BENCHMARK

admin@neon-loadtest:[~/neon]: ./target/release/pagebench get-page-latest-lsn --page-service-connstring 'postgres://postgres@localhost:15004' --mgmt-api-endpoint http://localhost:15005 --limit-to-first-n-targets 1 --num-clients 128 --runtime 10s
2024-03-22T21:39:21.716695Z  INFO timelines:
[TenantTimelineId { tenant_id: 0020aefd92542de004d4003bbc25153f, timeline_id: b854dccb715605809e98bb1106a171fa }]
2024-03-22T21:39:21.716718Z  INFO number of timelines:
1
2024-03-22T21:39:21.745957Z  INFO filter duration: 0.000499642
2024-03-22T21:39:21.745978Z  INFO spawning workers
2024-03-22T21:39:21.755971Z  INFO waiting for everything to become ready
2024-03-22T21:39:21.756341Z  INFO work started
2024-03-22T21:39:22.756691Z  INFO RPS: 16894   MISSED: 0
2024-03-22T21:39:23.757933Z  INFO RPS: 15817   MISSED: 0
2024-03-22T21:39:24.758885Z  INFO RPS: 16826   MISSED: 0
2024-03-22T21:39:25.759676Z  INFO RPS: 16750   MISSED: 0
2024-03-22T21:39:26.760710Z  INFO RPS: 17200   MISSED: 0
2024-03-22T21:39:27.761706Z  INFO RPS: 16952   MISSED: 0
2024-03-22T21:39:28.762708Z  INFO RPS: 15921   MISSED: 0
2024-03-22T21:39:29.763674Z  INFO RPS: 16960   MISSED: 0
2024-03-22T21:39:30.764704Z  INFO RPS: 17080   MISSED: 0
2024-03-22T21:39:31.756681Z  INFO runtime over, signalling cancellation
2024-03-22T21:39:31.766285Z  INFO RPS: 16799   MISSED: 0
2024-03-22T21:39:31.768556Z  INFO work sender exited
{
  "total": {
    "request_count": 167356,
    "latency_mean": "7ms 651us",
    "latency_percentiles": {
      "p95": "13ms 671us",
      "p99": "16ms 863us",
      "p99.9": "21ms 567us",
      "p99.99": "26ms 543us"
    }
  }
}

.dockerignore

@@ -22,7 +22,6 @@
 !s3_scrubber/
 !safekeeper/
 !storage_broker/
-!storage_controller/
 !trace/
 !vendor/postgres-*/
 !workspace_hack/

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

@@ -150,7 +150,7 @@ runs:
 
         # Use aws s3 cp (instead of aws s3 sync) to keep files from previous runs to make old URLs work,
         # and to keep files on the host to upload them to the database
-        time s5cmd --log error cp "${WORKDIR}/report/*" "s3://${BUCKET}/${REPORT_PREFIX}/${GITHUB_RUN_ID}/"
+        time aws s3 cp --recursive --only-show-errors "${WORKDIR}/report" "s3://${BUCKET}/${REPORT_PREFIX}/${GITHUB_RUN_ID}"
 
         # Generate redirect
         cat <<EOF > ${WORKDIR}/index.html

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

@@ -10,7 +10,7 @@ inputs:
     required: true
   api_host:
     desctiption: 'Neon API host'
-    default: console-stage.neon.build
+    default: console.stage.neon.tech
 outputs:
   dsn:
     description: 'Created Branch DSN (for main database)'

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

@@ -13,7 +13,7 @@ inputs:
     required: true
   api_host:
     desctiption: 'Neon API host'
-    default: console-stage.neon.build
+    default: console.stage.neon.tech
 
 runs:
   using: "composite"

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

@@ -13,7 +13,7 @@ inputs:
     default: 15
   api_host:
     desctiption: 'Neon API host'
-    default: console-stage.neon.build
+    default: console.stage.neon.tech
   provisioner:
     desctiption: 'k8s-pod or k8s-neonvm'
     default: 'k8s-pod'

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

@@ -10,7 +10,7 @@ inputs:
     required: true
   api_host:
     desctiption: 'Neon API host'
-    default: console-stage.neon.build
+    default: console.stage.neon.tech
 
 runs:
   using: "composite"

.github/workflows/approved-for-ci-run.yml

@@ -18,7 +18,6 @@ on:
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.event.pull_request.number }}
-  cancel-in-progress: false
 
 env:
   GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/benchmarking.yml

@@ -147,16 +147,15 @@ jobs:
             "neonvm-captest-new"
           ],
           "db_size": [ "10gb" ],
-          "include": [{ "platform": "neon-captest-freetier",         "db_size": "3gb"  },
-                      { "platform": "neon-captest-new",              "db_size": "50gb" },
-                      { "platform": "neonvm-captest-freetier",       "db_size": "3gb"  },
-                      { "platform": "neonvm-captest-new",            "db_size": "50gb" },
-                      { "platform": "neonvm-captest-sharding-reuse", "db_size": "50gb" }]
+          "include": [{ "platform": "neon-captest-freetier",   "db_size": "3gb"  },
+                      { "platform": "neon-captest-new",        "db_size": "50gb" },
+                      { "platform": "neonvm-captest-freetier", "db_size": "3gb"  },
+                      { "platform": "neonvm-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"}]')
+                                                   { "platform": "rds-aurora",   "db_size": "50gb"}]')
         fi
 
         echo "matrix=$(echo "$matrix" | jq --compact-output '.')" >> $GITHUB_OUTPUT
@@ -172,7 +171,7 @@ jobs:
 
         if [ "$(date +%A)" = "Saturday" ] || [ ${RUN_AWS_RDS_AND_AURORA} = "true" ]; then
           matrix=$(echo "$matrix" | jq '.include += [{ "platform": "rds-postgres" },
-                                                     { "platform": "rds-aurora"   }]')
+                                                   { "platform": "rds-aurora"   }]')
         fi
 
         echo "matrix=$(echo "$matrix" | jq --compact-output '.')" >> $GITHUB_OUTPUT
@@ -191,7 +190,7 @@ jobs:
 
         if [ "$(date +%A)" = "Saturday" ] || [ ${RUN_AWS_RDS_AND_AURORA} = "true" ]; then
           matrix=$(echo "$matrix" | jq '.include += [{ "platform": "rds-postgres", "scale": "10" },
-                                                     { "platform": "rds-aurora",   "scale": "10" }]')
+                                                    { "platform": "rds-aurora",   "scale": "10" }]')
         fi
 
         echo "matrix=$(echo "$matrix" | jq --compact-output '.')" >> $GITHUB_OUTPUT
@@ -254,9 +253,6 @@ jobs:
           neon-captest-reuse)
             CONNSTR=${{ secrets.BENCHMARK_CAPTEST_CONNSTR }}
             ;;
-          neonvm-captest-sharding-reuse)
-            CONNSTR=${{ secrets.BENCHMARK_CAPTEST_SHARDING_CONNSTR }}
-            ;;
           neon-captest-new | neon-captest-freetier | neonvm-captest-new | neonvm-captest-freetier)
             CONNSTR=${{ steps.create-neon-project.outputs.dsn }}
             ;;
@@ -274,15 +270,11 @@ jobs:
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERIES=("SELECT version()")
+        QUERY="SELECT version();"
         if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERIES+=("SHOW neon.tenant_id")
-          QUERIES+=("SHOW neon.timeline_id")
+          QUERY="${QUERY} SHOW neon.tenant_id; SHOW neon.timeline_id;"
         fi
-
-        for q in "${QUERIES[@]}"; do
-          psql ${CONNSTR} -c "${q}"
-        done
+        psql ${CONNSTR} -c "${QUERY}"
 
     - name: Benchmark init
       uses: ./.github/actions/run-python-test-set
@@ -409,15 +401,11 @@ jobs:
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERIES=("SELECT version()")
+        QUERY="SELECT version();"
         if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERIES+=("SHOW neon.tenant_id")
-          QUERIES+=("SHOW neon.timeline_id")
+          QUERY="${QUERY} SHOW neon.tenant_id; SHOW neon.timeline_id;"
         fi
-
-        for q in "${QUERIES[@]}"; do
-          psql ${CONNSTR} -c "${q}"
-        done
+        psql ${CONNSTR} -c "${QUERY}"
 
     - name: ClickBench benchmark
       uses: ./.github/actions/run-python-test-set
@@ -519,15 +507,11 @@ jobs:
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERIES=("SELECT version()")
+        QUERY="SELECT version();"
         if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERIES+=("SHOW neon.tenant_id")
-          QUERIES+=("SHOW neon.timeline_id")
+          QUERY="${QUERY} SHOW neon.tenant_id; SHOW neon.timeline_id;"
         fi
-
-        for q in "${QUERIES[@]}"; do
-          psql ${CONNSTR} -c "${q}"
-        done
+        psql ${CONNSTR} -c "${QUERY}"
 
     - name: Run TPC-H benchmark
       uses: ./.github/actions/run-python-test-set
@@ -613,15 +597,11 @@ jobs:
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERIES=("SELECT version()")
+        QUERY="SELECT version();"
         if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERIES+=("SHOW neon.tenant_id")
-          QUERIES+=("SHOW neon.timeline_id")
+          QUERY="${QUERY} SHOW neon.tenant_id; SHOW neon.timeline_id;"
         fi
-
-        for q in "${QUERIES[@]}"; do
-          psql ${CONNSTR} -c "${q}"
-        done
+        psql ${CONNSTR} -c "${QUERY}"
 
     - name: Run user examples
       uses: ./.github/actions/run-python-test-set

.github/workflows/build-build-tools-image.yml

@@ -21,7 +21,6 @@ defaults:
 
 concurrency:
   group: build-build-tools-image-${{ inputs.image-tag }}
-  cancel-in-progress: false
 
 # No permission for GITHUB_TOKEN by default; the **minimal required** set of permissions should be granted in each job.
 permissions: {}

.github/workflows/build_and_test.yml

@@ -735,7 +735,7 @@ jobs:
         run: |
           mkdir -p .docker-custom
           echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
-      - uses: docker/setup-buildx-action@v2
+      - uses: docker/setup-buildx-action@v3
 
       - uses: docker/login-action@v3
         with:
@@ -792,7 +792,7 @@ jobs:
         run: |
           mkdir -p .docker-custom
           echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
-      - uses: docker/setup-buildx-action@v2
+      - uses: docker/setup-buildx-action@v3
         with:
           # Disable parallelism for docker buildkit.
           # As we already build everything with `make -j$(nproc)`, running it in additional level of parallelisam blows up the Runner.
@@ -865,7 +865,7 @@ jobs:
       run:
         shell: sh -eu {0}
     env:
-      VM_BUILDER_VERSION: v0.28.1
+      VM_BUILDER_VERSION: v0.23.2
 
     steps:
       - name: Checkout
@@ -1121,34 +1121,18 @@ jobs:
         run: |
           if [[ "$GITHUB_REF_NAME" == "main" ]]; then
             gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}} -f deployPreprodRegion=false
+
+            # TODO: move deployPreprodRegion to release (`"$GITHUB_REF_NAME" == "release"` block), once Staging support different compute tag prefixes for different regions
+            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}} -f deployPreprodRegion=true
           elif [[ "$GITHUB_REF_NAME" == "release" ]]; then
-            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main \
+            gh workflow --repo neondatabase/aws run deploy-prod.yml --ref main \
               -f deployPgSniRouter=false \
               -f deployProxy=false \
               -f deployStorage=true \
               -f deployStorageBroker=true \
-              -f deployStorageController=true \
-              -f branch=main \
-              -f dockerTag=${{needs.tag.outputs.build-tag}} \
-              -f deployPreprodRegion=true
-
-            gh workflow --repo neondatabase/aws run deploy-prod.yml --ref main \
-              -f deployStorage=true \
-              -f deployStorageBroker=true \
-              -f deployStorageController=true \
               -f branch=main \
               -f dockerTag=${{needs.tag.outputs.build-tag}}
           elif [[ "$GITHUB_REF_NAME" == "release-proxy" ]]; then
-            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main \
-              -f deployPgSniRouter=true \
-              -f deployProxy=true \
-              -f deployStorage=false \
-              -f deployStorageBroker=false \
-              -f deployStorageController=false \
-              -f branch=main \
-              -f dockerTag=${{needs.tag.outputs.build-tag}} \
-              -f deployPreprodRegion=true
-
             gh workflow --repo neondatabase/aws run deploy-proxy-prod.yml --ref main \
               -f deployPgSniRouter=true \
               -f deployProxy=true \

.github/workflows/check-build-tools-image.yml

@@ -28,9 +28,7 @@ jobs:
       - name: Get build-tools image tag for the current commit
         id: get-build-tools-tag
         env:
-          # Usually, for COMMIT_SHA, we use `github.event.pull_request.head.sha || github.sha`, but here, even for PRs,
-          # we want to use `github.sha` i.e. point to a phantom merge commit to determine the image tag correctly.
-          COMMIT_SHA: ${{ github.sha }}
+          COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         run: |
           LAST_BUILD_TOOLS_SHA=$(

.github/workflows/pin-build-tools-image.yml

@@ -20,7 +20,6 @@ defaults:
 
 concurrency:
   group: pin-build-tools-image-${{ inputs.from-tag }}
-  cancel-in-progress: false
 
 permissions: {}
 

.github/workflows/trigger-e2e-tests.yml

@@ -62,14 +62,14 @@ jobs:
 
   trigger-e2e-tests:
     needs: [ tag ]
-    runs-on: ubuntu-latest
+    runs-on: [ self-hosted, gen3, small ]
     env:
       TAG: ${{ needs.tag.outputs.build-tag }}
+    container:
+      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/base:pinned
+      options: --init
     steps:
       - name: check if ecr image are present
-        env:
-          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
         run: |
           for REPO in neon compute-tools compute-node-v14 vm-compute-node-v14 compute-node-v15 vm-compute-node-v15 compute-node-v16 vm-compute-node-v16; do
             OUTPUT=$(aws ecr describe-images --repository-name ${REPO} --region eu-central-1 --query "imageDetails[?imageTags[?contains(@, '${TAG}')]]" --output text)
@@ -79,55 +79,41 @@ jobs:
             fi
           done
 
-      - name: Set e2e-platforms
-        id: e2e-platforms
-        env:
-          PR_NUMBER: ${{ github.event.pull_request.number }}
-          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          # Default set of platforms to run e2e tests on
-          platforms='["docker", "k8s"]'
-
-          # If the PR changes vendor/, pgxn/ or libs/vm_monitor/ directories, or Dockerfile.compute-node, add k8s-neonvm to the list of platforms.
-          # If the workflow run is not a pull request, add k8s-neonvm to the list.
-          if [ "$GITHUB_EVENT_NAME" == "pull_request" ]; then
-            for f in $(gh api "/repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}/files" --paginate --jq '.[].filename'); do
-              case "$f" in
-                vendor/*|pgxn/*|libs/vm_monitor/*|Dockerfile.compute-node)
-                  platforms=$(echo "${platforms}" | jq --compact-output '. += ["k8s-neonvm"] | unique')
-                  ;;
-                *)
-                  # no-op
-                  ;;
-              esac
-            done
-          else
-            platforms=$(echo "${platforms}" | jq --compact-output '. += ["k8s-neonvm"] | unique')
-          fi
-
-          echo "e2e-platforms=${platforms}" | tee -a $GITHUB_OUTPUT
-
       - name: Set PR's status to pending and request a remote CI test
-        env:
-          E2E_PLATFORMS: ${{ steps.e2e-platforms.outputs.e2e-platforms }}
-          COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
-          GH_TOKEN: ${{ secrets.CI_ACCESS_TOKEN }}
         run: |
-          REMOTE_REPO="${GITHUB_REPOSITORY_OWNER}/cloud"
+          # For pull requests, GH Actions set "github.sha" variable to point at a fake merge commit
+          # but we need to use a real sha of a latest commit in the PR's branch for the e2e job,
+          # to place a job run status update later.
+          COMMIT_SHA=${{ github.event.pull_request.head.sha }}
+          # For non-PR kinds of runs, the above will produce an empty variable, pick the original sha value for those
+          COMMIT_SHA=${COMMIT_SHA:-${{ github.sha }}}
 
-          gh api "/repos/${GITHUB_REPOSITORY}/statuses/${COMMIT_SHA}" \
-            --method POST \
-            --raw-field "state=pending" \
-            --raw-field "description=[$REMOTE_REPO] Remote CI job is about to start" \
-            --raw-field "context=neon-cloud-e2e"
+          REMOTE_REPO="${{ github.repository_owner }}/cloud"
 
-          gh workflow --repo ${REMOTE_REPO} \
-            run testing.yml \
-              --ref "main" \
-              --raw-field "ci_job_name=neon-cloud-e2e" \
-              --raw-field "commit_hash=$COMMIT_SHA" \
-              --raw-field "remote_repo=${GITHUB_REPOSITORY}" \
-              --raw-field "storage_image_tag=${TAG}" \
-              --raw-field "compute_image_tag=${TAG}" \
-              --raw-field "concurrency_group=${E2E_CONCURRENCY_GROUP}" \
-              --raw-field "e2e-platforms=${E2E_PLATFORMS}"
+          curl -f -X POST \
+          https://api.github.com/repos/${{ github.repository }}/statuses/$COMMIT_SHA \
+          -H "Accept: application/vnd.github.v3+json" \
+          --user "${{ secrets.CI_ACCESS_TOKEN }}" \
+          --data \
+            "{
+              \"state\": \"pending\",
+              \"context\": \"neon-cloud-e2e\",
+              \"description\": \"[$REMOTE_REPO] Remote CI job is about to start\"
+            }"
+
+          curl -f -X POST \
+          https://api.github.com/repos/$REMOTE_REPO/actions/workflows/testing.yml/dispatches \
+          -H "Accept: application/vnd.github.v3+json" \
+          --user "${{ secrets.CI_ACCESS_TOKEN }}" \
+          --data \
+            "{
+              \"ref\": \"main\",
+              \"inputs\": {
+                \"ci_job_name\": \"neon-cloud-e2e\",
+                \"commit_hash\": \"$COMMIT_SHA\",
+                \"remote_repo\": \"${{ github.repository }}\",
+                \"storage_image_tag\": \"${TAG}\",
+                \"compute_image_tag\": \"${TAG}\",
+                \"concurrency_group\": \"${{ env.E2E_CONCURRENCY_GROUP }}\"
+              }
+            }"

CODEOWNERS

@@ -1,5 +1,5 @@
 /compute_tools/ @neondatabase/control-plane @neondatabase/compute
-/storage_controller @neondatabase/storage
+/control_plane/attachment_service @neondatabase/storage
 /libs/pageserver_api/ @neondatabase/storage
 /libs/postgres_ffi/ @neondatabase/compute @neondatabase/safekeepers
 /libs/remote_storage/ @neondatabase/storage

Cargo.toml

@@ -3,7 +3,7 @@ resolver = "2"
 members = [
     "compute_tools",
     "control_plane",
-    "control_plane/storcon_cli",
+    "control_plane/attachment_service",
     "pageserver",
     "pageserver/compaction",
     "pageserver/ctl",
@@ -12,7 +12,6 @@ members = [
     "proxy",
     "safekeeper",
     "storage_broker",
-    "storage_controller",
     "s3_scrubber",
     "workspace_hack",
     "trace",
@@ -44,7 +43,6 @@ license = "Apache-2.0"
 anyhow = { version = "1.0", features = ["backtrace"] }
 arc-swap = "1.6"
 async-compression = { version = "0.4.0", features = ["tokio", "gzip", "zstd"] }
-atomic-take = "1.1.0"
 azure_core = "0.18"
 azure_identity = "0.18"
 azure_storage = "0.18"
@@ -54,6 +52,7 @@ async-stream = "0.3"
 async-trait = "0.1"
 aws-config = { version = "1.1.4", default-features = false, features=["rustls"] }
 aws-sdk-s3 = "1.14"
+aws-sdk-secretsmanager = { version = "1.14.0" }
 aws-sdk-iam = "1.15.0"
 aws-smithy-async = { version = "1.1.4", default-features = false, features=["rt-tokio"] }
 aws-smithy-types = "1.1.4"
@@ -80,7 +79,6 @@ either = "1.8"
 enum-map = "2.4.2"
 enumset = "1.0.12"
 fail = "0.5.0"
-fallible-iterator = "0.2"
 fs2 = "0.4.3"
 futures = "0.3"
 futures-core = "0.3"
@@ -98,7 +96,7 @@ http-types = { version = "2", default-features = false }
 humantime = "2.1"
 humantime-serde = "1.1.1"
 hyper = "0.14"
-hyper-tungstenite = "0.13.0"
+hyper-tungstenite = "0.11"
 inotify = "0.10.2"
 ipnet = "2.9.0"
 itertools = "0.10"
@@ -107,8 +105,7 @@ lasso = "0.7"
 leaky-bucket = "1.0.1"
 libc = "0.2"
 md5 = "0.7.0"
-measured = { version = "0.0.21", features=["lasso"] }
-measured-process = { version = "0.0.21" }
+measured = { version = "0.0.13", features=["default", "lasso"] }
 memoffset = "0.8"
 native-tls = "0.2"
 nix = { version = "0.27", features = ["fs", "process", "socket", "signal", "poll"] }
@@ -157,12 +154,11 @@ socket2 = "0.5"
 strum = "0.24"
 strum_macros = "0.24"
 "subtle"  = "2.5.0"
-# https://github.com/nical/rust_debug/pull/4
-svg_fmt = { git = "https://github.com/neondatabase/fork--nical--rust_debug", branch = "neon" }
+svg_fmt = "0.4.1"
 sync_wrapper = "0.1.2"
 tar = "0.4"
 task-local-extensions = "0.1.4"
-test-context = "0.3"
+test-context = "0.1"
 thiserror = "1.0"
 tikv-jemallocator = "0.5"
 tikv-jemalloc-ctl = "0.5"

Dockerfile.build-tools

@@ -58,12 +58,6 @@ RUN curl -fsSL "https://github.com/protocolbuffers/protobuf/releases/download/v$
     && mv protoc/include/google /usr/local/include/google \
     && rm -rf protoc.zip protoc
 
-# s5cmd
-ENV S5CMD_VERSION=2.2.2
-RUN curl -sL "https://github.com/peak/s5cmd/releases/download/v${S5CMD_VERSION}/s5cmd_${S5CMD_VERSION}_Linux-$(uname -m | sed 's/x86_64/64bit/g' | sed 's/aarch64/arm64/g').tar.gz" | tar zxvf - s5cmd \
-    && chmod +x s5cmd \
-    && mv s5cmd /usr/local/bin/s5cmd
-
 # LLVM
 ENV LLVM_VERSION=17
 RUN curl -fsSL 'https://apt.llvm.org/llvm-snapshot.gpg.key' | apt-key add - \

Dockerfile.compute-node

@@ -944,9 +944,6 @@ RUN mkdir /var/db && useradd -m -d /var/db/postgres postgres && \
 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
 
-# Create remote extension download directory
-RUN mkdir /usr/local/download_extensions && chown -R postgres:postgres /usr/local/download_extensions
-
 # Install:
 # libreadline8 for psql
 # libicu67, locales for collations (including ICU and plpgsql_check)

compute_tools/src/compute.rs

@@ -818,15 +818,9 @@ impl ComputeNode {
                         Client::connect(zenith_admin_connstr.as_str(), NoTls)
                             .context("broken cloud_admin credential: tried connecting with cloud_admin but could not authenticate, and zenith_admin does not work either")?;
                     // Disable forwarding so that users don't get a cloud_admin role
-
-                    let mut func = || {
-                        client.simple_query("SET neon.forward_ddl = false")?;
-                        client.simple_query("CREATE USER cloud_admin WITH SUPERUSER")?;
-                        client.simple_query("GRANT zenith_admin TO cloud_admin")?;
-                        Ok::<_, anyhow::Error>(())
-                    };
-                    func().context("apply_config setup cloud_admin")?;
-
+                    client.simple_query("SET neon.forward_ddl = false")?;
+                    client.simple_query("CREATE USER cloud_admin WITH SUPERUSER")?;
+                    client.simple_query("GRANT zenith_admin TO cloud_admin")?;
                     drop(client);
 
                     // reconnect with connstring with expected name
@@ -838,29 +832,24 @@ impl ComputeNode {
         };
 
         // Disable DDL forwarding because control plane already knows about these roles/databases.
-        client
-            .simple_query("SET neon.forward_ddl = false")
-            .context("apply_config SET neon.forward_ddl = false")?;
+        client.simple_query("SET neon.forward_ddl = false")?;
 
         // Proceed with post-startup configuration. Note, that order of operations is important.
         let spec = &compute_state.pspec.as_ref().expect("spec must be set").spec;
-        create_neon_superuser(spec, &mut client).context("apply_config create_neon_superuser")?;
-        cleanup_instance(&mut client).context("apply_config cleanup_instance")?;
-        handle_roles(spec, &mut client).context("apply_config handle_roles")?;
-        handle_databases(spec, &mut client).context("apply_config handle_databases")?;
-        handle_role_deletions(spec, connstr.as_str(), &mut client)
-            .context("apply_config handle_role_deletions")?;
+        create_neon_superuser(spec, &mut client)?;
+        cleanup_instance(&mut client)?;
+        handle_roles(spec, &mut client)?;
+        handle_databases(spec, &mut client)?;
+        handle_role_deletions(spec, connstr.as_str(), &mut client)?;
         handle_grants(
             spec,
             &mut client,
             connstr.as_str(),
             self.has_feature(ComputeFeature::AnonExtension),
-        )
-        .context("apply_config handle_grants")?;
-        handle_extensions(spec, &mut client).context("apply_config handle_extensions")?;
-        handle_extension_neon(&mut client).context("apply_config handle_extension_neon")?;
-        create_availability_check_data(&mut client)
-            .context("apply_config create_availability_check_data")?;
+        )?;
+        handle_extensions(spec, &mut client)?;
+        handle_extension_neon(&mut client)?;
+        create_availability_check_data(&mut client)?;
 
         // 'Close' connection
         drop(client);
@@ -868,7 +857,7 @@ impl ComputeNode {
         // Run migrations separately to not hold up cold starts
         thread::spawn(move || {
             let mut client = Client::connect(connstr.as_str(), NoTls)?;
-            handle_migrations(&mut client).context("apply_config handle_migrations")
+            handle_migrations(&mut client)
         });
         Ok(())
     }
@@ -1273,12 +1262,10 @@ LIMIT 100",
         .await
         .map_err(DownloadError::Other);
 
-        if download_size.is_ok() {
-            self.ext_download_progress
-                .write()
-                .expect("bad lock")
-                .insert(ext_archive_name.to_string(), (download_start, true));
-        }
+        self.ext_download_progress
+            .write()
+            .expect("bad lock")
+            .insert(ext_archive_name.to_string(), (download_start, true));
 
         download_size
     }

compute_tools/src/config.rs

@@ -6,8 +6,8 @@ use std::path::Path;
 use anyhow::Result;
 
 use crate::pg_helpers::escape_conf_value;
-use crate::pg_helpers::{GenericOptionExt, PgOptionsSerialize};
-use compute_api::spec::{ComputeMode, ComputeSpec, GenericOption};
+use crate::pg_helpers::PgOptionsSerialize;
+use compute_api::spec::{ComputeMode, ComputeSpec};
 
 /// Check that `line` is inside a text file and put it there if it is not.
 /// Create file if it doesn't exist.
@@ -92,27 +92,6 @@ pub fn write_postgres_conf(
         }
     }
 
-    if cfg!(target_os = "linux") {
-        // Check /proc/sys/vm/overcommit_memory -- if it equals 2 (i.e. linux memory overcommit is
-        // disabled), then the control plane has enabled swap and we should set
-        // dynamic_shared_memory_type = 'mmap'.
-        //
-        // This is (maybe?) temporary - for more, see https://github.com/neondatabase/cloud/issues/12047.
-        let overcommit_memory_contents = std::fs::read_to_string("/proc/sys/vm/overcommit_memory")
-            // ignore any errors - they may be expected to occur under certain situations (e.g. when
-            // not running in Linux).
-            .unwrap_or_else(|_| String::new());
-        if overcommit_memory_contents.trim() == "2" {
-            let opt = GenericOption {
-                name: "dynamic_shared_memory_type".to_owned(),
-                value: Some("mmap".to_owned()),
-                vartype: "enum".to_owned(),
-            };
-
-            write!(file, "{}", opt.to_pg_setting())?;
-        }
-    }
-
     // If there are any extra options in the 'settings' field, append those
     if spec.cluster.settings.is_some() {
         writeln!(file, "# Managed by compute_ctl: begin")?;

compute_tools/src/pg_helpers.rs

@@ -44,7 +44,7 @@ pub fn escape_conf_value(s: &str) -> String {
     format!("'{}'", res)
 }
 
-pub trait GenericOptionExt {
+trait GenericOptionExt {
     fn to_pg_option(&self) -> String;
     fn to_pg_setting(&self) -> String;
 }

compute_tools/src/spec.rs

@@ -2,7 +2,7 @@ use std::fs::File;
 use std::path::Path;
 use std::str::FromStr;
 
-use anyhow::{anyhow, bail, Context, Result};
+use anyhow::{anyhow, bail, Result};
 use postgres::config::Config;
 use postgres::{Client, NoTls};
 use reqwest::StatusCode;
@@ -302,9 +302,9 @@ pub fn handle_roles(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
             RoleAction::Create => {
                 // This branch only runs when roles are created through the console, so it is
                 // safe to add more permissions here. BYPASSRLS and REPLICATION are inherited
-                // from neon_superuser.
+                // from neon_superuser. (NOTE: REPLICATION has been removed from here for now).
                 let mut query: String = format!(
-                    "CREATE ROLE {} INHERIT CREATEROLE CREATEDB BYPASSRLS REPLICATION IN ROLE neon_superuser",
+                    "CREATE ROLE {} INHERIT CREATEROLE CREATEDB BYPASSRLS IN ROLE neon_superuser",
                     name.pg_quote()
                 );
                 info!("running role create query: '{}'", &query);
@@ -698,8 +698,7 @@ pub fn handle_grants(
 
         // it is important to run this after all grants
         if enable_anon_extension {
-            handle_extension_anon(spec, &db.owner, &mut db_client, false)
-                .context("handle_grants handle_extension_anon")?;
+            handle_extension_anon(spec, &db.owner, &mut db_client, false)?;
         }
     }
 
@@ -744,24 +743,21 @@ pub fn handle_extension_neon(client: &mut Client) -> Result<()> {
     // which may happen in two cases:
     // - extension was just installed
     // - extension was already installed and is up to date
-    let query = "ALTER EXTENSION neon UPDATE";
-    info!("update neon extension version with query: {}", query);
-    if let Err(e) = client.simple_query(query) {
-        error!(
-            "failed to upgrade neon extension during `handle_extension_neon`: {}",
-            e
-        );
-    }
+    // DISABLED due to compute node unpinning epic
+    // let query = "ALTER EXTENSION neon UPDATE";
+    // info!("update neon extension version with query: {}", query);
+    // client.simple_query(query)?;
 
     Ok(())
 }
 
 #[instrument(skip_all)]
-pub fn handle_neon_extension_upgrade(client: &mut Client) -> Result<()> {
-    info!("handle neon extension upgrade");
-    let query = "ALTER EXTENSION neon UPDATE";
-    info!("update neon extension version with query: {}", query);
-    client.simple_query(query)?;
+pub fn handle_neon_extension_upgrade(_client: &mut Client) -> Result<()> {
+    info!("handle neon extension upgrade (not really)");
+    // DISABLED due to compute node unpinning epic
+    // let query = "ALTER EXTENSION neon UPDATE";
+    // info!("update neon extension version with query: {}", query);
+    // client.simple_query(query)?;
 
     Ok(())
 }
@@ -810,40 +806,43 @@ $$;"#,
         "",
         "",
         "",
-        "",
         // Add new migrations below.
+        r#"
+DO $$
+DECLARE
+    role_name TEXT;
+BEGIN
+    FOR role_name IN SELECT rolname FROM pg_roles WHERE rolreplication IS TRUE
+    LOOP
+        RAISE NOTICE 'EXECUTING ALTER ROLE % NOREPLICATION', quote_ident(role_name);
+        EXECUTE 'ALTER ROLE ' || quote_ident(role_name) || ' NOREPLICATION';
+    END LOOP;
+END
+$$;"#,
     ];
 
-    let mut func = || {
-        let query = "CREATE SCHEMA IF NOT EXISTS neon_migration";
-        client.simple_query(query)?;
+    let mut query = "CREATE SCHEMA IF NOT EXISTS neon_migration";
+    client.simple_query(query)?;
 
-        let query = "CREATE TABLE IF NOT EXISTS neon_migration.migration_id (key INT NOT NULL PRIMARY KEY, id bigint NOT NULL DEFAULT 0)";
-        client.simple_query(query)?;
+    query = "CREATE TABLE IF NOT EXISTS neon_migration.migration_id (key INT NOT NULL PRIMARY KEY, id bigint NOT NULL DEFAULT 0)";
+    client.simple_query(query)?;
 
-        let query = "INSERT INTO neon_migration.migration_id VALUES (0, 0) ON CONFLICT DO NOTHING";
-        client.simple_query(query)?;
+    query = "INSERT INTO neon_migration.migration_id VALUES (0, 0) ON CONFLICT DO NOTHING";
+    client.simple_query(query)?;
 
-        let query = "ALTER SCHEMA neon_migration OWNER TO cloud_admin";
-        client.simple_query(query)?;
+    query = "ALTER SCHEMA neon_migration OWNER TO cloud_admin";
+    client.simple_query(query)?;
 
-        let query = "REVOKE ALL ON SCHEMA neon_migration FROM PUBLIC";
-        client.simple_query(query)?;
-        Ok::<_, anyhow::Error>(())
-    };
-    func().context("handle_migrations prepare")?;
+    query = "REVOKE ALL ON SCHEMA neon_migration FROM PUBLIC";
+    client.simple_query(query)?;
 
-    let query = "SELECT id FROM neon_migration.migration_id";
-    let row = client
-        .query_one(query, &[])
-        .context("handle_migrations get migration_id")?;
+    query = "SELECT id FROM neon_migration.migration_id";
+    let row = client.query_one(query, &[])?;
     let mut current_migration: usize = row.get::<&str, i64>("id") as usize;
     let starting_migration_id = current_migration;
 
-    let query = "BEGIN";
-    client
-        .simple_query(query)
-        .context("handle_migrations begin")?;
+    query = "BEGIN";
+    client.simple_query(query)?;
 
     while current_migration < migrations.len() {
         let migration = &migrations[current_migration];
@@ -851,9 +850,7 @@ $$;"#,
             info!("Skip migration id={}", current_migration);
         } else {
             info!("Running migration:\n{}\n", migration);
-            client.simple_query(migration).with_context(|| {
-                format!("handle_migrations current_migration={}", current_migration)
-            })?;
+            client.simple_query(migration)?;
         }
         current_migration += 1;
     }
@@ -861,14 +858,10 @@ $$;"#,
         "UPDATE neon_migration.migration_id SET id={}",
         migrations.len()
     );
-    client
-        .simple_query(&setval)
-        .context("handle_migrations update id")?;
+    client.simple_query(&setval)?;
 
-    let query = "COMMIT";
-    client
-        .simple_query(query)
-        .context("handle_migrations commit")?;
+    query = "COMMIT";
+    client.simple_query(query)?;
 
     info!(
         "Ran {} migrations",

control_plane/attachment_service/Cargo.toml

@@ -1,5 +1,5 @@
 [package]
-name = "storage_controller"
+name = "attachment_service"
 version = "0.1.0"
 edition.workspace = true
 license.workspace = true
@@ -16,6 +16,7 @@ testing = []
 [dependencies]
 anyhow.workspace = true
 aws-config.workspace = true
+aws-sdk-secretsmanager.workspace = true
 bytes.workspace = true
 camino.workspace = true
 clap.workspace = true
@@ -25,7 +26,6 @@ git-version.workspace = true
 hex.workspace = true
 hyper.workspace = true
 humantime.workspace = true
-itertools.workspace = true
 lasso.workspace = true
 once_cell.workspace = true
 pageserver_api.workspace = true
@@ -45,8 +45,8 @@ diesel = { version = "2.1.4", features = ["serde_json", "postgres", "r2d2"] }
 diesel_migrations = { version = "2.1.0" }
 r2d2 = { version = "0.8.10" }
 
-utils = { path = "../libs/utils/" }
-metrics = { path = "../libs/metrics/" }
-control_plane = { path = "../control_plane" }
-workspace_hack = { version = "0.1", path = "../workspace_hack" }
+utils = { path = "../../libs/utils/" }
+metrics = { path = "../../libs/metrics/" }
+control_plane = { path = ".." }
+workspace_hack = { version = "0.1", path = "../../workspace_hack" }
 

control_plane/attachment_service/src/compute_hook.rs

@@ -1,4 +1,3 @@
-use std::sync::Arc;
 use std::{collections::HashMap, time::Duration};
 
 use control_plane::endpoint::{ComputeControlPlane, EndpointStatus};
@@ -15,32 +14,19 @@ use utils::{
 
 use crate::service::Config;
 
+const BUSY_DELAY: Duration = Duration::from_secs(1);
 const SLOWDOWN_DELAY: Duration = Duration::from_secs(5);
 
-const NOTIFY_REQUEST_TIMEOUT: Duration = Duration::from_secs(10);
-
 pub(crate) const API_CONCURRENCY: usize = 32;
 
-struct UnshardedComputeHookTenant {
-    // Which node is this tenant attached to
-    node_id: NodeId,
-
-    // Must hold this lock to send a notification.
-    send_lock: Arc<tokio::sync::Mutex<Option<ComputeHookNotifyRequest>>>,
-}
 struct ShardedComputeHookTenant {
     stripe_size: ShardStripeSize,
     shard_count: ShardCount,
     shards: Vec<(ShardNumber, NodeId)>,
-
-    // Must hold this lock to send a notification.  The contents represent
-    // the last successfully sent notification, and are used to coalesce multiple
-    // updates by only sending when there is a chance since our last successful send.
-    send_lock: Arc<tokio::sync::Mutex<Option<ComputeHookNotifyRequest>>>,
 }
 
 enum ComputeHookTenant {
-    Unsharded(UnshardedComputeHookTenant),
+    Unsharded(NodeId),
     Sharded(ShardedComputeHookTenant),
 }
 
@@ -52,20 +38,9 @@ impl ComputeHookTenant {
                 shards: vec![(tenant_shard_id.shard_number, node_id)],
                 stripe_size,
                 shard_count: tenant_shard_id.shard_count,
-                send_lock: Arc::default(),
             })
         } else {
-            Self::Unsharded(UnshardedComputeHookTenant {
-                node_id,
-                send_lock: Arc::default(),
-            })
-        }
-    }
-
-    fn get_send_lock(&self) -> &Arc<tokio::sync::Mutex<Option<ComputeHookNotifyRequest>>> {
-        match self {
-            Self::Unsharded(unsharded_tenant) => &unsharded_tenant.send_lock,
-            Self::Sharded(sharded_tenant) => &sharded_tenant.send_lock,
+            Self::Unsharded(node_id)
         }
     }
 
@@ -78,8 +53,8 @@ impl ComputeHookTenant {
         node_id: NodeId,
     ) {
         match self {
-            Self::Unsharded(unsharded_tenant) if tenant_shard_id.shard_count.count() == 1 => {
-                unsharded_tenant.node_id = node_id
+            Self::Unsharded(existing_node_id) if tenant_shard_id.shard_count.count() == 1 => {
+                *existing_node_id = node_id
             }
             Self::Sharded(sharded_tenant)
                 if sharded_tenant.stripe_size == stripe_size
@@ -106,14 +81,14 @@ impl ComputeHookTenant {
     }
 }
 
-#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
+#[derive(Serialize, Deserialize, Debug)]
 struct ComputeHookNotifyRequestShard {
     node_id: NodeId,
     shard_number: ShardNumber,
 }
 
 /// Request body that we send to the control plane to notify it of where a tenant is attached
-#[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
+#[derive(Serialize, Deserialize, Debug)]
 struct ComputeHookNotifyRequest {
     tenant_id: TenantId,
     stripe_size: Option<ShardStripeSize>,
@@ -146,44 +121,14 @@ pub(crate) enum NotifyError {
     Fatal(StatusCode),
 }
 
