo
+        let original_active = TenantInfo {
+            id: TenantId::generate(),
+            state: TenantState::Active,
+            current_physical_size: Some(42),
+            has_in_progress_downloads: Some(false),
+        };
+        let expected_active = json!({
+            "id": original_active.id.to_string(),
+            "state": {
+                "slug": "Active",
+            },
+            "current_physical_size": 42,
+            "has_in_progress_downloads": false,
+        });
+
+        let original_broken = TenantInfo {
+            id: TenantId::generate(),
+            state: TenantState::Broken {
+                reason: "reason".into(),
+                backtrace: "backtrace info".into(),
+            },
+            current_physical_size: Some(42),
+            has_in_progress_downloads: Some(false),
+        };
+        let expected_broken = json!({
+            "id": original_broken.id.to_string(),
+            "state": {
+                "slug": "Broken",
+                "data": {
+                    "backtrace": "backtrace info",
+                    "reason": "reason",
+                }
+            },
+            "current_physical_size": 42,
+            "has_in_progress_downloads": false,
+        });
+
+        assert_eq!(
+            serde_json::to_value(&original_active).unwrap(),
+            expected_active
+        );
+
+        assert_eq!(
+            serde_json::to_value(&original_broken).unwrap(),
+            expected_broken
+        );
+        assert!(format!("{:?}", &original_broken.state).contains("reason"));
+        assert!(format!("{:?}", &original_broken.state).contains("backtrace info"));
+    }
 }

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/postgres_backend/Cargo.toml

@@ -0,0 +1,26 @@
+[package]
+name = "postgres_backend"
+version = "0.1.0"
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+async-trait.workspace = true
+anyhow.workspace = true
+bytes.workspace = true
+futures.workspace = true
+rustls.workspace = true
+serde.workspace = true
+thiserror.workspace = true
+tokio.workspace = true
+tokio-rustls.workspace = true
+tracing.workspace = true
+
+pq_proto.workspace = true
+workspace_hack.workspace = true
+
+[dev-dependencies]
+once_cell.workspace = true
+rustls-pemfile.workspace = true
+tokio-postgres.workspace = true
+tokio-postgres-rustls.workspace = true

libs/postgres_backend/src/lib.rs

@@ -0,0 +1,956 @@
+//! Server-side asynchronous Postgres connection, as limited as we need.
+//! To use, create PostgresBackend and run() it, passing the Handler
+//! implementation determining how to process the queries. Currently its API
+//! is rather narrow, but we can extend it once required.
+use anyhow::Context;
+use bytes::Bytes;
+use futures::pin_mut;
+use serde::{Deserialize, Serialize};
+use std::io::ErrorKind;
+use std::net::SocketAddr;
+use std::pin::Pin;
+use std::sync::Arc;
+use std::task::{ready, Poll};
+use std::{fmt, io};
+use std::{future::Future, str::FromStr};
+use tokio::io::{AsyncRead, AsyncWrite};
+use tokio_rustls::TlsAcceptor;
+use tracing::{debug, error, info, trace};
+
+use pq_proto::framed::{ConnectionError, Framed, FramedReader, FramedWriter};
+use pq_proto::{
+    BeMessage, FeMessage, FeStartupPacket, ProtocolError, SQLSTATE_INTERNAL_ERROR,
+    SQLSTATE_SUCCESSFUL_COMPLETION,
+};
+
+/// An error, occurred during query processing:
+/// either during the connection ([`ConnectionError`]) or before/after it.
+#[derive(thiserror::Error, Debug)]
+pub enum QueryError {
+    /// The connection was lost while processing the query.
+    #[error(transparent)]
+    Disconnected(#[from] ConnectionError),
+    /// Some other error
+    #[error(transparent)]
+    Other(#[from] anyhow::Error),
+}
+
+impl From<io::Error> for QueryError {
+    fn from(e: io::Error) -> Self {
+        Self::Disconnected(ConnectionError::Io(e))
+    }
+}
+
+impl QueryError {
+    pub fn pg_error_code(&self) -> &'static [u8; 5] {
+        match self {
+            Self::Disconnected(_) => b"08006",         // connection failure
+            Self::Other(_) => SQLSTATE_INTERNAL_ERROR, // internal error
+        }
+    }
+}
+
+pub fn is_expected_io_error(e: &io::Error) -> bool {
+    use io::ErrorKind::*;
+    matches!(
+        e.kind(),
+        ConnectionRefused | ConnectionAborted | ConnectionReset | TimedOut
+    )
+}
+
+#[async_trait::async_trait]
+pub trait Handler<IO> {
+    /// Handle single query.
+    /// postgres_backend will issue ReadyForQuery after calling this (this
+    /// might be not what we want after CopyData streaming, but currently we don't
+    /// care). It will also flush out the output buffer.
+    async fn process_query(
+        &mut self,
+        pgb: &mut PostgresBackend<IO>,
+        query_string: &str,
+    ) -> Result<(), QueryError>;
+
+    /// Called on startup packet receival, allows to process params.
+    ///
+    /// If Ok(false) is returned postgres_backend will skip auth -- that is needed for new users
+    /// creation is the proxy code. That is quite hacky and ad-hoc solution, may be we could allow
+    /// to override whole init logic in implementations.
+    fn startup(
+        &mut self,
+        _pgb: &mut PostgresBackend<IO>,
+        _sm: &FeStartupPacket,
+    ) -> Result<(), QueryError> {
+        Ok(())
+    }
+
+    /// Check auth jwt
+    fn check_auth_jwt(
+        &mut self,
+        _pgb: &mut PostgresBackend<IO>,
+        _jwt_response: &[u8],
+    ) -> Result<(), QueryError> {
+        Err(QueryError::Other(anyhow::anyhow!("JWT auth failed")))
+    }
+}
+
+/// PostgresBackend protocol state.
+/// XXX: The order of the constructors matters.
+#[derive(Clone, Copy, PartialEq, Eq, PartialOrd)]
+pub enum ProtoState {
+    /// Nothing happened yet.
+    Initialization,
+    /// Encryption handshake is done; waiting for encrypted Startup message.
+    Encrypted,
+    /// Waiting for password (auth token).
+    Authentication,
+    /// Performed handshake and auth, ReadyForQuery is issued.
+    Established,
+    Closed,
+}
+
+#[derive(Clone, Copy)]
+pub enum ProcessMsgResult {
+    Continue,
+    Break,
+}
+
+/// Either plain TCP stream or encrypted one, implementing AsyncRead + AsyncWrite.
+pub enum MaybeTlsStream<IO> {
+    Unencrypted(IO),
+    Tls(Box<tokio_rustls::server::TlsStream<IO>>),
+}
+
+impl<IO: AsyncRead + AsyncWrite + Unpin> AsyncWrite for MaybeTlsStream<IO> {
+    fn poll_write(
+        self: Pin<&mut Self>,
+        cx: &mut std::task::Context<'_>,
+        buf: &[u8],
+    ) -> Poll<io::Result<usize>> {
+        match self.get_mut() {
+            Self::Unencrypted(stream) => Pin::new(stream).poll_write(cx, buf),
+            Self::Tls(stream) => Pin::new(stream).poll_write(cx, buf),
+        }
+    }
+    fn poll_flush(self: Pin<&mut Self>, cx: &mut std::task::Context<'_>) -> Poll<io::Result<()>> {
+        match self.get_mut() {
+            Self::Unencrypted(stream) => Pin::new(stream).poll_flush(cx),
+            Self::Tls(stream) => Pin::new(stream).poll_flush(cx),
+        }
+    }
+    fn poll_shutdown(
+        self: Pin<&mut Self>,
+        cx: &mut std::task::Context<'_>,
+    ) -> Poll<io::Result<()>> {
+        match self.get_mut() {
+            Self::Unencrypted(stream) => Pin::new(stream).poll_shutdown(cx),
+            Self::Tls(stream) => Pin::new(stream).poll_shutdown(cx),
+        }
+    }
+}
+impl<IO: AsyncRead + AsyncWrite + Unpin> AsyncRead for MaybeTlsStream<IO> {
+    fn poll_read(
+        self: Pin<&mut Self>,
+        cx: &mut std::task::Context<'_>,
+        buf: &mut tokio::io::ReadBuf<'_>,
+    ) -> Poll<io::Result<()>> {
+        match self.get_mut() {
+            Self::Unencrypted(stream) => Pin::new(stream).poll_read(cx, buf),
+            Self::Tls(stream) => Pin::new(stream).poll_read(cx, buf),
+        }
+    }
+}
+
+#[derive(Debug, PartialEq, Eq, Clone, Copy, Serialize, Deserialize)]
+pub enum AuthType {
+    Trust,
+    // This mimics postgres's AuthenticationCleartextPassword but instead of password expects JWT
+    NeonJWT,
+}
+
+impl FromStr for AuthType {
+    type Err = anyhow::Error;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        match s {
+            "Trust" => Ok(Self::Trust),
+            "NeonJWT" => Ok(Self::NeonJWT),
+            _ => anyhow::bail!("invalid value \"{s}\" for auth type"),
+        }
+    }
+}
+
+impl fmt::Display for AuthType {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        f.write_str(match self {
+            AuthType::Trust => "Trust",
+            AuthType::NeonJWT => "NeonJWT",
+        })
+    }
+}
+
+/// Either full duplex Framed or write only half; the latter is left in
+/// PostgresBackend after call to `split`. In principle we could always store a
+/// pair of splitted handles, but that would force to to pay splitting price
+/// (Arc and kinda mutex inside polling) for all uses (e.g. pageserver).
+enum MaybeWriteOnly<IO> {
+    Full(Framed<MaybeTlsStream<IO>>),
+    WriteOnly(FramedWriter<MaybeTlsStream<IO>>),
+    Broken, // temporary value palmed off during the split
+}
+
+impl<IO: AsyncRead + AsyncWrite + Unpin> MaybeWriteOnly<IO> {
+    async fn read_startup_message(&mut self) -> Result<Option<FeStartupPacket>, ConnectionError> {
+        match self {
+            MaybeWriteOnly::Full(framed) => framed.read_startup_message().await,
+            MaybeWriteOnly::WriteOnly(_) => {
+                Err(io::Error::new(ErrorKind::Other, "reading from write only half").into())
+            }
+            MaybeWriteOnly::Broken => panic!("IO on invalid MaybeWriteOnly"),
+        }
+    }
+
+    async fn read_message(&mut self) -> Result<Option<FeMessage>, ConnectionError> {
+        match self {
+            MaybeWriteOnly::Full(framed) => framed.read_message().await,
+            MaybeWriteOnly::WriteOnly(_) => {
+                Err(io::Error::new(ErrorKind::Other, "reading from write only half").into())
+            }
+            MaybeWriteOnly::Broken => panic!("IO on invalid MaybeWriteOnly"),
+        }
+    }
+
+    fn write_message_noflush(&mut self, msg: &BeMessage<'_>) -> Result<(), ProtocolError> {
+        match self {
+            MaybeWriteOnly::Full(framed) => framed.write_message(msg),
+            MaybeWriteOnly::WriteOnly(framed_writer) => framed_writer.write_message_noflush(msg),
+            MaybeWriteOnly::Broken => panic!("IO on invalid MaybeWriteOnly"),
+        }
+    }
+
+    async fn flush(&mut self) -> io::Result<()> {
+        match self {
+            MaybeWriteOnly::Full(framed) => framed.flush().await,
+            MaybeWriteOnly::WriteOnly(framed_writer) => framed_writer.flush().await,
+            MaybeWriteOnly::Broken => panic!("IO on invalid MaybeWriteOnly"),
+        }
+    }
+
+    async fn shutdown(&mut self) -> io::Result<()> {
+        match self {
+            MaybeWriteOnly::Full(framed) => framed.shutdown().await,
+            MaybeWriteOnly::WriteOnly(framed_writer) => framed_writer.shutdown().await,
+            MaybeWriteOnly::Broken => panic!("IO on invalid MaybeWriteOnly"),
+        }
+    }
+}
+
+pub struct PostgresBackend<IO> {
+    framed: MaybeWriteOnly<IO>,
+
+    pub state: ProtoState,
+
+    auth_type: AuthType,
+
+    peer_addr: SocketAddr,
+    pub tls_config: Option<Arc<rustls::ServerConfig>>,
+}
+
+pub type PostgresBackendTCP = PostgresBackend<tokio::net::TcpStream>;
+
+pub fn query_from_cstring(query_string: Bytes) -> Vec<u8> {
+    let mut query_string = query_string.to_vec();
+    if let Some(ch) = query_string.last() {
+        if *ch == 0 {
+            query_string.pop();
+        }
+    }
+    query_string
+}
+
+/// Cast a byte slice to a string slice, dropping null terminator if there's one.
+fn cstr_to_str(bytes: &[u8]) -> anyhow::Result<&str> {
+    let without_null = bytes.strip_suffix(&[0]).unwrap_or(bytes);
+    std::str::from_utf8(without_null).map_err(|e| e.into())
+}
+
+impl PostgresBackend<tokio::net::TcpStream> {
+    pub fn new(
+        socket: tokio::net::TcpStream,
+        auth_type: AuthType,
+        tls_config: Option<Arc<rustls::ServerConfig>>,
+    ) -> io::Result<Self> {
+        let peer_addr = socket.peer_addr()?;
+        let stream = MaybeTlsStream::Unencrypted(socket);
+
+        Ok(Self {
+            framed: MaybeWriteOnly::Full(Framed::new(stream)),
+            state: ProtoState::Initialization,
+            auth_type,
+            tls_config,
+            peer_addr,
+        })
+    }
+}
+
+impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
+    pub fn new_from_io(
+        socket: IO,
+        peer_addr: SocketAddr,
+        auth_type: AuthType,
+        tls_config: Option<Arc<rustls::ServerConfig>>,
+    ) -> io::Result<Self> {
+        let stream = MaybeTlsStream::Unencrypted(socket);
+
+        Ok(Self {
+            framed: MaybeWriteOnly::Full(Framed::new(stream)),
+            state: ProtoState::Initialization,
+            auth_type,
+            tls_config,
+            peer_addr,
+        })
+    }
+
+    pub fn get_peer_addr(&self) -> &SocketAddr {
+        &self.peer_addr
+    }
+
+    /// Read full message or return None if connection is cleanly closed with no
+    /// unprocessed data.
+    pub async fn read_message(&mut self) -> Result<Option<FeMessage>, ConnectionError> {
+        if let ProtoState::Closed = self.state {
+            Ok(None)
+        } else {
+            match self.framed.read_message().await {
+                Ok(m) => {
+                    trace!("read msg {:?}", m);
+                    Ok(m)
+                }
+                Err(e) => {
+                    // remember not to try to read anymore
+                    self.state = ProtoState::Closed;
+                    Err(e)
+                }
+            }
+        }
+    }
+
+    /// Write message into internal output buffer, doesn't flush it. Technically
+    /// error type can be only ProtocolError here (if, unlikely, serialization
+    /// fails), but callers typically wrap it anyway.
+    pub fn write_message_noflush(
+        &mut self,
+        message: &BeMessage<'_>,
+    ) -> Result<&mut Self, ConnectionError> {
+        self.framed.write_message_noflush(message)?;
+        trace!("wrote msg {:?}", message);
+        Ok(self)
+    }
+
+    /// Flush output buffer into the socket.
+    pub async fn flush(&mut self) -> io::Result<()> {
+        self.framed.flush().await
+    }
+
+    /// Polling version of `flush()`, saves the caller need to pin.
+    pub fn poll_flush(
+        &mut self,
+        cx: &mut std::task::Context<'_>,
+    ) -> Poll<Result<(), std::io::Error>> {
+        let flush_fut = self.flush();
+        pin_mut!(flush_fut);
+        flush_fut.poll(cx)
+    }
+
+    /// Write message into internal output buffer and flush it to the stream.
+    pub async fn write_message(
+        &mut self,
+        message: &BeMessage<'_>,
+    ) -> Result<&mut Self, ConnectionError> {
+        self.write_message_noflush(message)?;
+        self.flush().await?;
+        Ok(self)
+    }
+
+    /// Returns an AsyncWrite implementation that wraps all the data written
+    /// to it in CopyData messages, and writes them to the connection
+    ///
+    /// The caller is responsible for sending CopyOutResponse and CopyDone messages.
+    pub fn copyout_writer(&mut self) -> CopyDataWriter<IO> {
+        CopyDataWriter { pgb: self }
+    }
+
+    /// Wrapper for run_message_loop() that shuts down socket when we are done
+    pub async fn run<F, S>(
+        mut self,
+        handler: &mut impl Handler<IO>,
+        shutdown_watcher: F,
+    ) -> Result<(), QueryError>
+    where
+        F: Fn() -> S,
+        S: Future,
+    {
+        let ret = self.run_message_loop(handler, shutdown_watcher).await;
+        // socket might be already closed, e.g. if previously received error,
+        // so ignore result.
+        self.framed.shutdown().await.ok();
+        ret
+    }
+
+    async fn run_message_loop<F, S>(
+        &mut self,
+        handler: &mut impl Handler<IO>,
+        shutdown_watcher: F,
+    ) -> Result<(), QueryError>
+    where
+        F: Fn() -> S,
+        S: Future,
+    {
+        trace!("postgres backend to {:?} started", self.peer_addr);
+
+        tokio::select!(
+            biased;
+
+            _ = shutdown_watcher() => {
+                // We were requested to shut down.
+                tracing::info!("shutdown request received during handshake");
+                return Ok(())
+            },
+
+            result = self.handshake(handler) => {
+                // Handshake complete.
+                result?;
+                if self.state == ProtoState::Closed {
+                    return Ok(()); // EOF during handshake
+                }
+            }
+        );
+
+        // Authentication completed
+        let mut query_string = Bytes::new();
+        while let Some(msg) = tokio::select!(
+            biased;
+            _ = shutdown_watcher() => {
+                // We were requested to shut down.
+                tracing::info!("shutdown request received in run_message_loop");
+                Ok(None)
+            },
+            msg = self.read_message() => { msg },
+        )? {
+            trace!("got message {:?}", msg);
+
+            let result = self.process_message(handler, msg, &mut query_string).await;
+            self.flush().await?;
+            match result? {
+                ProcessMsgResult::Continue => {
+                    self.flush().await?;
+                    continue;
+                }
+                ProcessMsgResult::Break => break,
+            }
+        }
+
+        trace!("postgres backend to {:?} exited", self.peer_addr);
+        Ok(())
+    }
+
+    /// Try to upgrade MaybeTlsStream into actual TLS one, performing handshake.
+    async fn tls_upgrade(
+        src: MaybeTlsStream<IO>,
+        tls_config: Arc<rustls::ServerConfig>,
+    ) -> anyhow::Result<MaybeTlsStream<IO>> {
+        match src {
+            MaybeTlsStream::Unencrypted(s) => {
+                let acceptor = TlsAcceptor::from(tls_config);
+                let tls_stream = acceptor.accept(s).await?;
+                Ok(MaybeTlsStream::Tls(Box::new(tls_stream)))
+            }
+            MaybeTlsStream::Tls(_) => {
+                anyhow::bail!("TLS already started");
+            }
+        }
+    }
+
+    async fn start_tls(&mut self) -> anyhow::Result<()> {
+        // temporary replace stream with fake to cook TLS one, Indiana Jones style
+        match std::mem::replace(&mut self.framed, MaybeWriteOnly::Broken) {
+            MaybeWriteOnly::Full(framed) => {
+                let tls_config = self
+                    .tls_config
+                    .as_ref()
+                    .context("start_tls called without conf")?
+                    .clone();
+                let tls_framed = framed
+                    .map_stream(|s| PostgresBackend::tls_upgrade(s, tls_config))
+                    .await?;
+                // push back ready TLS stream
+                self.framed = MaybeWriteOnly::Full(tls_framed);
+                Ok(())
+            }
+            MaybeWriteOnly::WriteOnly(_) => {
+                anyhow::bail!("TLS upgrade attempt in split state")
+            }
+            MaybeWriteOnly::Broken => panic!("TLS upgrade on framed in invalid state"),
+        }
+    }
+
+    /// Split off owned read part from which messages can be read in different
+    /// task/thread.
+    pub fn split(&mut self) -> anyhow::Result<PostgresBackendReader<IO>> {
+        // temporary replace stream with fake to cook split one, Indiana Jones style
+        match std::mem::replace(&mut self.framed, MaybeWriteOnly::Broken) {
+            MaybeWriteOnly::Full(framed) => {
+                let (reader, writer) = framed.split();
+                self.framed = MaybeWriteOnly::WriteOnly(writer);
+                Ok(PostgresBackendReader {
+                    reader,
+                    closed: false,
+                })
+            }
+            MaybeWriteOnly::WriteOnly(_) => {
+                anyhow::bail!("PostgresBackend is already split")
+            }
+            MaybeWriteOnly::Broken => panic!("split on framed in invalid state"),
+        }
+    }
+
+    /// Join read part back.
+    pub fn unsplit(&mut self, reader: PostgresBackendReader<IO>) -> anyhow::Result<()> {
+        // temporary replace stream with fake to cook joined one, Indiana Jones style
+        match std::mem::replace(&mut self.framed, MaybeWriteOnly::Broken) {
+            MaybeWriteOnly::Full(_) => {
+                anyhow::bail!("PostgresBackend is not split")
+            }
+            MaybeWriteOnly::WriteOnly(writer) => {
+                let joined = Framed::unsplit(reader.reader, writer);
+                self.framed = MaybeWriteOnly::Full(joined);
+                // if reader encountered connection error, do not attempt reading anymore
+                if reader.closed {
+                    self.state = ProtoState::Closed;
+                }
+                Ok(())
+            }
+            MaybeWriteOnly::Broken => panic!("unsplit on framed in invalid state"),
+        }
+    }
+
+    /// Perform handshake with the client, transitioning to Established.
+    /// In case of EOF during handshake logs this, sets state to Closed and returns Ok(()).
+    async fn handshake(&mut self, handler: &mut impl Handler<IO>) -> Result<(), QueryError> {
+        while self.state < ProtoState::Authentication {
+            match self.framed.read_startup_message().await? {
+                Some(msg) => {
+                    self.process_startup_message(handler, msg).await?;
+                }
+                None => {
+                    trace!(
+                        "postgres backend to {:?} received EOF during handshake",
+                        self.peer_addr
+                    );
+                    self.state = ProtoState::Closed;
+                    return Ok(());
+                }
+            }
+        }
+
+        // Perform auth, if needed.
+        if self.state == ProtoState::Authentication {
+            match self.framed.read_message().await? {
+                Some(FeMessage::PasswordMessage(m)) => {
+                    assert!(self.auth_type == AuthType::NeonJWT);
+
+                    let (_, jwt_response) = m.split_last().context("protocol violation")?;
+
+                    if let Err(e) = handler.check_auth_jwt(self, jwt_response) {
+                        self.write_message_noflush(&BeMessage::ErrorResponse(
+                            &e.to_string(),
+                            Some(e.pg_error_code()),
+                        ))?;
+                        return Err(e);
+                    }
+
+                    self.write_message_noflush(&BeMessage::AuthenticationOk)?
+                        .write_message_noflush(&BeMessage::CLIENT_ENCODING)?
+                        .write_message(&BeMessage::ReadyForQuery)
+                        .await?;
+                    self.state = ProtoState::Established;
+                }
+                Some(m) => {
+                    return Err(QueryError::Other(anyhow::anyhow!(
+                        "Unexpected message {:?} while waiting for handshake",
+                        m
+                    )));
+                }
+                None => {
+                    trace!(
+                        "postgres backend to {:?} received EOF during auth",
+                        self.peer_addr
+                    );
+                    self.state = ProtoState::Closed;
+                    return Ok(());
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    /// Process startup packet:
+    /// - transition to Established if auth type is trust
+    /// - transition to Authentication if auth type is NeonJWT.
+    /// - or perform TLS handshake -- then need to call this again to receive
+    ///   actual startup packet.
+    async fn process_startup_message(
+        &mut self,
+        handler: &mut impl Handler<IO>,
+        msg: FeStartupPacket,
+    ) -> Result<(), QueryError> {
+        assert!(self.state < ProtoState::Authentication);
+        let have_tls = self.tls_config.is_some();
+        match msg {
+            FeStartupPacket::SslRequest => {
+                debug!("SSL requested");
+
+                self.write_message(&BeMessage::EncryptionResponse(have_tls))
+                    .await?;
+
+                if have_tls {
+                    self.start_tls().await?;
+                    self.state = ProtoState::Encrypted;
+                }
+            }
+            FeStartupPacket::GssEncRequest => {
+                debug!("GSS requested");
+                self.write_message(&BeMessage::EncryptionResponse(false))
+                    .await?;
+            }
+            FeStartupPacket::StartupMessage { .. } => {
+                if have_tls && !matches!(self.state, ProtoState::Encrypted) {
+                    self.write_message(&BeMessage::ErrorResponse("must connect with TLS", None))
+                        .await?;
+                    return Err(QueryError::Other(anyhow::anyhow!(
+                        "client did not connect with TLS"
+                    )));
+                }
+
+                // NB: startup() may change self.auth_type -- we are using that in proxy code
+                // to bypass auth for new users.
+                handler.startup(self, &msg)?;
+
+                match self.auth_type {
+                    AuthType::Trust => {
+                        self.write_message_noflush(&BeMessage::AuthenticationOk)?
+                            .write_message_noflush(&BeMessage::CLIENT_ENCODING)?
+                            .write_message_noflush(&BeMessage::INTEGER_DATETIMES)?
+                            // The async python driver requires a valid server_version
+                            .write_message_noflush(&BeMessage::server_version("14.1"))?
+                            .write_message(&BeMessage::ReadyForQuery)
+                            .await?;
+                        self.state = ProtoState::Established;
+                    }
+                    AuthType::NeonJWT => {
+                        self.write_message(&BeMessage::AuthenticationCleartextPassword)
+                            .await?;
+                        self.state = ProtoState::Authentication;
+                    }
+                }
+            }
+            FeStartupPacket::CancelRequest { .. } => {
+                return Err(QueryError::Other(anyhow::anyhow!(
+                    "Unexpected CancelRequest message during handshake"
+                )));
+            }
+        }
+        Ok(())
+    }
+
+    async fn process_message(
+        &mut self,
+        handler: &mut impl Handler<IO>,
+        msg: FeMessage,
+        unnamed_query_string: &mut Bytes,
+    ) -> Result<ProcessMsgResult, QueryError> {
+        // Allow only startup and password messages during auth. Otherwise client would be able to bypass auth
+        // TODO: change that to proper top-level match of protocol state with separate message handling for each state
+        assert!(self.state == ProtoState::Established);
+
+        match msg {
+            FeMessage::Query(body) => {
+                // remove null terminator
+                let query_string = cstr_to_str(&body)?;
+
+                trace!("got query {query_string:?}");
+                if let Err(e) = handler.process_query(self, query_string).await {
+                    log_query_error(query_string, &e);
+                    let short_error = short_error(&e);
+                    self.write_message_noflush(&BeMessage::ErrorResponse(
+                        &short_error,
+                        Some(e.pg_error_code()),
+                    ))?;
+                }
+                self.write_message_noflush(&BeMessage::ReadyForQuery)?;
+            }
+
+            FeMessage::Parse(m) => {
+                *unnamed_query_string = m.query_string;
+                self.write_message_noflush(&BeMessage::ParseComplete)?;
+            }
+
+            FeMessage::Describe(_) => {
+                self.write_message_noflush(&BeMessage::ParameterDescription)?
+                    .write_message_noflush(&BeMessage::NoData)?;
+            }
+
+            FeMessage::Bind(_) => {
+                self.write_message_noflush(&BeMessage::BindComplete)?;
+            }
+
+            FeMessage::Close(_) => {
+                self.write_message_noflush(&BeMessage::CloseComplete)?;
+            }
+
+            FeMessage::Execute(_) => {
+                let query_string = cstr_to_str(unnamed_query_string)?;
+                trace!("got execute {query_string:?}");
+                if let Err(e) = handler.process_query(self, query_string).await {
+                    log_query_error(query_string, &e);
+                    self.write_message_noflush(&BeMessage::ErrorResponse(
+                        &e.to_string(),
+                        Some(e.pg_error_code()),
+                    ))?;
+                }
+                // NOTE there is no ReadyForQuery message. This handler is used
+                // for basebackup and it uses CopyOut which doesn't require
+                // ReadyForQuery message and backend just switches back to
+                // processing mode after sending CopyDone or ErrorResponse.
+            }
+
+            FeMessage::Sync => {
+                self.write_message_noflush(&BeMessage::ReadyForQuery)?;
+            }
+
+            FeMessage::Terminate => {
+                return Ok(ProcessMsgResult::Break);
+            }
+
+            // We prefer explicit pattern matching to wildcards, because
+            // this helps us spot the places where new variants are missing
+            FeMessage::CopyData(_)
+            | FeMessage::CopyDone
+            | FeMessage::CopyFail
+            | FeMessage::PasswordMessage(_) => {
+                return Err(QueryError::Other(anyhow::anyhow!(
+                    "unexpected message type: {msg:?}",
+                )));
+            }
+        }
+
+        Ok(ProcessMsgResult::Continue)
+    }
+
+    /// Log as info/error result of handling COPY stream and send back
+    /// ErrorResponse if that makes sense. Shutdown the stream if we got
+    /// Terminate. TODO: transition into waiting for Sync msg if we initiate the
+    /// close.
+    pub async fn handle_copy_stream_end(&mut self, end: CopyStreamHandlerEnd) {
+        use CopyStreamHandlerEnd::*;
+
+        let expected_end = match &end {
+            ServerInitiated(_) | CopyDone | CopyFail | Terminate | EOF => true,
+            CopyStreamHandlerEnd::Disconnected(ConnectionError::Io(io_error))
+                if is_expected_io_error(io_error) =>
+            {
+                true
+            }
+            _ => false,
+        };
+        if expected_end {
+            info!("terminated: {:#}", end);
+        } else {
+            error!("terminated: {:?}", end);
+        }
+
+        // Note: no current usages ever send this
+        if let CopyDone = &end {
+            if let Err(e) = self.write_message(&BeMessage::CopyDone).await {
+                error!("failed to send CopyDone: {}", e);
+            }
+        }
+
+        if let Terminate = &end {
+            self.state = ProtoState::Closed;
+        }
+
+        let err_to_send_and_errcode = match &end {
+            ServerInitiated(_) => Some((end.to_string(), SQLSTATE_SUCCESSFUL_COMPLETION)),
+            Other(_) => Some((format!("{end:#}"), SQLSTATE_INTERNAL_ERROR)),
+            // Note: CopyFail in duplex copy is somewhat unexpected (at least to
+            // PG walsender; evidently and per my docs reading client should
+            // finish it with CopyDone). It is not a problem to recover from it
+            // finishing the stream in both directions like we do, but note that
+            // sync rust-postgres client (which we don't use anymore) hangs if
+            // socket is not closed here.
+            // https://github.com/sfackler/rust-postgres/issues/755
+            // https://github.com/neondatabase/neon/issues/935
+            //
+            // Currently, the version of tokio_postgres replication patch we use
+            // sends this when it closes the stream (e.g. pageserver decided to
+            // switch conn to another safekeeper and client gets dropped).
+            // Moreover, seems like 'connection' task errors with 'unexpected
+            // message from server' when it receives ErrorResponse (anything but
+            // CopyData/CopyDone) back.
+            CopyFail => Some((end.to_string(), SQLSTATE_SUCCESSFUL_COMPLETION)),
+            _ => None,
+        };
+        if let Some((err, errcode)) = err_to_send_and_errcode {
+            if let Err(ee) = self
+                .write_message(&BeMessage::ErrorResponse(&err, Some(errcode)))
+                .await
+            {
+                error!("failed to send ErrorResponse: {}", ee);
+            }
+        }
+    }
+}
+
+pub struct PostgresBackendReader<IO> {
+    reader: FramedReader<MaybeTlsStream<IO>>,
+    closed: bool, // true if received error closing the connection
+}
+
+impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackendReader<IO> {
+    /// Read full message or return None if connection is cleanly closed with no
+    /// unprocessed data.
+    pub async fn read_message(&mut self) -> Result<Option<FeMessage>, ConnectionError> {
+        match self.reader.read_message().await {
+            Ok(m) => {
+                trace!("read msg {:?}", m);
+                Ok(m)
+            }
+            Err(e) => {
+                self.closed = true;
+                Err(e)
+            }
+        }
+    }
+
+    /// Get CopyData contents of the next message in COPY stream or error
+    /// closing it. The error type is wider than actual errors which can happen
+    /// here -- it includes 'Other' and 'ServerInitiated', but that's ok for
+    /// current callers.
+    pub async fn read_copy_message(&mut self) -> Result<Bytes, CopyStreamHandlerEnd> {
+        match self.read_message().await? {
+            Some(msg) => match msg {
+                FeMessage::CopyData(m) => Ok(m),
+                FeMessage::CopyDone => Err(CopyStreamHandlerEnd::CopyDone),
+                FeMessage::CopyFail => Err(CopyStreamHandlerEnd::CopyFail),
+                FeMessage::Terminate => Err(CopyStreamHandlerEnd::Terminate),
+                _ => Err(CopyStreamHandlerEnd::from(ConnectionError::Protocol(
+                    ProtocolError::Protocol(format!("unexpected message in COPY stream {:?}", msg)),
+                ))),
+            },
+            None => Err(CopyStreamHandlerEnd::EOF),
+        }
+    }
+}
+
+///
+/// A futures::AsyncWrite implementation that wraps all data written to it in CopyData
+/// messages.
+///
+
+pub struct CopyDataWriter<'a, IO> {
+    pgb: &'a mut PostgresBackend<IO>,
+}
+
+impl<'a, IO: AsyncRead + AsyncWrite + Unpin> AsyncWrite for CopyDataWriter<'a, IO> {
+    fn poll_write(
+        self: Pin<&mut Self>,
+        cx: &mut std::task::Context<'_>,
+        buf: &[u8],
+    ) -> Poll<Result<usize, std::io::Error>> {
+        let this = self.get_mut();
+
+        // It's not strictly required to flush between each message, but makes it easier
+        // to view in wireshark, and usually the messages that the callers write are
+        // decently-sized anyway.
+        if let Err(err) = ready!(this.pgb.poll_flush(cx)) {
+            return Poll::Ready(Err(err));
+        }
+
+        // CopyData
+        // XXX: if the input is large, we should split it into multiple messages.
+        // Not sure what the threshold should be, but the ultimate hard limit is that
+        // the length cannot exceed u32.
+        this.pgb
+            .write_message_noflush(&BeMessage::CopyData(buf))
+            // write_message only writes to the buffer, so it can fail iff the
+            // message is invaid, but CopyData can't be invalid.
+            .map_err(|_| io::Error::new(ErrorKind::Other, "failed to serialize CopyData"))?;
+
+        Poll::Ready(Ok(buf.len()))
+    }
+
+    fn poll_flush(
+        self: Pin<&mut Self>,
+        cx: &mut std::task::Context<'_>,
+    ) -> Poll<Result<(), std::io::Error>> {
+        let this = self.get_mut();
+        this.pgb.poll_flush(cx)
+    }
+
+    fn poll_shutdown(
+        self: Pin<&mut Self>,
+        cx: &mut std::task::Context<'_>,
+    ) -> Poll<Result<(), std::io::Error>> {
+        let this = self.get_mut();
+        this.pgb.poll_flush(cx)
+    }
+}
+
+pub fn short_error(e: &QueryError) -> String {
+    match e {
+        QueryError::Disconnected(connection_error) => connection_error.to_string(),
+        QueryError::Other(e) => format!("{e:#}"),
+    }
+}
+
+fn log_query_error(query: &str, e: &QueryError) {
+    match e {
+        QueryError::Disconnected(ConnectionError::Io(io_error)) => {
+            if is_expected_io_error(io_error) {
+                info!("query handler for '{query}' failed with expected io error: {io_error}");
+            } else {
+                error!("query handler for '{query}' failed with io error: {io_error}");
+            }
+        }
+        QueryError::Disconnected(other_connection_error) => {
+            error!("query handler for '{query}' failed with connection error: {other_connection_error:?}")
+        }
+        QueryError::Other(e) => {
+            error!("query handler for '{query}' failed: {e:?}");
+        }
+    }
+}
+
+/// Something finishing handling of COPY stream, see handle_copy_stream_end.
+/// This is not always a real error, but it allows to use ? and thiserror impls.
+#[derive(thiserror::Error, Debug)]
+pub enum CopyStreamHandlerEnd {
+    /// Handler initiates the end of streaming.
+    #[error("{0}")]
+    ServerInitiated(String),
+    #[error("received CopyDone")]
+    CopyDone,
+    #[error("received CopyFail")]
+    CopyFail,
+    #[error("received Terminate")]
+    Terminate,
+    #[error("EOF on COPY stream")]
+    EOF,
+    /// The connection was lost
+    #[error("connection error: {0}")]
+    Disconnected(#[from] ConnectionError),
+    /// Some other error
+    #[error(transparent)]
+    Other(#[from] anyhow::Error),
+}