-enum MaybeSendResult {
-    // Please send this request while holding the lock, and if you succeed then write
-    // the request into the lock.
-    Transmit(
-        (
-            ComputeHookNotifyRequest,
-            tokio::sync::OwnedMutexGuard<Option<ComputeHookNotifyRequest>>,
-        ),
-    ),
-    // Something requires sending, but you must wait for a current sender then call again
-    AwaitLock(Arc<tokio::sync::Mutex<Option<ComputeHookNotifyRequest>>>),
-    // Nothing requires sending
-    Noop,
-}
-
 impl ComputeHookTenant {
-    fn maybe_send(
-        &self,
-        tenant_id: TenantId,
-        lock: Option<tokio::sync::OwnedMutexGuard<Option<ComputeHookNotifyRequest>>>,
-    ) -> MaybeSendResult {
-        let locked = match lock {
-            Some(already_locked) => already_locked,
-            None => {
-                // Lock order: this _must_ be only a try_lock, because we are called inside of the [`ComputeHook::state`] lock.
-                let Ok(locked) = self.get_send_lock().clone().try_lock_owned() else {
-                    return MaybeSendResult::AwaitLock(self.get_send_lock().clone());
-                };
-                locked
-            }
-        };
-
-        let request = match self {
-            Self::Unsharded(unsharded_tenant) => Some(ComputeHookNotifyRequest {
+    fn maybe_reconfigure(&self, tenant_id: TenantId) -> Option<ComputeHookNotifyRequest> {
+        match self {
+            Self::Unsharded(node_id) => Some(ComputeHookNotifyRequest {
                 tenant_id,
                 shards: vec![ComputeHookNotifyRequestShard {
                     shard_number: ShardNumber(0),
-                    node_id: unsharded_tenant.node_id,
+                    node_id: *node_id,
                 }],
                 stripe_size: None,
             }),
@@ -207,25 +152,12 @@ impl ComputeHookTenant {
                 // Sharded tenant doesn't yet have information for all its shards
 
                 tracing::info!(
-                    "ComputeHookTenant::maybe_send: not enough shards ({}/{})",
+                    "ComputeHookTenant::maybe_reconfigure: not enough shards ({}/{})",
                     sharded_tenant.shards.len(),
                     sharded_tenant.shard_count.count()
                 );
                 None
             }
-        };
-
-        match request {
-            None => {
-                // Not yet ready to emit a notification
-                tracing::info!("Tenant isn't yet ready to emit a notification");
-                MaybeSendResult::Noop
-            }
-            Some(request) if Some(&request) == locked.as_ref() => {
-                // No change from the last value successfully sent
-                MaybeSendResult::Noop
-            }
-            Some(request) => MaybeSendResult::Transmit((request, locked)),
         }
     }
 }
@@ -235,19 +167,8 @@ impl ComputeHookTenant {
 /// the compute connection string.
 pub(super) struct ComputeHook {
     config: Config,
-    state: std::sync::Mutex<HashMap<TenantId, ComputeHookTenant>>,
+    state: tokio::sync::Mutex<HashMap<TenantId, ComputeHookTenant>>,
     authorization_header: Option<String>,
-
-    // Concurrency limiter, so that we do not overload the cloud control plane when updating
-    // large numbers of tenants (e.g. when failing over after a node failure)
-    api_concurrency: tokio::sync::Semaphore,
-
-    // This lock is only used in testing enviroments, to serialize calls into neon_lock
-    neon_local_lock: tokio::sync::Mutex<()>,
-
-    // We share a client across all notifications to enable connection re-use etc when
-    // sending large numbers of notifications
-    client: reqwest::Client,
 }
 
 impl ComputeHook {
@@ -257,30 +178,18 @@ impl ComputeHook {
             .clone()
             .map(|jwt| format!("Bearer {}", jwt));
 
-        let client = reqwest::ClientBuilder::new()
-            .timeout(NOTIFY_REQUEST_TIMEOUT)
-            .build()
-            .expect("Failed to construct HTTP client");
-
         Self {
             state: Default::default(),
             config,
             authorization_header,
-            neon_local_lock: Default::default(),
-            api_concurrency: tokio::sync::Semaphore::new(API_CONCURRENCY),
-            client,
         }
     }
 
     /// For test environments: use neon_local's LocalEnv to update compute
     async fn do_notify_local(
         &self,
-        reconfigure_request: &ComputeHookNotifyRequest,
+        reconfigure_request: ComputeHookNotifyRequest,
     ) -> anyhow::Result<()> {
-        // neon_local updates are not safe to call concurrently, use a lock to serialize
-        // all calls to this function
-        let _locked = self.neon_local_lock.lock().await;
-
         let env = match LocalEnv::load_config() {
             Ok(e) => e,
             Err(e) => {
@@ -297,7 +206,7 @@ impl ComputeHook {
         } = reconfigure_request;
 
         let compute_pageservers = shards
-            .iter()
+            .into_iter()
             .map(|shard| {
                 let ps_conf = env
                     .get_pageserver_conf(shard.node_id)
@@ -309,10 +218,10 @@ impl ComputeHook {
             .collect::<Vec<_>>();
 
         for (endpoint_name, endpoint) in &cplane.endpoints {
-            if endpoint.tenant_id == *tenant_id && endpoint.status() == EndpointStatus::Running {
+            if endpoint.tenant_id == tenant_id && endpoint.status() == EndpointStatus::Running {
                 tracing::info!("Reconfiguring endpoint {}", endpoint_name,);
                 endpoint
-                    .reconfigure(compute_pageservers.clone(), *stripe_size)
+                    .reconfigure(compute_pageservers.clone(), stripe_size)
                     .await?;
             }
         }
@@ -322,11 +231,12 @@ impl ComputeHook {
 
     async fn do_notify_iteration(
         &self,
+        client: &reqwest::Client,
         url: &String,
         reconfigure_request: &ComputeHookNotifyRequest,
         cancel: &CancellationToken,
     ) -> Result<(), NotifyError> {
-        let req = self.client.request(Method::PUT, url);
+        let req = client.request(Method::PUT, url);
         let req = if let Some(value) = &self.authorization_header {
             req.header(reqwest::header::AUTHORIZATION, value)
         } else {
@@ -370,10 +280,11 @@ impl ComputeHook {
                 Err(NotifyError::SlowDown)
             }
             StatusCode::LOCKED => {
-                // We consider this fatal, because it's possible that the operation blocking the control one is
-                // also the one that is waiting for this reconcile.  We should let the reconciler calling
-                // this hook fail, to give control plane a chance to un-lock.
-                tracing::info!("Control plane reports tenant is locked, dropping out of notify");
+                // Delay our retry if busy: the usual fast exponential backoff in backoff::retry
+                // is not appropriate
+                tokio::time::timeout(BUSY_DELAY, cancel.cancelled())
+                    .await
+                    .ok();
                 Err(NotifyError::Busy)
             }
             StatusCode::SERVICE_UNAVAILABLE
@@ -389,27 +300,13 @@ impl ComputeHook {
     async fn do_notify(
         &self,
         url: &String,
-        reconfigure_request: &ComputeHookNotifyRequest,
+        reconfigure_request: ComputeHookNotifyRequest,
         cancel: &CancellationToken,
     ) -> Result<(), NotifyError> {
-        // We hold these semaphore units across all retries, rather than only across each
-        // HTTP request: this is to preserve fairness and avoid a situation where a retry might
-        // time out waiting for a semaphore.
-        let _units = self
-            .api_concurrency
-            .acquire()
-            .await
-            // Interpret closed semaphore as shutdown
-            .map_err(|_| NotifyError::ShuttingDown)?;
-
+        let client = reqwest::Client::new();
         backoff::retry(
-            || self.do_notify_iteration(url, reconfigure_request, cancel),
-            |e| {
-                matches!(
-                    e,
-                    NotifyError::Fatal(_) | NotifyError::Unexpected(_) | NotifyError::Busy
-                )
-            },
+            || self.do_notify_iteration(&client, url, &reconfigure_request, cancel),
+            |e| matches!(e, NotifyError::Fatal(_) | NotifyError::Unexpected(_)),
             3,
             10,
             "Send compute notification",
@@ -443,70 +340,42 @@ impl ComputeHook {
         stripe_size: ShardStripeSize,
         cancel: &CancellationToken,
     ) -> Result<(), NotifyError> {
-        let maybe_send_result = {
-            let mut state_locked = self.state.lock().unwrap();
+        let mut locked = self.state.lock().await;
 
-            use std::collections::hash_map::Entry;
-            let tenant = match state_locked.entry(tenant_shard_id.tenant_id) {
-                Entry::Vacant(e) => e.insert(ComputeHookTenant::new(
-                    tenant_shard_id,
-                    stripe_size,
-                    node_id,
-                )),
-                Entry::Occupied(e) => {
-                    let tenant = e.into_mut();
-                    tenant.update(tenant_shard_id, stripe_size, node_id);
-                    tenant
-                }
-            };
-            tenant.maybe_send(tenant_shard_id.tenant_id, None)
+        use std::collections::hash_map::Entry;
+        let tenant = match locked.entry(tenant_shard_id.tenant_id) {
+            Entry::Vacant(e) => e.insert(ComputeHookTenant::new(
+                tenant_shard_id,
+                stripe_size,
+                node_id,
+            )),
+            Entry::Occupied(e) => {
+                let tenant = e.into_mut();
+                tenant.update(tenant_shard_id, stripe_size, node_id);
+                tenant
+            }
         };
 
-        // Process result: we may get an update to send, or we may have to wait for a lock
-        // before trying again.
-        let (request, mut send_lock_guard) = match maybe_send_result {
-            MaybeSendResult::Noop => {
-                return Ok(());
-            }
-            MaybeSendResult::AwaitLock(send_lock) => {
-                let send_locked = send_lock.lock_owned().await;
-
-                // Lock order: maybe_send is called within the `[Self::state]` lock, and takes the send lock, but here
-                // we have acquired the send lock and take `[Self::state]` lock.  This is safe because maybe_send only uses
-                // try_lock.
-                let state_locked = self.state.lock().unwrap();
-                let Some(tenant) = state_locked.get(&tenant_shard_id.tenant_id) else {
-                    return Ok(());
-                };
-                match tenant.maybe_send(tenant_shard_id.tenant_id, Some(send_locked)) {
-                    MaybeSendResult::AwaitLock(_) => {
-                        unreachable!("We supplied lock guard")
-                    }
-                    MaybeSendResult::Noop => {
-                        return Ok(());
-                    }
-                    MaybeSendResult::Transmit((request, lock)) => (request, lock),
-                }
-            }
-            MaybeSendResult::Transmit((request, lock)) => (request, lock),
+        let reconfigure_request = tenant.maybe_reconfigure(tenant_shard_id.tenant_id);
+        let Some(reconfigure_request) = reconfigure_request else {
+            // The tenant doesn't yet have pageservers for all its shards: we won't notify anything
+            // until it does.
+            tracing::info!("Tenant isn't yet ready to emit a notification");
+            return Ok(());
         };
 
-        let result = if let Some(notify_url) = &self.config.compute_hook_url {
-            self.do_notify(notify_url, &request, cancel).await
+        if let Some(notify_url) = &self.config.compute_hook_url {
+            self.do_notify(notify_url, reconfigure_request, cancel)
+                .await
         } else {
-            self.do_notify_local(&request).await.map_err(|e| {
-                // This path is for testing only, so munge the error into our prod-style error type.
-                tracing::error!("Local notification hook failed: {e}");
-                NotifyError::Fatal(StatusCode::INTERNAL_SERVER_ERROR)
-            })
-        };
-
-        if result.is_ok() {
-            // Before dropping the send lock, stash the request we just sent so that
-            // subsequent callers can avoid redundantly re-sending the same thing.
-            *send_lock_guard = Some(request);
+            self.do_notify_local(reconfigure_request)
+                .await
+                .map_err(|e| {
+                    // This path is for testing only, so munge the error into our prod-style error type.
+                    tracing::error!("Local notification hook failed: {e}");
+                    NotifyError::Fatal(StatusCode::INTERNAL_SERVER_ERROR)
+                })
         }
-        result
     }
 }
 
@@ -530,22 +399,21 @@ pub(crate) mod tests {
             NodeId(1),
         );
 
-        // An unsharded tenant is always ready to emit a notification, but won't
-        // send the same one twice
-        let send_result = tenant_state.maybe_send(tenant_id, None);
-        let MaybeSendResult::Transmit((request, mut guard)) = send_result else {
-            anyhow::bail!("Wrong send result");
-        };
-        assert_eq!(request.shards.len(), 1);
-        assert!(request.stripe_size.is_none());
-
-        // Simulate successful send
-        *guard = Some(request);
-        drop(guard);
-
-        // Try asking again: this should be a no-op
-        let send_result = tenant_state.maybe_send(tenant_id, None);
-        assert!(matches!(send_result, MaybeSendResult::Noop));
+        // An unsharded tenant is always ready to emit a notification
+        assert!(tenant_state.maybe_reconfigure(tenant_id).is_some());
+        assert_eq!(
+            tenant_state
+                .maybe_reconfigure(tenant_id)
+                .unwrap()
+                .shards
+                .len(),
+            1
+        );
+        assert!(tenant_state
+            .maybe_reconfigure(tenant_id)
+            .unwrap()
+            .stripe_size
+            .is_none());
 
         // Writing the first shard of a multi-sharded situation (i.e. in a split)
         // resets the tenant state and puts it in an non-notifying state (need to
@@ -559,10 +427,7 @@ pub(crate) mod tests {
             ShardStripeSize(32768),
             NodeId(1),
         );
-        assert!(matches!(
-            tenant_state.maybe_send(tenant_id, None),
-            MaybeSendResult::Noop
-        ));
+        assert!(tenant_state.maybe_reconfigure(tenant_id).is_none());
 
         // Writing the second shard makes it ready to notify
         tenant_state.update(
@@ -575,16 +440,22 @@ pub(crate) mod tests {
             NodeId(1),
         );
 
-        let send_result = tenant_state.maybe_send(tenant_id, None);
-        let MaybeSendResult::Transmit((request, mut guard)) = send_result else {
-            anyhow::bail!("Wrong send result");
-        };
-        assert_eq!(request.shards.len(), 2);
-        assert_eq!(request.stripe_size, Some(ShardStripeSize(32768)));
-
-        // Simulate successful send
-        *guard = Some(request);
-        drop(guard);
+        assert!(tenant_state.maybe_reconfigure(tenant_id).is_some());
+        assert_eq!(
+            tenant_state
+                .maybe_reconfigure(tenant_id)
+                .unwrap()
+                .shards
+                .len(),
+            2
+        );
+        assert_eq!(
+            tenant_state
+                .maybe_reconfigure(tenant_id)
+                .unwrap()
+                .stripe_size,
+            Some(ShardStripeSize(32768))
+        );
 
         Ok(())
     }

control_plane/attachment_service/src/heartbeater.rs

@@ -139,7 +139,7 @@ impl HeartbeaterTask {
                         .with_client_retries(
                             |client| async move { client.get_utilization().await },
                             &jwt_token,
-                            3,
+                            2,
                             3,
                             Duration::from_secs(1),
                             &cancel,

control_plane/attachment_service/src/http.rs

@@ -8,7 +8,6 @@ use futures::Future;
 use hyper::header::CONTENT_TYPE;
 use hyper::{Body, Request, Response};
 use hyper::{StatusCode, Uri};
-use metrics::{BuildInfo, NeonMetrics};
 use pageserver_api::models::{
     TenantConfigRequest, TenantCreateRequest, TenantLocationConfigRequest, TenantShardSplitRequest,
     TenantTimeTravelRequest, TimelineCreateRequest,
@@ -35,8 +34,7 @@ use utils::{
 };
 
 use pageserver_api::controller_api::{
-    NodeAvailability, NodeConfigureRequest, NodeRegisterRequest, TenantPolicyRequest,
-    TenantShardMigrateRequest,
+    NodeAvailability, NodeConfigureRequest, NodeRegisterRequest, TenantShardMigrateRequest,
 };
 use pageserver_api::upcall_api::{ReAttachRequest, ValidateRequest};
 
@@ -45,19 +43,15 @@ use control_plane::storage_controller::{AttachHookRequest, InspectRequest};
 use routerify::Middleware;
 
 /// State available to HTTP request handlers
+#[derive(Clone)]
 pub struct HttpState {
     service: Arc<crate::service::Service>,
     auth: Option<Arc<SwappableJwtAuth>>,
-    neon_metrics: NeonMetrics,
     allowlist_routes: Vec<Uri>,
 }
 
 impl HttpState {
-    pub fn new(
-        service: Arc<crate::service::Service>,
-        auth: Option<Arc<SwappableJwtAuth>>,
-        build_info: BuildInfo,
-    ) -> Self {
+    pub fn new(service: Arc<crate::service::Service>, auth: Option<Arc<SwappableJwtAuth>>) -> Self {
         let allowlist_routes = ["/status", "/ready", "/metrics"]
             .iter()
             .map(|v| v.parse().unwrap())
@@ -65,7 +59,6 @@ impl HttpState {
         Self {
             service,
             auth,
-            neon_metrics: NeonMetrics::new(build_info),
             allowlist_routes,
         }
     }
@@ -405,15 +398,6 @@ async fn handle_tenant_describe(
     json_response(StatusCode::OK, service.tenant_describe(tenant_id)?)
 }
 
-async fn handle_tenant_list(
-    service: Arc<Service>,
-    req: Request<Body>,
-) -> Result<Response<Body>, ApiError> {
-    check_permissions(&req, Scope::Admin)?;
-
-    json_response(StatusCode::OK, service.tenant_list())
-}
-
 async fn handle_node_register(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
     check_permissions(&req, Scope::Admin)?;
 
@@ -427,10 +411,7 @@ async fn handle_node_list(req: Request<Body>) -> Result<Response<Body>, ApiError
     check_permissions(&req, Scope::Admin)?;
 
     let state = get_state(&req);
-    let nodes = state.service.node_list().await?;
-    let api_nodes = nodes.into_iter().map(|n| n.describe()).collect::<Vec<_>>();
-
-    json_response(StatusCode::OK, api_nodes)
+    json_response(StatusCode::OK, state.service.node_list().await?)
 }
 
 async fn handle_node_drop(req: Request<Body>) -> Result<Response<Body>, ApiError> {
@@ -497,22 +478,6 @@ async fn handle_tenant_shard_migrate(
     )
 }
 
-async fn handle_tenant_update_policy(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
-    check_permissions(&req, Scope::Admin)?;
-
-    let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
-    let update_req = json_request::<TenantPolicyRequest>(&mut req).await?;
-    let state = get_state(&req);
-
-    json_response(
-        StatusCode::OK,
-        state
-            .service
-            .tenant_update_policy(tenant_id, update_req)
-            .await?,
-    )
-}
-
 async fn handle_tenant_drop(req: Request<Body>) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
     check_permissions(&req, Scope::PageServerApi)?;
@@ -544,14 +509,6 @@ async fn handle_consistency_check(req: Request<Body>) -> Result<Response<Body>,
     json_response(StatusCode::OK, state.service.consistency_check().await?)
 }
 
-async fn handle_reconcile_all(req: Request<Body>) -> Result<Response<Body>, ApiError> {
-    check_permissions(&req, Scope::Admin)?;
-
-    let state = get_state(&req);
-
-    json_response(StatusCode::OK, state.service.reconcile_all_now().await?)
-}
-
 /// Status endpoint is just used for checking that our HTTP listener is up
 async fn handle_status(_req: Request<Body>) -> Result<Response<Body>, ApiError> {
     json_response(StatusCode::OK, ())
@@ -608,17 +565,9 @@ where
     .await
 }
 
-/// Check if the required scope is held in the request's token, or if the request has
-/// a token with 'admin' scope then always permit it.
 fn check_permissions(request: &Request<Body>, required_scope: Scope) -> Result<(), ApiError> {
     check_permission_with(request, |claims| {
-        match crate::auth::check_permission(claims, required_scope) {
-            Err(e) => match crate::auth::check_permission(claims, Scope::Admin) {
-                Ok(()) => Ok(()),
-                Err(_) => Err(e),
-            },
-            Ok(()) => Ok(()),
-        }
+        crate::auth::check_permission(claims, required_scope)
     })
 }
 
@@ -678,11 +627,10 @@ fn epilogue_metrics_middleware<B: hyper::body::HttpBody + Send + Sync + 'static>
     })
 }
 
-pub async fn measured_metrics_handler(req: Request<Body>) -> Result<Response<Body>, ApiError> {
+pub async fn measured_metrics_handler(_req: Request<Body>) -> Result<Response<Body>, ApiError> {
     pub const TEXT_FORMAT: &str = "text/plain; version=0.0.4";
 
-    let state = get_state(&req);
-    let payload = crate::metrics::METRICS_REGISTRY.encode(&state.neon_metrics);
+    let payload = crate::metrics::METRICS_REGISTRY.encode();
     let response = Response::builder()
         .status(200)
         .header(CONTENT_TYPE, TEXT_FORMAT)
@@ -711,7 +659,6 @@ where
 pub fn make_router(
     service: Arc<Service>,
     auth: Option<Arc<SwappableJwtAuth>>,
-    build_info: BuildInfo,
 ) -> RouterBuilder<hyper::Body, ApiError> {
     let mut router = endpoint::make_router()
         .middleware(prologue_metrics_middleware())
@@ -728,7 +675,7 @@ pub fn make_router(
     }
 
     router
-        .data(Arc::new(HttpState::new(service, auth, build_info)))
+        .data(Arc::new(HttpState::new(service, auth)))
         .get("/metrics", |r| {
             named_request_span(r, measured_metrics_handler, RequestName("metrics"))
         })
@@ -779,9 +726,6 @@ pub fn make_router(
                 RequestName("debug_v1_consistency_check"),
             )
         })
-        .post("/debug/v1/reconcile_all", |r| {
-            request_span(r, handle_reconcile_all)
-        })
         .put("/debug/v1/failpoints", |r| {
             request_span(r, |r| failpoints_handler(r, CancellationToken::new()))
         })
@@ -821,16 +765,6 @@ pub fn make_router(
                 RequestName("control_v1_tenant_describe"),
             )
         })
-        .get("/control/v1/tenant", |r| {
-            tenant_service_handler(r, handle_tenant_list, RequestName("control_v1_tenant_list"))
-        })
-        .put("/control/v1/tenant/:tenant_id/policy", |r| {
-            named_request_span(
-                r,
-                handle_tenant_update_policy,
-                RequestName("control_v1_tenant_policy"),
-            )
-        })
         // Tenant operations
         // The ^/v1/ endpoints act as a "Virtual Pageserver", enabling shard-naive clients to call into
         // this service to manage tenants that actually consist of many tenant shards, as if they are a single entity.

control_plane/attachment_service/src/lib.rs

@@ -14,7 +14,7 @@ mod reconciler;
 mod scheduler;
 mod schema;
 pub mod service;
-mod tenant_shard;
+mod tenant_state;
 
 #[derive(Ord, PartialOrd, Eq, PartialEq, Copy, Clone, Serialize)]
 struct Sequence(u64);

control_plane/attachment_service/src/main.rs

@@ -1,21 +1,19 @@
 use anyhow::{anyhow, Context};
+use attachment_service::http::make_router;
+use attachment_service::metrics::preinitialize_metrics;
+use attachment_service::persistence::Persistence;
+use attachment_service::service::{Config, Service, MAX_UNAVAILABLE_INTERVAL_DEFAULT};
+use aws_config::{BehaviorVersion, Region};
 use camino::Utf8PathBuf;
 use clap::Parser;
 use diesel::Connection;
 use metrics::launch_timestamp::LaunchTimestamp;
-use metrics::BuildInfo;
 use std::sync::Arc;
-use std::time::Duration;
-use storage_controller::http::make_router;
-use storage_controller::metrics::preinitialize_metrics;
-use storage_controller::persistence::Persistence;
-use storage_controller::service::{Config, Service, MAX_UNAVAILABLE_INTERVAL_DEFAULT};
 use tokio::signal::unix::SignalKind;
 use tokio_util::sync::CancellationToken;
 use utils::auth::{JwtAuth, SwappableJwtAuth};
 use utils::logging::{self, LogFormat};
 
-use utils::sentry_init::init_sentry;
 use utils::{project_build_tag, project_git_version, tcp_listener};
 
 project_git_version!(GIT_VERSION);
@@ -53,35 +51,15 @@ struct Cli {
     #[arg(short, long)]
     path: Option<Utf8PathBuf>,
 
-    /// URL to connect to postgres, like postgresql://localhost:1234/storage_controller
+    /// URL to connect to postgres, like postgresql://localhost:1234/attachment_service
     #[arg(long)]
     database_url: Option<String>,
 
-    /// Flag to enable dev mode, which permits running without auth
-    #[arg(long, default_value = "false")]
-    dev: bool,
-
     /// Grace period before marking unresponsive pageserver offline
     #[arg(long)]
     max_unavailable_interval: Option<humantime::Duration>,
 }
 
-enum StrictMode {
-    /// In strict mode, we will require that all secrets are loaded, i.e. security features
-    /// may not be implicitly turned off by omitting secrets in the environment.
-    Strict,
-    /// In dev mode, secrets are optional, and omitting a particular secret will implicitly
-    /// disable the auth related to it (e.g. no pageserver jwt key -> send unauthenticated
-    /// requests, no public key -> don't authenticate incoming requests).
-    Dev,
-}
-
-impl Default for StrictMode {
-    fn default() -> Self {
-        Self::Strict
-    }
-}
-
 /// Secrets may either be provided on the command line (for testing), or loaded from AWS SecretManager: this
 /// type encapsulates the logic to decide which and do the loading.
 struct Secrets {
@@ -92,6 +70,13 @@ struct Secrets {
 }
 
 impl Secrets {
+    const DATABASE_URL_SECRET: &'static str = "rds-neon-storage-controller-url";
+    const PAGESERVER_JWT_TOKEN_SECRET: &'static str =
+        "neon-storage-controller-pageserver-jwt-token";
+    const CONTROL_PLANE_JWT_TOKEN_SECRET: &'static str =
+        "neon-storage-controller-control-plane-jwt-token";
+    const PUBLIC_KEY_SECRET: &'static str = "neon-storage-controller-public-key";
+
     const DATABASE_URL_ENV: &'static str = "DATABASE_URL";
     const PAGESERVER_JWT_TOKEN_ENV: &'static str = "PAGESERVER_JWT_TOKEN";
     const CONTROL_PLANE_JWT_TOKEN_ENV: &'static str = "CONTROL_PLANE_JWT_TOKEN";
@@ -102,41 +87,111 @@ impl Secrets {
     /// - Environment variables if DATABASE_URL is set.
     /// - AWS Secrets Manager secrets
     async fn load(args: &Cli) -> anyhow::Result<Self> {
-        let Some(database_url) =
-            Self::load_secret(&args.database_url, Self::DATABASE_URL_ENV).await
-        else {
-            anyhow::bail!(
-                "Database URL is not set (set `--database-url`, or `DATABASE_URL` environment)"
-            )
-        };
-
-        let public_key = match Self::load_secret(&args.public_key, Self::PUBLIC_KEY_ENV).await {
-            Some(v) => Some(JwtAuth::from_key(v).context("Loading public key")?),
-            None => None,
-        };
-
-        let this = Self {
-            database_url,
-            public_key,
-            jwt_token: Self::load_secret(&args.jwt_token, Self::PAGESERVER_JWT_TOKEN_ENV).await,
-            control_plane_jwt_token: Self::load_secret(
-                &args.control_plane_jwt_token,
-                Self::CONTROL_PLANE_JWT_TOKEN_ENV,
-            )
-            .await,
-        };
-
-        Ok(this)
+        match &args.database_url {
+            Some(url) => Self::load_cli(url, args),
+            None => match std::env::var(Self::DATABASE_URL_ENV) {
+                Ok(database_url) => Self::load_env(database_url),
+                Err(_) => Self::load_aws_sm().await,
+            },
+        }
     }
 
-    async fn load_secret(cli: &Option<String>, env_name: &str) -> Option<String> {
-        if let Some(v) = cli {
-            Some(v.clone())
-        } else if let Ok(v) = std::env::var(env_name) {
-            Some(v)
-        } else {
-            None
+    fn load_env(database_url: String) -> anyhow::Result<Self> {
+        let public_key = match std::env::var(Self::PUBLIC_KEY_ENV) {
+            Ok(public_key) => Some(JwtAuth::from_key(public_key).context("Loading public key")?),
+            Err(_) => None,
+        };
+        Ok(Self {
+            database_url,
+            public_key,
+            jwt_token: std::env::var(Self::PAGESERVER_JWT_TOKEN_ENV).ok(),
+            control_plane_jwt_token: std::env::var(Self::CONTROL_PLANE_JWT_TOKEN_ENV).ok(),
+        })
+    }
+
+    async fn load_aws_sm() -> anyhow::Result<Self> {
+        let Ok(region) = std::env::var("AWS_REGION") else {
+            anyhow::bail!("AWS_REGION is not set, cannot load secrets automatically: either set this, or use CLI args to supply secrets");
+        };
+        let config = aws_config::defaults(BehaviorVersion::v2023_11_09())
+            .region(Region::new(region.clone()))
+            .load()
+            .await;
+
+        let asm = aws_sdk_secretsmanager::Client::new(&config);
+
+        let Some(database_url) = asm
+            .get_secret_value()
+            .secret_id(Self::DATABASE_URL_SECRET)
+            .send()
+            .await?
+            .secret_string()
+            .map(str::to_string)
+        else {
+            anyhow::bail!(
+                "Database URL secret not found at {region}/{}",
+                Self::DATABASE_URL_SECRET
+            )
+        };
+
+        let jwt_token = asm
+            .get_secret_value()
+            .secret_id(Self::PAGESERVER_JWT_TOKEN_SECRET)
+            .send()
+            .await?
+            .secret_string()
+            .map(str::to_string);
+        if jwt_token.is_none() {
+            tracing::warn!("No pageserver JWT token set: this will only work if authentication is disabled on the pageserver");
         }
+
+        let control_plane_jwt_token = asm
+            .get_secret_value()
+            .secret_id(Self::CONTROL_PLANE_JWT_TOKEN_SECRET)
+            .send()
+            .await?
+            .secret_string()
+            .map(str::to_string);
+        if jwt_token.is_none() {
+            tracing::warn!("No control plane JWT token set: this will only work if authentication is disabled on the pageserver");
+        }
+
+        let public_key = asm
+            .get_secret_value()
+            .secret_id(Self::PUBLIC_KEY_SECRET)
+            .send()
+            .await?
+            .secret_string()
+            .map(str::to_string);
+        let public_key = match public_key {
+            Some(key) => Some(JwtAuth::from_key(key)?),
+            None => {
+                tracing::warn!(
+                    "No public key set: inccoming HTTP requests will not be authenticated"
+                );
+                None
+            }
+        };
+
+        Ok(Self {
+            database_url,
+            public_key,
+            jwt_token,
+            control_plane_jwt_token,
+        })
+    }
+
+    fn load_cli(database_url: &str, args: &Cli) -> anyhow::Result<Self> {
+        let public_key = match &args.public_key {
+            None => None,
+            Some(key) => Some(JwtAuth::from_key(key.clone()).context("Loading public key")?),
+        };
+        Ok(Self {
+            database_url: database_url.to_owned(),
+            public_key,
+            jwt_token: args.jwt_token.clone(),
+            control_plane_jwt_token: args.control_plane_jwt_token.clone(),
+        })
     }
 }
 
@@ -161,8 +216,6 @@ fn main() -> anyhow::Result<()> {
         std::process::exit(1);
     }));
 
-    let _sentry_guard = init_sentry(Some(GIT_VERSION.into()), &[]);
-
     tokio::runtime::Builder::new_current_thread()
         // We use spawn_blocking for database operations, so require approximately
         // as many blocking threads as we will open database connections.
@@ -194,47 +247,8 @@ async fn async_main() -> anyhow::Result<()> {
         args.listen
     );
 
-    let build_info = BuildInfo {
-        revision: GIT_VERSION,
-        build_tag: BUILD_TAG,
-    };
-
-    let strict_mode = if args.dev {
-        StrictMode::Dev
-    } else {
-        StrictMode::Strict
-    };
-
     let secrets = Secrets::load(&args).await?;
 
-    // Validate required secrets and arguments are provided in strict mode
-    match strict_mode {
-        StrictMode::Strict
-            if (secrets.public_key.is_none()
-                || secrets.jwt_token.is_none()
-                || secrets.control_plane_jwt_token.is_none()) =>
-        {
-            // Production systems should always have secrets configured: if public_key was not set
-            // then we would implicitly disable auth.
-            anyhow::bail!(
-                    "Insecure config!  One or more secrets is not set.  This is only permitted in `--dev` mode"
-                );
-        }
-        StrictMode::Strict if args.compute_hook_url.is_none() => {
-            // Production systems should always have a compute hook set, to prevent falling
-            // back to trying to use neon_local.
-            anyhow::bail!(
-                "`--compute-hook-url` is not set: this is only permitted in `--dev` mode"
-            );
-        }
-        StrictMode::Strict => {
-            tracing::info!("Starting in strict mode: configuration is OK.")
-        }
-        StrictMode::Dev => {
-            tracing::warn!("Starting in dev mode: this may be an insecure configuration.")
-        }
-    }
-
     let config = Config {
         jwt_token: secrets.jwt_token,
         control_plane_jwt_token: secrets.control_plane_jwt_token,
@@ -246,8 +260,6 @@ async fn async_main() -> anyhow::Result<()> {
     };
 
     // After loading secrets & config, but before starting anything else, apply database migrations
-    Persistence::await_connection(&secrets.database_url, Duration::from_secs(5)).await?;
-
     migration_run(&secrets.database_url)
         .await
         .context("Running database migrations")?;
@@ -262,7 +274,7 @@ async fn async_main() -> anyhow::Result<()> {
     let auth = secrets
         .public_key
         .map(|jwt_auth| Arc::new(SwappableJwtAuth::new(jwt_auth)));
-    let router = make_router(service.clone(), auth, build_info)
+    let router = make_router(service.clone(), auth)
         .build()
         .map_err(|err| anyhow!(err))?;
     let router_service = utils::http::RouterService::new(router).unwrap();

control_plane/attachment_service/src/metrics.rs

@@ -8,8 +8,10 @@
 //! The rest of the code defines label group types and deals with converting outer types to labels.
 //!
 use bytes::Bytes;
-use measured::{label::LabelValue, metric::histogram, FixedCardinalityLabel, MetricGroup};
-use metrics::NeonMetrics;
+use measured::{
+    label::{LabelValue, StaticLabelSet},
+    FixedCardinalityLabel, MetricGroup,
+};
 use once_cell::sync::Lazy;
 use std::sync::Mutex;
 
@@ -24,28 +26,21 @@ pub fn preinitialize_metrics() {
 
 pub(crate) struct StorageControllerMetrics {
     pub(crate) metrics_group: StorageControllerMetricGroup,
-    encoder: Mutex<measured::text::BufferedTextEncoder>,
+    encoder: Mutex<measured::text::TextEncoder>,
 }
 
 #[derive(measured::MetricGroup)]
-#[metric(new())]
 pub(crate) struct StorageControllerMetricGroup {
     /// Count of how many times we spawn a reconcile task
     pub(crate) storage_controller_reconcile_spawn: measured::Counter,
-
     /// Reconciler tasks completed, broken down by success/failure/cancelled
     pub(crate) storage_controller_reconcile_complete:
         measured::CounterVec<ReconcileCompleteLabelGroupSet>,
 
-    /// Count of how many times we make an optimization change to a tenant's scheduling
-    pub(crate) storage_controller_schedule_optimization: measured::Counter,
-
     /// HTTP request status counters for handled requests
     pub(crate) storage_controller_http_request_status:
         measured::CounterVec<HttpRequestStatusLabelGroupSet>,
-
     /// HTTP request handler latency across all status codes
-    #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
     pub(crate) storage_controller_http_request_latency:
         measured::HistogramVec<HttpRequestLatencyLabelGroupSet, 5>,
 
@@ -57,7 +52,6 @@ pub(crate) struct StorageControllerMetricGroup {
     /// Latency of HTTP requests to the pageserver, broken down by pageserver
     /// node id, request name and method. This include both successful and unsuccessful
     /// requests.
-    #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
     pub(crate) storage_controller_pageserver_request_latency:
         measured::HistogramVec<PageserverRequestLabelGroupSet, 5>,
 
@@ -69,7 +63,6 @@ pub(crate) struct StorageControllerMetricGroup {
     /// Latency of pass-through HTTP requests to the pageserver, broken down by pageserver
     /// node id, request name and method. This include both successful and unsuccessful
     /// requests.
-    #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
     pub(crate) storage_controller_passthrough_request_latency:
         measured::HistogramVec<PageserverRequestLabelGroupSet, 5>,
 
@@ -78,34 +71,75 @@ pub(crate) struct StorageControllerMetricGroup {
         measured::CounterVec<DatabaseQueryErrorLabelGroupSet>,
 
     /// Latency of database queries, broken down by operation.
-    #[metric(metadata = histogram::Thresholds::exponential_buckets(0.1, 2.0))]
     pub(crate) storage_controller_database_query_latency:
         measured::HistogramVec<DatabaseQueryLatencyLabelGroupSet, 5>,
 }
 
 impl StorageControllerMetrics {
-    pub(crate) fn encode(&self, neon_metrics: &NeonMetrics) -> Bytes {
+    pub(crate) fn encode(&self) -> Bytes {
         let mut encoder = self.encoder.lock().unwrap();
-        neon_metrics
-            .collect_group_into(&mut *encoder)
-            .unwrap_or_else(|infallible| match infallible {});
-        self.metrics_group
-            .collect_group_into(&mut *encoder)
-            .unwrap_or_else(|infallible| match infallible {});
+        self.metrics_group.collect_into(&mut *encoder);
         encoder.finish()
     }
 }
 
 impl Default for StorageControllerMetrics {
     fn default() -> Self {
-        let mut metrics_group = StorageControllerMetricGroup::new();
-        metrics_group
-            .storage_controller_reconcile_complete
-            .init_all_dense();
-
         Self {
-            metrics_group,
-            encoder: Mutex::new(measured::text::BufferedTextEncoder::new()),
+            metrics_group: StorageControllerMetricGroup::new(),
+            encoder: Mutex::new(measured::text::TextEncoder::new()),
+        }
+    }
+}
+
+impl StorageControllerMetricGroup {
+    pub(crate) fn new() -> Self {
+        Self {
+            storage_controller_reconcile_spawn: measured::Counter::new(),
+            storage_controller_reconcile_complete: measured::CounterVec::new(
+                ReconcileCompleteLabelGroupSet {
+                    status: StaticLabelSet::new(),
+                },
+            ),
+            storage_controller_http_request_status: measured::CounterVec::new(
+                HttpRequestStatusLabelGroupSet {
+                    path: lasso::ThreadedRodeo::new(),
+                    method: StaticLabelSet::new(),
+                    status: StaticLabelSet::new(),
+                },
+            ),
+            storage_controller_http_request_latency: measured::HistogramVec::new(
+                measured::metric::histogram::Thresholds::exponential_buckets(0.1, 2.0),
+            ),
+            storage_controller_pageserver_request_error: measured::CounterVec::new(
+                PageserverRequestLabelGroupSet {
+                    pageserver_id: lasso::ThreadedRodeo::new(),
+                    path: lasso::ThreadedRodeo::new(),
+                    method: StaticLabelSet::new(),
+                },
+            ),
+            storage_controller_pageserver_request_latency: measured::HistogramVec::new(
+                measured::metric::histogram::Thresholds::exponential_buckets(0.1, 2.0),
+            ),
+            storage_controller_passthrough_request_error: measured::CounterVec::new(
+                PageserverRequestLabelGroupSet {
+                    pageserver_id: lasso::ThreadedRodeo::new(),
+                    path: lasso::ThreadedRodeo::new(),
+                    method: StaticLabelSet::new(),
+                },
+            ),
+            storage_controller_passthrough_request_latency: measured::HistogramVec::new(
+                measured::metric::histogram::Thresholds::exponential_buckets(0.1, 2.0),
+            ),
+            storage_controller_database_query_error: measured::CounterVec::new(
+                DatabaseQueryErrorLabelGroupSet {
+                    operation: StaticLabelSet::new(),
+                    error_type: StaticLabelSet::new(),
+                },
+            ),
+            storage_controller_database_query_latency: measured::HistogramVec::new(
+                measured::metric::histogram::Thresholds::exponential_buckets(0.1, 2.0),
+            ),
         }
     }
 }
@@ -119,7 +153,7 @@ pub(crate) struct ReconcileCompleteLabelGroup {
 #[derive(measured::LabelGroup)]
 #[label(set = HttpRequestStatusLabelGroupSet)]
 pub(crate) struct HttpRequestStatusLabelGroup<'a> {
-    #[label(dynamic_with = lasso::ThreadedRodeo, default)]
+    #[label(dynamic_with = lasso::ThreadedRodeo)]
     pub(crate) path: &'a str,
     pub(crate) method: Method,
     pub(crate) status: StatusCode,
@@ -128,21 +162,40 @@ pub(crate) struct HttpRequestStatusLabelGroup<'a> {
 #[derive(measured::LabelGroup)]
 #[label(set = HttpRequestLatencyLabelGroupSet)]
 pub(crate) struct HttpRequestLatencyLabelGroup<'a> {
-    #[label(dynamic_with = lasso::ThreadedRodeo, default)]
+    #[label(dynamic_with = lasso::ThreadedRodeo)]
     pub(crate) path: &'a str,
     pub(crate) method: Method,
 }
 