libs/postgres_backend/tests/simple_select.rs

@@ -0,0 +1,140 @@
+/// Test postgres_backend_async with tokio_postgres
+use once_cell::sync::Lazy;
+use postgres_backend::{AuthType, Handler, PostgresBackend, QueryError};
+use pq_proto::{BeMessage, RowDescriptor};
+use std::io::Cursor;
+use std::{future, sync::Arc};
+use tokio::io::{AsyncRead, AsyncWrite};
+use tokio::net::{TcpListener, TcpStream};
+use tokio_postgres::config::SslMode;
+use tokio_postgres::tls::MakeTlsConnect;
+use tokio_postgres::{Config, NoTls, SimpleQueryMessage};
+use tokio_postgres_rustls::MakeRustlsConnect;
+
+// generate client, server test streams
+async fn make_tcp_pair() -> (TcpStream, TcpStream) {
+    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+    let addr = listener.local_addr().unwrap();
+    let client_stream = TcpStream::connect(addr).await.unwrap();
+    let (server_stream, _) = listener.accept().await.unwrap();
+    (client_stream, server_stream)
+}
+
+struct TestHandler {}
+
+#[async_trait::async_trait]
+impl<IO: AsyncRead + AsyncWrite + Unpin + Send> Handler<IO> for TestHandler {
+    // return single col 'hey' for any query
+    async fn process_query(
+        &mut self,
+        pgb: &mut PostgresBackend<IO>,
+        _query_string: &str,
+    ) -> Result<(), QueryError> {
+        pgb.write_message_noflush(&BeMessage::RowDescription(&[RowDescriptor::text_col(
+            b"hey",
+        )]))?
+        .write_message_noflush(&BeMessage::DataRow(&[Some("hey".as_bytes())]))?
+        .write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
+        Ok(())
+    }
+}
+
+// test that basic select works
+#[tokio::test]
+async fn simple_select() {
+    let (client_sock, server_sock) = make_tcp_pair().await;
+
+    // create and run pgbackend
+    let pgbackend =
+        PostgresBackend::new(server_sock, AuthType::Trust, None).expect("pgbackend creation");
+
+    tokio::spawn(async move {
+        let mut handler = TestHandler {};
+        pgbackend.run(&mut handler, future::pending::<()>).await
+    });
+
+    let conf = Config::new();
+    let (client, connection) = conf.connect_raw(client_sock, NoTls).await.expect("connect");
+    // The connection object performs the actual communication with the database,
+    // so spawn it off to run on its own.
+    tokio::spawn(async move {
+        if let Err(e) = connection.await {
+            eprintln!("connection error: {}", e);
+        }
+    });
+
+    let first_val = &(client.simple_query("SELECT 42;").await.expect("select"))[0];
+    if let SimpleQueryMessage::Row(row) = first_val {
+        let first_col = row.get(0).expect("first column");
+        assert_eq!(first_col, "hey");
+    } else {
+        panic!("expected SimpleQueryMessage::Row");
+    }
+}
+
+static KEY: Lazy<rustls::PrivateKey> = Lazy::new(|| {
+    let mut cursor = Cursor::new(include_bytes!("key.pem"));
+    rustls::PrivateKey(rustls_pemfile::rsa_private_keys(&mut cursor).unwrap()[0].clone())
+});
+
+static CERT: Lazy<rustls::Certificate> = Lazy::new(|| {
+    let mut cursor = Cursor::new(include_bytes!("cert.pem"));
+    rustls::Certificate(rustls_pemfile::certs(&mut cursor).unwrap()[0].clone())
+});
+
+// test that basic select with ssl works
+#[tokio::test]
+async fn simple_select_ssl() {
+    let (client_sock, server_sock) = make_tcp_pair().await;
+
+    let server_cfg = rustls::ServerConfig::builder()
+        .with_safe_defaults()
+        .with_no_client_auth()
+        .with_single_cert(vec![CERT.clone()], KEY.clone())
+        .unwrap();
+    let tls_config = Some(Arc::new(server_cfg));
+    let pgbackend =
+        PostgresBackend::new(server_sock, AuthType::Trust, tls_config).expect("pgbackend creation");
+
+    tokio::spawn(async move {
+        let mut handler = TestHandler {};
+        pgbackend.run(&mut handler, future::pending::<()>).await
+    });
+
+    let client_cfg = rustls::ClientConfig::builder()
+        .with_safe_defaults()
+        .with_root_certificates({
+            let mut store = rustls::RootCertStore::empty();
+            store.add(&CERT).unwrap();
+            store
+        })
+        .with_no_client_auth();
+    let mut make_tls_connect = tokio_postgres_rustls::MakeRustlsConnect::new(client_cfg);
+    let tls_connect = <MakeRustlsConnect as MakeTlsConnect<TcpStream>>::make_tls_connect(
+        &mut make_tls_connect,
+        "localhost",
+    )
+    .expect("make_tls_connect");
+
+    let mut conf = Config::new();
+    conf.ssl_mode(SslMode::Require);
+    let (client, connection) = conf
+        .connect_raw(client_sock, tls_connect)
+        .await
+        .expect("connect");
+    // The connection object performs the actual communication with the database,
+    // so spawn it off to run on its own.
+    tokio::spawn(async move {
+        if let Err(e) = connection.await {
+            eprintln!("connection error: {}", e);
+        }
+    });
+
+    let first_val = &(client.simple_query("SELECT 42;").await.expect("select"))[0];
+    if let SimpleQueryMessage::Row(row) = first_val {
+        let first_col = row.get(0).expect("first column");
+        assert_eq!(first_col, "hey");
+    } else {
+        panic!("expected SimpleQueryMessage::Row");
+    }
+}