+impl Default for HttpRequestLatencyLabelGroupSet {
+    fn default() -> Self {
+        Self {
+            path: lasso::ThreadedRodeo::new(),
+            method: StaticLabelSet::new(),
+        }
+    }
+}
+
 #[derive(measured::LabelGroup, Clone)]
 #[label(set = PageserverRequestLabelGroupSet)]
 pub(crate) struct PageserverRequestLabelGroup<'a> {
-    #[label(dynamic_with = lasso::ThreadedRodeo, default)]
+    #[label(dynamic_with = lasso::ThreadedRodeo)]
     pub(crate) pageserver_id: &'a str,
-    #[label(dynamic_with = lasso::ThreadedRodeo, default)]
+    #[label(dynamic_with = lasso::ThreadedRodeo)]
     pub(crate) path: &'a str,
     pub(crate) method: Method,
 }
 
+impl Default for PageserverRequestLabelGroupSet {
+    fn default() -> Self {
+        Self {
+            pageserver_id: lasso::ThreadedRodeo::new(),
+            path: lasso::ThreadedRodeo::new(),
+            method: StaticLabelSet::new(),
+        }
+    }
+}
+
 #[derive(measured::LabelGroup)]
 #[label(set = DatabaseQueryErrorLabelGroupSet)]
 pub(crate) struct DatabaseQueryErrorLabelGroup {
@@ -156,7 +209,7 @@ pub(crate) struct DatabaseQueryLatencyLabelGroup {
     pub(crate) operation: DatabaseOperation,
 }
 
-#[derive(FixedCardinalityLabel, Clone, Copy)]
+#[derive(FixedCardinalityLabel)]
 pub(crate) enum ReconcileOutcome {
     #[label(rename = "ok")]
     Success,
@@ -164,7 +217,7 @@ pub(crate) enum ReconcileOutcome {
     Cancel,
 }
 
-#[derive(FixedCardinalityLabel, Copy, Clone)]
+#[derive(FixedCardinalityLabel, Clone)]
 pub(crate) enum Method {
     Get,
     Put,
@@ -189,12 +242,11 @@ impl From<hyper::Method> for Method {
     }
 }
 
-#[derive(Clone, Copy)]
 pub(crate) struct StatusCode(pub(crate) hyper::http::StatusCode);
 
 impl LabelValue for StatusCode {
     fn visit<V: measured::label::LabelVisitor>(&self, v: V) -> V::Output {
-        v.write_int(self.0.as_u16() as i64)
+        v.write_int(self.0.as_u16() as u64)
     }
 }
 
@@ -212,7 +264,7 @@ impl FixedCardinalityLabel for StatusCode {
     }
 }
 
-#[derive(FixedCardinalityLabel, Clone, Copy)]
+#[derive(FixedCardinalityLabel)]
 pub(crate) enum DatabaseErrorLabel {
     Query,
     Connection,

control_plane/attachment_service/src/node.rs

@@ -3,8 +3,7 @@ use std::{str::FromStr, time::Duration};
 use hyper::StatusCode;
 use pageserver_api::{
     controller_api::{
-        NodeAvailability, NodeDescribeResponse, NodeRegisterRequest, NodeSchedulingPolicy,
-        TenantLocateResponseShard,
+        NodeAvailability, NodeRegisterRequest, NodeSchedulingPolicy, TenantLocateResponseShard,
     },
     shard::TenantShardId,
 };
@@ -257,19 +256,6 @@ impl Node {
         )
         .await
     }
-
-    /// Generate the simplified API-friendly description of a node's state
-    pub(crate) fn describe(&self) -> NodeDescribeResponse {
-        NodeDescribeResponse {
-            id: self.id,
-            availability: self.availability.into(),
-            scheduling: self.scheduling,
-            listen_http_addr: self.listen_http_addr.clone(),
-            listen_http_port: self.listen_http_port,
-            listen_pg_addr: self.listen_pg_addr.clone(),
-            listen_pg_port: self.listen_pg_port,
-        }
-    }
 }
 
 impl std::fmt::Display for Node {

control_plane/attachment_service/src/persistence.rs

@@ -2,7 +2,6 @@ pub(crate) mod split_state;
 use std::collections::HashMap;
 use std::str::FromStr;
 use std::time::Duration;
-use std::time::Instant;
 
 use self::split_state::SplitState;
 use camino::Utf8Path;
@@ -10,7 +9,6 @@ use camino::Utf8PathBuf;
 use diesel::pg::PgConnection;
 use diesel::prelude::*;
 use diesel::Connection;
-use pageserver_api::controller_api::ShardSchedulingPolicy;
 use pageserver_api::controller_api::{NodeSchedulingPolicy, PlacementPolicy};
 use pageserver_api::models::TenantConfig;
 use pageserver_api::shard::ShardConfigError;
@@ -80,7 +78,7 @@ pub(crate) enum DatabaseError {
     Logical(String),
 }
 
-#[derive(measured::FixedCardinalityLabel, Copy, Clone)]
+#[derive(measured::FixedCardinalityLabel, Clone)]
 pub(crate) enum DatabaseOperation {
     InsertNode,
     UpdateNode,
@@ -109,12 +107,6 @@ pub(crate) enum AbortShardSplitStatus {
 
 pub(crate) type DatabaseResult<T> = Result<T, DatabaseError>;
 
-/// Some methods can operate on either a whole tenant or a single shard
-pub(crate) enum TenantFilter {
-    Tenant(TenantId),
-    Shard(TenantShardId),
-}
-
 impl Persistence {
     // The default postgres connection limit is 100.  We use up to 99, to leave one free for a human admin under
     // normal circumstances.  This assumes we have exclusive use of the database cluster to which we connect.
@@ -145,41 +137,18 @@ impl Persistence {
         }
     }
 
-    /// A helper for use during startup, where we would like to tolerate concurrent restarts of the
-    /// database and the storage controller, therefore the database might not be available right away
-    pub async fn await_connection(
-        database_url: &str,
-        timeout: Duration,
-    ) -> Result<(), diesel::ConnectionError> {
-        let started_at = Instant::now();
-        loop {
-            match PgConnection::establish(database_url) {
-                Ok(_) => {
-                    tracing::info!("Connected to database.");
-                    return Ok(());
-                }
-                Err(e) => {
-                    if started_at.elapsed() > timeout {
-                        return Err(e);
-                    } else {
-                        tracing::info!("Database not yet available, waiting... ({e})");
-                        tokio::time::sleep(Duration::from_millis(100)).await;
-                    }
-                }
-            }
-        }
-    }
-
     /// Wraps `with_conn` in order to collect latency and error metrics
     async fn with_measured_conn<F, R>(&self, op: DatabaseOperation, func: F) -> DatabaseResult<R>
     where
-        F: FnOnce(&mut PgConnection) -> DatabaseResult<R> + Send + 'static,
+        F: Fn(&mut PgConnection) -> DatabaseResult<R> + Send + 'static,
         R: Send + 'static,
     {
         let latency = &METRICS_REGISTRY
             .metrics_group
             .storage_controller_database_query_latency;
-        let _timer = latency.start_timer(DatabaseQueryLatencyLabelGroup { operation: op });
+        let _timer = latency.start_timer(DatabaseQueryLatencyLabelGroup {
+            operation: op.clone(),
+        });
 
         let res = self.with_conn(func).await;
 
@@ -199,7 +168,7 @@ impl Persistence {
     /// Call the provided function in a tokio blocking thread, with a Diesel database connection.
     async fn with_conn<F, R>(&self, func: F) -> DatabaseResult<R>
     where
-        F: FnOnce(&mut PgConnection) -> DatabaseResult<R> + Send + 'static,
+        F: Fn(&mut PgConnection) -> DatabaseResult<R> + Send + 'static,
         R: Send + 'static,
     {
         let mut conn = self.connection_pool.get()?;
@@ -306,11 +275,6 @@ impl Persistence {
                 // Backward compat for test data after PR https://github.com/neondatabase/neon/pull/7165
                 shard.placement_policy = "{\"Attached\":0}".to_string();
             }
-
-            if shard.scheduling_policy.is_empty() {
-                shard.scheduling_policy =
-                    serde_json::to_string(&ShardSchedulingPolicy::default()).unwrap();
-            }
         }
 
         let tenants: Vec<TenantShardPersistence> = decoded.tenants.into_values().collect();
@@ -501,45 +465,59 @@ impl Persistence {
     /// that we only do the first time a tenant is set to an attached policy via /location_config.
     pub(crate) async fn update_tenant_shard(
         &self,
-        tenant: TenantFilter,
-        input_placement_policy: Option<PlacementPolicy>,
-        input_config: Option<TenantConfig>,
+        tenant_shard_id: TenantShardId,
+        input_placement_policy: PlacementPolicy,
+        input_config: TenantConfig,
         input_generation: Option<Generation>,
-        input_scheduling_policy: Option<ShardSchedulingPolicy>,
     ) -> DatabaseResult<()> {
         use crate::schema::tenant_shards::dsl::*;
 
         self.with_measured_conn(DatabaseOperation::UpdateTenantShard, move |conn| {
-            let query = match tenant {
-                TenantFilter::Shard(tenant_shard_id) => diesel::update(tenant_shards)
-                    .filter(tenant_id.eq(tenant_shard_id.tenant_id.to_string()))
-                    .filter(shard_number.eq(tenant_shard_id.shard_number.0 as i32))
-                    .filter(shard_count.eq(tenant_shard_id.shard_count.literal() as i32))
-                    .into_boxed(),
-                TenantFilter::Tenant(input_tenant_id) => diesel::update(tenant_shards)
-                    .filter(tenant_id.eq(input_tenant_id.to_string()))
-                    .into_boxed(),
-            };
+            let query = diesel::update(tenant_shards)
+                .filter(tenant_id.eq(tenant_shard_id.tenant_id.to_string()))
+                .filter(shard_number.eq(tenant_shard_id.shard_number.0 as i32))
+                .filter(shard_count.eq(tenant_shard_id.shard_count.literal() as i32));
 
-            #[derive(AsChangeset)]
-            #[diesel(table_name = crate::schema::tenant_shards)]
-            struct ShardUpdate {
-                generation: Option<i32>,
-                placement_policy: Option<String>,
-                config: Option<String>,
-                scheduling_policy: Option<String>,
+            if let Some(input_generation) = input_generation {
+                // Update includes generation column
+                query
+                    .set((
+                        generation.eq(Some(input_generation.into().unwrap() as i32)),
+                        placement_policy
+                            .eq(serde_json::to_string(&input_placement_policy).unwrap()),
+                        config.eq(serde_json::to_string(&input_config).unwrap()),
+                    ))
+                    .execute(conn)?;
+            } else {
+                // Update does not include generation column
+                query
+                    .set((
+                        placement_policy
+                            .eq(serde_json::to_string(&input_placement_policy).unwrap()),
+                        config.eq(serde_json::to_string(&input_config).unwrap()),
+                    ))
+                    .execute(conn)?;
             }
 
-            let update = ShardUpdate {
-                generation: input_generation.map(|g| g.into().unwrap() as i32),
-                placement_policy: input_placement_policy
-                    .map(|p| serde_json::to_string(&p).unwrap()),
-                config: input_config.map(|c| serde_json::to_string(&c).unwrap()),
-                scheduling_policy: input_scheduling_policy
-                    .map(|p| serde_json::to_string(&p).unwrap()),
-            };
+            Ok(())
+        })
+        .await?;
 
-            query.set(update).execute(conn)?;
+        Ok(())
+    }
+
+    pub(crate) async fn update_tenant_config(
+        &self,
+        input_tenant_id: TenantId,
+        input_config: TenantConfig,
+    ) -> DatabaseResult<()> {
+        use crate::schema::tenant_shards::dsl::*;
+
+        self.with_measured_conn(DatabaseOperation::UpdateTenantConfig, move |conn| {
+            diesel::update(tenant_shards)
+                .filter(tenant_id.eq(input_tenant_id.to_string()))
+                .set((config.eq(serde_json::to_string(&input_config).unwrap()),))
+                .execute(conn)?;
 
             Ok(())
         })
@@ -720,7 +698,7 @@ impl Persistence {
     }
 }
 
-/// Parts of [`crate::tenant_shard::TenantShard`] that are stored durably
+/// Parts of [`crate::tenant_state::TenantState`] that are stored durably
 #[derive(Queryable, Selectable, Insertable, Serialize, Deserialize, Clone, Eq, PartialEq)]
 #[diesel(table_name = crate::schema::tenant_shards)]
 pub(crate) struct TenantShardPersistence {
@@ -750,8 +728,6 @@ pub(crate) struct TenantShardPersistence {
     pub(crate) splitting: SplitState,
     #[serde(default)]
     pub(crate) config: String,
-    #[serde(default)]
-    pub(crate) scheduling_policy: String,
 }
 
 impl TenantShardPersistence {

control_plane/attachment_service/src/reconciler.rs

@@ -18,14 +18,14 @@ use utils::sync::gate::GateGuard;
 
 use crate::compute_hook::{ComputeHook, NotifyError};
 use crate::node::Node;
-use crate::tenant_shard::{IntentState, ObservedState, ObservedStateLocation};
+use crate::tenant_state::{IntentState, ObservedState, ObservedStateLocation};
 
 const DEFAULT_HEATMAP_PERIOD: &str = "60s";
 
 /// Object with the lifetime of the background reconcile task that is created
 /// for tenants which have a difference between their intent and observed states.
 pub(super) struct Reconciler {
-    /// See [`crate::tenant_shard::TenantShard`] for the meanings of these fields: they are a snapshot
+    /// See [`crate::tenant_state::TenantState`] for the meanings of these fields: they are a snapshot
     /// of a tenant's state from when we spawned a reconcile task.
     pub(super) tenant_shard_id: TenantShardId,
     pub(crate) shard: ShardIdentity,
@@ -48,11 +48,11 @@ pub(super) struct Reconciler {
 
     /// To avoid stalling if the cloud control plane is unavailable, we may proceed
     /// past failures in [`ComputeHook::notify`], but we _must_ remember that we failed
-    /// so that we can set [`crate::tenant_shard::TenantShard::pending_compute_notification`] to ensure a later retry.
+    /// so that we can set [`crate::tenant_state::TenantState::pending_compute_notification`] to ensure a later retry.
     pub(crate) compute_notify_failure: bool,
 
     /// A means to abort background reconciliation: it is essential to
-    /// call this when something changes in the original TenantShard that
+    /// call this when something changes in the original TenantState that
     /// will make this reconciliation impossible or unnecessary, for
     /// example when a pageserver node goes offline, or the PlacementPolicy for
     /// the tenant is changed.
@@ -66,7 +66,7 @@ pub(super) struct Reconciler {
     pub(crate) persistence: Arc<Persistence>,
 }
 
-/// This is a snapshot of [`crate::tenant_shard::IntentState`], but it does not do any
+/// This is a snapshot of [`crate::tenant_state::IntentState`], but it does not do any
 /// reference counting for Scheduler.  The IntentState is what the scheduler works with,
 /// and the TargetState is just the instruction for a particular Reconciler run.
 #[derive(Debug)]
@@ -487,7 +487,6 @@ impl Reconciler {
         while let Err(e) = self.compute_notify().await {
             match e {
                 NotifyError::Fatal(_) => return Err(ReconcileError::Notify(e)),
-                NotifyError::ShuttingDown => return Err(ReconcileError::Cancel),
                 _ => {
                     tracing::warn!(
                         "Live migration blocked by compute notification error, retrying: {e}"

control_plane/attachment_service/src/scheduler.rs

@@ -1,4 +1,4 @@
-use crate::{node::Node, tenant_shard::TenantShard};
+use crate::{node::Node, tenant_state::TenantState};
 use pageserver_api::controller_api::UtilizationScore;
 use serde::Serialize;
 use std::collections::HashMap;
@@ -27,7 +27,7 @@ pub enum MaySchedule {
 
 #[derive(Serialize)]
 struct SchedulerNode {
-    /// How many shards are currently scheduled on this node, via their [`crate::tenant_shard::IntentState`].
+    /// How many shards are currently scheduled on this node, via their [`crate::tenant_state::IntentState`].
     shard_count: usize,
 
     /// Whether this node is currently elegible to have new shards scheduled (this is derived
@@ -58,86 +58,6 @@ pub(crate) struct Scheduler {
     nodes: HashMap<NodeId, SchedulerNode>,
 }
 
-/// Score for soft constraint scheduling: lower scores are preferred to higher scores.
-///
-/// For example, we may set an affinity score based on the number of shards from the same
-/// tenant already on a node, to implicitly prefer to balance out shards.
-#[derive(Copy, Clone, Debug, Eq, PartialEq, PartialOrd, Ord)]
-pub(crate) struct AffinityScore(pub(crate) usize);
-
-impl AffinityScore {
-    /// If we have no anti-affinity at all toward a node, this is its score.  It means
-    /// the scheduler has a free choice amongst nodes with this score, and may pick a node
-    /// based on other information such as total utilization.
-    pub(crate) const FREE: Self = Self(0);
-
-    pub(crate) fn inc(&mut self) {
-        self.0 += 1;
-    }
-}
-
-impl std::ops::Add for AffinityScore {
-    type Output = Self;
-
-    fn add(self, rhs: Self) -> Self::Output {
-        Self(self.0 + rhs.0)
-    }
-}
-
-/// Hint for whether this is a sincere attempt to schedule, or a speculative
-/// check for where we _would_ schedule (done during optimization)
-#[derive(Debug)]
-pub(crate) enum ScheduleMode {
-    Normal,
-    Speculative,
-}
-
-impl Default for ScheduleMode {
-    fn default() -> Self {
-        Self::Normal
-    }
-}
-
-// For carrying state between multiple calls to [`TenantShard::schedule`], e.g. when calling
-// it for many shards in the same tenant.
-#[derive(Debug, Default)]
-pub(crate) struct ScheduleContext {
-    /// Sparse map of nodes: omitting a node implicitly makes its affinity [`AffinityScore::FREE`]
-    pub(crate) nodes: HashMap<NodeId, AffinityScore>,
-
-    /// Specifically how many _attached_ locations are on each node
-    pub(crate) attached_nodes: HashMap<NodeId, usize>,
-
-    pub(crate) mode: ScheduleMode,
-}
-
-impl ScheduleContext {
-    /// Input is a list of nodes we would like to avoid using again within this context.  The more
-    /// times a node is passed into this call, the less inclined we are to use it.
-    pub(crate) fn avoid(&mut self, nodes: &[NodeId]) {
-        for node_id in nodes {
-            let entry = self.nodes.entry(*node_id).or_insert(AffinityScore::FREE);
-            entry.inc()
-        }
-    }
-
-    pub(crate) fn push_attached(&mut self, node_id: NodeId) {
-        let entry = self.attached_nodes.entry(node_id).or_default();
-        *entry += 1;
-    }
-
-    pub(crate) fn get_node_affinity(&self, node_id: NodeId) -> AffinityScore {
-        self.nodes
-            .get(&node_id)
-            .copied()
-            .unwrap_or(AffinityScore::FREE)
-    }
-
-    pub(crate) fn get_node_attachments(&self, node_id: NodeId) -> usize {
-        self.attached_nodes.get(&node_id).copied().unwrap_or(0)
-    }
-}
-
 impl Scheduler {
     pub(crate) fn new<'a>(nodes: impl Iterator<Item = &'a Node>) -> Self {
         let mut scheduler_nodes = HashMap::new();
@@ -163,7 +83,7 @@ impl Scheduler {
     pub(crate) fn consistency_check<'a>(
         &self,
         nodes: impl Iterator<Item = &'a Node>,
-        shards: impl Iterator<Item = &'a TenantShard>,
+        shards: impl Iterator<Item = &'a TenantState>,
     ) -> anyhow::Result<()> {
         let mut expect_nodes: HashMap<NodeId, SchedulerNode> = HashMap::new();
         for node in nodes {
@@ -304,87 +224,53 @@ impl Scheduler {
         node.and_then(|(node_id, may_schedule)| if may_schedule { Some(node_id) } else { None })
     }
 
-    /// hard_exclude: it is forbidden to use nodes in this list, typically becacuse they
-    /// are already in use by this shard -- we use this to avoid picking the same node
-    /// as both attached and secondary location.  This is a hard constraint: if we cannot
-    /// find any nodes that aren't in this list, then we will return a [`ScheduleError::ImpossibleConstraint`].
-    ///
-    /// context: we prefer to avoid using nodes identified in the context, according
-    /// to their anti-affinity score.  We use this to prefeer to avoid placing shards in
-    /// the same tenant on the same node.  This is a soft constraint: the context will never
-    /// cause us to fail to schedule a shard.
-    pub(crate) fn schedule_shard(
-        &self,
-        hard_exclude: &[NodeId],
-        context: &ScheduleContext,
-    ) -> Result<NodeId, ScheduleError> {
+    pub(crate) fn schedule_shard(&self, hard_exclude: &[NodeId]) -> Result<NodeId, ScheduleError> {
         if self.nodes.is_empty() {
             return Err(ScheduleError::NoPageservers);
         }
 
-        let mut scores: Vec<(NodeId, AffinityScore, usize)> = self
+        let mut tenant_counts: Vec<(NodeId, usize)> = self
             .nodes
             .iter()
             .filter_map(|(k, v)| {
                 if hard_exclude.contains(k) || v.may_schedule == MaySchedule::No {
                     None
                 } else {
-                    Some((
-                        *k,
-                        context.nodes.get(k).copied().unwrap_or(AffinityScore::FREE),
-                        v.shard_count,
-                    ))
+                    Some((*k, v.shard_count))
                 }
             })
             .collect();
 
-        // Sort by, in order of precedence:
-        //  1st: Affinity score.  We should never pick a higher-score node if a lower-score node is available
-        //  2nd: Utilization.  Within nodes with the same affinity, use the least loaded nodes.
-        //  3rd: Node ID.  This is a convenience to make selection deterministic in tests and empty systems.
-        scores.sort_by_key(|i| (i.1, i.2, i.0));
+        // Sort by tenant count.  Nodes with the same tenant count are sorted by ID.
+        tenant_counts.sort_by_key(|i| (i.1, i.0));
 
-        if scores.is_empty() {
-            // After applying constraints, no pageservers were left.
-            if !matches!(context.mode, ScheduleMode::Speculative) {
-                // If this was not a speculative attempt, log details to understand why we couldn't
-                // schedule: this may help an engineer understand if some nodes are marked offline
-                // in a way that's preventing progress.
+        if tenant_counts.is_empty() {
+            // After applying constraints, no pageservers were left.  We log some detail about
+            // the state of nodes to help understand why this happened.  This is not logged as an error because
+            // it is legitimately possible for enough nodes to be Offline to prevent scheduling a shard.
+            tracing::info!("Scheduling failure, while excluding {hard_exclude:?}, node states:");
+            for (node_id, node) in &self.nodes {
                 tracing::info!(
-                    "Scheduling failure, while excluding {hard_exclude:?}, node states:"
+                    "Node {node_id}: may_schedule={} shards={}",
+                    node.may_schedule != MaySchedule::No,
+                    node.shard_count
                 );
-                for (node_id, node) in &self.nodes {
-                    tracing::info!(
-                        "Node {node_id}: may_schedule={} shards={}",
-                        node.may_schedule != MaySchedule::No,
-                        node.shard_count
-                    );
-                }
             }
+
             return Err(ScheduleError::ImpossibleConstraint);
         }
 
-        // Lowest score wins
-        let node_id = scores.first().unwrap().0;
-
-        if !matches!(context.mode, ScheduleMode::Speculative) {
-            tracing::info!(
-            "scheduler selected node {node_id} (elegible nodes {:?}, hard exclude: {hard_exclude:?}, soft exclude: {context:?})",
-            scores.iter().map(|i| i.0 .0).collect::<Vec<_>>()
+        let node_id = tenant_counts.first().unwrap().0;
+        tracing::info!(
+            "scheduler selected node {node_id} (elegible nodes {:?}, exclude: {hard_exclude:?})",
+            tenant_counts.iter().map(|i| i.0 .0).collect::<Vec<_>>()
         );
-        }
 
         // Note that we do not update shard count here to reflect the scheduling: that
         // is IntentState's job when the scheduled location is used.
 
         Ok(node_id)
     }
-
-    /// Unit test access to internal state
-    #[cfg(test)]
-    pub(crate) fn get_node_shard_count(&self, node_id: NodeId) -> usize {
-        self.nodes.get(&node_id).unwrap().shard_count
-    }
 }
 
 #[cfg(test)]
@@ -421,7 +307,7 @@ pub(crate) mod test_utils {
 mod tests {
     use super::*;
 
-    use crate::tenant_shard::IntentState;
+    use crate::tenant_state::IntentState;
     #[test]
     fn scheduler_basic() -> anyhow::Result<()> {
         let nodes = test_utils::make_test_nodes(2);
@@ -430,17 +316,15 @@ mod tests {
         let mut t1_intent = IntentState::new();
         let mut t2_intent = IntentState::new();
 
-        let context = ScheduleContext::default();
-
-        let scheduled = scheduler.schedule_shard(&[], &context)?;
+        let scheduled = scheduler.schedule_shard(&[])?;
         t1_intent.set_attached(&mut scheduler, Some(scheduled));
-        let scheduled = scheduler.schedule_shard(&[], &context)?;
+        let scheduled = scheduler.schedule_shard(&[])?;
         t2_intent.set_attached(&mut scheduler, Some(scheduled));
 
         assert_eq!(scheduler.nodes.get(&NodeId(1)).unwrap().shard_count, 1);
         assert_eq!(scheduler.nodes.get(&NodeId(2)).unwrap().shard_count, 1);
 
-        let scheduled = scheduler.schedule_shard(&t1_intent.all_pageservers(), &context)?;
+        let scheduled = scheduler.schedule_shard(&t1_intent.all_pageservers())?;
         t1_intent.push_secondary(&mut scheduler, scheduled);
 
         assert_eq!(scheduler.nodes.get(&NodeId(1)).unwrap().shard_count, 1);

control_plane/attachment_service/src/schema.rs

@@ -22,7 +22,6 @@ diesel::table! {
         placement_policy -> Varchar,
         splitting -> Int2,
         config -> Text,
-        scheduling_policy -> Varchar,
     }
 }
 

control_plane/attachment_service/src/tenant_state.rs

@@ -7,9 +7,8 @@ use std::{
 use crate::{
     metrics::{self, ReconcileCompleteLabelGroup, ReconcileOutcome},
     persistence::TenantShardPersistence,
-    scheduler::{AffinityScore, MaySchedule, ScheduleContext},
 };
-use pageserver_api::controller_api::{PlacementPolicy, ShardSchedulingPolicy};
+use pageserver_api::controller_api::PlacementPolicy;
 use pageserver_api::{
     models::{LocationConfig, LocationConfigMode, TenantConfig},
     shard::{ShardIdentity, TenantShardId},
@@ -50,7 +49,7 @@ where
 /// This struct implement Serialize for debugging purposes, but is _not_ persisted
 /// itself: see [`crate::persistence`] for the subset of tenant shard state that is persisted.
 #[derive(Serialize)]
-pub(crate) struct TenantShard {
+pub(crate) struct TenantState {
     pub(crate) tenant_shard_id: TenantShardId,
 
     pub(crate) shard: ShardIdentity,
@@ -117,10 +116,6 @@ pub(crate) struct TenantShard {
     /// sending it.  This is the mechanism by which compute notifications are included in the scope
     /// of state that we publish externally in an eventually consistent way.
     pub(crate) pending_compute_notification: bool,
-
-    // Support/debug tool: if something is going wrong or flapping with scheduling, this may
-    // be set to a non-active state to avoid making changes while the issue is fixed.
-    scheduling_policy: ShardSchedulingPolicy,
 }
 
 #[derive(Default, Clone, Debug, Serialize)]
@@ -251,13 +246,8 @@ impl IntentState {
 
 impl Drop for IntentState {
     fn drop(&mut self) {
-        // Must clear before dropping, to avoid leaving stale refcounts in the Scheduler.
-        // We do not check this while panicking, to avoid polluting unit test failures or
-        // other assertions with this assertion's output.  It's still wrong to leak these,
-        // but if we already have a panic then we don't need to independently flag this case.
-        if !(std::thread::panicking()) {
-            debug_assert!(self.attached.is_none() && self.secondary.is_empty());
-        }
+        // Must clear before dropping, to avoid leaving stale refcounts in the Scheduler
+        debug_assert!(self.attached.is_none() && self.secondary.is_empty());
     }
 }
 
@@ -302,26 +292,6 @@ pub enum ReconcileWaitError {
     Failed(TenantShardId, String),
 }
 
-#[derive(Eq, PartialEq, Debug)]
-pub(crate) struct ReplaceSecondary {
-    old_node_id: NodeId,
-    new_node_id: NodeId,
-}
-
-#[derive(Eq, PartialEq, Debug)]
-pub(crate) struct MigrateAttachment {
-    old_attached_node_id: NodeId,
-    new_attached_node_id: NodeId,
-}
-
-#[derive(Eq, PartialEq, Debug)]
-pub(crate) enum ScheduleOptimization {
-    // Replace one of our secondary locations with a different node
-    ReplaceSecondary(ReplaceSecondary),
-    // Migrate attachment to an existing secondary location
-    MigrateAttachment(MigrateAttachment),
-}
-
 impl ReconcilerWaiter {
     pub(crate) async fn wait_timeout(&self, timeout: Duration) -> Result<(), ReconcileWaitError> {
         tokio::select! {
@@ -354,7 +324,7 @@ pub(crate) struct ReconcilerHandle {
 }
 
 /// When a reconcile task completes, it sends this result object
-/// to be applied to the primary TenantShard.
+/// to be applied to the primary TenantState.
 pub(crate) struct ReconcileResult {
     pub(crate) sequence: Sequence,
     /// On errors, `observed` should be treated as an incompleted description
@@ -367,7 +337,7 @@ pub(crate) struct ReconcileResult {
     pub(crate) generation: Option<Generation>,
     pub(crate) observed: ObservedState,
 
-    /// Set [`TenantShard::pending_compute_notification`] from this flag
+    /// Set [`TenantState::pending_compute_notification`] from this flag
     pub(crate) pending_compute_notification: bool,
 }
 
@@ -379,7 +349,7 @@ impl ObservedState {
     }
 }
 
-impl TenantShard {
+impl TenantState {
     pub(crate) fn new(
         tenant_shard_id: TenantShardId,
         shard: ShardIdentity,
@@ -400,7 +370,6 @@ impl TenantShard {
             error_waiter: Arc::new(SeqWait::new(Sequence(0))),
             last_error: Arc::default(),
             pending_compute_notification: false,
-            scheduling_policy: ShardSchedulingPolicy::default(),
         }
     }
 
@@ -456,7 +425,6 @@ impl TenantShard {
     fn schedule_attached(
         &mut self,
         scheduler: &mut Scheduler,
-        context: &ScheduleContext,
     ) -> Result<(bool, NodeId), ScheduleError> {
         // No work to do if we already have an attached tenant
         if let Some(node_id) = self.intent.attached {
@@ -470,33 +438,14 @@ impl TenantShard {
             Ok((true, promote_secondary))
         } else {
             // Pick a fresh node: either we had no secondaries or none were schedulable
-            let node_id = scheduler.schedule_shard(&self.intent.secondary, context)?;
+            let node_id = scheduler.schedule_shard(&self.intent.secondary)?;
             tracing::debug!("Selected {} as attached", node_id);
             self.intent.set_attached(scheduler, Some(node_id));
             Ok((true, node_id))
         }
     }
 
-    pub(crate) fn schedule(
-        &mut self,
-        scheduler: &mut Scheduler,
-        context: &mut ScheduleContext,
-    ) -> Result<(), ScheduleError> {
-        let r = self.do_schedule(scheduler, context);
-
-        context.avoid(&self.intent.all_pageservers());
-        if let Some(attached) = self.intent.get_attached() {
-            context.push_attached(*attached);
-        }
-
-        r
-    }
-
-    pub(crate) fn do_schedule(
-        &mut self,
-        scheduler: &mut Scheduler,
-        context: &ScheduleContext,
-    ) -> Result<(), ScheduleError> {
+    pub(crate) fn schedule(&mut self, scheduler: &mut Scheduler) -> Result<(), ScheduleError> {
         // TODO: before scheduling new nodes, check if any existing content in
         // self.intent refers to pageservers that are offline, and pick other
         // pageservers if so.
@@ -504,16 +453,6 @@ impl TenantShard {
         // TODO: respect the splitting bit on tenants: if they are currently splitting then we may not
         // change their attach location.
 
-        match self.scheduling_policy {
-            ShardSchedulingPolicy::Active | ShardSchedulingPolicy::Essential => {}
-            ShardSchedulingPolicy::Pause | ShardSchedulingPolicy::Stop => {
-                // Warn to make it obvious why other things aren't happening/working, if we skip scheduling
-                tracing::warn!(tenant_id=%self.tenant_shard_id.tenant_id, shard_id=%self.tenant_shard_id.shard_slug(),
-                    "Scheduling is disabled by policy {:?}", self.scheduling_policy);
-                return Ok(());
-            }
-        }
-
         // Build the set of pageservers already in use by this tenant, to avoid scheduling
         // more work on the same pageservers we're already using.
         let mut modified = false;
@@ -540,13 +479,12 @@ impl TenantShard {
                 }
 
                 // Should have exactly one attached, and N secondaries
-                let (modified_attached, attached_node_id) =
-                    self.schedule_attached(scheduler, context)?;
+                let (modified_attached, attached_node_id) = self.schedule_attached(scheduler)?;
                 modified |= modified_attached;
 
                 let mut used_pageservers = vec![attached_node_id];
                 while self.intent.secondary.len() < secondary_count {
-                    let node_id = scheduler.schedule_shard(&used_pageservers, context)?;
+                    let node_id = scheduler.schedule_shard(&used_pageservers)?;
                     self.intent.push_secondary(scheduler, node_id);
                     used_pageservers.push(node_id);
                     modified = true;
@@ -559,7 +497,7 @@ impl TenantShard {
                     modified = true;
                 } else if self.intent.secondary.is_empty() {
                     // Populate secondary by scheduling a fresh node
-                    let node_id = scheduler.schedule_shard(&[], context)?;
+                    let node_id = scheduler.schedule_shard(&[])?;
                     self.intent.push_secondary(scheduler, node_id);
                     modified = true;
                 }
@@ -586,167 +524,6 @@ impl TenantShard {
         Ok(())
     }
 
-    /// Optimize attachments: if a shard has a secondary location that is preferable to
-    /// its primary location based on soft constraints, switch that secondary location
-    /// to be attached.
-    #[instrument(skip_all, fields(tenant_id=%self.tenant_shard_id.tenant_id, shard_id=%self.tenant_shard_id.shard_slug()))]
-    pub(crate) fn optimize_attachment(
-        &self,
-        nodes: &HashMap<NodeId, Node>,
-        schedule_context: &ScheduleContext,
-    ) -> Option<ScheduleOptimization> {
-        let attached = (*self.intent.get_attached())?;
-        if self.intent.secondary.is_empty() {
-            // We can only do useful work if we have both attached and secondary locations: this
-            // function doesn't schedule new locations, only swaps between attached and secondaries.
-            return None;
-        }
-
-        let current_affinity_score = schedule_context.get_node_affinity(attached);
-        let current_attachment_count = schedule_context.get_node_attachments(attached);
-
-        // Generate score for each node, dropping any un-schedulable nodes.
-        let all_pageservers = self.intent.all_pageservers();
-        let mut scores = all_pageservers
-            .iter()
-            .flat_map(|node_id| {
-                if matches!(
-                    nodes
-                        .get(node_id)
-                        .map(|n| n.may_schedule())
-                        .unwrap_or(MaySchedule::No),
-                    MaySchedule::No
-                ) {
-                    None
-                } else {
-                    let affinity_score = schedule_context.get_node_affinity(*node_id);
-                    let attachment_count = schedule_context.get_node_attachments(*node_id);
-                    Some((*node_id, affinity_score, attachment_count))
-                }
-            })
-            .collect::<Vec<_>>();
-
-        // Sort precedence:
-        //  1st - prefer nodes with the lowest total affinity score
-        //  2nd - prefer nodes with the lowest number of attachments in this context
-        //  3rd - if all else is equal, sort by node ID for determinism in tests.
-        scores.sort_by_key(|i| (i.1, i.2, i.0));
-
-        if let Some((preferred_node, preferred_affinity_score, preferred_attachment_count)) =
-            scores.first()
-        {
-            if attached != *preferred_node {
-                // The best alternative must be more than 1 better than us, otherwise we could end
-                // up flapping back next time we're called (e.g. there's no point migrating from
-                // a location with score 1 to a score zero, because on next location the situation
-                // would be the same, but in reverse).
-                if current_affinity_score > *preferred_affinity_score + AffinityScore(1)
-                    || current_attachment_count > *preferred_attachment_count + 1
-                {
-                    tracing::info!(
-                        "Identified optimization: migrate attachment {attached}->{preferred_node} (secondaries {:?})",
-                        self.intent.get_secondary()
-                    );
-                    return Some(ScheduleOptimization::MigrateAttachment(MigrateAttachment {
-                        old_attached_node_id: attached,
-                        new_attached_node_id: *preferred_node,
-                    }));
-                }
-            } else {
-                tracing::debug!(
-                    "Node {} is already preferred (score {:?})",
-                    preferred_node,
-                    preferred_affinity_score
-                );
-            }
-        }
-
-        // Fall-through: we didn't find an optimization
-        None
-    }
-
-    #[instrument(skip_all, fields(tenant_id=%self.tenant_shard_id.tenant_id, shard_id=%self.tenant_shard_id.shard_slug()))]
-    pub(crate) fn optimize_secondary(
-        &self,
-        scheduler: &Scheduler,
-        schedule_context: &ScheduleContext,
-    ) -> Option<ScheduleOptimization> {
-        if self.intent.secondary.is_empty() {
-            // We can only do useful work if we have both attached and secondary locations: this
-            // function doesn't schedule new locations, only swaps between attached and secondaries.
-            return None;
-        }
-
-        for secondary in self.intent.get_secondary() {
-            let Some(affinity_score) = schedule_context.nodes.get(secondary) else {
-                // We're already on a node unaffected any affinity constraints,
-                // so we won't change it.
-                continue;
-            };
-
-            // Let the scheduler suggest a node, where it would put us if we were scheduling afresh
-            // This implicitly limits the choice to nodes that are available, and prefers nodes
-            // with lower utilization.
-            let Ok(candidate_node) =
-                scheduler.schedule_shard(&self.intent.all_pageservers(), schedule_context)
-            else {
-                // A scheduling error means we have no possible candidate replacements
-                continue;
-            };
-
-            let candidate_affinity_score = schedule_context
-                .nodes
-                .get(&candidate_node)
-                .unwrap_or(&AffinityScore::FREE);
-
-            // The best alternative must be more than 1 better than us, otherwise we could end
-            // up flapping back next time we're called.
-            if *candidate_affinity_score + AffinityScore(1) < *affinity_score {
-                // If some other node is available and has a lower score than this node, then
-                // that other node is a good place to migrate to.
-                tracing::info!(
-                    "Identified optimization: replace secondary {secondary}->{candidate_node} (current secondaries {:?})",
-                    self.intent.get_secondary()
-                );
-                return Some(ScheduleOptimization::ReplaceSecondary(ReplaceSecondary {
-                    old_node_id: *secondary,
-                    new_node_id: candidate_node,
-                }));
-            }
-        }
-
-        None
-    }
-
-    pub(crate) fn apply_optimization(
-        &mut self,
-        scheduler: &mut Scheduler,
-        optimization: ScheduleOptimization,
-    ) {
-        metrics::METRICS_REGISTRY
-            .metrics_group
-            .storage_controller_schedule_optimization
-            .inc();
-
-        match optimization {
-            ScheduleOptimization::MigrateAttachment(MigrateAttachment {
-                old_attached_node_id,
-                new_attached_node_id,
-            }) => {
-                self.intent.demote_attached(old_attached_node_id);
-                self.intent
-                    .promote_attached(scheduler, new_attached_node_id);
-            }
-            ScheduleOptimization::ReplaceSecondary(ReplaceSecondary {
-                old_node_id,
-                new_node_id,
-            }) => {
-                self.intent.remove_secondary(scheduler, old_node_id);
-                self.intent.push_secondary(scheduler, new_node_id);
-            }
-        }
-    }
-
     /// Query whether the tenant's observed state for attached node matches its intent state, and if so,
     /// yield the node ID.  This is appropriate for emitting compute hook notifications: we are checking that
     /// the node in question is not only where we intend to attach, but that the tenant is indeed already attached there.
@@ -891,19 +668,6 @@ impl TenantShard {
             }
         }
 