libs/postgres_ffi/build.rs

@@ -5,7 +5,7 @@ use std::path::PathBuf;
 use std::process::Command;
 
 use anyhow::{anyhow, Context};
-use bindgen::callbacks::ParseCallbacks;
+use bindgen::callbacks::{DeriveInfo, ParseCallbacks};
 
 #[derive(Debug)]
 struct PostgresFfiCallbacks;
@@ -20,7 +20,7 @@ impl ParseCallbacks for PostgresFfiCallbacks {
 
     // Add any custom #[derive] attributes to the data structures that bindgen
     // creates.
-    fn add_derives(&self, name: &str) -> Vec<String> {
+    fn add_derives(&self, derive_info: &DeriveInfo) -> Vec<String> {
         // This is the list of data structures that we want to serialize/deserialize.
         let serde_list = [
             "XLogRecord",
@@ -31,7 +31,7 @@ impl ParseCallbacks for PostgresFfiCallbacks {
             "ControlFileData",
         ];
 
-        if serde_list.contains(&name) {
+        if serde_list.contains(&derive_info.name) {
             vec![
                 "Default".into(), // Default allows us to easily fill the padding fields with 0.
                 "Serialize".into(),
@@ -63,10 +63,7 @@ fn main() -> anyhow::Result<()> {
             pg_install_dir_versioned = cwd.join("..").join("..").join(pg_install_dir_versioned);
         }
 
-        let pg_config_bin = pg_install_dir_versioned
-            .join(pg_version)
-            .join("bin")
-            .join("pg_config");
+        let pg_config_bin = pg_install_dir_versioned.join("bin").join("pg_config");
         let inc_server_path: String = if pg_config_bin.exists() {
             let output = Command::new(pg_config_bin)
                 .arg("--includedir-server")

libs/pq_proto/Cargo.toml

@@ -5,8 +5,8 @@ edition.workspace = true
 license.workspace = true
 
 [dependencies]
-anyhow.workspace = true
 bytes.workspace = true
+byteorder.workspace = true
 pin-project-lite.workspace = true
 postgres-protocol.workspace = true
 rand.workspace = true

libs/pq_proto/src/framed.rs

@@ -0,0 +1,244 @@
+//! Provides `Framed` -- writing/flushing and reading Postgres messages to/from
+//! the async stream based on (and buffered with) BytesMut. All functions are
+//! cancellation safe.
+//!
+//! It is similar to what tokio_util::codec::Framed with appropriate codec
+//! provides, but `FramedReader` and `FramedWriter` read/write parts can be used
+//! separately without using split from futures::stream::StreamExt (which
+//! allocates box[1] in polling internally). tokio::io::split is used for splitting
+//! instead. Plus we customize error messages more than a single type for all io
+//! calls.
+//!
+//! [1] https://docs.rs/futures-util/0.3.26/src/futures_util/lock/bilock.rs.html#107
+use bytes::{Buf, BytesMut};
+use std::{
+    future::Future,
+    io::{self, ErrorKind},
+};
+use tokio::io::{AsyncRead, AsyncReadExt, AsyncWrite, AsyncWriteExt, ReadHalf, WriteHalf};
+
+use crate::{BeMessage, FeMessage, FeStartupPacket, ProtocolError};
+
+const INITIAL_CAPACITY: usize = 8 * 1024;
+
+/// Error on postgres connection: either IO (physical transport error) or
+/// protocol violation.
+#[derive(thiserror::Error, Debug)]
+pub enum ConnectionError {
+    #[error(transparent)]
+    Io(#[from] io::Error),
+    #[error(transparent)]
+    Protocol(#[from] ProtocolError),
+}
+
+impl ConnectionError {
+    /// Proxy stream.rs uses only io::Error; provide it.
+    pub fn into_io_error(self) -> io::Error {
+        match self {
+            ConnectionError::Io(io) => io,
+            ConnectionError::Protocol(pe) => io::Error::new(io::ErrorKind::Other, pe.to_string()),
+        }
+    }
+}
+
+/// Wraps async io `stream`, providing messages to write/flush + read Postgres
+/// messages.
+pub struct Framed<S> {
+    stream: S,
+    read_buf: BytesMut,
+    write_buf: BytesMut,
+}
+
+impl<S> Framed<S> {
+    pub fn new(stream: S) -> Self {
+        Self {
+            stream,
+            read_buf: BytesMut::with_capacity(INITIAL_CAPACITY),
+            write_buf: BytesMut::with_capacity(INITIAL_CAPACITY),
+        }
+    }
+
+    /// Get a shared reference to the underlying stream.
+    pub fn get_ref(&self) -> &S {
+        &self.stream
+    }
+
+    /// Deconstruct into the underlying stream and read buffer.
+    pub fn into_inner(self) -> (S, BytesMut) {
+        (self.stream, self.read_buf)
+    }
+
+    /// Return new Framed with stream type transformed by async f, for TLS
+    /// upgrade.
+    pub async fn map_stream<S2, E, F, Fut>(self, f: F) -> Result<Framed<S2>, E>
+    where
+        F: FnOnce(S) -> Fut,
+        Fut: Future<Output = Result<S2, E>>,
+    {
+        let stream = f(self.stream).await?;
+        Ok(Framed {
+            stream,
+            read_buf: self.read_buf,
+            write_buf: self.write_buf,
+        })
+    }
+}
+
+impl<S: AsyncRead + Unpin> Framed<S> {
+    pub async fn read_startup_message(
+        &mut self,
+    ) -> Result<Option<FeStartupPacket>, ConnectionError> {
+        read_message(&mut self.stream, &mut self.read_buf, FeStartupPacket::parse).await
+    }
+
+    pub async fn read_message(&mut self) -> Result<Option<FeMessage>, ConnectionError> {
+        read_message(&mut self.stream, &mut self.read_buf, FeMessage::parse).await
+    }
+}
+
+impl<S: AsyncWrite + Unpin> Framed<S> {
+    /// Write next message to the output buffer; doesn't flush.
+    pub fn write_message(&mut self, msg: &BeMessage<'_>) -> Result<(), ProtocolError> {
+        BeMessage::write(&mut self.write_buf, msg)
+    }
+
+    /// Flush out the buffer. This function is cancellation safe: it can be
+    /// interrupted and flushing will be continued in the next call.
+    pub async fn flush(&mut self) -> Result<(), io::Error> {
+        flush(&mut self.stream, &mut self.write_buf).await
+    }
+
+    /// Flush out the buffer and shutdown the stream.
+    pub async fn shutdown(&mut self) -> Result<(), io::Error> {
+        shutdown(&mut self.stream, &mut self.write_buf).await
+    }
+}
+
+impl<S: AsyncRead + AsyncWrite + Unpin> Framed<S> {
+    /// Split into owned read and write parts. Beware of potential issues with
+    /// using halves in different tasks on TLS stream:
+    /// https://github.com/tokio-rs/tls/issues/40
+    pub fn split(self) -> (FramedReader<S>, FramedWriter<S>) {
+        let (read_half, write_half) = tokio::io::split(self.stream);
+        let reader = FramedReader {
+            stream: read_half,
+            read_buf: self.read_buf,
+        };
+        let writer = FramedWriter {
+            stream: write_half,
+            write_buf: self.write_buf,
+        };
+        (reader, writer)
+    }
+
+    /// Join read and write parts back.
+    pub fn unsplit(reader: FramedReader<S>, writer: FramedWriter<S>) -> Self {
+        Self {
+            stream: reader.stream.unsplit(writer.stream),
+            read_buf: reader.read_buf,
+            write_buf: writer.write_buf,
+        }
+    }
+}
+
+/// Read-only version of `Framed`.
+pub struct FramedReader<S> {
+    stream: ReadHalf<S>,
+    read_buf: BytesMut,
+}
+
+impl<S: AsyncRead + Unpin> FramedReader<S> {
+    pub async fn read_message(&mut self) -> Result<Option<FeMessage>, ConnectionError> {
+        read_message(&mut self.stream, &mut self.read_buf, FeMessage::parse).await
+    }
+}
+
+/// Write-only version of `Framed`.
+pub struct FramedWriter<S> {
+    stream: WriteHalf<S>,
+    write_buf: BytesMut,
+}
+
+impl<S: AsyncWrite + Unpin> FramedWriter<S> {
+    /// Write next message to the output buffer; doesn't flush.
+    pub fn write_message_noflush(&mut self, msg: &BeMessage<'_>) -> Result<(), ProtocolError> {
+        BeMessage::write(&mut self.write_buf, msg)
+    }
+
+    /// Flush out the buffer. This function is cancellation safe: it can be
+    /// interrupted and flushing will be continued in the next call.
+    pub async fn flush(&mut self) -> Result<(), io::Error> {
+        flush(&mut self.stream, &mut self.write_buf).await
+    }
+
+    /// Flush out the buffer and shutdown the stream.
+    pub async fn shutdown(&mut self) -> Result<(), io::Error> {
+        shutdown(&mut self.stream, &mut self.write_buf).await
+    }
+}
+
+/// Read next message from the stream. Returns Ok(None), if EOF happened and we
+/// don't have remaining data in the buffer. This function is cancellation safe:
+/// you can drop future which is not yet complete and finalize reading message
+/// with the next call.
+///
+/// Parametrized to allow reading startup or usual message, having different
+/// format.
+async fn read_message<S: AsyncRead + Unpin, M, P>(
+    stream: &mut S,
+    read_buf: &mut BytesMut,
+    parse: P,
+) -> Result<Option<M>, ConnectionError>
+where
+    P: Fn(&mut BytesMut) -> Result<Option<M>, ProtocolError>,
+{
+    loop {
+        if let Some(msg) = parse(read_buf)? {
+            return Ok(Some(msg));
+        }
+        // If we can't build a frame yet, try to read more data and try again.
+        // Make sure we've got room for at least one byte to read to ensure
+        // that we don't get a spurious 0 that looks like EOF.
+        read_buf.reserve(1);
+        if stream.read_buf(read_buf).await? == 0 {
+            if read_buf.has_remaining() {
+                return Err(io::Error::new(
+                    ErrorKind::UnexpectedEof,
+                    "EOF with unprocessed data in the buffer",
+                )
+                .into());
+            } else {
+                return Ok(None); // clean EOF
+            }
+        }
+    }
+}
+
+async fn flush<S: AsyncWrite + Unpin>(
+    stream: &mut S,
+    write_buf: &mut BytesMut,
+) -> Result<(), io::Error> {
+    while write_buf.has_remaining() {
+        let bytes_written = stream.write(write_buf.chunk()).await?;
+        if bytes_written == 0 {
+            return Err(io::Error::new(
+                ErrorKind::WriteZero,
+                "failed to write message",
+            ));
+        }
+        // The advanced part will be garbage collected, likely during shifting
+        // data left on next attempt to write to buffer when free space is not
+        // enough.
+        write_buf.advance(bytes_written);
+    }
+    write_buf.clear();
+    stream.flush().await
+}
+
+async fn shutdown<S: AsyncWrite + Unpin>(
+    stream: &mut S,
+    write_buf: &mut BytesMut,
+) -> Result<(), io::Error> {
+    flush(stream, write_buf).await?;
+    stream.shutdown().await
+}

libs/pq_proto/src/lib.rs

@@ -2,24 +2,18 @@
 //! <https://www.postgresql.org/docs/devel/protocol-message-formats.html>
 //! on message formats.
 