-        // Pre-checks done: finally check whether we may actually do the work
-        match self.scheduling_policy {
-            ShardSchedulingPolicy::Active
-            | ShardSchedulingPolicy::Essential
-            | ShardSchedulingPolicy::Pause => {}
-            ShardSchedulingPolicy::Stop => {
-                // We only reach this point if there is work to do and we're going to skip
-                // doing it: warn it obvious why this tenant isn't doing what it ought to.
-                tracing::warn!("Skipping reconcile for policy {:?}", self.scheduling_policy);
-                return None;
-            }
-        }
-
         // Build list of nodes from which the reconciler should detach
         let mut detach = Vec::new();
         for node_id in self.observed.locations.keys() {
@@ -1040,22 +804,6 @@ impl TenantShard {
         })
     }
 
-    /// Get a waiter for any reconciliation in flight, but do not start reconciliation
-    /// if it is not already running
-    pub(crate) fn get_waiter(&self) -> Option<ReconcilerWaiter> {
-        if self.reconciler.is_some() {
-            Some(ReconcilerWaiter {
-                tenant_shard_id: self.tenant_shard_id,
-                seq_wait: self.waiter.clone(),
-                error_seq_wait: self.error_waiter.clone(),
-                error: self.last_error.clone(),
-                seq: self.sequence,
-            })
-        } else {
-            None
-        }
-    }
-
     /// Called when a ReconcileResult has been emitted and the service is updating
     /// our state: if the result is from a sequence >= my ReconcileHandle, then drop
     /// the handle to indicate there is no longer a reconciliation in progress.
@@ -1081,40 +829,6 @@ impl TenantShard {
         debug_assert!(!self.intent.all_pageservers().contains(&node_id));
     }
 
-    pub(crate) fn set_scheduling_policy(&mut self, p: ShardSchedulingPolicy) {
-        self.scheduling_policy = p;
-    }
-
-    pub(crate) fn get_scheduling_policy(&self) -> &ShardSchedulingPolicy {
-        &self.scheduling_policy
-    }
-
-    pub(crate) fn from_persistent(
-        tsp: TenantShardPersistence,
-        intent: IntentState,
-    ) -> anyhow::Result<Self> {
-        let tenant_shard_id = tsp.get_tenant_shard_id()?;
-        let shard_identity = tsp.get_shard_identity()?;
-
-        Ok(Self {
-            tenant_shard_id,
-            shard: shard_identity,
-            sequence: Sequence::initial(),
-            generation: tsp.generation.map(|g| Generation::new(g as u32)),
-            policy: serde_json::from_str(&tsp.placement_policy).unwrap(),
-            intent,
-            observed: ObservedState::new(),
-            config: serde_json::from_str(&tsp.config).unwrap(),
-            reconciler: None,
-            splitting: tsp.splitting,
-            waiter: Arc::new(SeqWait::new(Sequence::initial())),
-            error_waiter: Arc::new(SeqWait::new(Sequence::initial())),
-            last_error: Arc::default(),
-            pending_compute_notification: false,
-            scheduling_policy: serde_json::from_str(&tsp.scheduling_policy).unwrap(),
-        })
-    }
-
     pub(crate) fn to_persistent(&self) -> TenantShardPersistence {
         TenantShardPersistence {
             tenant_id: self.tenant_shard_id.tenant_id.to_string(),
@@ -1126,7 +840,6 @@ impl TenantShard {
             placement_policy: serde_json::to_string(&self.policy).unwrap(),
             config: serde_json::to_string(&self.config).unwrap(),
             splitting: SplitState::default(),
-            scheduling_policy: serde_json::to_string(&self.scheduling_policy).unwrap(),
         }
     }
 }
@@ -1143,7 +856,7 @@ pub(crate) mod tests {
 
     use super::*;
 
-    fn make_test_tenant_shard(policy: PlacementPolicy) -> TenantShard {
+    fn make_test_tenant_shard(policy: PlacementPolicy) -> TenantState {
         let tenant_id = TenantId::generate();
         let shard_number = ShardNumber(0);
         let shard_count = ShardCount::new(1);
@@ -1153,7 +866,7 @@ pub(crate) mod tests {
             shard_number,
             shard_count,
         };
-        TenantShard::new(
+        TenantState::new(
             tenant_shard_id,
             ShardIdentity::new(
                 shard_number,
@@ -1165,32 +878,6 @@ pub(crate) mod tests {
         )
     }
 
-    fn make_test_tenant(policy: PlacementPolicy, shard_count: ShardCount) -> Vec<TenantShard> {
-        let tenant_id = TenantId::generate();
-
-        (0..shard_count.count())
-            .map(|i| {
-                let shard_number = ShardNumber(i);
-
-                let tenant_shard_id = TenantShardId {
-                    tenant_id,
-                    shard_number,
-                    shard_count,
-                };
-                TenantShard::new(
-                    tenant_shard_id,
-                    ShardIdentity::new(
-                        shard_number,
-                        shard_count,
-                        pageserver_api::shard::ShardStripeSize(32768),
-                    )
-                    .unwrap(),
-                    policy.clone(),
-                )
-            })
-            .collect()
-    }
-
     /// Test the scheduling behaviors used when a tenant configured for HA is subject
     /// to nodes being marked offline.
     #[test]
@@ -1200,26 +887,25 @@ pub(crate) mod tests {
         let mut nodes = make_test_nodes(3);
 
         let mut scheduler = Scheduler::new(nodes.values());
-        let mut context = ScheduleContext::default();
 
-        let mut tenant_shard = make_test_tenant_shard(PlacementPolicy::Attached(1));
-        tenant_shard
-            .schedule(&mut scheduler, &mut context)
+        let mut tenant_state = make_test_tenant_shard(PlacementPolicy::Attached(1));
+        tenant_state
+            .schedule(&mut scheduler)
             .expect("we have enough nodes, scheduling should work");
 
         // Expect to initially be schedule on to different nodes
-        assert_eq!(tenant_shard.intent.secondary.len(), 1);
-        assert!(tenant_shard.intent.attached.is_some());
+        assert_eq!(tenant_state.intent.secondary.len(), 1);
+        assert!(tenant_state.intent.attached.is_some());
 
-        let attached_node_id = tenant_shard.intent.attached.unwrap();
-        let secondary_node_id = *tenant_shard.intent.secondary.iter().last().unwrap();
+        let attached_node_id = tenant_state.intent.attached.unwrap();
+        let secondary_node_id = *tenant_state.intent.secondary.iter().last().unwrap();
         assert_ne!(attached_node_id, secondary_node_id);
 
         // Notifying the attached node is offline should demote it to a secondary
-        let changed = tenant_shard.intent.demote_attached(attached_node_id);
+        let changed = tenant_state.intent.demote_attached(attached_node_id);
         assert!(changed);
-        assert!(tenant_shard.intent.attached.is_none());
-        assert_eq!(tenant_shard.intent.secondary.len(), 2);
+        assert!(tenant_state.intent.attached.is_none());
+        assert_eq!(tenant_state.intent.secondary.len(), 2);
 
         // Update the scheduler state to indicate the node is offline
         nodes
@@ -1229,18 +915,18 @@ pub(crate) mod tests {
         scheduler.node_upsert(nodes.get(&attached_node_id).unwrap());
 
         // Scheduling the node should promote the still-available secondary node to attached
-        tenant_shard
-            .schedule(&mut scheduler, &mut context)
+        tenant_state
+            .schedule(&mut scheduler)
             .expect("active nodes are available");
-        assert_eq!(tenant_shard.intent.attached.unwrap(), secondary_node_id);
+        assert_eq!(tenant_state.intent.attached.unwrap(), secondary_node_id);
 
         // The original attached node should have been retained as a secondary
         assert_eq!(
-            *tenant_shard.intent.secondary.iter().last().unwrap(),
+            *tenant_state.intent.secondary.iter().last().unwrap(),
             attached_node_id
         );
 
-        tenant_shard.intent.clear(&mut scheduler);
+        tenant_state.intent.clear(&mut scheduler);
 
         Ok(())
     }
@@ -1250,263 +936,48 @@ pub(crate) mod tests {
         let nodes = make_test_nodes(3);
         let mut scheduler = Scheduler::new(nodes.values());
 
-        let mut tenant_shard = make_test_tenant_shard(PlacementPolicy::Attached(1));
+        let mut tenant_state = make_test_tenant_shard(PlacementPolicy::Attached(1));
 
-        tenant_shard.observed.locations.insert(
+        tenant_state.observed.locations.insert(
             NodeId(3),
             ObservedStateLocation {
                 conf: Some(LocationConfig {
                     mode: LocationConfigMode::AttachedMulti,
                     generation: Some(2),
                     secondary_conf: None,
-                    shard_number: tenant_shard.shard.number.0,
-                    shard_count: tenant_shard.shard.count.literal(),
-                    shard_stripe_size: tenant_shard.shard.stripe_size.0,
+                    shard_number: tenant_state.shard.number.0,
+                    shard_count: tenant_state.shard.count.literal(),
+                    shard_stripe_size: tenant_state.shard.stripe_size.0,
                     tenant_conf: TenantConfig::default(),
                 }),
             },
         );
 
-        tenant_shard.observed.locations.insert(
+        tenant_state.observed.locations.insert(
             NodeId(2),
             ObservedStateLocation {
                 conf: Some(LocationConfig {
                     mode: LocationConfigMode::AttachedStale,
                     generation: Some(1),
                     secondary_conf: None,
-                    shard_number: tenant_shard.shard.number.0,
-                    shard_count: tenant_shard.shard.count.literal(),
-                    shard_stripe_size: tenant_shard.shard.stripe_size.0,
+                    shard_number: tenant_state.shard.number.0,
+                    shard_count: tenant_state.shard.count.literal(),
+                    shard_stripe_size: tenant_state.shard.stripe_size.0,
                     tenant_conf: TenantConfig::default(),
                 }),
             },
         );
 
-        tenant_shard.intent_from_observed(&mut scheduler);
+        tenant_state.intent_from_observed(&mut scheduler);
 
         // The highest generationed attached location gets used as attached
-        assert_eq!(tenant_shard.intent.attached, Some(NodeId(3)));
+        assert_eq!(tenant_state.intent.attached, Some(NodeId(3)));
         // Other locations get used as secondary
-        assert_eq!(tenant_shard.intent.secondary, vec![NodeId(2)]);
+        assert_eq!(tenant_state.intent.secondary, vec![NodeId(2)]);
 
-        scheduler.consistency_check(nodes.values(), [&tenant_shard].into_iter())?;
-
-        tenant_shard.intent.clear(&mut scheduler);
-        Ok(())
-    }
-
-    #[test]
-    fn scheduling_mode() -> anyhow::Result<()> {
-        let nodes = make_test_nodes(3);
-        let mut scheduler = Scheduler::new(nodes.values());
-
-        let mut tenant_shard = make_test_tenant_shard(PlacementPolicy::Attached(1));
-
-        // In pause mode, schedule() shouldn't do anything
-        tenant_shard.scheduling_policy = ShardSchedulingPolicy::Pause;
-        assert!(tenant_shard
-            .schedule(&mut scheduler, &mut ScheduleContext::default())
-            .is_ok());
-        assert!(tenant_shard.intent.all_pageservers().is_empty());
-
-        // In active mode, schedule() works
-        tenant_shard.scheduling_policy = ShardSchedulingPolicy::Active;
-        assert!(tenant_shard
-            .schedule(&mut scheduler, &mut ScheduleContext::default())
-            .is_ok());
-        assert!(!tenant_shard.intent.all_pageservers().is_empty());
-
-        tenant_shard.intent.clear(&mut scheduler);
-        Ok(())
-    }
-
-    #[test]
-    fn optimize_attachment() -> anyhow::Result<()> {
-        let nodes = make_test_nodes(3);
-        let mut scheduler = Scheduler::new(nodes.values());
-
-        let mut shard_a = make_test_tenant_shard(PlacementPolicy::Attached(1));
-        let mut shard_b = make_test_tenant_shard(PlacementPolicy::Attached(1));
-
-        // Initially: both nodes attached on shard 1, and both have secondary locations
-        // on different nodes.
-        shard_a.intent.set_attached(&mut scheduler, Some(NodeId(1)));
-        shard_a.intent.push_secondary(&mut scheduler, NodeId(2));
-        shard_b.intent.set_attached(&mut scheduler, Some(NodeId(1)));
-        shard_b.intent.push_secondary(&mut scheduler, NodeId(3));
-
-        let mut schedule_context = ScheduleContext::default();
-        schedule_context.avoid(&shard_a.intent.all_pageservers());
-        schedule_context.push_attached(shard_a.intent.get_attached().unwrap());
-        schedule_context.avoid(&shard_b.intent.all_pageservers());
-        schedule_context.push_attached(shard_b.intent.get_attached().unwrap());
-
-        let optimization_a = shard_a.optimize_attachment(&nodes, &schedule_context);
-
-        // Either shard should recognize that it has the option to switch to a secondary location where there
-        // would be no other shards from the same tenant, and request to do so.
-        assert_eq!(
-            optimization_a,
-            Some(ScheduleOptimization::MigrateAttachment(MigrateAttachment {
-                old_attached_node_id: NodeId(1),
-                new_attached_node_id: NodeId(2)
-            }))
-        );
-
-        // Note that these optimizing two shards in the same tenant with the same ScheduleContext is
-        // mutually exclusive (the optimization of one invalidates the stats) -- it is the responsibility
-        // of [`Service::optimize_all`] to avoid trying
-        // to do optimizations for multiple shards in the same tenant at the same time.  Generating
-        // both optimizations is just done for test purposes
-        let optimization_b = shard_b.optimize_attachment(&nodes, &schedule_context);
-        assert_eq!(
-            optimization_b,
-            Some(ScheduleOptimization::MigrateAttachment(MigrateAttachment {
-                old_attached_node_id: NodeId(1),
-                new_attached_node_id: NodeId(3)
-            }))
-        );
-
-        // Applying these optimizations should result in the end state proposed
-        shard_a.apply_optimization(&mut scheduler, optimization_a.unwrap());
-        assert_eq!(shard_a.intent.get_attached(), &Some(NodeId(2)));
-        assert_eq!(shard_a.intent.get_secondary(), &vec![NodeId(1)]);
-        shard_b.apply_optimization(&mut scheduler, optimization_b.unwrap());
-        assert_eq!(shard_b.intent.get_attached(), &Some(NodeId(3)));
-        assert_eq!(shard_b.intent.get_secondary(), &vec![NodeId(1)]);
-
-        shard_a.intent.clear(&mut scheduler);
-        shard_b.intent.clear(&mut scheduler);
-
-        Ok(())
-    }
-
-    #[test]
-    fn optimize_secondary() -> anyhow::Result<()> {
-        let nodes = make_test_nodes(4);
-        let mut scheduler = Scheduler::new(nodes.values());
-
-        let mut shard_a = make_test_tenant_shard(PlacementPolicy::Attached(1));
-        let mut shard_b = make_test_tenant_shard(PlacementPolicy::Attached(1));
-
-        // Initially: both nodes attached on shard 1, and both have secondary locations
-        // on different nodes.
-        shard_a.intent.set_attached(&mut scheduler, Some(NodeId(1)));
-        shard_a.intent.push_secondary(&mut scheduler, NodeId(3));
-        shard_b.intent.set_attached(&mut scheduler, Some(NodeId(2)));
-        shard_b.intent.push_secondary(&mut scheduler, NodeId(3));
-
-        let mut schedule_context = ScheduleContext::default();
-        schedule_context.avoid(&shard_a.intent.all_pageservers());
-        schedule_context.push_attached(shard_a.intent.get_attached().unwrap());
-        schedule_context.avoid(&shard_b.intent.all_pageservers());
-        schedule_context.push_attached(shard_b.intent.get_attached().unwrap());
-
-        let optimization_a = shard_a.optimize_secondary(&scheduler, &schedule_context);
-
-        // Since there is a node with no locations available, the node with two locations for the
-        // same tenant should generate an optimization to move one away
-        assert_eq!(
-            optimization_a,
-            Some(ScheduleOptimization::ReplaceSecondary(ReplaceSecondary {
-                old_node_id: NodeId(3),
-                new_node_id: NodeId(4)
-            }))
-        );
-
-        shard_a.apply_optimization(&mut scheduler, optimization_a.unwrap());
-        assert_eq!(shard_a.intent.get_attached(), &Some(NodeId(1)));
-        assert_eq!(shard_a.intent.get_secondary(), &vec![NodeId(4)]);
-
-        shard_a.intent.clear(&mut scheduler);
-        shard_b.intent.clear(&mut scheduler);
-
-        Ok(())
-    }
-
-    // Optimize til quiescent: this emulates what Service::optimize_all does, when
-    // called repeatedly in the background.
-    fn optimize_til_idle(
-        nodes: &HashMap<NodeId, Node>,
-        scheduler: &mut Scheduler,
-        shards: &mut [TenantShard],
-    ) {
-        let mut loop_n = 0;
-        loop {
-            let mut schedule_context = ScheduleContext::default();
-            let mut any_changed = false;
-
-            for shard in shards.iter() {
-                schedule_context.avoid(&shard.intent.all_pageservers());
-                if let Some(attached) = shard.intent.get_attached() {
-                    schedule_context.push_attached(*attached);
-                }
-            }
-
-            for shard in shards.iter_mut() {
-                let optimization = shard.optimize_attachment(nodes, &schedule_context);
-                if let Some(optimization) = optimization {
-                    shard.apply_optimization(scheduler, optimization);
-                    any_changed = true;
-                    break;
-                }
-
-                let optimization = shard.optimize_secondary(scheduler, &schedule_context);
-                if let Some(optimization) = optimization {
-                    shard.apply_optimization(scheduler, optimization);
-                    any_changed = true;
-                    break;
-                }
-            }
-
-            if !any_changed {
-                break;
-            }
-
-            // Assert no infinite loop
-            loop_n += 1;
-            assert!(loop_n < 1000);
-        }
-    }
-
-    /// Test the balancing behavior of shard scheduling: that it achieves a balance, and
-    /// that it converges.
-    #[test]
-    fn optimize_add_nodes() -> anyhow::Result<()> {
-        let nodes = make_test_nodes(4);
-
-        // Only show the scheduler a couple of nodes
-        let mut scheduler = Scheduler::new([].iter());
-        scheduler.node_upsert(nodes.get(&NodeId(1)).unwrap());
-        scheduler.node_upsert(nodes.get(&NodeId(2)).unwrap());
-
-        let mut shards = make_test_tenant(PlacementPolicy::Attached(1), ShardCount::new(4));
-        let mut schedule_context = ScheduleContext::default();
-        for shard in &mut shards {
-            assert!(shard
-                .schedule(&mut scheduler, &mut schedule_context)
-                .is_ok());
-        }
-
-        // We should see equal number of locations on the two nodes.
-        assert_eq!(scheduler.get_node_shard_count(NodeId(1)), 4);
-        assert_eq!(scheduler.get_node_shard_count(NodeId(2)), 4);
-
-        // Add another two nodes: we should see the shards spread out when their optimize
-        // methods are called
-        scheduler.node_upsert(nodes.get(&NodeId(3)).unwrap());
-        scheduler.node_upsert(nodes.get(&NodeId(4)).unwrap());
-        optimize_til_idle(&nodes, &mut scheduler, &mut shards);
-
-        assert_eq!(scheduler.get_node_shard_count(NodeId(1)), 2);
-        assert_eq!(scheduler.get_node_shard_count(NodeId(2)), 2);
-        assert_eq!(scheduler.get_node_shard_count(NodeId(3)), 2);
-        assert_eq!(scheduler.get_node_shard_count(NodeId(4)), 2);
-
-        for shard in shards.iter_mut() {
-            shard.intent.clear(&mut scheduler);
-        }
+        scheduler.consistency_check(nodes.values(), [&tenant_state].into_iter())?;
 
+        tenant_state.intent.clear(&mut scheduler);
         Ok(())
     }
 }

control_plane/src/background_process.rs

@@ -86,10 +86,7 @@ where
         .stdout(process_log_file)
         .stderr(same_file_for_stderr)
         .args(args);
-
-    let filled_cmd = fill_env_vars_prefixed_neon(fill_remote_storage_secrets_vars(
-        fill_rust_env_vars(background_command),
-    ));
+    let filled_cmd = fill_remote_storage_secrets_vars(fill_rust_env_vars(background_command));
     filled_cmd.envs(envs);
 
     let pid_file_to_check = match &initial_pid_file {
@@ -271,15 +268,6 @@ fn fill_remote_storage_secrets_vars(mut cmd: &mut Command) -> &mut Command {
     cmd
 }
 
-fn fill_env_vars_prefixed_neon(mut cmd: &mut Command) -> &mut Command {
-    for (var, val) in std::env::vars() {
-        if var.starts_with("NEON_PAGESERVER_") {
-            cmd = cmd.env(var, val);
-        }
-    }
-    cmd
-}
-
 /// Add a `pre_exec` to the cmd that, inbetween fork() and exec(),
 /// 1. Claims a pidfile with a fcntl lock on it and
 /// 2. Sets up the pidfile's file descriptor so that it (and the lock)
@@ -306,7 +294,7 @@ where
     //      is in state 'taken' but the thread that would unlock it is
     //      not there.
     //   2. A rust object that represented some external resource in the
-    //      parent now got implicitly copied by the fork, even though
+    //      parent now got implicitly copied by the the fork, even though
     //      the object's type is not `Copy`. The parent program may use
     //      non-copyability as way to enforce unique ownership of an
     //      external resource in the typesystem. The fork breaks that

control_plane/src/bin/neon_local.rs

@@ -14,7 +14,9 @@ use control_plane::pageserver::{PageServerNode, PAGESERVER_REMOTE_STORAGE_DIR};
 use control_plane::safekeeper::SafekeeperNode;
 use control_plane::storage_controller::StorageController;
 use control_plane::{broker, local_env};
-use pageserver_api::controller_api::PlacementPolicy;
+use pageserver_api::controller_api::{
+    NodeAvailability, NodeConfigureRequest, NodeSchedulingPolicy, PlacementPolicy,
+};
 use pageserver_api::models::{
     ShardParameters, TenantCreateRequest, TimelineCreateRequest, TimelineInfo,
 };
@@ -1058,6 +1060,21 @@ async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
             }
         }
 
+        Some(("set-state", subcommand_args)) => {
+            let pageserver = get_pageserver(env, subcommand_args)?;
+            let scheduling = subcommand_args.get_one("scheduling");
+            let availability = subcommand_args.get_one("availability");
+
+            let storage_controller = StorageController::from_env(env);
+            storage_controller
+                .node_configure(NodeConfigureRequest {
+                    node_id: pageserver.conf.id,
+                    scheduling: scheduling.cloned(),
+                    availability: availability.cloned(),
+                })
+                .await?;
+        }
+
         Some(("status", subcommand_args)) => {
             match get_pageserver(env, subcommand_args)?.check_status().await {
                 Ok(_) => println!("Page server is up and running"),
@@ -1231,7 +1248,7 @@ async fn try_stop_all(env: &local_env::LocalEnv, immediate: bool) {
     match ComputeControlPlane::load(env.clone()) {
         Ok(cplane) => {
             for (_k, node) in cplane.endpoints {
-                if let Err(e) = node.stop(if immediate { "immediate" } else { "fast" }, false) {
+                if let Err(e) = node.stop(if immediate { "immediate" } else { "fast " }, false) {
                     eprintln!("postgres stop failed: {e:#}");
                 }
             }
@@ -1417,7 +1434,6 @@ fn cli() -> Command {
         .subcommand(
             Command::new("timeline")
             .about("Manage timelines")
-            .arg_required_else_help(true)
             .subcommand(Command::new("list")
                 .about("List all timelines, available to this pageserver")
                 .arg(tenant_id_arg.clone()))
@@ -1499,6 +1515,12 @@ fn cli() -> Command {
                     .about("Restart local pageserver")
                     .arg(pageserver_config_args.clone())
                 )
+                .subcommand(Command::new("set-state")
+                    .arg(Arg::new("availability").value_parser(value_parser!(NodeAvailability)).long("availability").action(ArgAction::Set).help("Availability state: offline,active"))
+                    .arg(Arg::new("scheduling").value_parser(value_parser!(NodeSchedulingPolicy)).long("scheduling").action(ArgAction::Set).help("Scheduling state: draining,pause,filling,active"))
+                    .about("Set scheduling or availability state of pageserver node")
+                    .arg(pageserver_config_args.clone())
+                )
         )
         .subcommand(
             Command::new("storage_controller")

control_plane/src/endpoint.rs

@@ -12,7 +12,7 @@
 //!
 //! The endpoint is managed by the `compute_ctl` binary. When an endpoint is
 //! started, we launch `compute_ctl` It synchronizes the safekeepers, downloads
-//! the basebackup from the pageserver to initialize the data directory, and
+//! the basebackup from the pageserver to initialize the the data directory, and
 //! finally launches the PostgreSQL process. It watches the PostgreSQL process
 //! until it exits.
 //!

control_plane/src/local_env.rs

@@ -156,7 +156,6 @@ pub struct SafekeeperConf {
     pub remote_storage: Option<String>,
     pub backup_threads: Option<u32>,
     pub auth_enabled: bool,
-    pub listen_addr: Option<String>,
 }
 
 impl Default for SafekeeperConf {
@@ -170,7 +169,6 @@ impl Default for SafekeeperConf {
             remote_storage: None,
             backup_threads: None,
             auth_enabled: false,
-            listen_addr: None,
         }
     }
 }

control_plane/src/pageserver.rs

@@ -389,10 +389,6 @@ impl PageServerNode {
                 .remove("image_creation_threshold")
                 .map(|x| x.parse::<usize>())
                 .transpose()?,
-            image_layer_creation_check_threshold: settings
-                .remove("image_layer_creation_check_threshold")
-                .map(|x| x.parse::<u8>())
-                .transpose()?,
             pitr_interval: settings.remove("pitr_interval").map(|x| x.to_string()),
             walreceiver_connect_timeout: settings
                 .remove("walreceiver_connect_timeout")
@@ -505,12 +501,6 @@ impl PageServerNode {
                     .map(|x| x.parse::<usize>())
                     .transpose()
                     .context("Failed to parse 'image_creation_threshold' as non zero integer")?,
-                image_layer_creation_check_threshold: settings
-                    .remove("image_layer_creation_check_threshold")
-                    .map(|x| x.parse::<u8>())
-                    .transpose()
-                    .context("Failed to parse 'image_creation_check_threshold' as integer")?,
-
                 pitr_interval: settings.remove("pitr_interval").map(|x| x.to_string()),
                 walreceiver_connect_timeout: settings
                     .remove("walreceiver_connect_timeout")

control_plane/src/safekeeper.rs

@@ -70,31 +70,24 @@ pub struct SafekeeperNode {
     pub pg_connection_config: PgConnectionConfig,
     pub env: LocalEnv,
     pub http_client: reqwest::Client,
-    pub listen_addr: String,
     pub http_base_url: String,
 }
 
 impl SafekeeperNode {
     pub fn from_env(env: &LocalEnv, conf: &SafekeeperConf) -> SafekeeperNode {
-        let listen_addr = if let Some(ref listen_addr) = conf.listen_addr {
-            listen_addr.clone()
-        } else {
-            "127.0.0.1".to_string()
-        };
         SafekeeperNode {
             id: conf.id,
             conf: conf.clone(),
-            pg_connection_config: Self::safekeeper_connection_config(&listen_addr, conf.pg_port),
+            pg_connection_config: Self::safekeeper_connection_config(conf.pg_port),
             env: env.clone(),
             http_client: reqwest::Client::new(),
-            http_base_url: format!("http://{}:{}/v1", listen_addr, conf.http_port),
-            listen_addr,
+            http_base_url: format!("http://127.0.0.1:{}/v1", conf.http_port),
         }
     }
 
     /// Construct libpq connection string for connecting to this safekeeper.
-    fn safekeeper_connection_config(addr: &str, port: u16) -> PgConnectionConfig {
-        PgConnectionConfig::new_host_port(url::Host::parse(addr).unwrap(), port)
+    fn safekeeper_connection_config(port: u16) -> PgConnectionConfig {
+        PgConnectionConfig::new_host_port(url::Host::parse("127.0.0.1").unwrap(), port)
     }
 
     pub fn datadir_path_by_id(env: &LocalEnv, sk_id: NodeId) -> PathBuf {
@@ -118,8 +111,8 @@ impl SafekeeperNode {
         );
         io::stdout().flush().unwrap();
 
-        let listen_pg = format!("{}:{}", self.listen_addr, self.conf.pg_port);
-        let listen_http = format!("{}:{}", self.listen_addr, self.conf.http_port);
+        let listen_pg = format!("127.0.0.1:{}", self.conf.pg_port);
+        let listen_http = format!("127.0.0.1:{}", self.conf.http_port);
         let id = self.id;
         let datadir = self.datadir_path();
 
@@ -146,7 +139,7 @@ impl SafekeeperNode {
             availability_zone,
         ];
         if let Some(pg_tenant_only_port) = self.conf.pg_tenant_only_port {
-            let listen_pg_tenant_only = format!("{}:{}", self.listen_addr, pg_tenant_only_port);
+            let listen_pg_tenant_only = format!("127.0.0.1:{}", pg_tenant_only_port);
             args.extend(["--listen-pg-tenant-only".to_owned(), listen_pg_tenant_only]);
         }
         if !self.conf.sync {

control_plane/src/storage_controller.rs

@@ -279,7 +279,6 @@ impl StorageController {
             &self.listen,
             "-p",
             self.path.as_ref(),
-            "--dev",
             "--database-url",
             &database_url,
             "--max-unavailable-interval",

control_plane/storcon_cli/Cargo.toml

@@ -1,23 +0,0 @@
-[package]
-name = "storcon_cli"
-version = "0.1.0"
-edition.workspace = true
-license.workspace = true
-
-
-[dependencies]
-anyhow.workspace = true
-clap.workspace = true
-comfy-table.workspace = true
-hyper.workspace = true
-pageserver_api.workspace = true
-pageserver_client.workspace = true
-reqwest.workspace = true
-serde.workspace = true
-serde_json = { workspace = true, features = ["raw_value"] }
-thiserror.workspace = true
-tokio.workspace = true
-tracing.workspace = true
-utils.workspace = true
-workspace_hack.workspace = true
-

control_plane/storcon_cli/src/main.rs

@@ -1,681 +0,0 @@
-use std::{collections::HashMap, str::FromStr, time::Duration};
-
-use clap::{Parser, Subcommand};
-use hyper::{Method, StatusCode};
-use pageserver_api::{
-    controller_api::{
-        NodeAvailabilityWrapper, NodeDescribeResponse, ShardSchedulingPolicy,
-        TenantDescribeResponse, TenantPolicyRequest,
-    },
-    models::{
-        LocationConfigSecondary, ShardParameters, TenantConfig, TenantConfigRequest,
-        TenantCreateRequest, TenantShardSplitRequest, TenantShardSplitResponse,
-    },
-    shard::{ShardStripeSize, TenantShardId},
-};
-use pageserver_client::mgmt_api::{self, ResponseErrorMessageExt};
-use reqwest::Url;
-use serde::{de::DeserializeOwned, Serialize};
-use utils::id::{NodeId, TenantId};
-
-use pageserver_api::controller_api::{
-    NodeConfigureRequest, NodeRegisterRequest, NodeSchedulingPolicy, PlacementPolicy,
-    TenantLocateResponse, TenantShardMigrateRequest, TenantShardMigrateResponse,
-};
-
-#[derive(Subcommand, Debug)]
-enum Command {
-    /// Register a pageserver with the storage controller.  This shouldn't usually be necessary,
-    /// since pageservers auto-register when they start up
-    NodeRegister {
-        #[arg(long)]
-        node_id: NodeId,
-
-        #[arg(long)]
-        listen_pg_addr: String,
-        #[arg(long)]
-        listen_pg_port: u16,
-
-        #[arg(long)]
-        listen_http_addr: String,
-        #[arg(long)]
-        listen_http_port: u16,
-    },
-
-    /// Modify a node's configuration in the storage controller
-    NodeConfigure {
-        #[arg(long)]
-        node_id: NodeId,
-
-        /// Availability is usually auto-detected based on heartbeats.  Set 'offline' here to
-        /// manually mark a node offline
-        #[arg(long)]
-        availability: Option<NodeAvailabilityArg>,
-        /// Scheduling policy controls whether tenant shards may be scheduled onto this node.
-        #[arg(long)]
-        scheduling: Option<NodeSchedulingPolicy>,
-    },
-    /// Modify a tenant's policies in the storage controller
-    TenantPolicy {
-        #[arg(long)]
-        tenant_id: TenantId,
-        /// Placement policy controls whether a tenant is `detached`, has only a secondary location (`secondary`),
-        /// or is in the normal attached state with N secondary locations (`attached:N`)
-        #[arg(long)]
-        placement: Option<PlacementPolicyArg>,
-        /// Scheduling policy enables pausing the controller's scheduling activity involving this tenant.  `active` is normal,
-        /// `essential` disables optimization scheduling changes, `pause` disables all scheduling changes, and `stop` prevents
-        /// all reconciliation activity including for scheduling changes already made.  `pause` and `stop` can make a tenant
-        /// unavailable, and are only for use in emergencies.
-        #[arg(long)]
-        scheduling: Option<ShardSchedulingPolicyArg>,
-    },
-    /// List nodes known to the storage controller
-    Nodes {},
-    /// List tenants known to the storage controller
-    Tenants {},
-    /// Create a new tenant in the storage controller, and by extension on pageservers.
-    TenantCreate {
-        #[arg(long)]
-        tenant_id: TenantId,
-    },
-    /// Delete a tenant in the storage controller, and by extension on pageservers.
-    TenantDelete {
-        #[arg(long)]
-        tenant_id: TenantId,
-    },
-    /// Split an existing tenant into a higher number of shards than its current shard count.
-    TenantShardSplit {
-        #[arg(long)]
-        tenant_id: TenantId,
-        #[arg(long)]
-        shard_count: u8,
-        /// Optional, in 8kiB pages.  e.g. set 2048 for 16MB stripes.
-        #[arg(long)]
-        stripe_size: Option<u32>,
-    },
-    /// Migrate the attached location for a tenant shard to a specific pageserver.
-    TenantShardMigrate {
-        #[arg(long)]
-        tenant_shard_id: TenantShardId,
-        #[arg(long)]
-        node: NodeId,
-    },
-    /// Modify the pageserver tenant configuration of a tenant: this is the configuration structure
-    /// that is passed through to pageservers, and does not affect storage controller behavior.
-    TenantConfig {
-        #[arg(long)]
-        tenant_id: TenantId,
-        #[arg(long)]
-        config: String,
-    },
-    /// Attempt to balance the locations for a tenant across pageservers.  This is a client-side
-    /// alternative to the storage controller's scheduling optimization behavior.
-    TenantScatter {
-        #[arg(long)]
-        tenant_id: TenantId,
-    },
-    /// Print details about a particular tenant, including all its shards' states.
-    TenantDescribe {
-        #[arg(long)]
-        tenant_id: TenantId,
-    },
-    /// For a tenant which hasn't been onboarded to the storage controller yet, add it in secondary
-    /// mode so that it can warm up content on a pageserver.
-    TenantWarmup {
-        #[arg(long)]
-        tenant_id: TenantId,
-    },
-}
-
-#[derive(Parser)]
-#[command(
-    author,
-    version,
-    about,
-    long_about = "CLI for Storage Controller Support/Debug"
-)]
-#[command(arg_required_else_help(true))]
-struct Cli {
-    #[arg(long)]
-    /// URL to storage controller.  e.g. http://127.0.0.1:1234 when using `neon_local`
-    api: Url,
-
-    #[arg(long)]
-    /// JWT token for authenticating with storage controller.  Depending on the API used, this
-    /// should have either `pageserverapi` or `admin` scopes: for convenience, you should mint
-    /// a token with both scopes to use with this tool.
-    jwt: Option<String>,
-
-    #[command(subcommand)]
-    command: Command,
-}
-
-#[derive(Debug, Clone)]
-struct PlacementPolicyArg(PlacementPolicy);
-
-impl FromStr for PlacementPolicyArg {
-    type Err = anyhow::Error;
-
-    fn from_str(s: &str) -> Result<Self, Self::Err> {
-        match s {
-            "detached" => Ok(Self(PlacementPolicy::Detached)),
-            "secondary" => Ok(Self(PlacementPolicy::Secondary)),
-            _ if s.starts_with("attached:") => {
-                let mut splitter = s.split(':');
-                let _prefix = splitter.next().unwrap();
-                match splitter.next().and_then(|s| s.parse::<usize>().ok()) {
-                    Some(n) => Ok(Self(PlacementPolicy::Attached(n))),
-                    None => Err(anyhow::anyhow!(
-                        "Invalid format '{s}', a valid example is 'attached:1'"
-                    )),
-                }
-            }
-            _ => Err(anyhow::anyhow!(
-                "Unknown placement policy '{s}', try detached,secondary,attached:<n>"
-            )),
-        }
-    }
-}
-
-#[derive(Debug, Clone)]
-struct ShardSchedulingPolicyArg(ShardSchedulingPolicy);
-
-impl FromStr for ShardSchedulingPolicyArg {
-    type Err = anyhow::Error;
-
-    fn from_str(s: &str) -> Result<Self, Self::Err> {
-        match s {
-            "active" => Ok(Self(ShardSchedulingPolicy::Active)),
-            "essential" => Ok(Self(ShardSchedulingPolicy::Essential)),
-            "pause" => Ok(Self(ShardSchedulingPolicy::Pause)),
-            "stop" => Ok(Self(ShardSchedulingPolicy::Stop)),
-            _ => Err(anyhow::anyhow!(
-                "Unknown scheduling policy '{s}', try active,essential,pause,stop"
-            )),
-        }
-    }
-}
-
-#[derive(Debug, Clone)]
-struct NodeAvailabilityArg(NodeAvailabilityWrapper);
-
-impl FromStr for NodeAvailabilityArg {
-    type Err = anyhow::Error;
-
-    fn from_str(s: &str) -> Result<Self, Self::Err> {
-        match s {
-            "active" => Ok(Self(NodeAvailabilityWrapper::Active)),
-            "offline" => Ok(Self(NodeAvailabilityWrapper::Offline)),
-            _ => Err(anyhow::anyhow!("Unknown availability state '{s}'")),
-        }
-    }
-}
-
-struct Client {
-    base_url: Url,
-    jwt_token: Option<String>,
-    client: reqwest::Client,
-}
-
-impl Client {
-    fn new(base_url: Url, jwt_token: Option<String>) -> Self {
-        Self {
-            base_url,
-            jwt_token,
-            client: reqwest::ClientBuilder::new()
-                .build()
-                .expect("Failed to construct http client"),
-        }
-    }
-
-    /// Simple HTTP request wrapper for calling into storage controller
-    async fn dispatch<RQ, RS>(
-        &self,
-        method: hyper::Method,
-        path: String,
-        body: Option<RQ>,
-    ) -> mgmt_api::Result<RS>
-    where
-        RQ: Serialize + Sized,
-        RS: DeserializeOwned + Sized,
-    {
-        // The configured URL has the /upcall path prefix for pageservers to use: we will strip that out
-        // for general purpose API access.
-        let url = Url::from_str(&format!(
-            "http://{}:{}/{path}",
-            self.base_url.host_str().unwrap(),
-            self.base_url.port().unwrap()
-        ))
-        .unwrap();
-
-        let mut builder = self.client.request(method, url);
-        if let Some(body) = body {
-            builder = builder.json(&body)
-        }
-        if let Some(jwt_token) = &self.jwt_token {
-            builder = builder.header(
-                reqwest::header::AUTHORIZATION,
-                format!("Bearer {jwt_token}"),
-            );
-        }
-
-        let response = builder.send().await.map_err(mgmt_api::Error::ReceiveBody)?;
-        let response = response.error_from_body().await?;
-
-        response
-            .json()
-            .await
-            .map_err(pageserver_client::mgmt_api::Error::ReceiveBody)
-    }
-}
-
-#[tokio::main]
-async fn main() -> anyhow::Result<()> {
-    let cli = Cli::parse();
-
-    let storcon_client = Client::new(cli.api.clone(), cli.jwt.clone());
-
-    let mut trimmed = cli.api.to_string();
-    trimmed.pop();
-    let vps_client = mgmt_api::Client::new(trimmed, cli.jwt.as_deref());
-
-    match cli.command {
-        Command::NodeRegister {
-            node_id,
-            listen_pg_addr,
-            listen_pg_port,
-            listen_http_addr,
-            listen_http_port,
-        } => {
-            storcon_client
-                .dispatch::<_, ()>(
-                    Method::POST,
-                    "control/v1/node".to_string(),
-                    Some(NodeRegisterRequest {
-                        node_id,
-                        listen_pg_addr,
-                        listen_pg_port,
-                        listen_http_addr,
-                        listen_http_port,
-                    }),
-                )
-                .await?;
-        }
-        Command::TenantCreate { tenant_id } => {
-            vps_client
-                .tenant_create(&TenantCreateRequest {
-                    new_tenant_id: TenantShardId::unsharded(tenant_id),
-                    generation: None,
-                    shard_parameters: ShardParameters::default(),
-                    placement_policy: Some(PlacementPolicy::Attached(1)),
-                    config: TenantConfig::default(),
-                })
-                .await?;
-        }
-        Command::TenantDelete { tenant_id } => {
-            let status = vps_client
-                .tenant_delete(TenantShardId::unsharded(tenant_id))
-                .await?;
-            tracing::info!("Delete status: {}", status);
-        }
-        Command::Nodes {} => {
-            let resp = storcon_client
-                .dispatch::<(), Vec<NodeDescribeResponse>>(
-                    Method::GET,
-                    "control/v1/node".to_string(),
-                    None,
-                )
-                .await?;
-            let mut table = comfy_table::Table::new();
-            table.set_header(["Id", "Hostname", "Scheduling", "Availability"]);
-            for node in resp {
-                table.add_row([
-                    format!("{}", node.id),
-                    node.listen_http_addr,
-                    format!("{:?}", node.scheduling),
-                    format!("{:?}", node.availability),
-                ]);
-            }
-            println!("{table}");
-        }
-        Command::NodeConfigure {
-            node_id,
-            availability,
-            scheduling,
-        } => {
-            let req = NodeConfigureRequest {
-                node_id,
-                availability: availability.map(|a| a.0),
-                scheduling,
-            };
-            storcon_client
-                .dispatch::<_, ()>(
-                    Method::PUT,
-                    format!("control/v1/node/{node_id}/config"),
-                    Some(req),
-                )
-                .await?;
-        }
-        Command::Tenants {} => {
-            let resp = storcon_client
-                .dispatch::<(), Vec<TenantDescribeResponse>>(
-                    Method::GET,
-                    "control/v1/tenant".to_string(),
-                    None,
-                )
-                .await?;
-            let mut table = comfy_table::Table::new();
-            table.set_header([
-                "TenantId",
-                "ShardCount",
-                "StripeSize",
-                "Placement",
-                "Scheduling",
-            ]);
-            for tenant in resp {
-                let shard_zero = tenant.shards.into_iter().next().unwrap();
-                table.add_row([
-                    format!("{}", tenant.tenant_id),
-                    format!("{}", shard_zero.tenant_shard_id.shard_count.literal()),
-                    format!("{:?}", tenant.stripe_size),
-                    format!("{:?}", tenant.policy),
-                    format!("{:?}", shard_zero.scheduling_policy),
-                ]);
-            }
-
-            println!("{table}");
-        }
-        Command::TenantPolicy {
-            tenant_id,
-            placement,
-            scheduling,
-        } => {
-            let req = TenantPolicyRequest {
-                scheduling: scheduling.map(|s| s.0),
-                placement: placement.map(|p| p.0),
-            };
-            storcon_client
-                .dispatch::<_, ()>(
-                    Method::PUT,
-                    format!("control/v1/tenant/{tenant_id}/policy"),
-                    Some(req),
-                )
-                .await?;
-        }
-        Command::TenantShardSplit {
-            tenant_id,
-            shard_count,
-            stripe_size,
-        } => {
-            let req = TenantShardSplitRequest {
-                new_shard_count: shard_count,
-                new_stripe_size: stripe_size.map(ShardStripeSize),
-            };
-
-            let response = storcon_client
-                .dispatch::<TenantShardSplitRequest, TenantShardSplitResponse>(
-                    Method::PUT,
-                    format!("control/v1/tenant/{tenant_id}/shard_split"),
-                    Some(req),
-                )
-                .await?;
-            println!(
-                "Split tenant {} into {} shards: {}",
-                tenant_id,
-                shard_count,
-                response
-                    .new_shards
-                    .iter()
-                    .map(|s| format!("{:?}", s))
-                    .collect::<Vec<_>>()
-                    .join(",")
-            );
-        }
-        Command::TenantShardMigrate {
-            tenant_shard_id,
-            node,
-        } => {
-            let req = TenantShardMigrateRequest {
-                tenant_shard_id,
-                node_id: node,
-            };
-
-            storcon_client
-                .dispatch::<TenantShardMigrateRequest, TenantShardMigrateResponse>(
-                    Method::PUT,
-                    format!("control/v1/tenant/{tenant_shard_id}/migrate"),
-                    Some(req),
-                )
-                .await?;
-        }
-        Command::TenantConfig { tenant_id, config } => {
-            let tenant_conf = serde_json::from_str(&config)?;
-
-            vps_client
-                .tenant_config(&TenantConfigRequest {
-                    tenant_id,
-                    config: tenant_conf,
-                })
-                .await?;
-        }
-        Command::TenantScatter { tenant_id } => {
-            // Find the shards
-            let locate_response = storcon_client
-                .dispatch::<(), TenantLocateResponse>(
-                    Method::GET,
-                    format!("control/v1/tenant/{tenant_id}/locate"),
-                    None,
-                )
-                .await?;
-            let shards = locate_response.shards;
-
-            let mut node_to_shards: HashMap<NodeId, Vec<TenantShardId>> = HashMap::new();
-            let shard_count = shards.len();
-            for s in shards {
-                let entry = node_to_shards.entry(s.node_id).or_default();
-                entry.push(s.shard_id);
-            }
-
-            // Load list of available nodes
-            let nodes_resp = storcon_client
-                .dispatch::<(), Vec<NodeDescribeResponse>>(
-                    Method::GET,
-                    "control/v1/node".to_string(),
-                    None,
-                )
-                .await?;
-
-            for node in nodes_resp {
-                if matches!(node.availability, NodeAvailabilityWrapper::Active) {
-                    node_to_shards.entry(node.id).or_default();
-                }
-            }
-
-            let max_shard_per_node = shard_count / node_to_shards.len();
-
-            loop {
-                let mut migrate_shard = None;
-                for shards in node_to_shards.values_mut() {
-                    if shards.len() > max_shard_per_node {
-                        // Pick the emptiest
-                        migrate_shard = Some(shards.pop().unwrap());
-                    }
-                }
-                let Some(migrate_shard) = migrate_shard else {
-                    break;
-                };
-
-                // Pick the emptiest node to migrate to
-                let mut destinations = node_to_shards
-                    .iter()
-                    .map(|(k, v)| (k, v.len()))
-                    .collect::<Vec<_>>();
-                destinations.sort_by_key(|i| i.1);
-                let (destination_node, destination_count) = *destinations.first().unwrap();
-                if destination_count + 1 > max_shard_per_node {
-                    // Even the emptiest destination doesn't have space: we're done
-                    break;
-                }
-                let destination_node = *destination_node;
-
-                node_to_shards
-                    .get_mut(&destination_node)
-                    .unwrap()
-                    .push(migrate_shard);
-
-                println!("Migrate {} -> {} ...", migrate_shard, destination_node);
-
-                storcon_client
-                    .dispatch::<TenantShardMigrateRequest, TenantShardMigrateResponse>(
-                        Method::PUT,
-                        format!("control/v1/tenant/{migrate_shard}/migrate"),
-                        Some(TenantShardMigrateRequest {
-                            tenant_shard_id: migrate_shard,
-                            node_id: destination_node,
-                        }),
-                    )
-                    .await?;
-                println!("Migrate {} -> {} OK", migrate_shard, destination_node);
-            }
-
-            // Spread the shards across the nodes
-        }
-        Command::TenantDescribe { tenant_id } => {
-            let describe_response = storcon_client
-                .dispatch::<(), TenantDescribeResponse>(
-                    Method::GET,
-                    format!("control/v1/tenant/{tenant_id}"),
-                    None,
-                )
-                .await?;
-            let shards = describe_response.shards;
-            let mut table = comfy_table::Table::new();
-            table.set_header(["Shard", "Attached", "Secondary", "Last error", "status"]);
-            for shard in shards {
-                let secondary = shard
-                    .node_secondary
-                    .iter()
-                    .map(|n| format!("{}", n))
-                    .collect::<Vec<_>>()
-                    .join(",");
-
-                let mut status_parts = Vec::new();
-                if shard.is_reconciling {
-                    status_parts.push("reconciling");
-                }
-
-                if shard.is_pending_compute_notification {
-                    status_parts.push("pending_compute");
-                }
-
-                if shard.is_splitting {
-                    status_parts.push("splitting");
-                }
-                let status = status_parts.join(",");
-
-                table.add_row([
-                    format!("{}", shard.tenant_shard_id),
-                    shard
-                        .node_attached
-                        .map(|n| format!("{}", n))
-                        .unwrap_or(String::new()),
-                    secondary,
-                    shard.last_error,
-                    status,
-                ]);
-            }
-            println!("{table}");
-        }
-        Command::TenantWarmup { tenant_id } => {
-            let describe_response = storcon_client
-                .dispatch::<(), TenantDescribeResponse>(
-                    Method::GET,
-                    format!("control/v1/tenant/{tenant_id}"),
-                    None,
-                )
-                .await;
-            match describe_response {
-                Ok(describe) => {
-                    if matches!(describe.policy, PlacementPolicy::Secondary) {
-                        // Fine: it's already known to controller in secondary mode: calling
-                        // again to put it into secondary mode won't cause problems.
-                    } else {
-                        anyhow::bail!("Tenant already present with policy {:?}", describe.policy);
-                    }
-                }
-                Err(mgmt_api::Error::ApiError(StatusCode::NOT_FOUND, _)) => {
-                    // Fine: this tenant isn't know to the storage controller yet.
-                }
-                Err(e) => {
-                    // Unexpected API error
-                    return Err(e.into());
-                }
-            }
-
-            vps_client
-                .location_config(
-                    TenantShardId::unsharded(tenant_id),
-                    pageserver_api::models::LocationConfig {
-                        mode: pageserver_api::models::LocationConfigMode::Secondary,
-                        generation: None,
-                        secondary_conf: Some(LocationConfigSecondary { warm: true }),
-                        shard_number: 0,
-                        shard_count: 0,
-                        shard_stripe_size: ShardParameters::DEFAULT_STRIPE_SIZE.0,
-                        tenant_conf: TenantConfig::default(),
-                    },
-                    None,
-                    true,
-                )
-                .await?;
-
-            let describe_response = storcon_client
-                .dispatch::<(), TenantDescribeResponse>(
-                    Method::GET,
-                    format!("control/v1/tenant/{tenant_id}"),
-                    None,
-                )
-                .await?;
-
-            let secondary_ps_id = describe_response
-                .shards
-                .first()
-                .unwrap()
-                .node_secondary
-                .first()
-                .unwrap();
-
-            println!("Tenant {tenant_id} warming up on pageserver {secondary_ps_id}");
-            loop {
-                let (status, progress) = vps_client
-                    .tenant_secondary_download(
-                        TenantShardId::unsharded(tenant_id),
-                        Some(Duration::from_secs(10)),
-                    )
-                    .await?;
-                println!(
-                    "Progress: {}/{} layers, {}/{} bytes",
-                    progress.layers_downloaded,
-                    progress.layers_total,
-                    progress.bytes_downloaded,
-                    progress.bytes_total
-                );
-                match status {
-                    StatusCode::OK => {
-                        println!("Download complete");
-                        break;
-                    }
-                    StatusCode::ACCEPTED => {
-                        // Loop
-                    }
-                    _ => {
-                        anyhow::bail!("Unexpected download status: {status}");
-                    }
-                }
-            }
-        }
-    }
-
-    Ok(())
-}

diesel.toml

@@ -2,8 +2,8 @@
 # see https://diesel.rs/guides/configuring-diesel-cli
 
 [print_schema]
-file = "storage_controller/src/schema.rs"
+file = "control_plane/attachment_service/src/schema.rs"
 custom_type_derives = ["diesel::query_builder::QueryId"]
 
 [migrations_directory]
-dir = "storage_controller/migrations"
+dir = "control_plane/attachment_service/migrations"

docs/sourcetree.md

@@ -7,11 +7,6 @@ Below you will find a brief overview of each subdir in the source tree in alphab
 Neon storage broker, providing messaging between safekeepers and pageservers.
 [storage_broker.md](./storage_broker.md)
 
-`storage_controller`:
-
-Neon storage controller, manages a cluster of pageservers and exposes an API that enables
-managing a many-sharded tenant as a single entity.
-
 `/control_plane`:
 
 Local control plane.

docs/storage_controller.md

@@ -1,150 +0,0 @@
-# Storage Controller
-
-## Concepts
-
-The storage controller sits between administrative API clients and pageservers, and handles the details of mapping tenants to pageserver tenant shards. For example, creating a tenant is one API call to the storage controller,
-which is mapped into many API calls to many pageservers (for multiple shards, and for secondary locations).
-
-It implements a pageserver-compatible API that may be used for CRUD operations on tenants and timelines, translating these requests into appropriate operations on the shards within a tenant, which may be on many different pageservers. Using this API, the storage controller may be used in the same way as the pageserver's administrative HTTP API, hiding
-the underlying details of how data is spread across multiple nodes.
-
-The storage controller also manages generations, high availability (via secondary locations) and live migrations for tenants under its management. This is done with a reconciliation loop pattern, where tenants have an “intent” state and a “reconcile” task that tries to make the outside world match the intent.
-
-## APIs
-
-The storage controller’s HTTP server implements four logically separate APIs:
-
-- `/v1/...` path is the pageserver-compatible API. This has to be at the path root because that’s where clients expect to find it on a pageserver.
-- `/control/v1/...` path is the storage controller’s API, which enables operations such as registering and management pageservers, or executing shard splits.
-- `/debug/v1/...` path contains endpoints which are either exclusively used in tests, or are for use by engineers when supporting a deployed system.
-- `/upcall/v1/...` path contains endpoints that are called by pageservers. This includes the `/re-attach` and `/validate` APIs used by pageservers
-  to ensure data safety with generation numbers.
-
-The API is authenticated with a JWT token, and tokens must have scope `pageserverapi` (i.e. the same scope as pageservers’ APIs).
-
-See the `http.rs` file in the source for where the HTTP APIs are implemented.
-
-## Database
-
-The storage controller uses a postgres database to persist a subset of its state. Note that the storage controller does _not_ keep all its state in the database: this is a design choice to enable most operations to be done efficiently in memory, rather than having to read from the database. See `persistence.rs` for a more comprehensive comment explaining what we do and do not persist: a useful metaphor is that we persist objects like tenants and nodes, but we do not
-persist the _relationships_ between them: the attachment state of a tenant's shards to nodes is kept in memory and
-rebuilt on startup.
-
-The file `[persistence.rs](http://persistence.rs)` contains all the code for accessing the database, and has a large doc comment that goes into more detail about exactly what we persist and why.
-
-The `diesel` crate is used for defining models & migrations.
-
-Running a local cluster with `cargo neon` automatically starts a vanilla postgress process to host the storage controller’s database.
-
-### Diesel tip: migrations
-
-If you need to modify the database schema, here’s how to create a migration:
-
-- Install the diesel CLI with `cargo install diesel_cli`
-- Use `diesel migration generate <name>` to create a new migration
-- Populate the SQL files in the `migrations/` subdirectory
-- Use `DATABASE_URL=... diesel migration run` to apply the migration you just wrote: this will update the `[schema.rs](http://schema.rs)` file automatically.
-  - This requires a running database: the easiest way to do that is to just run `cargo neon init ; cargo neon start`, which will leave a database available at `postgresql://localhost:1235/attachment_service`
-- Commit the migration files and the changes to schema.rs
-- If you need to iterate, you can rewind migrations with `diesel migration revert -a` and then `diesel migration run` again.
-- The migrations are build into the storage controller binary, and automatically run at startup after it is deployed, so once you’ve committed a migration no further steps are needed.
-
-## storcon_cli
-
-The `storcon_cli` tool enables interactive management of the storage controller. This is usually
-only necessary for debug, but may also be used to manage nodes (e.g. marking a node as offline).
-
-`storcon_cli --help` includes details on commands.
-
-# Deploying
-
-This section is aimed at engineers deploying the storage controller outside of Neon's cloud platform, as
-part of a self-hosted system.
-
-_General note: since the default `neon_local` environment includes a storage controller, this is a useful
-reference when figuring out deployment._
-
-## Database
-
-It is **essential** that the database used by the storage controller is durable (**do not store it on ephemeral
-local disk**). This database contains pageserver generation numbers, which are essential to data safety on the pageserver.
-
-The resource requirements for the database are very low: a single CPU core and 1GiB of memory should work well for most deployments. The physical size of the database is typically under a gigabyte.
-
-Set the URL to the database using the `--database-url` CLI option.
-
-There is no need to run migrations manually: the storage controller automatically applies migrations
-when it starts up.
-
-## Configure pageservers to use the storage controller
-
-1. The pageserver `control_plane_api` and `control_plane_api_token` should be set in the `pageserver.toml` file. The API setting should
-   point to the "upcall" prefix, for example `http://127.0.0.1:1234/upcall/v1/` is used in neon_local clusters.
-2. Create a `metadata.json` file in the same directory as `pageserver.toml`: this enables the pageserver to automatically register itself
-   with the storage controller when it starts up. See the example below for the format of this file.
-
-### Example `metadata.json`
-
-```
-{"host":"acmehost.localdomain","http_host":"acmehost.localdomain","http_port":9898,"port":64000}
-```
-
-- `port` and `host` refer to the _postgres_ port and host, and these must be accessible from wherever
-  postgres runs.
-- `http_port` and `http_host` refer to the pageserver's HTTP api, this must be accessible from where
-  the storage controller runs.
-
-## Handle compute notifications.
-
-The storage controller independently moves tenant attachments between pageservers in response to
-changes such as a pageserver node becoming unavailable, or the tenant's shard count changing. To enable
-postgres clients to handle such changes, the storage controller calls an API hook when a tenant's pageserver
-location changes.
-
-The hook is configured using the storage controller's `--compute-hook-url` CLI option. If the hook requires
-JWT auth, the token may be provided with `--control-plane-jwt-token`. The hook will be invoked with a `PUT` request.
-
-In the Neon cloud service, this hook is implemented by Neon's internal cloud control plane. In `neon_local` systems
-the storage controller integrates directly with neon_local to reconfigure local postgres processes instead of calling
-the compute hook.
-
-When implementing an on-premise Neon deployment, you must implement a service that handles the compute hook. This is not complicated:
-the request body has format of the `ComputeHookNotifyRequest` structure, provided below for convenience.
-
-```
-struct ComputeHookNotifyRequestShard {
-    node_id: NodeId,
-    shard_number: ShardNumber,
-}
-
-struct ComputeHookNotifyRequest {
-    tenant_id: TenantId,
-    stripe_size: Option<ShardStripeSize>,
-    shards: Vec<ComputeHookNotifyRequestShard>,
-}
-```
-
-When a notification is received:
-
-1. Modify postgres configuration for this tenant:
-
-   - set `neon.pageserver_connstr` to a comma-separated list of postgres connection strings to pageservers according to the `shards` list. The
-     shards identified by `NodeId` must be converted to the address+port of the node.
-   - if stripe_size is not None, set `neon.stripe_size` to this value
-
-2. Send SIGHUP to postgres to reload configuration
-3. Respond with 200 to the notification request. Do not return success if postgres was not updated: if an error is returned, the controller
-   will retry the notification until it succeeds..
-
-### Example notification body
-
-```
-{
-  "tenant_id": "1f359dd625e519a1a4e8d7509690f6fc",
-  "stripe_size": 32768,
-  "shards": [
-      {"node_id": 344, "shard_number": 0},
-      {"node_id": 722, "shard_number": 1},
-  ],
-}
-```

libs/metrics/Cargo.toml

@@ -10,13 +10,11 @@ libc.workspace = true
 once_cell.workspace = true
 chrono.workspace = true
 twox-hash.workspace = true
-measured.workspace = true
 
 workspace_hack.workspace = true
 
 [target.'cfg(target_os = "linux")'.dependencies]
 procfs.workspace = true
-measured-process.workspace = true
 
 [dev-dependencies]
 rand = "0.8"

libs/metrics/src/hll.rs

@@ -7,19 +7,14 @@
 //! use significantly less memory than this, but can only approximate the cardinality.
 
 use std::{
-    hash::{BuildHasher, BuildHasherDefault, Hash},
-    sync::atomic::AtomicU8,
+    collections::HashMap,
+    hash::{BuildHasher, BuildHasherDefault, Hash, Hasher},
+    sync::{atomic::AtomicU8, Arc, RwLock},
 };
 
-use measured::{
-    label::{LabelGroupVisitor, LabelName, LabelValue, LabelVisitor},
-    metric::{
-        group::{Encoding, MetricValue},
-        name::MetricNameEncoder,
-        Metric, MetricType, MetricVec,
-    },
-    text::TextEncoder,
-    LabelGroup,
+use prometheus::{
+    core::{self, Describer},
+    proto, Opts,
 };
 use twox_hash::xxh3;
 
@@ -45,7 +40,7 @@ macro_rules! register_hll {
     }};
 
     ($N:literal, $NAME:expr, $HELP:expr $(,)?) => {{
-        $crate::register_hll!($N, $crate::opts!($NAME, $HELP))
+        $crate::register_hll!($N, $crate::opts!($NAME, $HELP), $LABELS_NAMES)
     }};
 }
 
@@ -98,25 +93,203 @@ macro_rules! register_hll {
 /// ```
 ///
 /// See <https://en.wikipedia.org/wiki/HyperLogLog#Practical_considerations> for estimates on alpha
-pub type HyperLogLogVec<L, const N: usize> = MetricVec<HyperLogLogState<N>, L>;
-pub type HyperLogLog<const N: usize> = Metric<HyperLogLogState<N>>;
-
-pub struct HyperLogLogState<const N: usize> {
-    shards: [AtomicU8; N],
+#[derive(Clone)]
+pub struct HyperLogLogVec<const N: usize> {
+    core: Arc<HyperLogLogVecCore<N>>,
 }
-impl<const N: usize> Default for HyperLogLogState<N> {
-    fn default() -> Self {
-        #[allow(clippy::declare_interior_mutable_const)]
-        const ZERO: AtomicU8 = AtomicU8::new(0);
-        Self { shards: [ZERO; N] }
+
+struct HyperLogLogVecCore<const N: usize> {
+    pub children: RwLock<HashMap<u64, HyperLogLog<N>, BuildHasherDefault<xxh3::Hash64>>>,
+    pub desc: core::Desc,
+    pub opts: Opts,
+}
+
+impl<const N: usize> core::Collector for HyperLogLogVec<N> {
+    fn desc(&self) -> Vec<&core::Desc> {
+        vec![&self.core.desc]
+    }
+
+    fn collect(&self) -> Vec<proto::MetricFamily> {
+        let mut m = proto::MetricFamily::default();
+        m.set_name(self.core.desc.fq_name.clone());
+        m.set_help(self.core.desc.help.clone());
+        m.set_field_type(proto::MetricType::GAUGE);
+
+        let mut metrics = Vec::new();
+        for child in self.core.children.read().unwrap().values() {
+            child.core.collect_into(&mut metrics);
+        }
+        m.set_metric(metrics);
+
+        vec![m]
     }
 }
 
-impl<const N: usize> MetricType for HyperLogLogState<N> {
-    type Metadata = ();
+impl<const N: usize> HyperLogLogVec<N> {
+    /// Create a new [`HyperLogLogVec`] based on the provided
+    /// [`Opts`] and partitioned by the given label names. At least one label name must be
+    /// provided.
+    pub fn new(opts: Opts, label_names: &[&str]) -> prometheus::Result<Self> {
+        assert!(N.is_power_of_two());
+        let variable_names = label_names.iter().map(|s| (*s).to_owned()).collect();
+        let opts = opts.variable_labels(variable_names);
+
+        let desc = opts.describe()?;
+        let v = HyperLogLogVecCore {
+            children: RwLock::new(HashMap::default()),
+            desc,
+            opts,
+        };
+
+        Ok(Self { core: Arc::new(v) })
+    }
+
+    /// `get_metric_with_label_values` returns the [`HyperLogLog<P>`] for the given slice
+    /// of label values (same order as the VariableLabels in Desc). If that combination of
+    /// label values is accessed for the first time, a new [`HyperLogLog<P>`] is created.
+    ///
+    /// An error is returned if the number of label values is not the same as the
+    /// number of VariableLabels in Desc.
+    pub fn get_metric_with_label_values(
+        &self,
+        vals: &[&str],
+    ) -> prometheus::Result<HyperLogLog<N>> {
+        self.core.get_metric_with_label_values(vals)
+    }
+
+    /// `with_label_values` works as `get_metric_with_label_values`, but panics if an error
+    /// occurs.
+    pub fn with_label_values(&self, vals: &[&str]) -> HyperLogLog<N> {
+        self.get_metric_with_label_values(vals).unwrap()
+    }
 }
 
-impl<const N: usize> HyperLogLogState<N> {
+impl<const N: usize> HyperLogLogVecCore<N> {
+    pub fn get_metric_with_label_values(
+        &self,
+        vals: &[&str],
+    ) -> prometheus::Result<HyperLogLog<N>> {
+        let h = self.hash_label_values(vals)?;
+
+        if let Some(metric) = self.children.read().unwrap().get(&h).cloned() {
+            return Ok(metric);
+        }
+
+        self.get_or_create_metric(h, vals)
+    }
+
+    pub(crate) fn hash_label_values(&self, vals: &[&str]) -> prometheus::Result<u64> {
+        if vals.len() != self.desc.variable_labels.len() {
+            return Err(prometheus::Error::InconsistentCardinality {
+                expect: self.desc.variable_labels.len(),
+                got: vals.len(),
+            });
+        }
+
+        let mut h = xxh3::Hash64::default();
+        for val in vals {
+            h.write(val.as_bytes());
+        }
+
+        Ok(h.finish())
+    }
+
+    fn get_or_create_metric(
+        &self,
+        hash: u64,
+        label_values: &[&str],
+    ) -> prometheus::Result<HyperLogLog<N>> {
+        let mut children = self.children.write().unwrap();
+        // Check exist first.
+        if let Some(metric) = children.get(&hash).cloned() {
+            return Ok(metric);
+        }
+
+        let metric = HyperLogLog::with_opts_and_label_values(&self.opts, label_values)?;
+        children.insert(hash, metric.clone());
+        Ok(metric)
+    }
+}
+
+/// HLL is a probabilistic cardinality measure.
+///
+/// How to use this time-series for a metric name `my_metrics_total_hll`:
+///
+/// ```promql
+/// # harmonic mean
+/// 1 / (
+///     sum (
+///         2 ^ -(
+///             # HLL merge operation
+///             max (my_metrics_total_hll{}) by (hll_shard, other_labels...)
+///         )
+///     ) without (hll_shard)
+/// )
+/// * alpha
+/// * shards_count
+/// * shards_count
+/// ```
+///
+/// If you want an estimate over time, you can use the following query:
+///
+/// ```promql
+/// # harmonic mean
+/// 1 / (
+///     sum (
+///         2 ^ -(
+///             # HLL merge operation
+///             max (
+///                 max_over_time(my_metrics_total_hll{}[$__rate_interval])
+///             ) by (hll_shard, other_labels...)
+///         )
+///     ) without (hll_shard)
+/// )
+/// * alpha
+/// * shards_count
+/// * shards_count
+/// ```
+///
+/// In the case of low cardinality, you might want to use the linear counting approximation:
+///
+/// ```promql
+/// # LinearCounting(m, V) = m log (m / V)
+/// shards_count * ln(shards_count /
+///     # calculate V = how many shards contain a 0
+///     count(max (proxy_connecting_endpoints{}) by (hll_shard, protocol) == 0) without (hll_shard)
+/// )
+/// ```
+///
+/// See <https://en.wikipedia.org/wiki/HyperLogLog#Practical_considerations> for estimates on alpha
+#[derive(Clone)]
+pub struct HyperLogLog<const N: usize> {
+    core: Arc<HyperLogLogCore<N>>,
+}
+
+impl<const N: usize> HyperLogLog<N> {
+    /// Create a [`HyperLogLog`] with the `name` and `help` arguments.
+    pub fn new<S1: Into<String>, S2: Into<String>>(name: S1, help: S2) -> prometheus::Result<Self> {
+        assert!(N.is_power_of_two());
+        let opts = Opts::new(name, help);
+        Self::with_opts(opts)
+    }
+
+    /// Create a [`HyperLogLog`] with the `opts` options.
+    pub fn with_opts(opts: Opts) -> prometheus::Result<Self> {
+        Self::with_opts_and_label_values(&opts, &[])
+    }
+
+    fn with_opts_and_label_values(opts: &Opts, label_values: &[&str]) -> prometheus::Result<Self> {
+        let desc = opts.describe()?;
+        let labels = make_label_pairs(&desc, label_values)?;
+
+        let v = HyperLogLogCore {
+            shards: [0; N].map(AtomicU8::new),
+            desc,
+            labels,
+        };
+        Ok(Self { core: Arc::new(v) })
+    }
+
     pub fn measure(&self, item: &impl Hash) {
         // changing the hasher will break compatibility with previous measurements.
         self.record(BuildHasherDefault::<xxh3::Hash64>::default().hash_one(item));
@@ -126,11 +299,42 @@ impl<const N: usize> HyperLogLogState<N> {
         let p = N.ilog2() as u8;
         let j = hash & (N as u64 - 1);
         let rho = (hash >> p).leading_zeros() as u8 + 1 - p;
-        self.shards[j as usize].fetch_max(rho, std::sync::atomic::Ordering::Relaxed);
+        self.core.shards[j as usize].fetch_max(rho, std::sync::atomic::Ordering::Relaxed);
+    }
+}
+
+struct HyperLogLogCore<const N: usize> {
+    shards: [AtomicU8; N],
+    desc: core::Desc,
+    labels: Vec<proto::LabelPair>,
+}
+
+impl<const N: usize> core::Collector for HyperLogLog<N> {
+    fn desc(&self) -> Vec<&core::Desc> {
+        vec![&self.core.desc]
     }
 
-    fn take_sample(&self) -> [u8; N] {
-        self.shards.each_ref().map(|x| {
+    fn collect(&self) -> Vec<proto::MetricFamily> {
+        let mut m = proto::MetricFamily::default();
+        m.set_name(self.core.desc.fq_name.clone());
+        m.set_help(self.core.desc.help.clone());
+        m.set_field_type(proto::MetricType::GAUGE);
+
+        let mut metrics = Vec::new();
+        self.core.collect_into(&mut metrics);
+        m.set_metric(metrics);
+
+        vec![m]
+    }
+}
+
+impl<const N: usize> HyperLogLogCore<N> {
+    fn collect_into(&self, metrics: &mut Vec<proto::Metric>) {
+        self.shards.iter().enumerate().for_each(|(i, x)| {
+            let mut shard_label = proto::LabelPair::default();
+            shard_label.set_name("hll_shard".to_owned());
+            shard_label.set_value(format!("{i}"));
+
             // We reset the counter to 0 so we can perform a cardinality measure over any time slice in prometheus.
 
             // This seems like it would be a race condition,
@@ -140,90 +344,85 @@ impl<const N: usize> HyperLogLogState<N> {
 
             // TODO: maybe we shouldn't reset this on every collect, instead, only after a time window.
             // this would mean that a dev port-forwarding the metrics url won't break the sampling.
-            x.swap(0, std::sync::atomic::Ordering::Relaxed)
+            let v = x.swap(0, std::sync::atomic::Ordering::Relaxed);
+
+            let mut m = proto::Metric::default();
+            let mut c = proto::Gauge::default();
+            c.set_value(v as f64);
+            m.set_gauge(c);
+
+            let mut labels = Vec::with_capacity(self.labels.len() + 1);
+            labels.extend_from_slice(&self.labels);
+            labels.push(shard_label);
+
+            m.set_label(labels);
+            metrics.push(m);
         })
     }
 }
-impl<W: std::io::Write, const N: usize> measured::metric::MetricEncoding<TextEncoder<W>>
-    for HyperLogLogState<N>
-{
-    fn write_type(
-        name: impl MetricNameEncoder,
-        enc: &mut TextEncoder<W>,
-    ) -> Result<(), std::io::Error> {
-        enc.write_type(&name, measured::text::MetricType::Gauge)
+
+fn make_label_pairs(
+    desc: &core::Desc,
+    label_values: &[&str],
+) -> prometheus::Result<Vec<proto::LabelPair>> {
+    if desc.variable_labels.len() != label_values.len() {
+        return Err(prometheus::Error::InconsistentCardinality {
+            expect: desc.variable_labels.len(),
+            got: label_values.len(),
+        });
     }
-    fn collect_into(
-        &self,
-        _: &(),
-        labels: impl LabelGroup,
-        name: impl MetricNameEncoder,
-        enc: &mut TextEncoder<W>,
-    ) -> Result<(), std::io::Error> {
-        struct I64(i64);
-        impl LabelValue for I64 {
-            fn visit<V: LabelVisitor>(&self, v: V) -> V::Output {
-                v.write_int(self.0)
-            }
-        }
 
-        struct HllShardLabel {
-            hll_shard: i64,
-        }
-
-        impl LabelGroup for HllShardLabel {
-            fn visit_values(&self, v: &mut impl LabelGroupVisitor) {
-                const LE: &LabelName = LabelName::from_str("hll_shard");
-                v.write_value(LE, &I64(self.hll_shard));
-            }
-        }
-
-        self.take_sample()
-            .into_iter()
-            .enumerate()
-            .try_for_each(|(hll_shard, val)| {
-                enc.write_metric_value(
-                    name.by_ref(),
-                    labels.by_ref().compose_with(HllShardLabel {
-                        hll_shard: hll_shard as i64,
-                    }),
-                    MetricValue::Int(val as i64),
-                )
-            })
+    let total_len = desc.variable_labels.len() + desc.const_label_pairs.len();
+    if total_len == 0 {
+        return Ok(vec![]);
     }
+
+    if desc.variable_labels.is_empty() {
+        return Ok(desc.const_label_pairs.clone());
+    }
+
+    let mut label_pairs = Vec::with_capacity(total_len);
+    for (i, n) in desc.variable_labels.iter().enumerate() {
+        let mut label_pair = proto::LabelPair::default();
+        label_pair.set_name(n.clone());
+        label_pair.set_value(label_values[i].to_owned());
+        label_pairs.push(label_pair);
+    }
+
+    for label_pair in &desc.const_label_pairs {
+        label_pairs.push(label_pair.clone());
+    }
+    label_pairs.sort();
+    Ok(label_pairs)
 }
 
 #[cfg(test)]
 mod tests {
     use std::collections::HashSet;
 
-    use measured::{label::StaticLabelSet, FixedCardinalityLabel};
+    use prometheus::{proto, Opts};
     use rand::{rngs::StdRng, Rng, SeedableRng};
     use rand_distr::{Distribution, Zipf};
 
     use crate::HyperLogLogVec;
 
-    #[derive(FixedCardinalityLabel, Clone, Copy)]
-    #[label(singleton = "x")]
-    enum Label {
-        A,
-        B,
+    fn collect(hll: &HyperLogLogVec<32>) -> Vec<proto::Metric> {
+        let mut metrics = vec![];
+        hll.core
+            .children
+            .read()
+            .unwrap()
+            .values()
+            .for_each(|c| c.core.collect_into(&mut metrics));
+        metrics
     }
-
-    fn collect(hll: &HyperLogLogVec<StaticLabelSet<Label>, 32>) -> ([u8; 32], [u8; 32]) {
-        // cannot go through the `hll.collect_family_into` interface yet...
-        // need to see if I can fix the conflicting impls problem in measured.
-        (
-            hll.get_metric(hll.with_labels(Label::A)).take_sample(),
-            hll.get_metric(hll.with_labels(Label::B)).take_sample(),
-        )
-    }
-
-    fn get_cardinality(samples: &[[u8; 32]]) -> f64 {
+    fn get_cardinality(metrics: &[proto::Metric], filter: impl Fn(&proto::Metric) -> bool) -> f64 {
         let mut buckets = [0.0; 32];
-        for &sample in samples {
-            for (i, m) in sample.into_iter().enumerate() {
-                buckets[i] = f64::max(buckets[i], m as f64);
+        for metric in metrics.chunks_exact(32) {
+            if filter(&metric[0]) {
+                for (i, m) in metric.iter().enumerate() {
+                    buckets[i] = f64::max(buckets[i], m.get_gauge().get_value());
+                }
             }
         }
 
@@ -238,7 +437,7 @@ mod tests {
     }
 
     fn test_cardinality(n: usize, dist: impl Distribution<f64>) -> ([usize; 3], [f64; 3]) {
-        let hll = HyperLogLogVec::<StaticLabelSet<Label>, 32>::new();
+        let hll = HyperLogLogVec::<32>::new(Opts::new("foo", "bar"), &["x"]).unwrap();
 
         let mut iter = StdRng::seed_from_u64(0x2024_0112).sample_iter(dist);
         let mut set_a = HashSet::new();
@@ -246,20 +445,18 @@ mod tests {
 
         for x in iter.by_ref().take(n) {
             set_a.insert(x.to_bits());
-            hll.get_metric(hll.with_labels(Label::A))
-                .measure(&x.to_bits());
+            hll.with_label_values(&["a"]).measure(&x.to_bits());
         }
         for x in iter.by_ref().take(n) {
             set_b.insert(x.to_bits());
-            hll.get_metric(hll.with_labels(Label::B))
-                .measure(&x.to_bits());
+            hll.with_label_values(&["b"]).measure(&x.to_bits());
         }
         let merge = &set_a | &set_b;
 
-        let (a, b) = collect(&hll);
-        let len = get_cardinality(&[a, b]);
-        let len_a = get_cardinality(&[a]);
-        let len_b = get_cardinality(&[b]);
+        let metrics = collect(&hll);
+        let len = get_cardinality(&metrics, |_| true);
+        let len_a = get_cardinality(&metrics, |l| l.get_label()[0].get_value() == "a");
+        let len_b = get_cardinality(&metrics, |l| l.get_label()[0].get_value() == "b");
 
         ([merge.len(), set_a.len(), set_b.len()], [len, len_a, len_b])
     }

libs/metrics/src/lib.rs

@@ -4,17 +4,6 @@
 //! a default registry.
 #![deny(clippy::undocumented_unsafe_blocks)]
 
-use measured::{
-    label::{LabelGroupSet, LabelGroupVisitor, LabelName, NoLabels},
-    metric::{
-        counter::CounterState,
-        gauge::GaugeState,
-        group::{Encoding, MetricValue},
-        name::{MetricName, MetricNameEncoder},
-        MetricEncoding, MetricFamilyEncoding,
-    },
-    FixedCardinalityLabel, LabelGroup, MetricGroup,
-};
 use once_cell::sync::Lazy;
 use prometheus::core::{
     Atomic, AtomicU64, Collector, GenericCounter, GenericCounterVec, GenericGauge, GenericGaugeVec,
@@ -22,7 +11,6 @@ use prometheus::core::{
 pub use prometheus::opts;
 pub use prometheus::register;
 pub use prometheus::Error;
-use prometheus::Registry;
 pub use prometheus::{core, default_registry, proto};
 pub use prometheus::{exponential_buckets, linear_buckets};
 pub use prometheus::{register_counter_vec, Counter, CounterVec};
@@ -35,12 +23,13 @@ pub use prometheus::{register_int_counter_vec, IntCounterVec};
 pub use prometheus::{register_int_gauge, IntGauge};
 pub use prometheus::{register_int_gauge_vec, IntGaugeVec};
 pub use prometheus::{Encoder, TextEncoder};
+use prometheus::{Registry, Result};
 
 pub mod launch_timestamp;
 mod wrappers;
 pub use wrappers::{CountedReader, CountedWriter};
 mod hll;
-pub use hll::{HyperLogLog, HyperLogLogState, HyperLogLogVec};
+pub use hll::{HyperLogLog, HyperLogLogVec};
 #[cfg(target_os = "linux")]
 pub mod more_process_metrics;
 
@@ -70,7 +59,7 @@ static INTERNAL_REGISTRY: Lazy<Registry> = Lazy::new(Registry::new);
 /// Register a collector in the internal registry. MUST be called before the first call to `gather()`.
 /// Otherwise, we can have a deadlock in the `gather()` call, trying to register a new collector
 /// while holding the lock.
-pub fn register_internal(c: Box<dyn Collector>) -> prometheus::Result<()> {
+pub fn register_internal(c: Box<dyn Collector>) -> Result<()> {
     INTERNAL_REGISTRY.register(c)
 }
 
@@ -107,127 +96,6 @@ pub const DISK_WRITE_SECONDS_BUCKETS: &[f64] = &[
     0.000_050, 0.000_100, 0.000_500, 0.001, 0.003, 0.005, 0.01, 0.05, 0.1, 0.3, 0.5,
 ];
 