-// Tools for calling certain async methods in sync contexts.
-pub mod sync;
+pub mod framed;
 
-use anyhow::{ensure, Context, Result};
+use byteorder::{BigEndian, ReadBytesExt};
 use bytes::{Buf, BufMut, Bytes, BytesMut};
 use postgres_protocol::PG_EPOCH;
 use serde::{Deserialize, Serialize};
 use std::{
     borrow::Cow,
     collections::HashMap,
-    fmt,
-    future::Future,
-    io::{self, Cursor},
-    str,
+    fmt, io, str,
     time::{Duration, SystemTime},
 };
-use sync::{AsyncishRead, SyncFuture};
-use tokio::io::AsyncReadExt;
 use tracing::{trace, warn};
 
 pub type Oid = u32;
@@ -31,7 +25,6 @@ pub const TEXT_OID: Oid = 25;
 
 #[derive(Debug)]
 pub enum FeMessage {
-    StartupPacket(FeStartupPacket),
     // Simple query.
     Query(Bytes),
     // Extended query protocol.
@@ -75,27 +68,36 @@ 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 +110,12 @@ 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.
@@ -179,260 +184,208 @@ pub struct FeExecuteMessage {
 #[derive(Debug)]
 pub struct FeCloseMessage;
 
-/// Retry a read on EINTR
-///
-/// This runs the enclosed expression, and if it returns
-/// Err(io::ErrorKind::Interrupted), retries it.
-macro_rules! retry_read {
-    ( $x:expr ) => {
-        loop {
-            match $x {
-                Err(e) if e.kind() == io::ErrorKind::Interrupted => continue,
-                res => break res,
-            }
-        }
-    };
-}
-
-/// An error occured during connection being open.
+/// An error occured while parsing or serializing raw stream into Postgres
+/// messages.
 #[derive(thiserror::Error, Debug)]
-pub enum ConnectionError {
-    /// IO error during writing to or reading from the connection socket.
-    #[error("Socket IO error: {0}")]
-    Socket(std::io::Error),
-    /// Invalid packet was received from client
+pub enum ProtocolError {
+    /// Invalid packet was received from the client (e.g. unexpected message
+    /// type or broken len).
     #[error("Protocol error: {0}")]
     Protocol(String),
-    /// Failed to parse a protocol mesage
+    /// Failed to parse or, (unlikely), serialize a protocol message.
     #[error("Message parse error: {0}")]
-    MessageParse(anyhow::Error),
+    BadMessage(String),
 }
 
-impl From<anyhow::Error> for ConnectionError {
-    fn from(e: anyhow::Error) -> Self {
-        Self::MessageParse(e)
-    }
-}
-
-impl ConnectionError {
+impl ProtocolError {
+    /// Proxy stream.rs uses only io::Error; provide it.
     pub fn into_io_error(self) -> io::Error {
-        match self {
-            ConnectionError::Socket(io) => io,
-            other => io::Error::new(io::ErrorKind::Other, other.to_string()),
-        }
+        io::Error::new(io::ErrorKind::Other, self.to_string())
     }
 }
 
 impl FeMessage {
-    /// Read one message from the stream.
-    /// This function returns `Ok(None)` in case of EOF.
-    /// One way to handle this properly:
+    /// Read and parse one message from the `buf` input buffer. If there is at
+    /// least one valid message, returns it, advancing `buf`; redundant copies
+    /// are avoided, as thanks to `bytes` crate ptrs in parsed message point
+    /// directly into the `buf` (processed data is garbage collected after
+    /// parsed message is dropped).
     ///
-    /// ```
-    /// # use std::io;
-    /// # use pq_proto::FeMessage;
-    /// #
-    /// # fn process_message(msg: FeMessage) -> anyhow::Result<()> {
-    /// #     Ok(())
-    /// # };
-    /// #
-    /// fn do_the_job(stream: &mut (impl io::Read + Unpin)) -> anyhow::Result<()> {
-    ///     while let Some(msg) = FeMessage::read(stream)? {
-    ///         process_message(msg)?;
-    ///     }
+    /// Returns None if `buf` doesn't contain enough data for a single message.
+    /// For efficiency, tries to reserve large enough space in `buf` for the
+    /// next message in this case to save the repeated calls.
     ///
-    ///     Ok(())
-    /// }
-    /// ```
-    #[inline(never)]
-    pub fn read(
-        stream: &mut (impl io::Read + Unpin),
-    ) -> Result<Option<FeMessage>, ConnectionError> {
-        Self::read_fut(&mut AsyncishRead(stream)).wait()
-    }
+    /// Returns Error if message is malformed, the only possible ErrorKind is
+    /// InvalidInput.
+    //
+    // Inspired by rust-postgres Message::parse.
+    pub fn parse(buf: &mut BytesMut) -> Result<Option<FeMessage>, ProtocolError> {
+        // Every message contains message type byte and 4 bytes len; can't do
+        // much without them.
+        if buf.len() < 5 {
+            let to_read = 5 - buf.len();
+            buf.reserve(to_read);
+            return Ok(None);
+        }
 
-    /// Read one message from the stream.
-    /// See documentation for `Self::read`.
-    pub fn read_fut<Reader>(
-        stream: &mut Reader,
-    ) -> SyncFuture<Reader, impl Future<Output = Result<Option<FeMessage>, ConnectionError>> + '_>
-    where
-        Reader: tokio::io::AsyncRead + Unpin,
-    {
-        // We return a Future that's sync (has a `wait` method) if and only if the provided stream is SyncProof.
-        // SyncFuture contract: we are only allowed to await on sync-proof futures, the AsyncRead and
-        // AsyncReadExt methods of the stream.
-        SyncFuture::new(async move {
-            // Each libpq message begins with a message type byte, followed by message length
-            // If the client closes the connection, return None. But if the client closes the
-            // connection in the middle of a message, we will return an error.
-            let tag = match retry_read!(stream.read_u8().await) {
-                Ok(b) => b,
-                Err(e) if e.kind() == io::ErrorKind::UnexpectedEof => return Ok(None),
-                Err(e) => return Err(ConnectionError::Socket(e)),
-            };
+        // We shouldn't advance `buf` as probably full message is not there yet,
+        // so can't directly use Bytes::get_u32 etc.
+        let tag = buf[0];
+        let len = (&buf[1..5]).read_u32::<BigEndian>().unwrap();
+        if len < 4 {
+            return Err(ProtocolError::Protocol(format!(
+                "invalid message length {}",
+                len
+            )));
+        }
 
-            // The message length includes itself, so it better be at least 4.
-            let len = retry_read!(stream.read_u32().await)
-                .map_err(ConnectionError::Socket)?
-                .checked_sub(4)
-                .ok_or_else(|| ConnectionError::Protocol("invalid message length".to_string()))?;
+        // length field includes itself, but not message type.
+        let total_len = len as usize + 1;
+        if buf.len() < total_len {
+            // Don't have full message yet.
+            let to_read = total_len - buf.len();
+            buf.reserve(to_read);
+            return Ok(None);
+        }
 
-            let body = {
-                let mut buffer = vec![0u8; len as usize];
-                stream
-                    .read_exact(&mut buffer)
-                    .await
-                    .map_err(ConnectionError::Socket)?;
-                Bytes::from(buffer)
-            };
+        // got the message, advance buffer
+        let mut msg = buf.split_to(total_len).freeze();
+        msg.advance(5); // consume message type and len
 
-            match tag {
-                b'Q' => Ok(Some(FeMessage::Query(body))),
-                b'P' => Ok(Some(FeParseMessage::parse(body)?)),
-                b'D' => Ok(Some(FeDescribeMessage::parse(body)?)),
-                b'E' => Ok(Some(FeExecuteMessage::parse(body)?)),
-                b'B' => Ok(Some(FeBindMessage::parse(body)?)),
-                b'C' => Ok(Some(FeCloseMessage::parse(body)?)),
-                b'S' => Ok(Some(FeMessage::Sync)),
-                b'X' => Ok(Some(FeMessage::Terminate)),
-                b'd' => Ok(Some(FeMessage::CopyData(body))),
-                b'c' => Ok(Some(FeMessage::CopyDone)),
-                b'f' => Ok(Some(FeMessage::CopyFail)),
-                b'p' => Ok(Some(FeMessage::PasswordMessage(body))),
-                tag => {
-                    return Err(ConnectionError::Protocol(format!(
-                        "unknown message tag: {tag},'{body:?}'"
-                    )))
-                }
-            }
-        })
+        match tag {
+            b'Q' => Ok(Some(FeMessage::Query(msg))),
+            b'P' => Ok(Some(FeParseMessage::parse(msg)?)),
+            b'D' => Ok(Some(FeDescribeMessage::parse(msg)?)),
+            b'E' => Ok(Some(FeExecuteMessage::parse(msg)?)),
+            b'B' => Ok(Some(FeBindMessage::parse(msg)?)),
+            b'C' => Ok(Some(FeCloseMessage::parse(msg)?)),
+            b'S' => Ok(Some(FeMessage::Sync)),
+            b'X' => Ok(Some(FeMessage::Terminate)),
+            b'd' => Ok(Some(FeMessage::CopyData(msg))),
+            b'c' => Ok(Some(FeMessage::CopyDone)),
+            b'f' => Ok(Some(FeMessage::CopyFail)),
+            b'p' => Ok(Some(FeMessage::PasswordMessage(msg))),
+            tag => Err(ProtocolError::Protocol(format!(
+                "unknown message tag: {tag},'{msg:?}'"
+            ))),
+        }
     }
 }
 
 impl FeStartupPacket {
-    /// Read startup message from the stream.
-    // XXX: It's tempting yet undesirable to accept `stream` by value,
-    // since such a change will cause user-supplied &mut references to be consumed
-    pub fn read(
-        stream: &mut (impl io::Read + Unpin),
-    ) -> Result<Option<FeMessage>, ConnectionError> {
-        Self::read_fut(&mut AsyncishRead(stream)).wait()
-    }
-
-    /// Read startup message from the stream.
-    // XXX: It's tempting yet undesirable to accept `stream` by value,
-    // since such a change will cause user-supplied &mut references to be consumed
-    pub fn read_fut<Reader>(
-        stream: &mut Reader,
-    ) -> SyncFuture<Reader, impl Future<Output = Result<Option<FeMessage>, ConnectionError>> + '_>
-    where
-        Reader: tokio::io::AsyncRead + Unpin,
-    {
+    /// Read and parse startup message from the `buf` input buffer. It is
+    /// different from [`FeMessage::parse`] because startup messages don't have
+    /// message type byte; otherwise, its comments apply.
+    pub fn parse(buf: &mut BytesMut) -> Result<Option<FeStartupPacket>, ProtocolError> {
         const MAX_STARTUP_PACKET_LENGTH: usize = 10000;
         const RESERVED_INVALID_MAJOR_VERSION: u32 = 1234;
         const CANCEL_REQUEST_CODE: u32 = 5678;
         const NEGOTIATE_SSL_CODE: u32 = 5679;
         const NEGOTIATE_GSS_CODE: u32 = 5680;
 
-        SyncFuture::new(async move {
-            // Read length. If the connection is closed before reading anything (or before
-            // reading 4 bytes, to be precise), return None to indicate that the connection
-            // was closed. This matches the PostgreSQL server's behavior, which avoids noise
-            // in the log if the client opens connection but closes it immediately.
-            let len = match retry_read!(stream.read_u32().await) {
-                Ok(len) => len as usize,
-                Err(e) if e.kind() == io::ErrorKind::UnexpectedEof => return Ok(None),
-                Err(e) => return Err(ConnectionError::Socket(e)),
-            };
+        // need at least 4 bytes with packet len
+        if buf.len() < 4 {
+            let to_read = 4 - buf.len();
+            buf.reserve(to_read);
+            return Ok(None);
+        }
 