-pub struct BuildInfo {
-    pub revision: &'static str,
-    pub build_tag: &'static str,
-}
-
-// todo: allow label group without the set
-impl LabelGroup for BuildInfo {
-    fn visit_values(&self, v: &mut impl LabelGroupVisitor) {
-        const REVISION: &LabelName = LabelName::from_str("revision");
-        v.write_value(REVISION, &self.revision);
-        const BUILD_TAG: &LabelName = LabelName::from_str("build_tag");
-        v.write_value(BUILD_TAG, &self.build_tag);
-    }
-}
-
-impl<T: Encoding> MetricFamilyEncoding<T> for BuildInfo
-where
-    GaugeState: MetricEncoding<T>,
-{
-    fn collect_family_into(
-        &self,
-        name: impl measured::metric::name::MetricNameEncoder,
-        enc: &mut T,
-    ) -> Result<(), T::Err> {
-        enc.write_help(&name, "Build/version information")?;
-        GaugeState::write_type(&name, enc)?;
-        GaugeState {
-            count: std::sync::atomic::AtomicI64::new(1),
-        }
-        .collect_into(&(), self, name, enc)
-    }
-}
-
-#[derive(MetricGroup)]
-#[metric(new(build_info: BuildInfo))]
-pub struct NeonMetrics {
-    #[cfg(target_os = "linux")]
-    #[metric(namespace = "process")]
-    #[metric(init = measured_process::ProcessCollector::for_self())]
-    process: measured_process::ProcessCollector,
-
-    #[metric(namespace = "libmetrics")]
-    #[metric(init = LibMetrics::new(build_info))]
-    libmetrics: LibMetrics,
-}
-
-#[derive(MetricGroup)]
-#[metric(new(build_info: BuildInfo))]
-pub struct LibMetrics {
-    #[metric(init = build_info)]
-    build_info: BuildInfo,
-
-    #[metric(flatten)]
-    rusage: Rusage,
-
-    serve_count: CollectionCounter,
-}
-
-fn write_gauge<Enc: Encoding>(
-    x: i64,
-    labels: impl LabelGroup,
-    name: impl MetricNameEncoder,
-    enc: &mut Enc,
-) -> Result<(), Enc::Err> {
-    enc.write_metric_value(name, labels, MetricValue::Int(x))
-}
-
-#[derive(Default)]
-struct Rusage;
-
-#[derive(FixedCardinalityLabel, Clone, Copy)]
-#[label(singleton = "io_operation")]
-enum IoOp {
-    Read,
-    Write,
-}
-
-impl<T: Encoding> MetricGroup<T> for Rusage
-where
-    GaugeState: MetricEncoding<T>,
-{
-    fn collect_group_into(&self, enc: &mut T) -> Result<(), T::Err> {
-        const DISK_IO: &MetricName = MetricName::from_str("disk_io_bytes_total");
-        const MAXRSS: &MetricName = MetricName::from_str("maxrss_kb");
-
-        let ru = get_rusage_stats();
-
-        enc.write_help(
-            DISK_IO,
-            "Bytes written and read from disk, grouped by the operation (read|write)",
-        )?;
-        GaugeState::write_type(DISK_IO, enc)?;
-        write_gauge(ru.ru_inblock * BYTES_IN_BLOCK, IoOp::Read, DISK_IO, enc)?;
-        write_gauge(ru.ru_oublock * BYTES_IN_BLOCK, IoOp::Write, DISK_IO, enc)?;
-
-        enc.write_help(MAXRSS, "Memory usage (Maximum Resident Set Size)")?;
-        GaugeState::write_type(MAXRSS, enc)?;
-        write_gauge(ru.ru_maxrss, IoOp::Read, MAXRSS, enc)?;
-
-        Ok(())
-    }
-}
-
-#[derive(Default)]
-struct CollectionCounter(CounterState);
-
-impl<T: Encoding> MetricFamilyEncoding<T> for CollectionCounter
-where
-    CounterState: MetricEncoding<T>,
-{
-    fn collect_family_into(
-        &self,
-        name: impl measured::metric::name::MetricNameEncoder,
-        enc: &mut T,
-    ) -> Result<(), T::Err> {
-        self.0.inc();
-        enc.write_help(&name, "Number of metric requests made")?;
-        self.0.collect_into(&(), NoLabels, name, enc)
-    }
-}
-
 pub fn set_build_info_metric(revision: &str, build_tag: &str) {
     let metric = register_int_gauge_vec!(
         "libmetrics_build_info",
@@ -237,7 +105,6 @@ pub fn set_build_info_metric(revision: &str, build_tag: &str) {
     .expect("Failed to register build info metric");
     metric.with_label_values(&[revision, build_tag]).set(1);
 }
-const BYTES_IN_BLOCK: i64 = 512;
 
 // Records I/O stats in a "cross-platform" way.
 // Compiles both on macOS and Linux, but current macOS implementation always returns 0 as values for I/O stats.
@@ -250,22 +117,14 @@ const BYTES_IN_BLOCK: i64 = 512;
 fn update_rusage_metrics() {
     let rusage_stats = get_rusage_stats();
 
+    const BYTES_IN_BLOCK: i64 = 512;
     DISK_IO_BYTES
         .with_label_values(&["read"])
         .set(rusage_stats.ru_inblock * BYTES_IN_BLOCK);
     DISK_IO_BYTES
         .with_label_values(&["write"])
         .set(rusage_stats.ru_oublock * BYTES_IN_BLOCK);
-
-    // On macOS, the unit of maxrss is bytes; on Linux, it's kilobytes. https://stackoverflow.com/a/59915669
-    #[cfg(target_os = "macos")]
-    {
-        MAXRSS_KB.set(rusage_stats.ru_maxrss / 1024);
-    }
-    #[cfg(not(target_os = "macos"))]
-    {
-        MAXRSS_KB.set(rusage_stats.ru_maxrss);
-    }
+    MAXRSS_KB.set(rusage_stats.ru_maxrss);
 }
 
 fn get_rusage_stats() -> libc::rusage {
@@ -292,7 +151,6 @@ macro_rules! register_int_counter_pair_vec {
         }
     }};
 }
-
 /// Create an [`IntCounterPair`] and registers to default registry.
 #[macro_export(local_inner_macros)]
 macro_rules! register_int_counter_pair {
@@ -330,10 +188,7 @@ impl<P: Atomic> GenericCounterPairVec<P> {
     ///
     /// An error is returned if the number of label values is not the same as the
     /// number of VariableLabels in Desc.
-    pub fn get_metric_with_label_values(
-        &self,
-        vals: &[&str],
-    ) -> prometheus::Result<GenericCounterPair<P>> {
+    pub fn get_metric_with_label_values(&self, vals: &[&str]) -> Result<GenericCounterPair<P>> {
         Ok(GenericCounterPair {
             inc: self.inc.get_metric_with_label_values(vals)?,
             dec: self.dec.get_metric_with_label_values(vals)?,
@@ -346,7 +201,7 @@ impl<P: Atomic> GenericCounterPairVec<P> {
         self.get_metric_with_label_values(vals).unwrap()
     }
 
-    pub fn remove_label_values(&self, res: &mut [prometheus::Result<()>; 2], vals: &[&str]) {
+    pub fn remove_label_values(&self, res: &mut [Result<()>; 2], vals: &[&str]) {
         res[0] = self.inc.remove_label_values(vals);
         res[1] = self.dec.remove_label_values(vals);
     }
@@ -430,171 +285,3 @@ pub type IntCounterPair = GenericCounterPair<AtomicU64>;
 
 /// A guard for [`IntCounterPair`] that will decrement the gauge on drop
 pub type IntCounterPairGuard = GenericCounterPairGuard<AtomicU64>;
-
-pub trait CounterPairAssoc {
-    const INC_NAME: &'static MetricName;
-    const DEC_NAME: &'static MetricName;
-
-    const INC_HELP: &'static str;
-    const DEC_HELP: &'static str;
-
-    type LabelGroupSet: LabelGroupSet;
-}
-
-pub struct CounterPairVec<A: CounterPairAssoc> {
-    vec: measured::metric::MetricVec<MeasuredCounterPairState, A::LabelGroupSet>,
-}
-
-impl<A: CounterPairAssoc> Default for CounterPairVec<A>
-where
-    A::LabelGroupSet: Default,
-{
-    fn default() -> Self {
-        Self {
-            vec: Default::default(),
-        }
-    }
-}
-
-impl<A: CounterPairAssoc> CounterPairVec<A> {
-    pub fn guard(
-        &self,
-        labels: <A::LabelGroupSet as LabelGroupSet>::Group<'_>,
-    ) -> MeasuredCounterPairGuard<'_, A> {
-        let id = self.vec.with_labels(labels);
-        self.vec.get_metric(id).inc.inc();
-        MeasuredCounterPairGuard { vec: &self.vec, id }
-    }
-    pub fn inc(&self, labels: <A::LabelGroupSet as LabelGroupSet>::Group<'_>) {
-        let id = self.vec.with_labels(labels);
-        self.vec.get_metric(id).inc.inc();
-    }
-    pub fn dec(&self, labels: <A::LabelGroupSet as LabelGroupSet>::Group<'_>) {
-        let id = self.vec.with_labels(labels);
-        self.vec.get_metric(id).dec.inc();
-    }
-    pub fn remove_metric(
-        &self,
-        labels: <A::LabelGroupSet as LabelGroupSet>::Group<'_>,
-    ) -> Option<MeasuredCounterPairState> {
-        let id = self.vec.with_labels(labels);
-        self.vec.remove_metric(id)
-    }
-}
-
-impl<T, A> ::measured::metric::group::MetricGroup<T> for CounterPairVec<A>
-where
-    T: ::measured::metric::group::Encoding,
-    A: CounterPairAssoc,
-    ::measured::metric::counter::CounterState: ::measured::metric::MetricEncoding<T>,
-{
-    fn collect_group_into(&self, enc: &mut T) -> Result<(), T::Err> {
-        // write decrement first to avoid a race condition where inc - dec < 0
-        T::write_help(enc, A::DEC_NAME, A::DEC_HELP)?;
-        self.vec
-            .collect_family_into(A::DEC_NAME, &mut Dec(&mut *enc))?;
-
-        T::write_help(enc, A::INC_NAME, A::INC_HELP)?;
-        self.vec
-            .collect_family_into(A::INC_NAME, &mut Inc(&mut *enc))?;
-
-        Ok(())
-    }
-}
-
-#[derive(MetricGroup, Default)]
-pub struct MeasuredCounterPairState {
-    pub inc: CounterState,
-    pub dec: CounterState,
-}
-
-impl measured::metric::MetricType for MeasuredCounterPairState {
-    type Metadata = ();
-}
-
-pub struct MeasuredCounterPairGuard<'a, A: CounterPairAssoc> {
-    vec: &'a measured::metric::MetricVec<MeasuredCounterPairState, A::LabelGroupSet>,
-    id: measured::metric::LabelId<A::LabelGroupSet>,
-}
-
-impl<A: CounterPairAssoc> Drop for MeasuredCounterPairGuard<'_, A> {
-    fn drop(&mut self) {
-        self.vec.get_metric(self.id).dec.inc();
-    }
-}
-
-/// [`MetricEncoding`] for [`MeasuredCounterPairState`] that only writes the inc counter to the inner encoder.
-struct Inc<T>(T);
-/// [`MetricEncoding`] for [`MeasuredCounterPairState`] that only writes the dec counter to the inner encoder.
-struct Dec<T>(T);
-
-impl<T: Encoding> Encoding for Inc<T> {
-    type Err = T::Err;
-
-    fn write_help(&mut self, name: impl MetricNameEncoder, help: &str) -> Result<(), Self::Err> {
-        self.0.write_help(name, help)
-    }
-
-    fn write_metric_value(
-        &mut self,
-        name: impl MetricNameEncoder,
-        labels: impl LabelGroup,
-        value: MetricValue,
-    ) -> Result<(), Self::Err> {
-        self.0.write_metric_value(name, labels, value)
-    }
-}
-
-impl<T: Encoding> MetricEncoding<Inc<T>> for MeasuredCounterPairState
-where
-    CounterState: MetricEncoding<T>,
-{
-    fn write_type(name: impl MetricNameEncoder, enc: &mut Inc<T>) -> Result<(), T::Err> {
-        CounterState::write_type(name, &mut enc.0)
-    }
-    fn collect_into(
-        &self,
-        metadata: &(),
-        labels: impl LabelGroup,
-        name: impl MetricNameEncoder,
-        enc: &mut Inc<T>,
-    ) -> Result<(), T::Err> {
-        self.inc.collect_into(metadata, labels, name, &mut enc.0)
-    }
-}
-
-impl<T: Encoding> Encoding for Dec<T> {
-    type Err = T::Err;
-
-    fn write_help(&mut self, name: impl MetricNameEncoder, help: &str) -> Result<(), Self::Err> {
-        self.0.write_help(name, help)
-    }
-
-    fn write_metric_value(
-        &mut self,
-        name: impl MetricNameEncoder,
-        labels: impl LabelGroup,
-        value: MetricValue,
-    ) -> Result<(), Self::Err> {
-        self.0.write_metric_value(name, labels, value)
-    }
-}
-
-/// Write the dec counter to the encoder
-impl<T: Encoding> MetricEncoding<Dec<T>> for MeasuredCounterPairState
-where
-    CounterState: MetricEncoding<T>,
-{
-    fn write_type(name: impl MetricNameEncoder, enc: &mut Dec<T>) -> Result<(), T::Err> {
-        CounterState::write_type(name, &mut enc.0)
-    }
-    fn collect_into(
-        &self,
-        metadata: &(),
-        labels: impl LabelGroup,
-        name: impl MetricNameEncoder,
-        enc: &mut Dec<T>,
-    ) -> Result<(), T::Err> {
-        self.dec.collect_into(metadata, labels, name, &mut enc.0)
-    }
-}

libs/pageserver_api/src/controller_api.rs

@@ -2,9 +2,9 @@ use std::str::FromStr;
 
 /// Request/response types for the storage controller
 /// API (`/control/v1` prefix).  Implemented by the server
-/// in [`storage_controller::http`]
+/// in [`attachment_service::http`]
 use serde::{Deserialize, Serialize};
-use utils::id::{NodeId, TenantId};
+use utils::id::NodeId;
 
 use crate::{
     models::{ShardParameters, TenantConfig},
@@ -42,12 +42,6 @@ pub struct NodeConfigureRequest {
     pub scheduling: Option<NodeSchedulingPolicy>,
 }
 
-#[derive(Serialize, Deserialize)]
-pub struct TenantPolicyRequest {
-    pub placement: Option<PlacementPolicy>,
-    pub scheduling: Option<ShardSchedulingPolicy>,
-}
-
 #[derive(Serialize, Deserialize, Debug)]
 pub struct TenantLocateResponseShard {
     pub shard_id: TenantShardId,
@@ -68,27 +62,12 @@ pub struct TenantLocateResponse {
 
 #[derive(Serialize, Deserialize)]
 pub struct TenantDescribeResponse {
-    pub tenant_id: TenantId,
     pub shards: Vec<TenantDescribeResponseShard>,
     pub stripe_size: ShardStripeSize,
     pub policy: PlacementPolicy,
     pub config: TenantConfig,
 }
 
-#[derive(Serialize, Deserialize)]
-pub struct NodeDescribeResponse {
-    pub id: NodeId,
-
-    pub availability: NodeAvailabilityWrapper,
-    pub scheduling: NodeSchedulingPolicy,
-
-    pub listen_http_addr: String,
-    pub listen_http_port: u16,
-
-    pub listen_pg_addr: String,
-    pub listen_pg_port: u16,
-}
-
 #[derive(Serialize, Deserialize)]
 pub struct TenantDescribeResponseShard {
     pub tenant_shard_id: TenantShardId,
@@ -104,8 +83,6 @@ pub struct TenantDescribeResponseShard {
     pub is_pending_compute_notification: bool,
     /// A shard split is currently underway
     pub is_splitting: bool,
-
-    pub scheduling_policy: ShardSchedulingPolicy,
 }
 
 /// Explicitly migrating a particular shard is a low level operation
@@ -120,7 +97,7 @@ pub struct TenantShardMigrateRequest {
 /// Utilisation score indicating how good a candidate a pageserver
 /// is for scheduling the next tenant. See [`crate::models::PageserverUtilization`].
 /// Lower values are better.
-#[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, PartialOrd, Ord, Debug)]
+#[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, PartialOrd, Ord)]
 pub struct UtilizationScore(pub u64);
 
 impl UtilizationScore {
@@ -129,7 +106,7 @@ impl UtilizationScore {
     }
 }
 
-#[derive(Serialize, Deserialize, Clone, Copy, Debug)]
+#[derive(Serialize, Clone, Copy)]
 #[serde(into = "NodeAvailabilityWrapper")]
 pub enum NodeAvailability {
     // Normal, happy state
@@ -152,7 +129,7 @@ impl Eq for NodeAvailability {}
 // This wrapper provides serde functionality and it should only be used to
 // communicate with external callers which don't know or care about the
 // utilisation score of the pageserver it is targeting.
-#[derive(Serialize, Deserialize, Clone, Copy, Debug)]
+#[derive(Serialize, Deserialize, Clone)]
 pub enum NodeAvailabilityWrapper {
     Active,
     Offline,
@@ -178,33 +155,22 @@ impl From<NodeAvailability> for NodeAvailabilityWrapper {
     }
 }
 
-#[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, Debug)]
-pub enum ShardSchedulingPolicy {
-    // Normal mode: the tenant's scheduled locations may be updated at will, including
-    // for non-essential optimization.
-    Active,
+impl FromStr for NodeAvailability {
+    type Err = anyhow::Error;
 
-    // Disable optimizations, but permit scheduling when necessary to fulfil the PlacementPolicy.
-    // For example, this still permits a node's attachment location to change to a secondary in
-    // response to a node failure, or to assign a new secondary if a node was removed.
-    Essential,
-
-    // No scheduling: leave the shard running wherever it currently is.  Even if the shard is
-    // unavailable, it will not be rescheduled to another node.
-    Pause,
-
-    // No reconciling: we will make no location_conf API calls to pageservers at all.  If the
-    // shard is unavailable, it stays that way.  If a node fails, this shard doesn't get failed over.
-    Stop,
-}
-
-impl Default for ShardSchedulingPolicy {
-    fn default() -> Self {
-        Self::Active
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        match s {
+            // This is used when parsing node configuration requests from neon-local.
+            // Assume the worst possible utilisation score
+            // and let it get updated via the heartbeats.
+            "active" => Ok(Self::Active(UtilizationScore::worst())),
+            "offline" => Ok(Self::Offline),
+            _ => Err(anyhow::anyhow!("Unknown availability state '{s}'")),
+        }
     }
 }
 
-#[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, Debug)]
+#[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq)]
 pub enum NodeSchedulingPolicy {
     Active,
     Filling,

libs/pageserver_api/src/key.rs

@@ -1,10 +1,8 @@
 use anyhow::{bail, Result};
 use byteorder::{ByteOrder, BE};
-use bytes::BufMut;
 use postgres_ffi::relfile_utils::{FSM_FORKNUM, VISIBILITYMAP_FORKNUM};
 use postgres_ffi::{Oid, TransactionId};
 use serde::{Deserialize, Serialize};
-use std::ops::RangeInclusive;
 use std::{fmt, ops::Range};
 
 use crate::reltag::{BlockNumber, RelTag, SlruKind};
@@ -23,81 +21,9 @@ pub struct Key {
     pub field6: u32,
 }
 
-/// The storage key size.
 pub const KEY_SIZE: usize = 18;
 
-/// The metadata key size. 2B fewer than the storage key size because field2 is not fully utilized.
-/// See [`Key::to_i128`] for more information on the encoding.
-pub const METADATA_KEY_SIZE: usize = 16;
-
-/// The key prefix start range for the metadata keys. All keys with the first byte >= 0x80 is a metadata key.
-pub const METADATA_KEY_BEGIN_PREFIX: u8 = 0x80;
-
-/// The (reserved) key prefix of relation sizes.
-pub const RELATION_SIZE_PREFIX: u8 = 0x81;
-
-/// The key prefix of AUX file keys.
-pub const AUX_KEY_PREFIX: u8 = 0x82;
-
-/// Check if the key falls in the range of metadata keys.
-pub const fn is_metadata_key_slice(key: &[u8]) -> bool {
-    key[0] >= METADATA_KEY_BEGIN_PREFIX
-}
-
 impl Key {
-    /// Check if the key falls in the range of metadata keys.
-    pub const fn is_metadata_key(&self) -> bool {
-        self.field1 >= METADATA_KEY_BEGIN_PREFIX
-    }
-
-    /// Encode a metadata key to a storage key.
-    pub fn from_metadata_key_fixed_size(key: &[u8; METADATA_KEY_SIZE]) -> Self {
-        assert!(is_metadata_key_slice(key), "key not in metadata key range");
-        Key {
-            field1: key[0],
-            field2: u16::from_be_bytes(key[1..3].try_into().unwrap()) as u32,
-            field3: u32::from_be_bytes(key[3..7].try_into().unwrap()),
-            field4: u32::from_be_bytes(key[7..11].try_into().unwrap()),
-            field5: key[11],
-            field6: u32::from_be_bytes(key[12..16].try_into().unwrap()),
-        }
-    }
-
-    /// Encode a metadata key to a storage key.
-    pub fn from_metadata_key(key: &[u8]) -> Self {
-        Self::from_metadata_key_fixed_size(key.try_into().expect("expect 16 byte metadata key"))
-    }
-
-    /// Extract a metadata key to a writer. The result should always be 16 bytes.
-    pub fn extract_metadata_key_to_writer(&self, mut writer: impl BufMut) {
-        writer.put_u8(self.field1);
-        assert!(self.field2 <= 0xFFFF);
-        writer.put_u16(self.field2 as u16);
-        writer.put_u32(self.field3);
-        writer.put_u32(self.field4);
-        writer.put_u8(self.field5);
-        writer.put_u32(self.field6);
-    }
-
-    /// Get the range of metadata keys.
-    pub fn metadata_key_range() -> RangeInclusive<Self> {
-        Key {
-            field1: METADATA_KEY_BEGIN_PREFIX,
-            field2: 0,
-            field3: 0,
-            field4: 0,
-            field5: 0,
-            field6: 0,
-        }..=Key {
-            field1: u8::MAX,
-            field2: u16::MAX as u32,
-            field3: u32::MAX,
-            field4: u32::MAX,
-            field5: u8::MAX,
-            field6: u32::MAX,
-        }
-    }
-
     /// 'field2' is used to store tablespaceid for relations and small enum numbers for other relish.
     /// As long as Neon does not support tablespace (because of lack of access to local file system),
     /// we can assume that only some predefined namespace OIDs are used which can fit in u16
@@ -122,11 +48,11 @@ impl Key {
         }
     }
 
-    pub const fn next(&self) -> Key {
+    pub fn next(&self) -> Key {
         self.add(1)
     }
 
-    pub const fn add(&self, x: u32) -> Key {
+    pub fn add(&self, x: u32) -> Key {
         let mut key = *self;
 
         let r = key.field6.overflowing_add(x);
@@ -155,8 +81,6 @@ impl Key {
         key
     }
 
-    /// Convert a 18B slice to a key. This function should not be used for metadata keys because field2 is handled differently.
-    /// Use [`Key::from_metadata_key`] instead.
     pub fn from_slice(b: &[u8]) -> Self {
         Key {
             field1: b[0],
@@ -168,8 +92,6 @@ impl Key {
         }
     }
 
-    /// Convert a key to a 18B slice. This function should not be used for metadata keys because field2 is handled differently.
-    /// Use [`Key::extract_metadata_key_to_writer`] instead.
     pub fn write_to_byte_slice(&self, buf: &mut [u8]) {
         buf[0] = self.field1;
         BE::write_u32(&mut buf[1..5], self.field2);
@@ -553,14 +475,12 @@ pub const AUX_FILES_KEY: Key = Key {
 // Reverse mappings for a few Keys.
 // These are needed by WAL redo manager.
 
-pub const NON_INHERITED_RANGE: Range<Key> = AUX_FILES_KEY..AUX_FILES_KEY.next();
-
 // AUX_FILES currently stores only data for logical replication (slots etc), and
 // we don't preserve these on a branch because safekeepers can't follow timeline
 // switch (and generally it likely should be optional), so ignore these.
 #[inline(always)]
 pub fn is_inherited_key(key: Key) -> bool {
-    !NON_INHERITED_RANGE.contains(&key)
+    key != AUX_FILES_KEY
 }
 
 #[inline(always)]
@@ -636,14 +556,11 @@ impl std::str::FromStr for Key {
 mod tests {
     use std::str::FromStr;
 
-    use crate::key::is_metadata_key_slice;
     use crate::key::Key;
 
     use rand::Rng;
     use rand::SeedableRng;
 
-    use super::AUX_KEY_PREFIX;
-
     #[test]
     fn display_fromstr_bijection() {
         let mut rng = rand::rngs::StdRng::seed_from_u64(42);
@@ -659,16 +576,4 @@ mod tests {
 
         assert_eq!(key, Key::from_str(&format!("{key}")).unwrap());
     }
-
-    #[test]
-    fn test_metadata_keys() {
-        let mut metadata_key = vec![AUX_KEY_PREFIX];
-        metadata_key.extend_from_slice(&[0xFF; 15]);
-        let encoded_key = Key::from_metadata_key(&metadata_key);
-        let mut output_key = Vec::new();
-        encoded_key.extract_metadata_key_to_writer(&mut output_key);
-        assert_eq!(metadata_key, output_key);
-        assert!(encoded_key.is_metadata_key());
-        assert!(is_metadata_key_slice(&metadata_key));
-    }
 }

libs/pageserver_api/src/keyspace.rs

@@ -94,13 +94,12 @@ impl KeySpace {
 
     /// Remove all keys in `other` from `self`.
     /// This can involve splitting or removing of existing ranges.
-    /// Returns the removed keyspace
-    pub fn remove_overlapping_with(&mut self, other: &KeySpace) -> KeySpace {
+    pub fn remove_overlapping_with(&mut self, other: &KeySpace) {
         let (self_start, self_end) = match (self.start(), self.end()) {
             (Some(start), Some(end)) => (start, end),
             _ => {
                 // self is empty
-                return KeySpace::default();
+                return;
             }
         };
 
@@ -113,37 +112,30 @@ impl KeySpace {
             .skip_while(|range| self_start >= range.end)
             .take_while(|range| self_end > range.start);
 
-        let mut removed_accum = KeySpaceRandomAccum::new();
         for range in other_ranges {
             while let Some(overlap_at) = self.overlaps_at(range) {
                 let overlapped = self.ranges[overlap_at].clone();
 
                 if overlapped.start < range.start && overlapped.end <= range.end {
                     // Higher part of the range is completely overlapped.
-                    removed_accum.add_range(range.start..self.ranges[overlap_at].end);
                     self.ranges[overlap_at].end = range.start;
                 }
                 if overlapped.start >= range.start && overlapped.end > range.end {
                     // Lower part of the range is completely overlapped.
-                    removed_accum.add_range(self.ranges[overlap_at].start..range.end);
                     self.ranges[overlap_at].start = range.end;
                 }
                 if overlapped.start < range.start && overlapped.end > range.end {
                     // Middle part of the range is overlapped.
-                    removed_accum.add_range(range.clone());
                     self.ranges[overlap_at].end = range.start;
                     self.ranges
                         .insert(overlap_at + 1, range.end..overlapped.end);
                 }
                 if overlapped.start >= range.start && overlapped.end <= range.end {
                     // Whole range is overlapped
-                    removed_accum.add_range(self.ranges[overlap_at].clone());
                     self.ranges.remove(overlap_at);
                 }
             }
         }
-
-        removed_accum.to_keyspace()
     }
 
     pub fn start(&self) -> Option<Key> {
@@ -162,10 +154,6 @@ impl KeySpace {
             .sum()
     }
 
-    pub fn is_empty(&self) -> bool {
-        self.total_size() == 0
-    }
-
     fn overlaps_at(&self, range: &Range<Key>) -> Option<usize> {
         match self.ranges.binary_search_by_key(&range.end, |r| r.start) {
             Ok(0) => None,
@@ -565,16 +553,7 @@ mod tests {
                 Key::from_i128(11)..Key::from_i128(13),
             ],
         };
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace {
-            ranges: vec![
-                Key::from_i128(2)..Key::from_i128(3),
-                Key::from_i128(6)..Key::from_i128(7),
-                Key::from_i128(11)..Key::from_i128(12),
-            ],
-        };
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![
@@ -604,17 +583,7 @@ mod tests {
                 Key::from_i128(14)..Key::from_i128(17),
             ],
         };
-
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace {
-            ranges: vec![
-                Key::from_i128(3)..Key::from_i128(5),
-                Key::from_i128(8)..Key::from_i128(10),
-                Key::from_i128(14)..Key::from_i128(15),
-            ],
-        };
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![
@@ -641,11 +610,7 @@ mod tests {
                 Key::from_i128(15)..Key::from_i128(17),
             ],
         };
-
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace::default();
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![
@@ -672,17 +637,7 @@ mod tests {
         let key_space2 = KeySpace {
             ranges: vec![Key::from_i128(9)..Key::from_i128(19)],
         };
-
-        let removed = key_space1.remove_overlapping_with(&key_space2);
-        let removed_expected = KeySpace {
-            ranges: vec![
-                Key::from_i128(9)..Key::from_i128(10),
-                Key::from_i128(12)..Key::from_i128(15),
-                Key::from_i128(17)..Key::from_i128(19),
-            ],
-        };
-        assert_eq!(removed, removed_expected);
-
+        key_space1.remove_overlapping_with(&key_space2);
         assert_eq!(
             key_space1.ranges,
             vec![

libs/pageserver_api/src/models.rs

@@ -20,7 +20,6 @@ use utils::{
     history_buffer::HistoryBufferWithDropCounter,
     id::{NodeId, TenantId, TimelineId},
     lsn::Lsn,
-    serde_system_time,
 };
 
 use crate::controller_api::PlacementPolicy;
@@ -302,7 +301,6 @@ pub struct TenantConfig {
     pub heatmap_period: Option<String>,
     pub lazy_slru_download: Option<bool>,
     pub timeline_get_throttle: Option<ThrottleConfig>,
-    pub image_layer_creation_check_threshold: Option<u8>,
 }
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
@@ -429,7 +427,6 @@ pub struct StatusResponse {
 #[derive(Serialize, Deserialize, Debug)]
 #[serde(deny_unknown_fields)]
 pub struct TenantLocationConfigRequest {
-    #[serde(skip_serializing_if = "Option::is_none")]
     pub tenant_id: Option<TenantShardId>,
     #[serde(flatten)]
     pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
@@ -748,18 +745,10 @@ pub struct TimelineGcRequest {
     pub gc_horizon: Option<u64>,
 }
 
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct WalRedoManagerProcessStatus {
-    pub pid: u32,
-    /// The strum-generated `into::<&'static str>()` for `pageserver::walredo::ProcessKind`.
-    /// `ProcessKind` are a transitory thing, so, they have no enum representation in `pageserver_api`.
-    pub kind: Cow<'static, str>,
-}
-
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct WalRedoManagerStatus {
     pub last_redo_at: Option<chrono::DateTime<chrono::Utc>>,
-    pub process: Option<WalRedoManagerProcessStatus>,
+    pub pid: Option<u32>,
 }
 
 /// The progress of a secondary tenant is mostly useful when doing a long running download: e.g. initiating
@@ -768,7 +757,11 @@ pub struct WalRedoManagerStatus {
 #[derive(Default, Debug, Serialize, Deserialize, Clone)]
 pub struct SecondaryProgress {
     /// The remote storage LastModified time of the heatmap object we last downloaded.
-    pub heatmap_mtime: Option<serde_system_time::SystemTime>,
+    #[serde(
+        serialize_with = "opt_ser_rfc3339_millis",
+        deserialize_with = "opt_deser_rfc3339_millis"
+    )]
+    pub heatmap_mtime: Option<SystemTime>,
 
     /// The number of layers currently on-disk
     pub layers_downloaded: usize,
@@ -781,6 +774,29 @@ pub struct SecondaryProgress {
     pub bytes_total: u64,
 }
 
+fn opt_ser_rfc3339_millis<S: serde::Serializer>(
+    ts: &Option<SystemTime>,
+    serializer: S,
+) -> Result<S::Ok, S::Error> {
+    match ts {
+        Some(ts) => serializer.collect_str(&humantime::format_rfc3339_millis(*ts)),
+        None => serializer.serialize_none(),
+    }
+}
+
+fn opt_deser_rfc3339_millis<'de, D>(deserializer: D) -> Result<Option<SystemTime>, D::Error>
+where
+    D: serde::de::Deserializer<'de>,
+{
+    let s: Option<String> = serde::de::Deserialize::deserialize(deserializer)?;
+    match s {
+        None => Ok(None),
+        Some(s) => humantime::parse_rfc3339(&s)
+            .map_err(serde::de::Error::custom)
+            .map(Some),
+    }
+}
+
 pub mod virtual_file {
     #[derive(
         Copy,

libs/pageserver_api/src/models/utilization.rs

@@ -1,4 +1,4 @@
-use utils::serde_system_time::SystemTime;
+use std::time::SystemTime;
 
 /// Pageserver current utilization and scoring for how good candidate the pageserver would be for
 /// the next tenant.
@@ -21,9 +21,28 @@ pub struct PageserverUtilization {
     /// When was this snapshot captured, pageserver local time.
     ///
     /// Use millis to give confidence that the value is regenerated often enough.
+    #[serde(
+        serialize_with = "ser_rfc3339_millis",
+        deserialize_with = "deser_rfc3339_millis"
+    )]
     pub captured_at: SystemTime,
 }
 
+fn ser_rfc3339_millis<S: serde::Serializer>(
+    ts: &SystemTime,
+    serializer: S,
+) -> Result<S::Ok, S::Error> {
+    serializer.collect_str(&humantime::format_rfc3339_millis(*ts))
+}
+
+fn deser_rfc3339_millis<'de, D>(deserializer: D) -> Result<SystemTime, D::Error>
+where
+    D: serde::de::Deserializer<'de>,
+{
+    let s: String = serde::de::Deserialize::deserialize(deserializer)?;
+    humantime::parse_rfc3339(&s).map_err(serde::de::Error::custom)
+}
+
 /// openapi knows only `format: int64`, so avoid outputting a non-parseable value by generated clients.
 ///
 /// Instead of newtype, use this because a newtype would get require handling deserializing values
@@ -50,9 +69,7 @@ mod tests {
             disk_usage_bytes: u64::MAX,
             free_space_bytes: 0,
             utilization_score: u64::MAX,
-            captured_at: SystemTime(
-                std::time::SystemTime::UNIX_EPOCH + Duration::from_secs(1708509779),
-            ),
+            captured_at: SystemTime::UNIX_EPOCH + Duration::from_secs(1708509779),
         };
 
         let s = serde_json::to_string(&doc).unwrap();

libs/pageserver_api/src/shard.rs

@@ -5,93 +5,15 @@ use crate::{
     models::ShardParameters,
 };
 use hex::FromHex;
-use postgres_ffi::relfile_utils::INIT_FORKNUM;
 use serde::{Deserialize, Serialize};
 use utils::id::TenantId;
 
-/// See docs/rfcs/031-sharding-static.md for an overview of sharding.
-///
-/// This module contains a variety of types used to represent the concept of sharding
-/// a Neon tenant across multiple physical shards.  Since there are quite a few of these,
-/// we provide an summary here.
-///
-/// Types used to describe shards:
-/// - [`ShardCount`] describes how many shards make up a tenant, plus the magic `unsharded` value
-///   which identifies a tenant which is not shard-aware.  This means its storage paths do not include
-///   a shard suffix.
-/// - [`ShardNumber`] is simply the zero-based index of a shard within a tenant.
-/// - [`ShardIndex`] is the 2-tuple of `ShardCount` and `ShardNumber`, it's just like a `TenantShardId`
-///   without the tenant ID.  This is useful for things that are implicitly scoped to a particular
-///   tenant, such as layer files.
-/// - [`ShardIdentity`]` is the full description of a particular shard's parameters, in sufficient
-///   detail to convert a [`Key`] to a [`ShardNumber`] when deciding where to write/read.
-/// - The [`ShardSlug`] is a terse formatter for ShardCount and ShardNumber, written as
-///   four hex digits.  An unsharded tenant is `0000`.
-/// - [`TenantShardId`] is the unique ID of a particular shard within a particular tenant
-///
-/// Types used to describe the parameters for data distribution in a sharded tenant:
-/// - [`ShardStripeSize`] controls how long contiguous runs of [`Key`]s (stripes) are when distributed across
-///   multiple shards.  Its value is given in 8kiB pages.
-/// - [`ShardLayout`] describes the data distribution scheme, and at time of writing is
-///   always zero: this is provided for future upgrades that might introduce different
-///   data distribution schemes.
-///
-/// Examples:
-/// - A legacy unsharded tenant has one shard with ShardCount(0), ShardNumber(0), and its slug is 0000
-/// - A single sharded tenant has one shard with ShardCount(1), ShardNumber(0), and its slug is 0001
-/// - In a tenant with 4 shards, each shard has ShardCount(N), ShardNumber(i) where i in 0..N-1 (inclusive),
-///   and their slugs are 0004, 0104, 0204, and 0304.
-
 #[derive(Ord, PartialOrd, Eq, PartialEq, Clone, Copy, Serialize, Deserialize, Debug, Hash)]
 pub struct ShardNumber(pub u8);
 
 #[derive(Ord, PartialOrd, Eq, PartialEq, Clone, Copy, Serialize, Deserialize, Debug, Hash)]
 pub struct ShardCount(u8);
 
-/// Combination of ShardNumber and ShardCount.  For use within the context of a particular tenant,
-/// when we need to know which shard we're dealing with, but do not need to know the full
-/// ShardIdentity (because we won't be doing any page->shard mapping), and do not need to know
-/// the fully qualified TenantShardId.
-#[derive(Eq, PartialEq, PartialOrd, Ord, Clone, Copy, Hash)]
-pub struct ShardIndex {
-    pub shard_number: ShardNumber,
-    pub shard_count: ShardCount,
-}
-
-/// The ShardIdentity contains enough information to map a [`Key`] to a [`ShardNumber`],
-/// and to check whether that [`ShardNumber`] is the same as the current shard.
-#[derive(Clone, Copy, Serialize, Deserialize, Eq, PartialEq, Debug)]
-pub struct ShardIdentity {
-    pub number: ShardNumber,
-    pub count: ShardCount,
-    pub stripe_size: ShardStripeSize,
-    layout: ShardLayout,
-}
-
-/// Formatting helper, for generating the `shard_id` label in traces.
-struct ShardSlug<'a>(&'a TenantShardId);
-
-/// TenantShardId globally identifies a particular shard in a particular tenant.
-///
-/// These are written as `<TenantId>-<ShardSlug>`, for example:
-///   # The second shard in a two-shard tenant
-///   072f1291a5310026820b2fe4b2968934-0102
-///
-/// If the `ShardCount` is _unsharded_, the `TenantShardId` is written without
-/// a shard suffix and is equivalent to the encoding of a `TenantId`: this enables
-/// an unsharded [`TenantShardId`] to be used interchangably with a [`TenantId`].
-///
-/// The human-readable encoding of an unsharded TenantShardId, such as used in API URLs,
-/// is both forward and backward compatible with TenantId: a legacy TenantId can be
-/// decoded as a TenantShardId, and when re-encoded it will be parseable
-/// as a TenantId.
-#[derive(Eq, PartialEq, PartialOrd, Ord, Clone, Copy, Hash)]
-pub struct TenantShardId {
-    pub tenant_id: TenantId,
-    pub shard_number: ShardNumber,
-    pub shard_count: ShardCount,
-}
-
 impl ShardCount {
     pub const MAX: Self = Self(u8::MAX);
 
@@ -116,7 +38,6 @@ impl ShardCount {
         self.0
     }
 
-    ///
     pub fn is_unsharded(&self) -> bool {
         self.0 == 0
     }
@@ -132,6 +53,33 @@ impl ShardNumber {
     pub const MAX: Self = Self(u8::MAX);
 }
 
+/// TenantShardId identify the units of work for the Pageserver.
+///
+/// These are written as `<tenant_id>-<shard number><shard-count>`, for example:
+///
+///   # The second shard in a two-shard tenant
+///   072f1291a5310026820b2fe4b2968934-0102
+///
+/// Historically, tenants could not have multiple shards, and were identified
+/// by TenantId.  To support this, TenantShardId has a special legacy
+/// mode where `shard_count` is equal to zero: this represents a single-sharded
+/// tenant which should be written as a TenantId with no suffix.
+///
+/// The human-readable encoding of TenantShardId, such as used in API URLs,
+/// is both forward and backward compatible: a legacy TenantId can be
+/// decoded as a TenantShardId, and when re-encoded it will be parseable
+/// as a TenantId.
+///
+/// Note that the binary encoding is _not_ backward compatible, because
+/// at the time sharding is introduced, there are no existing binary structures
+/// containing TenantId that we need to handle.
+#[derive(Eq, PartialEq, PartialOrd, Ord, Clone, Copy, Hash)]
+pub struct TenantShardId {
+    pub tenant_id: TenantId,
+    pub shard_number: ShardNumber,
+    pub shard_count: ShardCount,
+}
+
 impl TenantShardId {
     pub fn unsharded(tenant_id: TenantId) -> Self {
         Self {
@@ -163,13 +111,10 @@ impl TenantShardId {
     }
 
     /// Convenience for code that has special behavior on the 0th shard.
-    pub fn is_shard_zero(&self) -> bool {
+    pub fn is_zero(&self) -> bool {
         self.shard_number == ShardNumber(0)
     }
 
-    /// The "unsharded" value is distinct from simply having a single shard: it represents
-    /// a tenant which is not shard-aware at all, and whose storage paths will not include
-    /// a shard suffix.
     pub fn is_unsharded(&self) -> bool {
         self.shard_number == ShardNumber(0) && self.shard_count.is_unsharded()
     }
@@ -205,6 +150,9 @@ impl TenantShardId {
     }
 }
 
+/// Formatting helper
+struct ShardSlug<'a>(&'a TenantShardId);
+
 impl<'a> std::fmt::Display for ShardSlug<'a> {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         write!(
@@ -274,6 +222,16 @@ impl From<[u8; 18]> for TenantShardId {
     }
 }
 
+/// For use within the context of a particular tenant, when we need to know which
+/// shard we're dealing with, but do not need to know the full ShardIdentity (because
+/// we won't be doing any page->shard mapping), and do not need to know the fully qualified
+/// TenantShardId.
+#[derive(Eq, PartialEq, PartialOrd, Ord, Clone, Copy, Hash)]
+pub struct ShardIndex {
+    pub shard_number: ShardNumber,
+    pub shard_count: ShardCount,
+}
+
 impl ShardIndex {
     pub fn new(number: ShardNumber, count: ShardCount) -> Self {
         Self {
@@ -288,9 +246,6 @@ impl ShardIndex {
         }
     }
 
-    /// The "unsharded" value is distinct from simply having a single shard: it represents
-    /// a tenant which is not shard-aware at all, and whose storage paths will not include
-    /// a shard suffix.
     pub fn is_unsharded(&self) -> bool {
         self.shard_number == ShardNumber(0) && self.shard_count == ShardCount(0)
     }
@@ -358,8 +313,6 @@ impl Serialize for TenantShardId {
         if serializer.is_human_readable() {
             serializer.collect_str(self)
         } else {
-            // Note: while human encoding of [`TenantShardId`] is backward and forward
-            // compatible, this binary encoding is not.
             let mut packed: [u8; 18] = [0; 18];
             packed[0..16].clone_from_slice(&self.tenant_id.as_arr());
             packed[16] = self.shard_number.0;
@@ -437,6 +390,16 @@ const LAYOUT_BROKEN: ShardLayout = ShardLayout(255);
 /// Default stripe size in pages: 256MiB divided by 8kiB page size.
 const DEFAULT_STRIPE_SIZE: ShardStripeSize = ShardStripeSize(256 * 1024 / 8);
 
+/// The ShardIdentity contains the information needed for one member of map
+/// to resolve a key to a shard, and then check whether that shard is ==self.
+#[derive(Clone, Copy, Serialize, Deserialize, Eq, PartialEq, Debug)]
+pub struct ShardIdentity {
+    pub number: ShardNumber,
+    pub count: ShardCount,
+    pub stripe_size: ShardStripeSize,
+    layout: ShardLayout,
+}
+
 #[derive(thiserror::Error, Debug, PartialEq, Eq)]
 pub enum ShardConfigError {
     #[error("Invalid shard count")]
@@ -476,9 +439,6 @@ impl ShardIdentity {
         }
     }
 
-    /// The "unsharded" value is distinct from simply having a single shard: it represents
-    /// a tenant which is not shard-aware at all, and whose storage paths will not include
-    /// a shard suffix.
     pub fn is_unsharded(&self) -> bool {
         self.number == ShardNumber(0) && self.count == ShardCount(0)
     }
@@ -527,8 +487,6 @@ impl ShardIdentity {
     }
 
     /// Return true if the key should be ingested by this shard
-    ///
-    /// Shards must ingest _at least_ keys which return true from this check.
     pub fn is_key_local(&self, key: &Key) -> bool {
         assert!(!self.is_broken());
         if self.count < ShardCount(2) || (key_is_shard0(key) && self.number == ShardNumber(0)) {
@@ -538,28 +496,8 @@ impl ShardIdentity {
         }
     }
 
-    /// Special case for issue `<https://github.com/neondatabase/neon/issues/7451>`
-    ///
-    /// When we fail to read a forknum block, this function tells us whether we may ignore the error
-    /// as a symptom of that issue.
-    pub fn is_key_buggy_forknum(&self, key: &Key) -> bool {
-        if !is_rel_block_key(key) || key.field5 != INIT_FORKNUM {
-            return false;
-        }
-
-        let mut hash = murmurhash32(key.field4);
-        hash = hash_combine(hash, murmurhash32(key.field6 / self.stripe_size.0));
-        let mapped_shard = ShardNumber((hash % self.count.0 as u32) as u8);
-
-        // The key may be affected by issue #7454: it is an initfork and it would not
-        // have mapped to shard 0 until we fixed that issue.
-        mapped_shard != ShardNumber(0)
-    }
-
     /// Return true if the key should be discarded if found in this shard's
-    /// data store, e.g. during compaction after a split.
-    ///
-    /// Shards _may_ drop keys which return false here, but are not obliged to.
+    /// data store, e.g. during compaction after a split
     pub fn is_key_disposable(&self, key: &Key) -> bool {
         if key_is_shard0(key) {
             // Q: Why can't we dispose of shard0 content if we're not shard 0?
@@ -585,7 +523,7 @@ impl ShardIdentity {
 
     /// Convenience for checking if this identity is the 0th shard in a tenant,
     /// for special cases on shard 0 such as ingesting relation sizes.
-    pub fn is_shard_zero(&self) -> bool {
+    pub fn is_zero(&self) -> bool {
         self.number == ShardNumber(0)
     }
 }
@@ -668,13 +606,7 @@ fn key_is_shard0(key: &Key) -> bool {
     // relation pages are distributed to shards other than shard zero. Everything else gets
     // stored on shard 0.  This guarantees that shard 0 can independently serve basebackup
     // requests, and any request other than those for particular blocks in relations.
-    //
-    // The only exception to this rule is "initfork" data -- this relates to postgres's UNLOGGED table
-    // type. These are special relations, usually with only 0 or 1 blocks, and we store them on shard 0
-    // because they must be included in basebackups.
-    let is_initfork = key.field5 == INIT_FORKNUM;
-
-    !is_rel_block_key(key) || is_initfork
+    !is_rel_block_key(key)
 }
 
 /// Provide the same result as the function in postgres `hashfn.h` with the same name

libs/postgres_ffi/src/lib.rs

@@ -118,9 +118,7 @@ pub use v14::bindings::{TimeLineID, TimestampTz, XLogRecPtr, XLogSegNo};
 // Likewise for these, although the assumption that these don't change is a little more iffy.
 pub use v14::bindings::{MultiXactOffset, MultiXactStatus};
 pub use v14::bindings::{PageHeaderData, XLogRecord};
-pub use v14::xlog_utils::{
-    XLOG_SIZE_OF_XLOG_LONG_PHD, XLOG_SIZE_OF_XLOG_RECORD, XLOG_SIZE_OF_XLOG_SHORT_PHD,
-};
+pub use v14::xlog_utils::{XLOG_SIZE_OF_XLOG_RECORD, XLOG_SIZE_OF_XLOG_SHORT_PHD};
 
 pub use v14::bindings::{CheckPoint, ControlFileData};
 

libs/postgres_ffi/wal_craft/src/bin/wal_craft.rs

@@ -1,6 +1,5 @@
 use anyhow::*;
 use clap::{value_parser, Arg, ArgMatches, Command};
-use postgres::Client;
 use std::{path::PathBuf, str::FromStr};
 use wal_craft::*;
 
@@ -9,8 +8,8 @@ fn main() -> Result<()> {
         .init();
     let arg_matches = cli().get_matches();
 
-    let wal_craft = |arg_matches: &ArgMatches, client: &mut Client| {
-        let intermediate_lsns = match arg_matches
+    let wal_craft = |arg_matches: &ArgMatches, client| {
+        let (intermediate_lsns, end_of_wal_lsn) = match arg_matches
             .get_one::<String>("type")
             .map(|s| s.as_str())
             .context("'type' is required")?
@@ -26,7 +25,6 @@ fn main() -> Result<()> {
             LastWalRecordCrossingSegment::NAME => LastWalRecordCrossingSegment::craft(client)?,
             a => panic!("Unknown --type argument: {a}"),
         };
-        let end_of_wal_lsn = client.pg_current_wal_insert_lsn()?;
         for lsn in intermediate_lsns {
             println!("intermediate_lsn = {lsn}");
         }

libs/postgres_ffi/wal_craft/src/lib.rs

@@ -4,9 +4,8 @@ use log::*;
 use postgres::types::PgLsn;
 use postgres::Client;
 use postgres_ffi::{WAL_SEGMENT_SIZE, XLOG_BLCKSZ};
-use postgres_ffi::{
-    XLOG_SIZE_OF_XLOG_LONG_PHD, XLOG_SIZE_OF_XLOG_RECORD, XLOG_SIZE_OF_XLOG_SHORT_PHD,
-};
+use postgres_ffi::{XLOG_SIZE_OF_XLOG_RECORD, XLOG_SIZE_OF_XLOG_SHORT_PHD};
+use std::cmp::Ordering;
 use std::path::{Path, PathBuf};
 use std::process::Command;
 use std::time::{Duration, Instant};
@@ -233,62 +232,59 @@ pub fn ensure_server_config(client: &mut impl postgres::GenericClient) -> anyhow
 pub trait Crafter {
     const NAME: &'static str;
 
-    /// Generates WAL using the client `client`. Returns a vector of some valid
-    /// "interesting" intermediate LSNs which one may start reading from.
-    /// test_end_of_wal uses this to check various starting points.
-    ///
-    /// Note that postgres is generally keen about writing some WAL. While we
-    /// try to disable it (autovacuum, big wal_writer_delay, etc) it is always
-    /// possible, e.g. xl_running_xacts are dumped each 15s. So checks about
-    /// stable WAL end would be flaky unless postgres is shut down. For this
-    /// reason returning potential end of WAL here is pointless. Most of the
-    /// time this doesn't happen though, so it is reasonable to create needed
-    /// WAL structure and immediately kill postgres like test_end_of_wal does.
-    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<Vec<PgLsn>>;
+    /// Generates WAL using the client `client`. Returns a pair of:
+    /// * A vector of some valid "interesting" intermediate LSNs which one may start reading from.
+    ///   May include or exclude Lsn(0) and the end-of-wal.
+    /// * The expected end-of-wal LSN.
+    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<(Vec<PgLsn>, PgLsn)>;
 }
 
-/// Wraps some WAL craft function, providing current LSN to it before the
-/// insertion and flushing WAL afterwards. Also pushes initial LSN to the
-/// result.
 fn craft_internal<C: postgres::GenericClient>(
     client: &mut C,
-    f: impl Fn(&mut C, PgLsn) -> anyhow::Result<Vec<PgLsn>>,
-) -> anyhow::Result<Vec<PgLsn>> {
+    f: impl Fn(&mut C, PgLsn) -> anyhow::Result<(Vec<PgLsn>, Option<PgLsn>)>,
+) -> anyhow::Result<(Vec<PgLsn>, PgLsn)> {
     ensure_server_config(client)?;
 
     let initial_lsn = client.pg_current_wal_insert_lsn()?;
     info!("LSN initial = {}", initial_lsn);
 
-    let mut intermediate_lsns = f(client, initial_lsn)?;
+    let (mut intermediate_lsns, last_lsn) = f(client, initial_lsn)?;
+    let last_lsn = match last_lsn {
+        None => client.pg_current_wal_insert_lsn()?,
+        Some(last_lsn) => {
+            let insert_lsn = client.pg_current_wal_insert_lsn()?;
+            match last_lsn.cmp(&insert_lsn) {
+                Ordering::Less => bail!(
+                    "Some records were inserted after the crafted WAL: {} vs {}",
+                    last_lsn,
+                    insert_lsn
+                ),
+                Ordering::Equal => last_lsn,
+                Ordering::Greater => bail!("Reported LSN is greater than insert_lsn"),
+            }
+        }
+    };
     if !intermediate_lsns.starts_with(&[initial_lsn]) {
         intermediate_lsns.insert(0, initial_lsn);
     }
 
-    // Some records may be not flushed, e.g. non-transactional logical messages. Flush now.
-    //
-    // If the previous WAL record ended exactly at page boundary, pg_current_wal_insert_lsn
-    // returns the position just after the page header on the next page. That's where the next
-    // record will be inserted. But the page header hasn't actually been written to the WAL
-    // yet, and if you try to flush it, you get a "request to flush past end of generated WAL"
-    // error. Because of that, if the insert location is just after a page header, back off to
-    // previous page boundary.
-    let mut lsn = u64::from(client.pg_current_wal_insert_lsn()?);
-    if lsn % WAL_SEGMENT_SIZE as u64 == XLOG_SIZE_OF_XLOG_LONG_PHD as u64 {
-        lsn -= XLOG_SIZE_OF_XLOG_LONG_PHD as u64;
-    } else if lsn % XLOG_BLCKSZ as u64 == XLOG_SIZE_OF_XLOG_SHORT_PHD as u64 {
-        lsn -= XLOG_SIZE_OF_XLOG_SHORT_PHD as u64;
+    // Some records may be not flushed, e.g. non-transactional logical messages.
+    client.execute("select neon_xlogflush(pg_current_wal_insert_lsn())", &[])?;
+    match last_lsn.cmp(&client.pg_current_wal_flush_lsn()?) {
+        Ordering::Less => bail!("Some records were flushed after the crafted WAL"),
+        Ordering::Equal => {}
+        Ordering::Greater => bail!("Reported LSN is greater than flush_lsn"),
     }
-    client.execute("select neon_xlogflush($1)", &[&PgLsn::from(lsn)])?;
-    Ok(intermediate_lsns)
+    Ok((intermediate_lsns, last_lsn))
 }
 
 pub struct Simple;
 impl Crafter for Simple {
     const NAME: &'static str = "simple";
-    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<Vec<PgLsn>> {
+    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<(Vec<PgLsn>, PgLsn)> {
         craft_internal(client, |client, _| {
             client.execute("CREATE table t(x int)", &[])?;
-            Ok(Vec::new())
+            Ok((Vec::new(), None))
         })
     }
 }
@@ -296,85 +292,67 @@ impl Crafter for Simple {
 pub struct LastWalRecordXlogSwitch;
 impl Crafter for LastWalRecordXlogSwitch {
     const NAME: &'static str = "last_wal_record_xlog_switch";
-    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<Vec<PgLsn>> {
-        // Do not use craft_internal because here we end up with flush_lsn exactly on
+    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<(Vec<PgLsn>, PgLsn)> {
+        // Do not use generate_internal because here we end up with flush_lsn exactly on
         // the segment boundary and insert_lsn after the initial page header, which is unusual.
         ensure_server_config(client)?;
 
         client.execute("CREATE table t(x int)", &[])?;
         let before_xlog_switch = client.pg_current_wal_insert_lsn()?;
-        // pg_switch_wal returns end of last record of the switched segment,
-        // i.e. end of SWITCH itself.
-        let xlog_switch_record_end: PgLsn = client.query_one("SELECT pg_switch_wal()", &[])?.get(0);
-        let before_xlog_switch_u64 = u64::from(before_xlog_switch);
-        let next_segment = PgLsn::from(
-            before_xlog_switch_u64 - (before_xlog_switch_u64 % WAL_SEGMENT_SIZE as u64)
-                + WAL_SEGMENT_SIZE as u64,
-        );
+        let after_xlog_switch: PgLsn = client.query_one("SELECT pg_switch_wal()", &[])?.get(0);
+        let next_segment = PgLsn::from(0x0200_0000);
         ensure!(
-            xlog_switch_record_end <= next_segment,
-            "XLOG_SWITCH record ended after the expected segment boundary: {} > {}",
-            xlog_switch_record_end,
+            after_xlog_switch <= next_segment,
+            "XLOG_SWITCH message ended after the expected segment boundary: {} > {}",
+            after_xlog_switch,
             next_segment
         );
-        Ok(vec![before_xlog_switch, xlog_switch_record_end])
+        Ok((vec![before_xlog_switch, after_xlog_switch], next_segment))
     }
 }
 
 pub struct LastWalRecordXlogSwitchEndsOnPageBoundary;
-/// Craft xlog SWITCH record ending at page boundary.
 impl Crafter for LastWalRecordXlogSwitchEndsOnPageBoundary {
     const NAME: &'static str = "last_wal_record_xlog_switch_ends_on_page_boundary";
-    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<Vec<PgLsn>> {
+    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<(Vec<PgLsn>, PgLsn)> {
         // Do not use generate_internal because here we end up with flush_lsn exactly on
         // the segment boundary and insert_lsn after the initial page header, which is unusual.
         ensure_server_config(client)?;
 
         client.execute("CREATE table t(x int)", &[])?;
 
-        // Add padding so the XLOG_SWITCH record ends exactly on XLOG_BLCKSZ boundary.  We
-        // will use carefully-sized logical messages to advance WAL insert location such
-        // that there is just enough space on the page for the XLOG_SWITCH record.
-        loop {
-            // We start with measuring how much WAL it takes for one logical message,
-            // considering all alignments and headers.
+        // Add padding so the XLOG_SWITCH record ends exactly on XLOG_BLCKSZ boundary.
+        // We will use logical message as the padding. We start with detecting how much WAL
+        // it takes for one logical message, considering all alignments and headers.
+        let base_wal_advance = {
             let before_lsn = client.pg_current_wal_insert_lsn()?;
+            // Small non-empty message bigger than few bytes is more likely than an empty
+            // message to have the same format as the big padding message.
             client.execute(
                 "SELECT pg_logical_emit_message(false, 'swch', REPEAT('a', 10))",
                 &[],
             )?;
-            let after_lsn = client.pg_current_wal_insert_lsn()?;
-
-            // Did the record cross a page boundary? If it did, start over. Crossing a
-            // page boundary adds to the apparent size of the record because of the page
-            // header, which throws off the calculation.
-            if u64::from(before_lsn) / XLOG_BLCKSZ as u64
-                != u64::from(after_lsn) / XLOG_BLCKSZ as u64
-            {
-                continue;
-            }
-            // base_size is the size of a logical message without the payload
-            let base_size = u64::from(after_lsn) - u64::from(before_lsn) - 10;
-
-            // Is there enough space on the page for another logical message and an
-            // XLOG_SWITCH? If not, start over.
-            let page_remain = XLOG_BLCKSZ as u64 - u64::from(after_lsn) % XLOG_BLCKSZ as u64;
-            if page_remain < base_size - XLOG_SIZE_OF_XLOG_RECORD as u64 {
-                continue;
-            }
-
-            // We will write another logical message, such that after the logical message
-            // record, there will be space for exactly one XLOG_SWITCH. How large should
-            // the logical message's payload be? An XLOG_SWITCH record has no data => its
-            // size is exactly XLOG_SIZE_OF_XLOG_RECORD.
-            let repeats = page_remain - base_size - XLOG_SIZE_OF_XLOG_RECORD as u64;
-
-            client.execute(
-                "SELECT pg_logical_emit_message(false, 'swch', REPEAT('a', $1))",
-                &[&(repeats as i32)],
-            )?;
-            break;
+            // The XLOG_SWITCH record has no data => its size is exactly XLOG_SIZE_OF_XLOG_RECORD.
+            (u64::from(client.pg_current_wal_insert_lsn()?) - u64::from(before_lsn)) as usize
+                + XLOG_SIZE_OF_XLOG_RECORD
+        };
+        let mut remaining_lsn =
+            XLOG_BLCKSZ - u64::from(client.pg_current_wal_insert_lsn()?) as usize % XLOG_BLCKSZ;
+        if remaining_lsn < base_wal_advance {
+            remaining_lsn += XLOG_BLCKSZ;
         }
+        let repeats = 10 + remaining_lsn - base_wal_advance;
+        info!(
+            "current_wal_insert_lsn={}, remaining_lsn={}, base_wal_advance={}, repeats={}",
+            client.pg_current_wal_insert_lsn()?,
+            remaining_lsn,
+            base_wal_advance,
+            repeats
+        );
+        client.execute(
+            "SELECT pg_logical_emit_message(false, 'swch', REPEAT('a', $1))",
+            &[&(repeats as i32)],
+        )?;
         info!(
             "current_wal_insert_lsn={}, XLOG_SIZE_OF_XLOG_RECORD={}",
             client.pg_current_wal_insert_lsn()?,
@@ -383,29 +361,28 @@ impl Crafter for LastWalRecordXlogSwitchEndsOnPageBoundary {
 
         // Emit the XLOG_SWITCH
         let before_xlog_switch = client.pg_current_wal_insert_lsn()?;
-        let xlog_switch_record_end: PgLsn = client.query_one("SELECT pg_switch_wal()", &[])?.get(0);
+        let after_xlog_switch: PgLsn = client.query_one("SELECT pg_switch_wal()", &[])?.get(0);
         let next_segment = PgLsn::from(0x0200_0000);
         ensure!(
-            xlog_switch_record_end < next_segment,
-            "XLOG_SWITCH record ended on or after the expected segment boundary: {} > {}",
-            xlog_switch_record_end,
+            after_xlog_switch < next_segment,
+            "XLOG_SWITCH message ended on or after the expected segment boundary: {} > {}",
+            after_xlog_switch,
             next_segment
         );
         ensure!(
-            u64::from(xlog_switch_record_end) as usize % XLOG_BLCKSZ == XLOG_SIZE_OF_XLOG_SHORT_PHD,
+            u64::from(after_xlog_switch) as usize % XLOG_BLCKSZ == XLOG_SIZE_OF_XLOG_SHORT_PHD,
             "XLOG_SWITCH message ended not on page boundary: {}, offset = {}",
-            xlog_switch_record_end,
-            u64::from(xlog_switch_record_end) as usize % XLOG_BLCKSZ
+            after_xlog_switch,
+            u64::from(after_xlog_switch) as usize % XLOG_BLCKSZ
         );
-        Ok(vec![before_xlog_switch, xlog_switch_record_end])
+        Ok((vec![before_xlog_switch, after_xlog_switch], next_segment))
     }
 }
 
-/// Write ~16MB logical message; it should cross WAL segment.
-fn craft_seg_size_logical_message(
+fn craft_single_logical_message(
     client: &mut impl postgres::GenericClient,
     transactional: bool,
-) -> anyhow::Result<Vec<PgLsn>> {
+) -> anyhow::Result<(Vec<PgLsn>, PgLsn)> {
     craft_internal(client, |client, initial_lsn| {
         ensure!(
             initial_lsn < PgLsn::from(0x0200_0000 - 1024 * 1024),
@@ -428,24 +405,34 @@ fn craft_seg_size_logical_message(
             "Logical message crossed two segments"
         );
 
-        Ok(vec![message_lsn])
+        if transactional {
+            // Transactional logical messages are part of a transaction, so the one above is
+            // followed by a small COMMIT record.
+
+            let after_message_lsn = client.pg_current_wal_insert_lsn()?;
+            ensure!(
+                message_lsn < after_message_lsn,
+                "No record found after the emitted message"
+            );
+            Ok((vec![message_lsn], Some(after_message_lsn)))
+        } else {
+            Ok((Vec::new(), Some(message_lsn)))
+        }
     })
 }
 
 pub struct WalRecordCrossingSegmentFollowedBySmallOne;
 impl Crafter for WalRecordCrossingSegmentFollowedBySmallOne {
     const NAME: &'static str = "wal_record_crossing_segment_followed_by_small_one";
-    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<Vec<PgLsn>> {
-        // Transactional message crossing WAL segment will be followed by small
-        // commit record.
-        craft_seg_size_logical_message(client, true)
+    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<(Vec<PgLsn>, PgLsn)> {
+        craft_single_logical_message(client, true)
     }
 }
 
 pub struct LastWalRecordCrossingSegment;
 impl Crafter for LastWalRecordCrossingSegment {
     const NAME: &'static str = "last_wal_record_crossing_segment";
-    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<Vec<PgLsn>> {
-        craft_seg_size_logical_message(client, false)
+    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<(Vec<PgLsn>, PgLsn)> {
+        craft_single_logical_message(client, false)
     }
 }

libs/postgres_ffi/wal_craft/src/xlog_utils_test.rs

@@ -11,15 +11,13 @@ use utils::const_assert;
 use utils::lsn::Lsn;
 
 fn init_logging() {
-    let _ = env_logger::Builder::from_env(env_logger::Env::default().default_filter_or(format!(
-        "crate=info,postgres_ffi::{PG_MAJORVERSION}::xlog_utils=trace"
-    )))
+    let _ = env_logger::Builder::from_env(env_logger::Env::default().default_filter_or(
+        format!("crate=info,postgres_ffi::{PG_MAJORVERSION}::xlog_utils=trace"),
+    ))
     .is_test(true)
     .try_init();
 }
 
-/// Test that find_end_of_wal returns the same results as pg_dump on various
-/// WALs created by Crafter.
 fn test_end_of_wal<C: crate::Crafter>(test_name: &str) {
     use crate::*;
 
@@ -40,13 +38,13 @@ fn test_end_of_wal<C: crate::Crafter>(test_name: &str) {
     }
     cfg.initdb().unwrap();
     let srv = cfg.start_server().unwrap();
-    let intermediate_lsns = C::craft(&mut srv.connect_with_timeout().unwrap()).unwrap();
+    let (intermediate_lsns, expected_end_of_wal_partial) =
+        C::craft(&mut srv.connect_with_timeout().unwrap()).unwrap();
     let intermediate_lsns: Vec<Lsn> = intermediate_lsns
         .iter()
         .map(|&lsn| u64::from(lsn).into())
         .collect();
-    // Kill postgres. Note that it might have inserted to WAL something after
-    // 'craft' did its job.
+    let expected_end_of_wal: Lsn = u64::from(expected_end_of_wal_partial).into();
     srv.kill();
 
     // Check find_end_of_wal on the initial WAL
@@ -58,7 +56,7 @@ fn test_end_of_wal<C: crate::Crafter>(test_name: &str) {
         .filter(|fname| IsXLogFileName(fname))
         .max()
         .unwrap();
-    let expected_end_of_wal = find_pg_waldump_end_of_wal(&cfg, &last_segment);
+    check_pg_waldump_end_of_wal(&cfg, &last_segment, expected_end_of_wal);
     for start_lsn in intermediate_lsns
         .iter()
         .chain(std::iter::once(&expected_end_of_wal))
@@ -93,7 +91,11 @@ fn test_end_of_wal<C: crate::Crafter>(test_name: &str) {
     }
 }
 
-fn find_pg_waldump_end_of_wal(cfg: &crate::Conf, last_segment: &str) -> Lsn {
+fn check_pg_waldump_end_of_wal(
+    cfg: &crate::Conf,
+    last_segment: &str,
+    expected_end_of_wal: Lsn,
+) {
     // Get the actual end of WAL by pg_waldump
     let waldump_output = cfg
         .pg_waldump("000000010000000000000001", last_segment)
@@ -111,8 +113,11 @@ fn find_pg_waldump_end_of_wal(cfg: &crate::Conf, last_segment: &str) -> Lsn {
         }
     };
     let waldump_wal_end = Lsn::from_str(caps.get(1).unwrap().as_str()).unwrap();
-    info!("waldump erred on {}", waldump_wal_end);
-    waldump_wal_end
+    info!(
+        "waldump erred on {}, expected wal end at {}",
+        waldump_wal_end, expected_end_of_wal
+    );
+    assert_eq!(waldump_wal_end, expected_end_of_wal);
 }
 
 fn check_end_of_wal(
@@ -205,9 +210,9 @@ pub fn test_update_next_xid() {
 #[test]
 pub fn test_encode_logical_message() {
     let expected = [
-        64, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 21, 0, 0, 170, 34, 166, 227, 255, 38,
-        0, 0, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0, 0, 112, 114, 101, 102,
-        105, 120, 0, 109, 101, 115, 115, 97, 103, 101,
+        64, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 21, 0, 0, 170, 34, 166, 227, 255,
+        38, 0, 0, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0, 0, 112, 114,
+        101, 102, 105, 120, 0, 109, 101, 115, 115, 97, 103, 101,
     ];
     let actual = encode_logical_message("prefix", "message");
     assert_eq!(expected, actual[..]);

libs/remote_storage/src/lib.rs

@@ -21,13 +21,11 @@ use std::{
     fmt::Debug,
     num::{NonZeroU32, NonZeroUsize},
     pin::Pin,
-    str::FromStr,
     sync::Arc,
     time::{Duration, SystemTime},
 };
 
 use anyhow::{bail, Context};
-use aws_sdk_s3::types::StorageClass;
 use camino::{Utf8Path, Utf8PathBuf};
 
 use bytes::Bytes;
@@ -136,11 +134,6 @@ impl RemotePath {
     pub fn strip_prefix(&self, p: &RemotePath) -> Result<&Utf8Path, std::path::StripPrefixError> {
         self.0.strip_prefix(&p.0)
     }
-
-    pub fn add_trailing_slash(&self) -> Self {
-        // Unwrap safety inputs are guararnteed to be valid UTF-8
-        Self(format!("{}/", self.0).try_into().unwrap())
-    }
 }
 
 /// We don't need callers to be able to pass arbitrary delimiters: just control
@@ -164,21 +157,47 @@ pub struct Listing {
 /// providing basic CRUD operations for storage files.
 #[allow(async_fn_in_trait)]
 pub trait RemoteStorage: Send + Sync + 'static {
-    /// List objects in remote storage, with semantics matching AWS S3's ListObjectsV2.
-    /// (see `<https://docs.aws.amazon.com/AmazonS3/latest/API/API_ListObjectsV2.html>`)
-    ///
-    /// Note that the prefix is relative to any `prefix_in_bucket` configured for the client, not
-    /// from the absolute root of the bucket.
-    ///
-    /// `mode` configures whether to use a delimiter.  Without a delimiter all keys
-    /// within the prefix are listed in the `keys` of the result.  With a delimiter, any "directories" at the top level of
-    /// the prefix are returned in the `prefixes` of the result, and keys in the top level of the prefix are
-    /// returned in `keys` ().
-    ///
-    /// `max_keys` controls the maximum number of keys that will be returned.  If this is None, this function
-    /// will iteratively call listobjects until it runs out of keys.  Note that this is not safe to use on
-    /// unlimted size buckets, as the full list of objects is allocated into a monolithic data structure.
+    /// 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)
+    /// so this method doesnt need to.
+    async fn list_prefixes(
+        &self,
+        prefix: Option<&RemotePath>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        let result = self
+            .list(prefix, ListingMode::WithDelimiter, None, cancel)
+            .await?
+            .prefixes;
+        Ok(result)
+    }
+    /// Lists all files in directory "recursively"
+    /// (not really recursively, because AWS has a flat namespace)
+    /// Note: This is subtely different than list_prefixes,
+    /// because it is for listing files instead of listing
+    /// names sharing common prefixes.
+    /// For example,
+    /// list_files("foo/bar") = ["foo/bar/cat123.txt",
+    /// "foo/bar/cat567.txt", "foo/bar/dog123.txt", "foo/bar/dog456.txt"]
+    /// whereas,
+    /// list_prefixes("foo/bar/") = ["cat", "dog"]
+    /// See `test_real_s3.rs` for more details.
     ///
+    /// max_keys limits max number of keys returned; None means unlimited.
+    async fn list_files(
+        &self,
+        prefix: Option<&RemotePath>,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        let result = self
+            .list(prefix, ListingMode::NoDelimiter, max_keys, cancel)
+            .await?
+            .keys;
+        Ok(result)
+    }
+
     async fn list(
         &self,
         prefix: Option<&RemotePath>,
@@ -317,6 +336,41 @@ impl<Other: RemoteStorage> GenericRemoteStorage<Arc<Other>> {
         }
     }
 
+    // A function for listing all the files in a "directory"
+    // Example:
+    // list_files("foo/bar") = ["foo/bar/a.txt", "foo/bar/b.txt"]
+    //
+    // max_keys limits max number of keys returned; None means unlimited.
+    pub async fn list_files(
+        &self,
+        folder: Option<&RemotePath>,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        match self {
+            Self::LocalFs(s) => s.list_files(folder, max_keys, cancel).await,
+            Self::AwsS3(s) => s.list_files(folder, max_keys, cancel).await,
+            Self::AzureBlob(s) => s.list_files(folder, max_keys, cancel).await,
+            Self::Unreliable(s) => s.list_files(folder, max_keys, cancel).await,
+        }
+    }
+
+    // lists common *prefixes*, if any of files
+    // Example:
+    // list_prefixes("foo123","foo567","bar123","bar432") = ["foo", "bar"]
+    pub async fn list_prefixes(
+        &self,
+        prefix: Option<&RemotePath>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        match self {
+            Self::LocalFs(s) => s.list_prefixes(prefix, cancel).await,
+            Self::AwsS3(s) => s.list_prefixes(prefix, cancel).await,
+            Self::AzureBlob(s) => s.list_prefixes(prefix, cancel).await,
+            Self::Unreliable(s) => s.list_prefixes(prefix, cancel).await,
+        }
+    }
+
     /// See [`RemoteStorage::upload`]
     pub async fn upload(
         &self,
@@ -511,16 +565,6 @@ impl GenericRemoteStorage {
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct StorageMetadata(HashMap<String, String>);
 
-impl<const N: usize> From<[(&str, &str); N]> for StorageMetadata {
-    fn from(arr: [(&str, &str); N]) -> Self {
-        let map: HashMap<String, String> = arr
-            .iter()
-            .map(|(k, v)| (k.to_string(), v.to_string()))
-            .collect();
-        Self(map)
-    }
-}
-
 /// External backup storage configuration, enough for creating a client for that storage.
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct RemoteStorageConfig {
@@ -565,7 +609,6 @@ pub struct S3Config {
     /// See [`DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT`] for more details.
     pub concurrency_limit: NonZeroUsize,
     pub max_keys_per_list_response: Option<i32>,
-    pub upload_storage_class: Option<StorageClass>,
 }
 
 impl Debug for S3Config {
@@ -694,18 +737,6 @@ impl RemoteStorageConfig {
                     endpoint,
                     concurrency_limit,
                     max_keys_per_list_response,
-                    upload_storage_class: toml
-                        .get("upload_storage_class")
-                        .map(|prefix_in_bucket| -> anyhow::Result<_> {
-                            let s = parse_toml_string("upload_storage_class", prefix_in_bucket)?;
-                            let storage_class = StorageClass::from_str(&s).expect("infallible");
-                            #[allow(deprecated)]
-                            if matches!(storage_class, StorageClass::Unknown(_)) {
-                                bail!("Specified storage class unknown to SDK: '{s}'. Allowed values: {:?}", StorageClass::values());
-                            }
-                            Ok(storage_class)
-                        })
-                        .transpose()?,
                 })
             }
             (_, _, _, Some(_), None) => {

libs/remote_storage/src/local_fs.rs

@@ -5,9 +5,11 @@
 //! volume is mounted to the local FS.
 
 use std::{
-    collections::HashSet,
+    borrow::Cow,
+    future::Future,
     io::ErrorKind,
     num::NonZeroU32,
+    pin::Pin,
     time::{Duration, SystemTime, UNIX_EPOCH},
 };
 
@@ -20,11 +22,11 @@ use tokio::{
     io::{self, AsyncReadExt, AsyncSeekExt, AsyncWriteExt},
 };
 use tokio_util::{io::ReaderStream, sync::CancellationToken};
-use utils::crashsafe::path_with_suffix_extension;
+use tracing::*;
+use utils::{crashsafe::path_with_suffix_extension, fs_ext::is_directory_empty};
 
 use crate::{
     Download, DownloadError, Listing, ListingMode, RemotePath, TimeTravelError, TimeoutOrCancel,
-    REMOTE_STORAGE_PREFIX_SEPARATOR,
 };
 
 use super::{RemoteStorage, StorageMetadata};
@@ -91,47 +93,7 @@ impl LocalFs {
 
     #[cfg(test)]
     async fn list_all(&self) -> anyhow::Result<Vec<RemotePath>> {
-        use std::{future::Future, pin::Pin};
-        fn get_all_files<'a, P>(
-            directory_path: P,
-        ) -> Pin<Box<dyn Future<Output = anyhow::Result<Vec<Utf8PathBuf>>> + Send + Sync + 'a>>
-        where
-            P: AsRef<Utf8Path> + Send + Sync + 'a,
-        {
-            Box::pin(async move {
-                let directory_path = directory_path.as_ref();
-                if directory_path.exists() {
-                    if directory_path.is_dir() {
-                        let mut paths = Vec::new();
-                        let mut dir_contents = fs::read_dir(directory_path).await?;
-                        while let Some(dir_entry) = dir_contents.next_entry().await? {
-                            let file_type = dir_entry.file_type().await?;
-                            let entry_path =
-                                Utf8PathBuf::from_path_buf(dir_entry.path()).map_err(|pb| {
-                                    anyhow::Error::msg(format!(
-                                        "non-Unicode path: {}",
-                                        pb.to_string_lossy()
-                                    ))
-                                })?;
-                            if file_type.is_symlink() {
-                                tracing::debug!("{entry_path:?} is a symlink, skipping")
-                            } else if file_type.is_dir() {
-                                paths.extend(get_all_files(&entry_path).await?.into_iter())
-                            } else {
-                                paths.push(entry_path);
-                            }
-                        }
-                        Ok(paths)
-                    } else {
-                        bail!("Path {directory_path:?} is not a directory")
-                    }
-                } else {
-                    Ok(Vec::new())
-                }
-            })
-        }
-
-        Ok(get_all_files(&self.storage_root)
+        Ok(get_all_files(&self.storage_root, true)
             .await?
             .into_iter()
             .map(|path| {
@@ -158,14 +120,6 @@ impl LocalFs {
         // S3 object list prefixes can be arbitrary strings, but when reading
         // the local filesystem we need a directory to start calling read_dir on.
         let mut initial_dir = full_path.clone();
-
-        // If there's no trailing slash, we have to start looking from one above: even if
-        // `initial_dir` is a directory, we should still list any prefixes in the parent
-        // that start with the same string.
-        if !full_path.to_string().ends_with('/') {
-            initial_dir.pop();
-        }
-
         loop {
             // Did we make it to the root?
             if initial_dir.parent().is_none() {
@@ -341,66 +295,61 @@ impl RemoteStorage for LocalFs {
         let op = async {
             let mut result = Listing::default();
 
-            // Filter out directories: in S3 directories don't exist, only the keys within them do.
-            let keys = self
-                .list_recursive(prefix)
+            if let ListingMode::NoDelimiter = mode {
+                let keys = self
+                    .list_recursive(prefix)
+                    .await
+                    .map_err(DownloadError::Other)?;
+
+                result.keys = keys
+                    .into_iter()
+                    .filter(|k| {
+                        let path = k.with_base(&self.storage_root);
+                        !path.is_dir()
+                    })
+                    .collect();
+
+                if let Some(max_keys) = max_keys {
+                    result.keys.truncate(max_keys.get() as usize);
+                }
+
+                return Ok(result);
+            }
+
+            let path = match prefix {
+                Some(prefix) => Cow::Owned(prefix.with_base(&self.storage_root)),
+                None => Cow::Borrowed(&self.storage_root),
+            };
+
+            let prefixes_to_filter = get_all_files(path.as_ref(), false)
                 .await
                 .map_err(DownloadError::Other)?;
-            let keys = keys
-                .into_iter()
-                .filter(|k| {
-                    let path = k.with_base(&self.storage_root);
-                    !path.is_dir()
-                })
-                .collect();
 