-            #[allow(clippy::manual_range_contains)]
-            if len < 4 || len > MAX_STARTUP_PACKET_LENGTH {
-                return Err(ConnectionError::Protocol(format!(
-                    "invalid message length {len}"
+        // We shouldn't advance `buf` as probably full message is not there yet,
+        // so can't directly use Bytes::get_u32 etc.
+        let len = (&buf[0..4]).read_u32::<BigEndian>().unwrap() as usize;
+        // The proposed replacement is `!(4..=MAX_STARTUP_PACKET_LENGTH).contains(&len)`
+        // which is less readable
+        #[allow(clippy::manual_range_contains)]
+        if len < 4 || len > MAX_STARTUP_PACKET_LENGTH {
+            return Err(ProtocolError::Protocol(format!(
+                "invalid startup packet message length {}",
+                len
+            )));
+        }
+
+        if buf.len() < len {
+            // Don't have full message yet.
+            let to_read = len - buf.len();
+            buf.reserve(to_read);
+            return Ok(None);
+        }
+
+        // got the message, advance buffer
+        let mut msg = buf.split_to(len).freeze();
+        msg.advance(4); // consume len
+
+        let request_code = msg.get_u32();
+        let req_hi = request_code >> 16;
+        let req_lo = request_code & ((1 << 16) - 1);
+        // StartupMessage, CancelRequest, SSLRequest etc are differentiated by request code.
+        let message = match (req_hi, req_lo) {
+            (RESERVED_INVALID_MAJOR_VERSION, CANCEL_REQUEST_CODE) => {
+                if msg.remaining() != 8 {
+                    return Err(ProtocolError::BadMessage(
+                        "CancelRequest message is malformed, backend PID / secret key missing"
+                            .to_owned(),
+                    ));
+                }
+                FeStartupPacket::CancelRequest(CancelKeyData {
+                    backend_pid: msg.get_i32(),
+                    cancel_key: msg.get_i32(),
+                })
+            }
+            (RESERVED_INVALID_MAJOR_VERSION, NEGOTIATE_SSL_CODE) => {
+                // Requested upgrade to SSL (aka TLS)
+                FeStartupPacket::SslRequest
+            }
+            (RESERVED_INVALID_MAJOR_VERSION, NEGOTIATE_GSS_CODE) => {
+                // Requested upgrade to GSSAPI
+                FeStartupPacket::GssEncRequest
+            }
+            (RESERVED_INVALID_MAJOR_VERSION, unrecognized_code) => {
+                return Err(ProtocolError::Protocol(format!(
+                    "Unrecognized request code {unrecognized_code}"
                 )));
             }
+            // TODO bail if protocol major_version is not 3?
+            (major_version, minor_version) => {
+                // StartupMessage
 
-            let request_code =
-                retry_read!(stream.read_u32().await).map_err(ConnectionError::Socket)?;
+                // Parse pairs of null-terminated strings (key, value).
+                // See `postgres: ProcessStartupPacket, build_startup_packet`.
+                let mut tokens = str::from_utf8(&msg)
+                    .map_err(|_e| {
+                        ProtocolError::BadMessage("StartupMessage params: invalid utf-8".to_owned())
+                    })?
+                    .strip_suffix('\0') // drop packet's own null
+                    .ok_or_else(|| {
+                        ProtocolError::Protocol(
+                            "StartupMessage params: missing null terminator".to_string(),
+                        )
+                    })?
+                    .split_terminator('\0');
 
-            // the rest of startup packet are params
-            let params_len = len - 8;
-            let mut params_bytes = vec![0u8; params_len];
-            stream
-                .read_exact(params_bytes.as_mut())
-                .await
-                .map_err(ConnectionError::Socket)?;
+                let mut params = HashMap::new();
+                while let Some(name) = tokens.next() {
+                    let value = tokens.next().ok_or_else(|| {
+                        ProtocolError::Protocol(
+                            "StartupMessage params: key without value".to_string(),
+                        )
+                    })?;
 
-            // Parse params depending on request code
-            let req_hi = request_code >> 16;
-            let req_lo = request_code & ((1 << 16) - 1);
-            let message = match (req_hi, req_lo) {
-                (RESERVED_INVALID_MAJOR_VERSION, CANCEL_REQUEST_CODE) => {
-                    if params_len != 8 {
-                        return Err(ConnectionError::Protocol(
-                            "expected 8 bytes for CancelRequest params".to_string(),
-                        ));
-                    }
-                    let mut cursor = Cursor::new(params_bytes);
-                    FeStartupPacket::CancelRequest(CancelKeyData {
-                        backend_pid: cursor.read_i32().await.map_err(ConnectionError::Socket)?,
-                        cancel_key: cursor.read_i32().await.map_err(ConnectionError::Socket)?,
-                    })
+                    params.insert(name.to_owned(), value.to_owned());
                 }
-                (RESERVED_INVALID_MAJOR_VERSION, NEGOTIATE_SSL_CODE) => {
-                    // Requested upgrade to SSL (aka TLS)
-                    FeStartupPacket::SslRequest
-                }
-                (RESERVED_INVALID_MAJOR_VERSION, NEGOTIATE_GSS_CODE) => {
-                    // Requested upgrade to GSSAPI
-                    FeStartupPacket::GssEncRequest
-                }
-                (RESERVED_INVALID_MAJOR_VERSION, unrecognized_code) => {
-                    return Err(ConnectionError::Protocol(format!(
-                        "Unrecognized request code {unrecognized_code}"
-                    )));
-                }
-                // TODO bail if protocol major_version is not 3?
-                (major_version, minor_version) => {
-                    // Parse pairs of null-terminated strings (key, value).
-                    // See `postgres: ProcessStartupPacket, build_startup_packet`.
-                    let mut tokens = str::from_utf8(&params_bytes)
-                        .context("StartupMessage params: invalid utf-8")?
-                        .strip_suffix('\0') // drop packet's own null
-                        .ok_or_else(|| {
-                            ConnectionError::Protocol(
-                                "StartupMessage params: missing null terminator".to_string(),
-                            )
-                        })?
-                        .split_terminator('\0');
 
-                    let mut params = HashMap::new();
-                    while let Some(name) = tokens.next() {
-                        let value = tokens.next().ok_or_else(|| {
-                            ConnectionError::Protocol(
-                                "StartupMessage params: key without value".to_string(),
-                            )
-                        })?;
-
-                        params.insert(name.to_owned(), value.to_owned());
-                    }
-
-                    FeStartupPacket::StartupMessage {
-                        major_version,
-                        minor_version,
-                        params: StartupMessageParams { params },
-                    }
+                FeStartupPacket::StartupMessage {
+                    major_version,
+                    minor_version,
+                    params: StartupMessageParams { params },
                 }
-            };
-
-            Ok(Some(FeMessage::StartupPacket(message)))
-        })
+            }
+        };
+        Ok(Some(message))
     }
 }
 
 impl FeParseMessage {
-    fn parse(mut buf: Bytes) -> anyhow::Result<FeMessage> {
+    fn parse(mut buf: Bytes) -> Result<FeMessage, ProtocolError> {
         // FIXME: the rust-postgres driver uses a named prepared statement
         // for copy_out(). We're not prepared to handle that correctly. For
         // now, just ignore the statement name, assuming that the client never
@@ -440,55 +393,82 @@ impl FeParseMessage {
 
         let _pstmt_name = read_cstr(&mut buf)?;
         let query_string = read_cstr(&mut buf)?;
+        if buf.remaining() < 2 {
+            return Err(ProtocolError::BadMessage(
+                "Parse message is malformed, nparams missing".to_string(),
+            ));
+        }
         let nparams = buf.get_i16();
 
-        ensure!(nparams == 0, "query params not implemented");
+        if nparams != 0 {
+            return Err(ProtocolError::BadMessage(
+                "query params not implemented".to_string(),
+            ));
+        }
 
         Ok(FeMessage::Parse(FeParseMessage { query_string }))
     }
 }
 
 impl FeDescribeMessage {
-    fn parse(mut buf: Bytes) -> anyhow::Result<FeMessage> {
+    fn parse(mut buf: Bytes) -> Result<FeMessage, ProtocolError> {
         let kind = buf.get_u8();
         let _pstmt_name = read_cstr(&mut buf)?;
 
         // FIXME: see FeParseMessage::parse
-        ensure!(
-            kind == b'S',
-            "only prepared statemement Describe is implemented"
-        );
+        if kind != b'S' {
+            return Err(ProtocolError::BadMessage(
+                "only prepared statemement Describe is implemented".to_string(),
+            ));
+        }
 
         Ok(FeMessage::Describe(FeDescribeMessage { kind }))
     }
 }
 
 impl FeExecuteMessage {
-    fn parse(mut buf: Bytes) -> anyhow::Result<FeMessage> {
+    fn parse(mut buf: Bytes) -> Result<FeMessage, ProtocolError> {
         let portal_name = read_cstr(&mut buf)?;
+        if buf.remaining() < 4 {
+            return Err(ProtocolError::BadMessage(
+                "FeExecuteMessage message is malformed, maxrows missing".to_string(),
+            ));
+        }
         let maxrows = buf.get_i32();
 
-        ensure!(portal_name.is_empty(), "named portals not implemented");
-        ensure!(maxrows == 0, "row limit in Execute message not implemented");
+        if !portal_name.is_empty() {
+            return Err(ProtocolError::BadMessage(
+                "named portals not implemented".to_string(),
+            ));
+        }
+        if maxrows != 0 {
+            return Err(ProtocolError::BadMessage(
+                "row limit in Execute message not implemented".to_string(),
+            ));
+        }
 
         Ok(FeMessage::Execute(FeExecuteMessage { maxrows }))
     }
 }
 
 impl FeBindMessage {
-    fn parse(mut buf: Bytes) -> anyhow::Result<FeMessage> {
+    fn parse(mut buf: Bytes) -> Result<FeMessage, ProtocolError> {
         let portal_name = read_cstr(&mut buf)?;
         let _pstmt_name = read_cstr(&mut buf)?;
 
         // FIXME: see FeParseMessage::parse
-        ensure!(portal_name.is_empty(), "named portals not implemented");
+        if !portal_name.is_empty() {
+            return Err(ProtocolError::BadMessage(
+                "named portals not implemented".to_string(),
+            ));
+        }
 
         Ok(FeMessage::Bind(FeBindMessage))
     }
 }
 
 impl FeCloseMessage {
-    fn parse(mut buf: Bytes) -> anyhow::Result<FeMessage> {
+    fn parse(mut buf: Bytes) -> Result<FeMessage, ProtocolError> {
         let _kind = buf.get_u8();
         let _pstmt_or_portal_name = read_cstr(&mut buf)?;
 
@@ -517,6 +497,7 @@ pub enum BeMessage<'a> {
     CloseComplete,
     // None means column is NULL
     DataRow(&'a [Option<&'a [u8]>]),
+    // None errcode means internal_error will be sent.
     ErrorResponse(&'a str, Option<&'a [u8; 5]>),
     /// Single byte - used in response to SSLRequest/GSSENCRequest.
     EncryptionResponse(bool),
@@ -547,6 +528,11 @@ impl<'a> BeMessage<'a> {
         value: b"UTF8",
     };
 
+    pub const INTEGER_DATETIMES: Self = Self::ParameterStatus {
+        name: b"integer_datetimes",
+        value: b"on",
+    };
+
     /// Build a [`BeMessage::ParameterStatus`] holding the server version.
     pub fn server_version(version: &'a str) -> Self {
         Self::ParameterStatus {
@@ -625,7 +611,7 @@ impl RowDescriptor<'_> {
 #[derive(Debug)]
 pub struct XLogDataBody<'a> {
     pub wal_start: u64,
-    pub wal_end: u64,
+    pub wal_end: u64, // current end of WAL on the server
     pub timestamp: i64,
     pub data: &'a [u8],
 }
@@ -665,12 +651,11 @@ fn write_body<R>(buf: &mut BytesMut, f: impl FnOnce(&mut BytesMut) -> R) -> R {
 }
 
 /// Safe write of s into buf as cstring (String in the protocol).
-fn write_cstr(s: impl AsRef<[u8]>, buf: &mut BytesMut) -> io::Result<()> {
+fn write_cstr(s: impl AsRef<[u8]>, buf: &mut BytesMut) -> Result<(), ProtocolError> {
     let bytes = s.as_ref();
     if bytes.contains(&0) {
-        return Err(io::Error::new(
-            io::ErrorKind::InvalidInput,
-            "string contains embedded null",
+        return Err(ProtocolError::BadMessage(
+            "string contains embedded null".to_owned(),
         ));
     }
     buf.put_slice(bytes);
@@ -678,22 +663,27 @@ fn write_cstr(s: impl AsRef<[u8]>, buf: &mut BytesMut) -> io::Result<()> {
     Ok(())
 }
 
-fn read_cstr(buf: &mut Bytes) -> anyhow::Result<Bytes> {
-    let pos = buf.iter().position(|x| *x == 0);
-    let result = buf.split_to(pos.context("missing terminator")?);
+/// Read cstring from buf, advancing it.
+fn read_cstr(buf: &mut Bytes) -> Result<Bytes, ProtocolError> {
+    let pos = buf
+        .iter()
+        .position(|x| *x == 0)
+        .ok_or_else(|| ProtocolError::BadMessage("missing cstring terminator".to_owned()))?;
+    let result = buf.split_to(pos);
     buf.advance(1); // drop the null terminator
     Ok(result)
 }
 
 pub const SQLSTATE_INTERNAL_ERROR: &[u8; 5] = b"XX000";
+pub const SQLSTATE_SUCCESSFUL_COMPLETION: &[u8; 5] = b"00000";
 
 impl<'a> BeMessage<'a> {
-    /// Write message to the given buf.
-    // Unlike the reading side, we use BytesMut
-    // here as msg len precedes its body and it is handy to write it down first
-    // and then fill the length. With Write we would have to either calc it
-    // manually or have one more buffer.
-    pub fn write(buf: &mut BytesMut, message: &BeMessage) -> io::Result<()> {
+    /// Serialize `message` to the given `buf`.
+    /// Apart from smart memory managemet, BytesMut is good here as msg len
+    /// precedes its body and it is handy to write it down first and then fill
+    /// the length. With Write we would have to either calc it manually or have
+    /// one more buffer.
+    pub fn write(buf: &mut BytesMut, message: &BeMessage) -> Result<(), ProtocolError> {
         match message {
             BeMessage::AuthenticationOk => {
                 buf.put_u8(b'R');
@@ -738,7 +728,7 @@ impl<'a> BeMessage<'a> {
                             buf.put_slice(extra);
                         }
                     }
-                    Ok::<_, io::Error>(())
+                    Ok(())
                 })?;
             }
 
@@ -842,7 +832,7 @@ impl<'a> BeMessage<'a> {
                     write_cstr(error_msg, buf)?;
 
                     buf.put_u8(0); // terminator
-                    Ok::<_, io::Error>(())
+                    Ok(())
                 })?;
             }
 
@@ -865,7 +855,7 @@ impl<'a> BeMessage<'a> {
                     write_cstr(error_msg.as_bytes(), buf)?;
 
                     buf.put_u8(0); // terminator
-                    Ok::<_, io::Error>(())
+                    Ok(())
                 })?;
             }
 
@@ -920,7 +910,7 @@ impl<'a> BeMessage<'a> {
                         buf.put_i32(-1); /* typmod */
                         buf.put_i16(0); /* format code */
                     }
-                    Ok::<_, io::Error>(())
+                    Ok(())
                 })?;
             }
 
@@ -949,35 +939,40 @@ impl<'a> BeMessage<'a> {
     }
 }
 
-// Neon extension of postgres replication protocol
-// See NEON_STATUS_UPDATE_TAG_BYTE
+/// Feedback pageserver sends to safekeeper and safekeeper resends to compute.
+/// Serialized in custom flexible key/value format. In replication protocol, it
+/// is marked with NEON_STATUS_UPDATE_TAG_BYTE to differentiate from postgres
+/// Standby status update / Hot standby feedback messages.
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
-pub struct ReplicationFeedback {
-    // Last known size of the timeline. Used to enforce timeline size limit.
+pub struct PageserverFeedback {
+    /// Last known size of the timeline. Used to enforce timeline size limit.
     pub current_timeline_size: u64,
-    // Parts of StandbyStatusUpdate we resend to compute via safekeeper
-    pub ps_writelsn: u64,
-    pub ps_applylsn: u64,
-    pub ps_flushlsn: u64,
-    pub ps_replytime: SystemTime,
+    /// LSN last received and ingested by the pageserver.
+    pub last_received_lsn: u64,
+    /// LSN up to which data is persisted by the pageserver to its local disc.
+    pub disk_consistent_lsn: u64,
+    /// LSN up to which data is persisted by the pageserver on s3; safekeepers
+    /// consider WAL before it can be removed.
+    pub remote_consistent_lsn: u64,
+    pub replytime: SystemTime,
 }
 
-// NOTE: Do not forget to increment this number when adding new fields to ReplicationFeedback.
+// NOTE: Do not forget to increment this number when adding new fields to PageserverFeedback.
 // Do not remove previously available fields because this might be backwards incompatible.
-pub const REPLICATION_FEEDBACK_FIELDS_NUMBER: u8 = 5;
+pub const PAGESERVER_FEEDBACK_FIELDS_NUMBER: u8 = 5;
 
-impl ReplicationFeedback {
-    pub fn empty() -> ReplicationFeedback {
-        ReplicationFeedback {
+impl PageserverFeedback {
+    pub fn empty() -> PageserverFeedback {
+        PageserverFeedback {
             current_timeline_size: 0,
-            ps_writelsn: 0,
-            ps_applylsn: 0,
-            ps_flushlsn: 0,
-            ps_replytime: SystemTime::now(),
+            last_received_lsn: 0,
+            remote_consistent_lsn: 0,
+            disk_consistent_lsn: 0,
+            replytime: SystemTime::now(),
         }
     }
 
-    // Serialize ReplicationFeedback using custom format
+    // Serialize PageserverFeedback using custom format
     // to support protocol extensibility.
     //
     // Following layout is used:
@@ -987,24 +982,26 @@ impl ReplicationFeedback {
     // null-terminated string - key,
     // uint32 - value length in bytes
     // value itself
-    pub fn serialize(&self, buf: &mut BytesMut) -> Result<()> {
-        buf.put_u8(REPLICATION_FEEDBACK_FIELDS_NUMBER); // # of keys
+    //
+    // TODO: change serialized fields names once all computes migrate to rename.
+    pub fn serialize(&self, buf: &mut BytesMut) {
+        buf.put_u8(PAGESERVER_FEEDBACK_FIELDS_NUMBER); // # of keys
         buf.put_slice(b"current_timeline_size\0");
         buf.put_i32(8);
         buf.put_u64(self.current_timeline_size);
 
         buf.put_slice(b"ps_writelsn\0");
         buf.put_i32(8);
-        buf.put_u64(self.ps_writelsn);
+        buf.put_u64(self.last_received_lsn);
         buf.put_slice(b"ps_flushlsn\0");
         buf.put_i32(8);
-        buf.put_u64(self.ps_flushlsn);
+        buf.put_u64(self.disk_consistent_lsn);
         buf.put_slice(b"ps_applylsn\0");
         buf.put_i32(8);
-        buf.put_u64(self.ps_applylsn);
+        buf.put_u64(self.remote_consistent_lsn);
 
         let timestamp = self
-            .ps_replytime
+            .replytime
             .duration_since(*PG_EPOCH)
             .expect("failed to serialize pg_replytime earlier than PG_EPOCH")
             .as_micros() as i64;
@@ -1012,12 +1009,12 @@ impl ReplicationFeedback {
         buf.put_slice(b"ps_replytime\0");
         buf.put_i32(8);
         buf.put_i64(timestamp);
-        Ok(())
     }
 
-    // Deserialize ReplicationFeedback message
-    pub fn parse(mut buf: Bytes) -> ReplicationFeedback {
-        let mut rf = ReplicationFeedback::empty();
+    // Deserialize PageserverFeedback message
+    // TODO: change serialized fields names once all computes migrate to rename.
+    pub fn parse(mut buf: Bytes) -> PageserverFeedback {
+        let mut rf = PageserverFeedback::empty();
         let nfields = buf.get_u8();
         for _ in 0..nfields {
             let key = read_cstr(&mut buf).unwrap();
@@ -1030,39 +1027,39 @@ impl ReplicationFeedback {
                 b"ps_writelsn" => {
                     let len = buf.get_i32();
                     assert_eq!(len, 8);
-                    rf.ps_writelsn = buf.get_u64();
+                    rf.last_received_lsn = buf.get_u64();
                 }
                 b"ps_flushlsn" => {
                     let len = buf.get_i32();
                     assert_eq!(len, 8);
-                    rf.ps_flushlsn = buf.get_u64();
+                    rf.disk_consistent_lsn = buf.get_u64();
                 }
                 b"ps_applylsn" => {
                     let len = buf.get_i32();
                     assert_eq!(len, 8);
-                    rf.ps_applylsn = buf.get_u64();
+                    rf.remote_consistent_lsn = buf.get_u64();
                 }
                 b"ps_replytime" => {
                     let len = buf.get_i32();
                     assert_eq!(len, 8);
                     let raw_time = buf.get_i64();
                     if raw_time > 0 {
-                        rf.ps_replytime = *PG_EPOCH + Duration::from_micros(raw_time as u64);
+                        rf.replytime = *PG_EPOCH + Duration::from_micros(raw_time as u64);
                     } else {
-                        rf.ps_replytime = *PG_EPOCH - Duration::from_micros(-raw_time as u64);
+                        rf.replytime = *PG_EPOCH - Duration::from_micros(-raw_time as u64);
                     }
                 }
                 _ => {
                     let len = buf.get_i32();
                     warn!(
-                        "ReplicationFeedback parse. unknown key {} of len {len}. Skip it.",
+                        "PageserverFeedback parse. unknown key {} of len {len}. Skip it.",
                         String::from_utf8_lossy(key.as_ref())
                     );
                     buf.advance(len as usize);
                 }
             }
         }
-        trace!("ReplicationFeedback parsed is {:?}", rf);
+        trace!("PageserverFeedback parsed is {:?}", rf);
         rf
     }
 }
@@ -1073,33 +1070,33 @@ mod tests {
 
     #[test]
     fn test_replication_feedback_serialization() {
-        let mut rf = ReplicationFeedback::empty();
+        let mut rf = PageserverFeedback::empty();
         // Fill rf with some values
         rf.current_timeline_size = 12345678;
         // Set rounded time to be able to compare it with deserialized value,
         // because it is rounded up to microseconds during serialization.
-        rf.ps_replytime = *PG_EPOCH + Duration::from_secs(100_000_000);
+        rf.replytime = *PG_EPOCH + Duration::from_secs(100_000_000);
         let mut data = BytesMut::new();
-        rf.serialize(&mut data).unwrap();
+        rf.serialize(&mut data);
 
-        let rf_parsed = ReplicationFeedback::parse(data.freeze());
+        let rf_parsed = PageserverFeedback::parse(data.freeze());
         assert_eq!(rf, rf_parsed);
     }
 
     #[test]
     fn test_replication_feedback_unknown_key() {
-        let mut rf = ReplicationFeedback::empty();
+        let mut rf = PageserverFeedback::empty();
         // Fill rf with some values
         rf.current_timeline_size = 12345678;
         // Set rounded time to be able to compare it with deserialized value,
         // because it is rounded up to microseconds during serialization.
-        rf.ps_replytime = *PG_EPOCH + Duration::from_secs(100_000_000);
+        rf.replytime = *PG_EPOCH + Duration::from_secs(100_000_000);
         let mut data = BytesMut::new();
-        rf.serialize(&mut data).unwrap();
+        rf.serialize(&mut data);
 
         // Add an extra field to the buffer and adjust number of keys
         if let Some(first) = data.first_mut() {
-            *first = REPLICATION_FEEDBACK_FIELDS_NUMBER + 1;
+            *first = PAGESERVER_FEEDBACK_FIELDS_NUMBER + 1;
         }
 
         data.put_slice(b"new_field_one\0");
@@ -1107,7 +1104,7 @@ mod tests {
         data.put_u64(42);
 
         // Parse serialized data and check that new field is not parsed
-        let rf_parsed = ReplicationFeedback::parse(data.freeze());
+        let rf_parsed = PageserverFeedback::parse(data.freeze());
         assert_eq!(rf, rf_parsed);
     }
 
@@ -1137,15 +1134,6 @@ mod tests {
         let params = make_params("foo\\ bar \\ \\\\ baz\\  lol");
         assert_eq!(split_options(&params), ["foo bar", " \\", "baz ", "lol"]);
     }
-
-    // Make sure that `read` is sync/async callable
-    async fn _assert(stream: &mut (impl tokio::io::AsyncRead + Unpin)) {
-        let _ = FeMessage::read(&mut [].as_ref());
-        let _ = FeMessage::read_fut(stream).await;
-
-        let _ = FeStartupPacket::read(&mut [].as_ref());
-        let _ = FeStartupPacket::read_fut(stream).await;
-    }
 }
 
 fn terminate_code(code: &[u8; 5]) -> [u8; 6] {

libs/pq_proto/src/sync.rs

@@ -1,179 +0,0 @@
-use pin_project_lite::pin_project;
-use std::future::Future;
-use std::marker::PhantomData;
-use std::pin::Pin;
-use std::{io, task};
-
-pin_project! {
-    /// We use this future to mark certain methods
-    /// as callable in both sync and async modes.
-    #[repr(transparent)]
-    pub struct SyncFuture<S, T: Future> {
-        #[pin]
-        inner: T,
-        _marker: PhantomData<S>,
-    }
-}
-
-/// This wrapper lets us synchronously wait for inner future's completion
-/// (see [`SyncFuture::wait`]) **provided that `S` implements [`SyncProof`]**.
-/// For instance, `S` may be substituted with types implementing
-/// [`tokio::io::AsyncRead`], but it's not the only viable option.
-impl<S, T: Future> SyncFuture<S, T> {
-    /// NOTE: caller should carefully pick a type for `S`,
-    /// because we don't want to enable [`SyncFuture::wait`] when
-    /// it's in fact impossible to run the future synchronously.
-    /// Violation of this contract will not cause UB, but
-    /// panics and async event loop freezes won't please you.
-    ///
-    /// Example:
-    ///
-    /// ```
-    /// # use pq_proto::sync::SyncFuture;
-    /// # use std::future::Future;
-    /// # use tokio::io::AsyncReadExt;
-    /// #
-    /// // Parse a pair of numbers from a stream
-    /// pub fn parse_pair<Reader>(
-    ///     stream: &mut Reader,
-    /// ) -> SyncFuture<Reader, impl Future<Output = anyhow::Result<(u32, u64)>> + '_>
-    /// where
-    ///     Reader: tokio::io::AsyncRead + Unpin,
-    /// {
-    ///     // If `Reader` is a `SyncProof`, this will give caller
-    ///     // an opportunity to use `SyncFuture::wait`, because
-    ///     // `.await` will always result in `Poll::Ready`.
-    ///     SyncFuture::new(async move {
-    ///         let x = stream.read_u32().await?;
-    ///         let y = stream.read_u64().await?;
-    ///         Ok((x, y))
-    ///     })
-    /// }
-    /// ```
-    pub fn new(inner: T) -> Self {
-        Self {
-            inner,
-            _marker: PhantomData,
-        }
-    }
-}
-
-impl<S, T: Future> Future for SyncFuture<S, T> {
-    type Output = T::Output;
-
-    /// In async code, [`SyncFuture`] behaves like a regular wrapper.
-    #[inline(always)]
-    fn poll(self: Pin<&mut Self>, cx: &mut task::Context<'_>) -> task::Poll<Self::Output> {
-        self.project().inner.poll(cx)
-    }
-}
-
-/// Postulates that we can call [`SyncFuture::wait`].
-/// If implementer is also a [`Future`], it should always
-/// return [`task::Poll::Ready`] from [`Future::poll`].
-///
-/// Each implementation should document which futures
-/// specifically are being declared sync-proof.
-pub trait SyncPostulate {}
-
-impl<T: SyncPostulate> SyncPostulate for &T {}
-impl<T: SyncPostulate> SyncPostulate for &mut T {}
-
-impl<P: SyncPostulate, T: Future> SyncFuture<P, T> {
-    /// Synchronously wait for future completion.
-    pub fn wait(mut self) -> T::Output {
-        const RAW_WAKER: task::RawWaker = task::RawWaker::new(
-            std::ptr::null(),
-            &task::RawWakerVTable::new(
-                |_| RAW_WAKER,
-                |_| panic!("SyncFuture: failed to wake"),
-                |_| panic!("SyncFuture: failed to wake by ref"),
-                |_| { /* drop is no-op */ },
-            ),
-        );
-
-        // SAFETY: We never move `self` during this call;
-        // furthermore, it will be dropped in the end regardless of panics
-        let this = unsafe { Pin::new_unchecked(&mut self) };
-
-        // SAFETY: This waker doesn't do anything apart from panicking
-        let waker = unsafe { task::Waker::from_raw(RAW_WAKER) };
-        let context = &mut task::Context::from_waker(&waker);
-
-        match this.poll(context) {
-            task::Poll::Ready(res) => res,
-            _ => panic!("SyncFuture: unexpected pending!"),
-        }
-    }
-}
-
-/// This wrapper turns any [`std::io::Read`] into a blocking [`tokio::io::AsyncRead`],
-/// which lets us abstract over sync & async readers in methods returning [`SyncFuture`].
-/// NOTE: you **should not** use this in async code.
-#[repr(transparent)]
-pub struct AsyncishRead<T: io::Read + Unpin>(pub T);
-
-/// This lets us call [`SyncFuture<AsyncishRead<_>, _>::wait`],
-/// and allows the future to await on any of the [`AsyncRead`]
-/// and [`AsyncReadExt`] methods on `AsyncishRead`.
-impl<T: io::Read + Unpin> SyncPostulate for AsyncishRead<T> {}
-
-impl<T: io::Read + Unpin> tokio::io::AsyncRead for AsyncishRead<T> {
-    #[inline(always)]
-    fn poll_read(
-        mut self: Pin<&mut Self>,
-        _cx: &mut task::Context<'_>,
-        buf: &mut tokio::io::ReadBuf<'_>,
-    ) -> task::Poll<io::Result<()>> {
-        task::Poll::Ready(
-            // `Read::read` will block, meaning we don't need a real event loop!
-            self.0
-                .read(buf.initialize_unfilled())
-                .map(|sz| buf.advance(sz)),
-        )
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use tokio::io::{AsyncReadExt, AsyncWriteExt};
-
-    // async helper(stream: &mut impl AsyncRead) -> io::Result<u32>
-    fn bytes_add<Reader>(
-        stream: &mut Reader,
-    ) -> SyncFuture<Reader, impl Future<Output = io::Result<u32>> + '_>
-    where
-        Reader: tokio::io::AsyncRead + Unpin,
-    {
-        SyncFuture::new(async move {
-            let a = stream.read_u32().await?;
-            let b = stream.read_u32().await?;
-            Ok(a + b)
-        })
-    }
-
-    #[test]
-    fn test_sync() {
-        let bytes = [100u32.to_be_bytes(), 200u32.to_be_bytes()].concat();
-        let res = bytes_add(&mut AsyncishRead(&mut &bytes[..]))
-            .wait()
-            .unwrap();
-        assert_eq!(res, 300);
-    }
-
-    // We need a single-threaded executor for this test
-    #[tokio::test(flavor = "current_thread")]
-    async fn test_async() {
-        let (mut tx, mut rx) = tokio::net::UnixStream::pair().unwrap();
-
-        let write = async move {
-            tx.write_u32(100).await?;
-            tx.write_u32(200).await?;
-            Ok(())
-        };
-
-        let (res, ()) = tokio::try_join!(bytes_add(&mut rx), write).unwrap();
-        assert_eq!(res, 300);
-    }
-}

libs/remote_storage/Cargo.toml

@@ -21,8 +21,9 @@ toml_edit.workspace = true
 tracing.workspace = true
 metrics.workspace = true
 utils.workspace = true
-
+pin-project-lite.workspace = true
 workspace_hack.workspace = true
 
 [dev-dependencies]
 tempfile.workspace = true
+test-context.workspace = true

libs/remote_storage/src/lib.rs

@@ -13,7 +13,6 @@ use std::{
     collections::HashMap,
     fmt::Debug,
     num::{NonZeroU32, NonZeroUsize},
-    ops::Deref,
     path::{Path, PathBuf},
     pin::Pin,
     sync::Arc,
@@ -39,6 +38,9 @@ pub const DEFAULT_REMOTE_STORAGE_MAX_SYNC_ERRORS: u32 = 10;
 /// ~3500 PUT/COPY/POST/DELETE or 5500 GET/HEAD S3 requests
 /// https://aws.amazon.com/premiumsupport/knowledge-center/s3-request-limit-avoid-throttling/
 pub const DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT: usize = 100;
+/// No limits on the client side, which currenltly means 1000 for AWS S3.
+/// https://docs.aws.amazon.com/AmazonS3/latest/API/API_ListObjectsV2.html#API_ListObjectsV2_RequestSyntax
+pub const DEFAULT_MAX_KEYS_PER_LIST_RESPONSE: Option<i32> = None;
 
 const REMOTE_STORAGE_PREFIX_SEPARATOR: char = '/';
 
@@ -64,6 +66,10 @@ impl RemotePath {
     pub fn object_name(&self) -> Option<&str> {
         self.0.file_name().and_then(|os_str| os_str.to_str())
     }
+
+    pub fn join(&self, segment: &Path) -> Self {
+        Self(self.0.join(segment))
+    }
 }
 
 /// Storage (potentially remote) API to manage its state.
@@ -71,9 +77,6 @@ impl RemotePath {
 /// providing basic CRUD operations for storage files.
 #[async_trait::async_trait]
 pub trait RemoteStorage: Send + Sync + 'static {
-    /// Lists all items the storage has right now.
-    async fn list(&self) -> anyhow::Result<Vec<RemotePath>>;
-
     /// Lists all top level subdirectories for a given prefix
     /// Note: here we assume that if the prefix is passed it was obtained via remote_object_id
     /// which already takes into account any kind of global prefix (prefix_in_bucket for S3 or storage_root for LocalFS)
@@ -86,7 +89,7 @@ pub trait RemoteStorage: Send + Sync + 'static {
     /// Streams the local file contents into remote into the remote storage entry.
     async fn upload(
         &self,
-        data: Box<(dyn io::AsyncRead + Unpin + Send + Sync + 'static)>,
+        from: impl io::AsyncRead + Unpin + Send + Sync + 'static,
         // S3 PUT request requires the content length to be specified,
         // otherwise it starts to fail with the concurrent connection count increasing.
         data_size_bytes: usize,
@@ -111,7 +114,7 @@ pub trait RemoteStorage: Send + Sync + 'static {
 }
 
 pub struct Download {
-    pub download_stream: Pin<Box<dyn io::AsyncRead + Unpin + Send>>,
+    pub download_stream: Pin<Box<dyn io::AsyncRead + Unpin + Send + Sync>>,
     /// Extra key-value data, associated with the current remote file.
     pub metadata: Option<StorageMetadata>,
 }
@@ -157,14 +160,67 @@ pub enum GenericRemoteStorage {
     Unreliable(Arc<UnreliableWrapper>),
 }
 
-impl Deref for GenericRemoteStorage {
-    type Target = dyn RemoteStorage;
-
-    fn deref(&self) -> &Self::Target {
+impl GenericRemoteStorage {
+    pub async fn list_prefixes(
+        &self,
+        prefix: Option<&RemotePath>,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
         match self {
-            GenericRemoteStorage::LocalFs(local_fs) => local_fs,
-            GenericRemoteStorage::AwsS3(s3_bucket) => s3_bucket.as_ref(),
-            GenericRemoteStorage::Unreliable(s) => s.as_ref(),
+            Self::LocalFs(s) => s.list_prefixes(prefix).await,
+            Self::AwsS3(s) => s.list_prefixes(prefix).await,
+            Self::Unreliable(s) => s.list_prefixes(prefix).await,
+        }
+    }
+
+    pub async fn upload(
+        &self,
+        from: impl io::AsyncRead + Unpin + Send + Sync + 'static,
+        data_size_bytes: usize,
+        to: &RemotePath,
+        metadata: Option<StorageMetadata>,
+    ) -> anyhow::Result<()> {
+        match self {
+            Self::LocalFs(s) => s.upload(from, data_size_bytes, to, metadata).await,
+            Self::AwsS3(s) => s.upload(from, data_size_bytes, to, metadata).await,
+            Self::Unreliable(s) => s.upload(from, data_size_bytes, to, metadata).await,
+        }
+    }
+
+    pub async fn download(&self, from: &RemotePath) -> Result<Download, DownloadError> {
+        match self {
+            Self::LocalFs(s) => s.download(from).await,
+            Self::AwsS3(s) => s.download(from).await,
+            Self::Unreliable(s) => s.download(from).await,
+        }
+    }
+
+    pub async fn download_byte_range(
+        &self,
+        from: &RemotePath,
+        start_inclusive: u64,
+        end_exclusive: Option<u64>,
+    ) -> Result<Download, DownloadError> {
+        match self {
+            Self::LocalFs(s) => {
+                s.download_byte_range(from, start_inclusive, end_exclusive)
+                    .await
+            }
+            Self::AwsS3(s) => {
+                s.download_byte_range(from, start_inclusive, end_exclusive)
+                    .await
+            }
+            Self::Unreliable(s) => {
+                s.download_byte_range(from, start_inclusive, end_exclusive)
+                    .await
+            }
+        }
+    }
+
+    pub async fn delete(&self, path: &RemotePath) -> anyhow::Result<()> {
+        match self {
+            Self::LocalFs(s) => s.delete(path).await,
+            Self::AwsS3(s) => s.delete(path).await,
+            Self::Unreliable(s) => s.delete(path).await,
         }
     }
 }
@@ -195,7 +251,7 @@ impl GenericRemoteStorage {
     /// this path is used for the remote object id conversion only.
     pub async fn upload_storage_object(
         &self,
-        from: Box<dyn tokio::io::AsyncRead + Unpin + Send + Sync + 'static>,
+        from: impl tokio::io::AsyncRead + Unpin + Send + Sync + 'static,
         from_size_bytes: usize,
         to: &RemotePath,
     ) -> anyhow::Result<()> {
@@ -266,6 +322,7 @@ pub struct S3Config {
     /// AWS S3 has various limits on its API calls, we need not to exceed those.
     /// See [`DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT`] for more details.
     pub concurrency_limit: NonZeroUsize,
+    pub max_keys_per_list_response: Option<i32>,
 }
 
 impl Debug for S3Config {
@@ -275,6 +332,10 @@ impl Debug for S3Config {
             .field("bucket_region", &self.bucket_region)
             .field("prefix_in_bucket", &self.prefix_in_bucket)
             .field("concurrency_limit", &self.concurrency_limit)
+            .field(
+                "max_keys_per_list_response",
+                &self.max_keys_per_list_response,
+            )
             .finish()
     }
 }
@@ -303,6 +364,11 @@ impl RemoteStorageConfig {
         )
         .context("Failed to parse 'concurrency_limit' as a positive integer")?;
 
+        let max_keys_per_list_response =
+            parse_optional_integer::<i32, _>("max_keys_per_list_response", toml)
+                .context("Failed to parse 'max_keys_per_list_response' as a positive integer")?
+                .or(DEFAULT_MAX_KEYS_PER_LIST_RESPONSE);
+
         let storage = match (local_path, bucket_name, bucket_region) {
             // no 'local_path' nor 'bucket_name' options are provided, consider this remote storage disabled
             (None, None, None) => return Ok(None),
@@ -324,6 +390,7 @@ impl RemoteStorageConfig {
                     .map(|endpoint| parse_toml_string("endpoint", endpoint))
                     .transpose()?,
                 concurrency_limit,
+                max_keys_per_list_response,
             }),
             (Some(local_path), None, None) => RemoteStorageKind::LocalFs(PathBuf::from(
                 parse_toml_string("local_path", local_path)?,

libs/remote_storage/src/local_fs.rs

@@ -73,10 +73,8 @@ impl LocalFs {
             Ok(None)
         }
     }
-}
 
-#[async_trait::async_trait]
-impl RemoteStorage for LocalFs {
+    #[cfg(test)]
     async fn list(&self) -> anyhow::Result<Vec<RemotePath>> {
         Ok(get_all_files(&self.storage_root, true)
             .await?
@@ -91,7 +89,10 @@ impl RemoteStorage for LocalFs {
             })
             .collect())
     }
+}
 
+#[async_trait::async_trait]
+impl RemoteStorage for LocalFs {
     async fn list_prefixes(
         &self,
         prefix: Option<&RemotePath>,
@@ -117,7 +118,7 @@ impl RemoteStorage for LocalFs {
 
     async fn upload(
         &self,
-        data: Box<(dyn io::AsyncRead + Unpin + Send + Sync + 'static)>,
+        data: impl io::AsyncRead + Unpin + Send + Sync + 'static,
         data_size_bytes: usize,
         to: &RemotePath,
         metadata: Option<StorageMetadata>,