-            if let ListingMode::NoDelimiter = mode {
-                result.keys = keys;
-            } else {
-                let mut prefixes = HashSet::new();
-                for key in keys {
-                    // If the part after the prefix includes a "/", take only the first part and put it in `prefixes`.
-                    let relative_key = if let Some(prefix) = prefix {
-                        let mut prefix = prefix.clone();
-                        // We only strip the dirname of the prefix, so that when we strip it from the start of keys we
-                        // end up with full file/dir names.
-                        let prefix_full_local_path = prefix.with_base(&self.storage_root);
-                        let has_slash = prefix.0.to_string().ends_with('/');
-                        let strip_prefix = if prefix_full_local_path.is_dir() && has_slash {
-                            prefix
-                        } else {
-                            prefix.0.pop();
-                            prefix
-                        };
-
-                        RemotePath::new(key.strip_prefix(&strip_prefix).unwrap()).unwrap()
-                    } else {
-                        key
-                    };
-
-                    let relative_key = format!("{}", relative_key);
-                    if relative_key.contains(REMOTE_STORAGE_PREFIX_SEPARATOR) {
-                        let first_part = relative_key
-                            .split(REMOTE_STORAGE_PREFIX_SEPARATOR)
-                            .next()
-                            .unwrap()
-                            .to_owned();
-                        prefixes.insert(first_part);
-                    } else {
-                        result
-                            .keys
-                            .push(RemotePath::from_string(&relative_key).unwrap());
-                    }
+            // filter out empty directories to mirror s3 behavior.
+            for prefix in prefixes_to_filter {
+                if prefix.is_dir()
+                    && is_directory_empty(&prefix)
+                        .await
+                        .map_err(DownloadError::Other)?
+                {
+                    continue;
+                }
+
+                let stripped = prefix
+                    .strip_prefix(&self.storage_root)
+                    .context("Failed to strip prefix")
+                    .and_then(RemotePath::new)
+                    .expect(
+                        "We list files for storage root, hence should be able to remote the prefix",
+                    );
+
+                if prefix.is_dir() {
+                    result.prefixes.push(stripped);
+                } else {
+                    result.keys.push(stripped);
                 }
-                result.prefixes = prefixes
-                    .into_iter()
-                    .map(|s| RemotePath::from_string(&s).unwrap())
-                    .collect();
             }
 
-            if let Some(max_keys) = max_keys {
-                result.keys.truncate(max_keys.get() as usize);
-            }
             Ok(result)
         };
 
@@ -611,6 +560,50 @@ fn storage_metadata_path(original_path: &Utf8Path) -> Utf8PathBuf {
     path_with_suffix_extension(original_path, "metadata")
 }
 
+fn get_all_files<'a, P>(
+    directory_path: P,
+    recursive: bool,
+) -> Pin<Box<dyn Future<Output = anyhow::Result<Vec<Utf8PathBuf>>> + Send + Sync + 'a>>
+where
+    P: AsRef<Utf8Path> + Send + Sync + 'a,
+{
+    Box::pin(async move {
+        let directory_path = directory_path.as_ref();
+        if directory_path.exists() {
+            if directory_path.is_dir() {
+                let mut paths = Vec::new();
+                let mut dir_contents = fs::read_dir(directory_path).await?;
+                while let Some(dir_entry) = dir_contents.next_entry().await? {
+                    let file_type = dir_entry.file_type().await?;
+                    let entry_path =
+                        Utf8PathBuf::from_path_buf(dir_entry.path()).map_err(|pb| {
+                            anyhow::Error::msg(format!(
+                                "non-Unicode path: {}",
+                                pb.to_string_lossy()
+                            ))
+                        })?;
+                    if file_type.is_symlink() {
+                        debug!("{entry_path:?} is a symlink, skipping")
+                    } else if file_type.is_dir() {
+                        if recursive {
+                            paths.extend(get_all_files(&entry_path, true).await?.into_iter())
+                        } else {
+                            paths.push(entry_path)
+                        }
+                    } else {
+                        paths.push(entry_path);
+                    }
+                }
+                Ok(paths)
+            } else {
+                bail!("Path {directory_path:?} is not a directory")
+            }
+        } else {
+            Ok(Vec::new())
+        }
+    })
+}
+
 async fn create_target_directory(target_file_path: &Utf8Path) -> anyhow::Result<()> {
     let target_dir = match target_file_path.parent() {
         Some(parent_dir) => parent_dir,
@@ -930,18 +923,13 @@ mod fs_tests {
         // No delimiter: should recursively list everything
         let (storage, cancel) = create_storage()?;
         let child = upload_dummy_file(&storage, "grandparent/parent/child", None, &cancel).await?;
-        let child_sibling =
-            upload_dummy_file(&storage, "grandparent/parent/child_sibling", None, &cancel).await?;
         let uncle = upload_dummy_file(&storage, "grandparent/uncle", None, &cancel).await?;
 
         let listing = storage
             .list(None, ListingMode::NoDelimiter, None, &cancel)
             .await?;
         assert!(listing.prefixes.is_empty());
-        assert_eq!(
-            listing.keys.into_iter().collect::<HashSet<_>>(),
-            HashSet::from([uncle.clone(), child.clone(), child_sibling.clone()])
-        );
+        assert_eq!(listing.keys, [uncle.clone(), child.clone()].to_vec());
 
         // Delimiter: should only go one deep
         let listing = storage
@@ -954,25 +942,7 @@ mod fs_tests {
         );
         assert!(listing.keys.is_empty());
 
-        // Delimiter & prefix with a trailing slash
-        let listing = storage
-            .list(
-                Some(&RemotePath::from_string("timelines/some_timeline/grandparent/").unwrap()),
-                ListingMode::WithDelimiter,
-                None,
-                &cancel,
-            )
-            .await?;
-        assert_eq!(
-            listing.keys,
-            [RemotePath::from_string("uncle").unwrap()].to_vec()
-        );
-        assert_eq!(
-            listing.prefixes,
-            [RemotePath::from_string("parent").unwrap()].to_vec()
-        );
-
-        // Delimiter and prefix without a trailing slash
+        // Delimiter & prefix
         let listing = storage
             .list(
                 Some(&RemotePath::from_string("timelines/some_timeline/grandparent").unwrap()),
@@ -981,66 +951,12 @@ mod fs_tests {
                 &cancel,
             )
             .await?;
-        assert_eq!(listing.keys, [].to_vec());
         assert_eq!(
             listing.prefixes,
-            [RemotePath::from_string("grandparent").unwrap()].to_vec()
-        );
-
-        // Delimiter and prefix that's partway through a path component
-        let listing = storage
-            .list(
-                Some(&RemotePath::from_string("timelines/some_timeline/grandp").unwrap()),
-                ListingMode::WithDelimiter,
-                None,
-                &cancel,
-            )
-            .await?;
-        assert_eq!(listing.keys, [].to_vec());
-        assert_eq!(
-            listing.prefixes,
-            [RemotePath::from_string("grandparent").unwrap()].to_vec()
-        );
-
-        Ok(())
-    }
-
-    #[tokio::test]
-    async fn list_part_component() -> anyhow::Result<()> {
-        // No delimiter: should recursively list everything
-        let (storage, cancel) = create_storage()?;
-
-        // Imitates what happens in a tenant path when we have an unsharded path and a sharded path, and do a listing
-        // of the unsharded path: although there is a "directory" at the unsharded path, it should be handled as
-        // a freeform prefix.
-        let _child_a =
-            upload_dummy_file(&storage, "grandparent/tenant-01/child", None, &cancel).await?;
-        let _child_b =
-            upload_dummy_file(&storage, "grandparent/tenant/child", None, &cancel).await?;
-
-        // Delimiter and prefix that's partway through a path component
-        let listing = storage
-            .list(
-                Some(
-                    &RemotePath::from_string("timelines/some_timeline/grandparent/tenant").unwrap(),
-                ),
-                ListingMode::WithDelimiter,
-                None,
-                &cancel,
-            )
-            .await?;
-        assert_eq!(listing.keys, [].to_vec());
-
-        let mut found_prefixes = listing.prefixes.clone();
-        found_prefixes.sort();
-        assert_eq!(
-            found_prefixes,
-            [
-                RemotePath::from_string("tenant").unwrap(),
-                RemotePath::from_string("tenant-01").unwrap(),
-            ]
-            .to_vec()
+            [RemotePath::from_string("timelines/some_timeline/grandparent/parent").unwrap()]
+                .to_vec()
         );
+        assert_eq!(listing.keys, [uncle.clone()].to_vec());
 
         Ok(())
     }

libs/remote_storage/src/s3_bucket.rs

@@ -30,7 +30,7 @@ use aws_sdk_s3::{
     config::{AsyncSleep, Builder, IdentityCache, Region, SharedAsyncSleep},
     error::SdkError,
     operation::get_object::GetObjectError,
-    types::{Delete, DeleteMarkerEntry, ObjectIdentifier, ObjectVersion, StorageClass},
+    types::{Delete, DeleteMarkerEntry, ObjectIdentifier, ObjectVersion},
     Client,
 };
 use aws_smithy_async::rt::sleep::TokioSleep;
@@ -62,7 +62,6 @@ pub struct S3Bucket {
     bucket_name: String,
     prefix_in_bucket: Option<String>,
     max_keys_per_list_response: Option<i32>,
-    upload_storage_class: Option<StorageClass>,
     concurrency_limiter: ConcurrencyLimiter,
     // Per-request timeout. Accessible for tests.
     pub timeout: Duration,
@@ -155,7 +154,6 @@ impl S3Bucket {
             max_keys_per_list_response: aws_config.max_keys_per_list_response,
             prefix_in_bucket,
             concurrency_limiter: ConcurrencyLimiter::new(aws_config.concurrency_limit.get()),
-            upload_storage_class: aws_config.upload_storage_class.clone(),
             timeout,
         })
     }
@@ -180,7 +178,10 @@ impl S3Bucket {
 
     pub fn relative_path_to_s3_object(&self, path: &RemotePath) -> String {
         assert_eq!(std::path::MAIN_SEPARATOR, REMOTE_STORAGE_PREFIX_SEPARATOR);
-        let path_string = path.get_path().as_str();
+        let path_string = path
+            .get_path()
+            .as_str()
+            .trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR);
         match &self.prefix_in_bucket {
             Some(prefix) => prefix.clone() + "/" + path_string,
             None => path_string.to_string(),
@@ -470,11 +471,16 @@ impl RemoteStorage for S3Bucket {
         // get the passed prefix or if it is not set use prefix_in_bucket value
         let list_prefix = prefix
             .map(|p| self.relative_path_to_s3_object(p))
-            .or_else(|| {
-                self.prefix_in_bucket.clone().map(|mut s| {
-                    s.push(REMOTE_STORAGE_PREFIX_SEPARATOR);
-                    s
-                })
+            .or_else(|| self.prefix_in_bucket.clone())
+            .map(|mut p| {
+                // required to end with a separator
+                // otherwise request will return only the entry of a prefix
+                if matches!(mode, ListingMode::WithDelimiter)
+                    && !p.ends_with(REMOTE_STORAGE_PREFIX_SEPARATOR)
+                {
+                    p.push(REMOTE_STORAGE_PREFIX_SEPARATOR);
+                }
+                p
             });
 
         let _permit = self.permit(kind, cancel).await?;
@@ -543,15 +549,11 @@ impl RemoteStorage for S3Bucket {
                 }
             }
 
-            // S3 gives us prefixes like "foo/", we return them like "foo"
-            result.prefixes.extend(prefixes.iter().filter_map(|o| {
-                Some(
-                    self.s3_object_to_relative_path(
-                        o.prefix()?
-                            .trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR),
-                    ),
-                )
-            }));
+            result.prefixes.extend(
+                prefixes
+                    .iter()
+                    .filter_map(|o| Some(self.s3_object_to_relative_path(o.prefix()?))),
+            );
 
             continuation_token = match response.next_continuation_token {
                 Some(new_token) => Some(new_token),
@@ -584,7 +586,6 @@ impl RemoteStorage for S3Bucket {
             .bucket(self.bucket_name.clone())
             .key(self.relative_path_to_s3_object(to))
             .set_metadata(metadata.map(|m| m.0))
-            .set_storage_class(self.upload_storage_class.clone())
             .content_length(from_size_bytes.try_into()?)
             .body(bytes_stream)
             .send();
@@ -636,7 +637,6 @@ impl RemoteStorage for S3Bucket {
             .copy_object()
             .bucket(self.bucket_name.clone())
             .key(self.relative_path_to_s3_object(to))
-            .set_storage_class(self.upload_storage_class.clone())
             .copy_source(copy_source)
             .send();
 
@@ -894,7 +894,6 @@ impl RemoteStorage for S3Bucket {
                                     .copy_object()
                                     .bucket(self.bucket_name.clone())
                                     .key(key)
-                                    .set_storage_class(self.upload_storage_class.clone())
                                     .copy_source(&source_id)
                                     .send();
 
@@ -1051,22 +1050,22 @@ mod tests {
             Some("/test/prefix/"),
         ];
         let expected_outputs = [
-            vec!["", "some/path", "some/path/"],
-            vec!["/", "/some/path", "/some/path/"],
+            vec!["", "some/path", "some/path"],
+            vec!["/", "/some/path", "/some/path"],
             vec![
                 "test/prefix/",
                 "test/prefix/some/path",
-                "test/prefix/some/path/",
+                "test/prefix/some/path",
             ],
             vec![
                 "test/prefix/",
                 "test/prefix/some/path",
-                "test/prefix/some/path/",
+                "test/prefix/some/path",
             ],
             vec![
                 "test/prefix/",
                 "test/prefix/some/path",
-                "test/prefix/some/path/",
+                "test/prefix/some/path",
             ],
         ];
 
@@ -1078,7 +1077,6 @@ mod tests {
                 endpoint: None,
                 concurrency_limit: NonZeroUsize::new(100).unwrap(),
                 max_keys_per_list_response: Some(5),
-                upload_storage_class: None,
             };
             let storage =
                 S3Bucket::new(&config, std::time::Duration::ZERO).expect("remote storage init");

libs/remote_storage/src/simulate_failures.rs

@@ -107,6 +107,27 @@ impl UnreliableWrapper {
 type VoidStorage = crate::LocalFs;
 
 impl RemoteStorage for UnreliableWrapper {
+    async fn list_prefixes(
+        &self,
+        prefix: Option<&RemotePath>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        self.attempt(RemoteOp::ListPrefixes(prefix.cloned()))
+            .map_err(DownloadError::Other)?;
+        self.inner.list_prefixes(prefix, cancel).await
+    }
+
+    async fn list_files(
+        &self,
+        folder: Option<&RemotePath>,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> Result<Vec<RemotePath>, DownloadError> {
+        self.attempt(RemoteOp::ListPrefixes(folder.cloned()))
+            .map_err(DownloadError::Other)?;
+        self.inner.list_files(folder, max_keys, cancel).await
+    }
+
     async fn list(
         &self,
         prefix: Option<&RemotePath>,

libs/remote_storage/tests/common/tests.rs

@@ -1,6 +1,5 @@
 use anyhow::Context;
 use camino::Utf8Path;
-use remote_storage::ListingMode;
 use remote_storage::RemotePath;
 use std::sync::Arc;
 use std::{collections::HashSet, num::NonZeroU32};
@@ -55,9 +54,9 @@ async fn pagination_should_work(ctx: &mut MaybeEnabledStorageWithTestBlobs) -> a
     let base_prefix = RemotePath::new(Utf8Path::new(ctx.enabled.base_prefix))
         .context("common_prefix construction")?;
     let root_remote_prefixes = test_client
-        .list(None, ListingMode::WithDelimiter, None, &cancel)
-        .await?
-        .prefixes
+        .list_prefixes(None, &cancel)
+        .await
+        .context("client list root prefixes failure")?
         .into_iter()
         .collect::<HashSet<_>>();
     assert_eq!(
@@ -66,14 +65,9 @@ async fn pagination_should_work(ctx: &mut MaybeEnabledStorageWithTestBlobs) -> a
     );
 
     let nested_remote_prefixes = test_client
-        .list(
-            Some(&base_prefix.add_trailing_slash()),
-            ListingMode::WithDelimiter,
-            None,
-            &cancel,
-        )
-        .await?
-        .prefixes
+        .list_prefixes(Some(&base_prefix), &cancel)
+        .await
+        .context("client list nested prefixes failure")?
         .into_iter()
         .collect::<HashSet<_>>();
     let remote_only_prefixes = nested_remote_prefixes
@@ -96,13 +90,11 @@ async fn pagination_should_work(ctx: &mut MaybeEnabledStorageWithTestBlobs) -> a
 ///
 /// First, create a set of S3 objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_remote_data`]
 /// Then performs the following queries:
-///    1. `list(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
-///    2. `list("folder1")`.  This  should return all files `random_prefix/folder1/blob_{i}.txt`
+///    1. `list_files(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
+///    2. `list_files("folder1")`.  This  should return all files `random_prefix/folder1/blob_{i}.txt`
 #[test_context(MaybeEnabledStorageWithSimpleTestBlobs)]
 #[tokio::test]
-async fn list_no_delimiter_works(
-    ctx: &mut MaybeEnabledStorageWithSimpleTestBlobs,
-) -> anyhow::Result<()> {
+async fn list_files_works(ctx: &mut MaybeEnabledStorageWithSimpleTestBlobs) -> anyhow::Result<()> {
     let ctx = match ctx {
         MaybeEnabledStorageWithSimpleTestBlobs::Enabled(ctx) => ctx,
         MaybeEnabledStorageWithSimpleTestBlobs::Disabled => return Ok(()),
@@ -115,36 +107,29 @@ async fn list_no_delimiter_works(
     let base_prefix =
         RemotePath::new(Utf8Path::new("folder1")).context("common_prefix construction")?;
     let root_files = test_client
-        .list(None, ListingMode::NoDelimiter, None, &cancel)
+        .list_files(None, None, &cancel)
         .await
         .context("client list root files failure")?
-        .keys
         .into_iter()
         .collect::<HashSet<_>>();
     assert_eq!(
         root_files,
         ctx.remote_blobs.clone(),
-        "remote storage list on root mismatches with the uploads."
+        "remote storage list_files on root mismatches with the uploads."
     );
 
     // Test that max_keys limit works. In total there are about 21 files (see
     // upload_simple_remote_data call in test_real_s3.rs).
     let limited_root_files = test_client
-        .list(
-            None,
-            ListingMode::NoDelimiter,
-            Some(NonZeroU32::new(2).unwrap()),
-            &cancel,
-        )
+        .list_files(None, Some(NonZeroU32::new(2).unwrap()), &cancel)
         .await
         .context("client list root files failure")?;
-    assert_eq!(limited_root_files.keys.len(), 2);
+    assert_eq!(limited_root_files.len(), 2);
 
     let nested_remote_files = test_client
-        .list(Some(&base_prefix), ListingMode::NoDelimiter, None, &cancel)
+        .list_files(Some(&base_prefix), None, &cancel)
         .await
         .context("client list nested files failure")?
-        .keys
         .into_iter()
         .collect::<HashSet<_>>();
     let trim_remote_blobs: HashSet<_> = ctx
@@ -156,7 +141,7 @@ async fn list_no_delimiter_works(
         .collect();
     assert_eq!(
         nested_remote_files, trim_remote_blobs,
-        "remote storage list on subdirrectory mismatches with the uploads."
+        "remote storage list_files on subdirrectory mismatches with the uploads."
     );
     Ok(())
 }
@@ -214,11 +199,7 @@ async fn delete_objects_works(ctx: &mut MaybeEnabledStorage) -> anyhow::Result<(
 
     ctx.client.delete_objects(&[path1, path2], &cancel).await?;
 
-    let prefixes = ctx
-        .client
-        .list(None, ListingMode::WithDelimiter, None, &cancel)
-        .await?
-        .prefixes;
+    let prefixes = ctx.client.list_prefixes(None, &cancel).await?;
 
     assert_eq!(prefixes.len(), 1);
 

libs/remote_storage/tests/test_real_azure.rs

@@ -57,6 +57,7 @@ enum MaybeEnabledStorage {
     Disabled,
 }
 
+#[async_trait::async_trait]
 impl AsyncTestContext for MaybeEnabledStorage {
     async fn setup() -> Self {
         ensure_logging_ready();
@@ -85,6 +86,7 @@ struct AzureWithTestBlobs {
     remote_blobs: HashSet<RemotePath>,
 }
 
+#[async_trait::async_trait]
 impl AsyncTestContext for MaybeEnabledStorageWithTestBlobs {
     async fn setup() -> Self {
         ensure_logging_ready();
@@ -132,6 +134,10 @@ impl AsyncTestContext for MaybeEnabledStorageWithTestBlobs {
     }
 }
 
+// NOTE: the setups for the list_prefixes test and the list_files test are very similar
+// However, they are not idential. The list_prefixes function is concerned with listing prefixes,
+// whereas the list_files function is concerned with listing files.
+// See `RemoteStorage::list_files` documentation for more details
 enum MaybeEnabledStorageWithSimpleTestBlobs {
     Enabled(AzureWithSimpleTestBlobs),
     Disabled,
@@ -142,6 +148,7 @@ struct AzureWithSimpleTestBlobs {
     remote_blobs: HashSet<RemotePath>,
 }
 
+#[async_trait::async_trait]
 impl AsyncTestContext for MaybeEnabledStorageWithSimpleTestBlobs {
     async fn setup() -> Self {
         ensure_logging_ready();

libs/remote_storage/tests/test_real_s3.rs

@@ -12,8 +12,8 @@ use anyhow::Context;
 use camino::Utf8Path;
 use futures_util::StreamExt;
 use remote_storage::{
-    DownloadError, GenericRemoteStorage, ListingMode, RemotePath, RemoteStorageConfig,
-    RemoteStorageKind, S3Config,
+    DownloadError, GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind,
+    S3Config,
 };
 use test_context::test_context;
 use test_context::AsyncTestContext;
@@ -75,14 +75,11 @@ async fn s3_time_travel_recovery_works(ctx: &mut MaybeEnabledStorage) -> anyhow:
         client: &Arc<GenericRemoteStorage>,
         cancel: &CancellationToken,
     ) -> anyhow::Result<HashSet<RemotePath>> {
-        Ok(
-            retry(|| client.list(None, ListingMode::NoDelimiter, None, cancel))
-                .await
-                .context("list root files failure")?
-                .keys
-                .into_iter()
-                .collect::<HashSet<_>>(),
-        )
+        Ok(retry(|| client.list_files(None, None, cancel))
+            .await
+            .context("list root files failure")?
+            .into_iter()
+            .collect::<HashSet<_>>())
     }
 
     let cancel = CancellationToken::new();
@@ -222,6 +219,7 @@ enum MaybeEnabledStorage {
     Disabled,
 }
 
+#[async_trait::async_trait]
 impl AsyncTestContext for MaybeEnabledStorage {
     async fn setup() -> Self {
         ensure_logging_ready();
@@ -250,6 +248,7 @@ struct S3WithTestBlobs {
     remote_blobs: HashSet<RemotePath>,
 }
 
+#[async_trait::async_trait]
 impl AsyncTestContext for MaybeEnabledStorageWithTestBlobs {
     async fn setup() -> Self {
         ensure_logging_ready();
@@ -297,6 +296,10 @@ impl AsyncTestContext for MaybeEnabledStorageWithTestBlobs {
     }
 }
 
+// NOTE: the setups for the list_prefixes test and the list_files test are very similar
+// However, they are not idential. The list_prefixes function is concerned with listing prefixes,
+// whereas the list_files function is concerned with listing files.
+// See `RemoteStorage::list_files` documentation for more details
 enum MaybeEnabledStorageWithSimpleTestBlobs {
     Enabled(S3WithSimpleTestBlobs),
     Disabled,
@@ -307,6 +310,7 @@ struct S3WithSimpleTestBlobs {
     remote_blobs: HashSet<RemotePath>,
 }
 
+#[async_trait::async_trait]
 impl AsyncTestContext for MaybeEnabledStorageWithSimpleTestBlobs {
     async fn setup() -> Self {
         ensure_logging_ready();
@@ -380,7 +384,6 @@ fn create_s3_client(
             endpoint: None,
             concurrency_limit: NonZeroUsize::new(100).unwrap(),
             max_keys_per_list_response,
-            upload_storage_class: None,
         }),
         timeout: RemoteStorageConfig::DEFAULT_TIMEOUT,
     };

libs/tenant_size_model/tests/tests.rs

@@ -247,7 +247,7 @@ fn scenario_4() {
     //
     // This is in total 5000 + 1000 + 5000 + 1000 = 12000
     //
-    // (If we used the method from the previous scenario, and
+    // (If we used the the method from the previous scenario, and
     // kept only snapshot at the branch point, we'd need to keep
     // all the WAL between 10000-18000 on the main branch, so
     // the total size would be 5000 + 1000 + 8000 = 14000. The

libs/utils/Cargo.toml

@@ -22,7 +22,6 @@ camino.workspace = true
 chrono.workspace = true
 heapless.workspace = true
 hex = { workspace = true, features = ["serde"] }
-humantime.workspace = true
 hyper = { workspace = true, features = ["full"] }
 fail.workspace = true
 futures = { workspace = true}

libs/utils/src/env.rs

@@ -1,21 +0,0 @@
-//! Wrapper around `std::env::var` for parsing environment variables.
-
-use std::{fmt::Display, str::FromStr};
-
-pub fn var<V, E>(varname: &str) -> Option<V>
-where
-    V: FromStr<Err = E>,
-    E: Display,
-{
-    match std::env::var(varname) {
-        Ok(s) => Some(
-            s.parse()
-                .map_err(|e| format!("failed to parse env var {varname}: {e:#}"))
-                .unwrap(),
-        ),
-        Err(std::env::VarError::NotPresent) => None,
-        Err(std::env::VarError::NotUnicode(_)) => {
-            panic!("env var {varname} is not unicode")
-        }
-    }
-}

libs/utils/src/lib.rs

@@ -63,7 +63,6 @@ pub mod measured_stream;
 
 pub mod serde_percent;
 pub mod serde_regex;
-pub mod serde_system_time;
 
 pub mod pageserver_feedback;
 
@@ -90,10 +89,6 @@ pub mod yielding_loop;
 
 pub mod zstd;
 
-pub mod env;
-
-pub mod poison;
-
 /// This is a shortcut to embed git sha into binaries and avoid copying the same build script to all packages
 ///
 /// we have several cases:

libs/utils/src/poison.rs

@@ -1,121 +0,0 @@
-//!  Protect a piece of state from reuse after it is left in an inconsistent state.
-//!
-//!  # Example
-//!
-//!  ```
-//!  # tokio_test::block_on(async {
-//!  use utils::poison::Poison;
-//!  use std::time::Duration;
-//!
-//!  struct State {
-//!    clean: bool,
-//!  }
-//!  let state = tokio::sync::Mutex::new(Poison::new("mystate", State { clean: true }));
-//!
-//!  let mut mutex_guard = state.lock().await;
-//!  let mut poison_guard = mutex_guard.check_and_arm()?;
-//!  let state = poison_guard.data_mut();
-//!  state.clean = false;
-//!  // If we get cancelled at this await point, subsequent check_and_arm() calls will fail.
-//!  tokio::time::sleep(Duration::from_secs(10)).await;
-//!  state.clean = true;
-//!  poison_guard.disarm();
-//!  # Ok::<(), utils::poison::Error>(())
-//!  # });
-//!  ```
-
-use tracing::warn;
-
-pub struct Poison<T> {
-    what: &'static str,
-    state: State,
-    data: T,
-}
-
-#[derive(Clone, Copy)]
-enum State {
-    Clean,
-    Armed,
-    Poisoned { at: chrono::DateTime<chrono::Utc> },
-}
-
-impl<T> Poison<T> {
-    /// We log `what` `warning!` level if the [`Guard`] gets dropped without being [`Guard::disarm`]ed.
-    pub fn new(what: &'static str, data: T) -> Self {
-        Self {
-            what,
-            state: State::Clean,
-            data,
-        }
-    }
-
-    /// Check for poisoning and return a [`Guard`] that provides access to the wrapped state.
-    pub fn check_and_arm(&mut self) -> Result<Guard<T>, Error> {
-        match self.state {
-            State::Clean => {
-                self.state = State::Armed;
-                Ok(Guard(self))
-            }
-            State::Armed => unreachable!("transient state"),
-            State::Poisoned { at } => Err(Error::Poisoned {
-                what: self.what,
-                at,
-            }),
-        }
-    }
-}
-
-/// Use [`Self::data`] and [`Self::data_mut`] to access the wrapped state.
-/// Once modifications are done, use [`Self::disarm`].
-/// If [`Guard`] gets dropped instead of calling [`Self::disarm`], the state is poisoned
-/// and subsequent calls to [`Poison::check_and_arm`] will fail with an error.
-pub struct Guard<'a, T>(&'a mut Poison<T>);
-
-impl<'a, T> Guard<'a, T> {
-    pub fn data(&self) -> &T {
-        &self.0.data
-    }
-    pub fn data_mut(&mut self) -> &mut T {
-        &mut self.0.data
-    }
-
-    pub fn disarm(self) {
-        match self.0.state {
-            State::Clean => unreachable!("we set it to Armed in check_and_arm()"),
-            State::Armed => {
-                self.0.state = State::Clean;
-            }
-            State::Poisoned { at } => {
-                unreachable!("we fail check_and_arm() if it's in that state: {at}")
-            }
-        }
-    }
-}
-
-impl<'a, T> Drop for Guard<'a, T> {
-    fn drop(&mut self) {
-        match self.0.state {
-            State::Clean => {
-                // set by disarm()
-            }
-            State::Armed => {
-                // still armed => poison it
-                let at = chrono::Utc::now();
-                self.0.state = State::Poisoned { at };
-                warn!(at=?at, "poisoning {}", self.0.what);
-            }
-            State::Poisoned { at } => {
-                unreachable!("we fail check_and_arm() if it's in that state: {at}")
-            }
-        }
-    }
-}
-
-#[derive(thiserror::Error, Debug)]
-pub enum Error {
-    #[error("poisoned at {at}: {what}")]
-    Poisoned {
-        what: &'static str,
-        at: chrono::DateTime<chrono::Utc>,
-    },
-}

libs/utils/src/seqwait.rs

@@ -182,18 +182,6 @@ where
         }
     }
 
-    /// Check if [`Self::wait_for`] or [`Self::wait_for_timeout`] would wait if called with `num`.
-    pub fn would_wait_for(&self, num: V) -> Result<(), V> {
-        let internal = self.internal.lock().unwrap();
-        let cnt = internal.current.cnt_value();
-        drop(internal);
-        if cnt >= num {
-            Ok(())
-        } else {
-            Err(cnt)
-        }
-    }
-
     /// Register and return a channel that will be notified when a number arrives,
     /// or None, if it has already arrived.
     fn queue_for_wait(&self, num: V) -> Result<Option<Receiver<()>>, SeqWaitError> {

libs/utils/src/serde_system_time.rs

@@ -1,55 +0,0 @@
-//! A `serde::{Deserialize,Serialize}` type for SystemTime with RFC3339 format and millisecond precision.
-
-#[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord, serde::Serialize, serde::Deserialize)]
-#[serde(transparent)]
-pub struct SystemTime(
-    #[serde(
-        deserialize_with = "deser_rfc3339_millis",
-        serialize_with = "ser_rfc3339_millis"
-    )]
-    pub std::time::SystemTime,
-);
-
-fn ser_rfc3339_millis<S: serde::ser::Serializer>(
-    ts: &std::time::SystemTime,
-    serializer: S,
-) -> Result<S::Ok, S::Error> {
-    serializer.collect_str(&humantime::format_rfc3339_millis(*ts))
-}
-
-fn deser_rfc3339_millis<'de, D>(deserializer: D) -> Result<std::time::SystemTime, D::Error>
-where
-    D: serde::de::Deserializer<'de>,
-{
-    let s: String = serde::de::Deserialize::deserialize(deserializer)?;
-    humantime::parse_rfc3339(&s).map_err(serde::de::Error::custom)
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    /// Helper function to make a SystemTime have millisecond precision by truncating additional nanoseconds.
-    fn to_millisecond_precision(time: SystemTime) -> SystemTime {
-        match time.0.duration_since(std::time::SystemTime::UNIX_EPOCH) {
-            Ok(duration) => {
-                let total_millis = duration.as_secs() * 1_000 + u64::from(duration.subsec_millis());
-                SystemTime(
-                    std::time::SystemTime::UNIX_EPOCH
-                        + std::time::Duration::from_millis(total_millis),
-                )
-            }
-            Err(_) => time,
-        }
-    }
-
-    #[test]
-    fn test_serialize_deserialize() {
-        let input = SystemTime(std::time::SystemTime::now());
-        let expected_serialized = format!("\"{}\"", humantime::format_rfc3339_millis(input.0));
-        let serialized = serde_json::to_string(&input).unwrap();
-        assert_eq!(expected_serialized, serialized);
-        let deserialized: SystemTime = serde_json::from_str(&expected_serialized).unwrap();
-        assert_eq!(to_millisecond_precision(input), deserialized);
-    }
-}

libs/utils/src/sync/heavier_once_cell.rs

@@ -192,14 +192,6 @@ impl<T> OnceCell<T> {
         }
     }
 
-    /// Like [`Guard::take_and_deinit`], but will return `None` if this OnceCell was never
-    /// initialized.
-    pub fn take_and_deinit(&mut self) -> Option<(T, InitPermit)> {
-        let inner = self.inner.get_mut().unwrap();
-
-        inner.take_and_deinit()
-    }
-
     /// Return the number of [`Self::get_or_init`] calls waiting for initialization to complete.
     pub fn initializer_count(&self) -> usize {
         self.initializers.load(Ordering::Relaxed)
@@ -254,23 +246,15 @@ impl<'a, T> Guard<'a, T> {
     /// The permit will be on a semaphore part of the new internal value, and any following
     /// [`OnceCell::get_or_init`] will wait on it to complete.
     pub fn take_and_deinit(mut self) -> (T, InitPermit) {
-        self.0
-            .take_and_deinit()
-            .expect("guard is not created unless value has been initialized")
-    }
-}
-
-impl<T> Inner<T> {
-    pub fn take_and_deinit(&mut self) -> Option<(T, InitPermit)> {
-        let value = self.value.take()?;
-
         let mut swapped = Inner::default();
         let sem = swapped.init_semaphore.clone();
         // acquire and forget right away, moving the control over to InitPermit
         sem.try_acquire().expect("we just created this").forget();
-        let permit = InitPermit(sem);
-        std::mem::swap(self, &mut swapped);
-        Some((value, permit))
+        std::mem::swap(&mut *self.0, &mut swapped);
+        swapped
+            .value
+            .map(|v| (v, InitPermit(sem)))
+            .expect("guard is not created unless value has been initialized")
     }
 }
 
@@ -279,13 +263,6 @@ impl<T> Inner<T> {
 /// On drop, this type will return the permit.
 pub struct InitPermit(Arc<tokio::sync::Semaphore>);
 
-impl std::fmt::Debug for InitPermit {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        let ptr = Arc::as_ptr(&self.0) as *const ();
-        f.debug_tuple("InitPermit").field(&ptr).finish()
-    }
-}
-
 impl Drop for InitPermit {
     fn drop(&mut self) {
         assert_eq!(
@@ -582,22 +559,4 @@ mod tests {
 
         assert_eq!(*target.get().unwrap(), 11);
     }
-
-    #[tokio::test]
-    async fn take_and_deinit_on_mut() {
-        use std::convert::Infallible;
-
-        let mut target = OnceCell::<u32>::default();
-        assert!(target.take_and_deinit().is_none());
-
-        target
-            .get_or_init(|permit| async move { Ok::<_, Infallible>((42, permit)) })
-            .await
-            .unwrap();
-
-        let again = target.take_and_deinit();
-        assert!(matches!(again, Some((42, _))), "{again:?}");
-
-        assert!(target.take_and_deinit().is_none());
-    }
 }

libs/vm_monitor/src/runner.rs

@@ -69,7 +69,7 @@ pub struct Config {
     /// should be removed once we have a better solution there.
     sys_buffer_bytes: u64,
 
-    /// Minimum fraction of total system memory reserved *before* the cgroup threshold; in
+    /// Minimum fraction of total system memory reserved *before* the the cgroup threshold; in
     /// other words, providing a ceiling for the highest value of the threshold by enforcing that
     /// there's at least `cgroup_min_overhead_fraction` of the total memory remaining beyond the
     /// threshold.

pageserver/Cargo.toml

@@ -59,7 +59,6 @@ signal-hook.workspace = true
 smallvec = { workspace = true, features = ["write"] }
 svg_fmt.workspace = true
 sync_wrapper.workspace = true
-sysinfo.workspace = true
 tokio-tar.workspace = true
 thiserror.workspace = true
 tokio = { workspace = true, features = ["process", "sync", "fs", "rt", "io-util", "time"] }
@@ -70,7 +69,6 @@ tokio-stream.workspace = true
 tokio-util.workspace = true
 toml_edit = { workspace = true, features = [ "serde" ] }
 tracing.workspace = true
-twox-hash.workspace = true
 url.workspace = true
 walkdir.workspace = true
 metrics.workspace = true

pageserver/benches/bench_walredo.rs

@@ -27,50 +27,30 @@
 //!
 //! # Reference Numbers
 //!
-//! 2024-04-15 on i3en.3xlarge
+//! 2024-03-20 on i3en.3xlarge
 //!
 //! ```text
-//! async-short/1           time:   [24.584 µs 24.737 µs 24.922 µs]
-//! async-short/2           time:   [33.479 µs 33.660 µs 33.888 µs]
-//! async-short/4           time:   [42.713 µs 43.046 µs 43.440 µs]
-//! async-short/8           time:   [71.814 µs 72.478 µs 73.240 µs]
-//! async-short/16          time:   [132.73 µs 134.45 µs 136.22 µs]
-//! async-short/32          time:   [258.31 µs 260.73 µs 263.27 µs]
-//! async-short/64          time:   [511.61 µs 514.44 µs 517.51 µs]
-//! async-short/128         time:   [992.64 µs 998.23 µs 1.0042 ms]
-//! async-medium/1          time:   [110.11 µs 110.50 µs 110.96 µs]
-//! async-medium/2          time:   [153.06 µs 153.85 µs 154.99 µs]
-//! async-medium/4          time:   [317.51 µs 319.92 µs 322.85 µs]
-//! async-medium/8          time:   [638.30 µs 644.68 µs 652.12 µs]
-//! async-medium/16         time:   [1.2651 ms 1.2773 ms 1.2914 ms]
-//! async-medium/32         time:   [2.5117 ms 2.5410 ms 2.5720 ms]
-//! async-medium/64         time:   [4.8088 ms 4.8555 ms 4.9047 ms]
-//! async-medium/128        time:   [8.8311 ms 8.9849 ms 9.1263 ms]
-//! sync-short/1            time:   [25.503 µs 25.626 µs 25.771 µs]
-//! sync-short/2            time:   [30.850 µs 31.013 µs 31.208 µs]
-//! sync-short/4            time:   [45.543 µs 45.856 µs 46.193 µs]
-//! sync-short/8            time:   [84.114 µs 84.639 µs 85.220 µs]
-//! sync-short/16           time:   [185.22 µs 186.15 µs 187.13 µs]
-//! sync-short/32           time:   [377.43 µs 378.87 µs 380.46 µs]
-//! sync-short/64           time:   [756.49 µs 759.04 µs 761.70 µs]
-//! sync-short/128          time:   [1.4825 ms 1.4874 ms 1.4923 ms]
-//! sync-medium/1           time:   [105.66 µs 106.01 µs 106.43 µs]
-//! sync-medium/2           time:   [153.10 µs 153.84 µs 154.72 µs]
-//! sync-medium/4           time:   [327.13 µs 329.44 µs 332.27 µs]
-//! sync-medium/8           time:   [654.26 µs 658.73 µs 663.63 µs]
-//! sync-medium/16          time:   [1.2682 ms 1.2748 ms 1.2816 ms]
-//! sync-medium/32          time:   [2.4456 ms 2.4595 ms 2.4731 ms]
-//! sync-medium/64          time:   [4.6523 ms 4.6890 ms 4.7256 ms]
-//! sync-medium/128         time:   [8.7215 ms 8.8323 ms 8.9344 ms]
+//! short/1                 time:   [26.483 µs 26.614 µs 26.767 µs]
+//! short/2                 time:   [32.223 µs 32.465 µs 32.767 µs]
+//! short/4                 time:   [47.203 µs 47.583 µs 47.984 µs]
+//! short/8                 time:   [89.135 µs 89.612 µs 90.139 µs]
+//! short/16                time:   [190.12 µs 191.52 µs 192.88 µs]
+//! short/32                time:   [380.96 µs 382.63 µs 384.20 µs]
+//! short/64                time:   [736.86 µs 741.07 µs 745.03 µs]
+//! short/128               time:   [1.4106 ms 1.4206 ms 1.4294 ms]
+//! medium/1                time:   [111.81 µs 112.25 µs 112.79 µs]
+//! medium/2                time:   [158.26 µs 159.13 µs 160.21 µs]
+//! medium/4                time:   [334.65 µs 337.14 µs 340.07 µs]
+//! medium/8                time:   [675.32 µs 679.91 µs 685.25 µs]
+//! medium/16               time:   [1.2929 ms 1.2996 ms 1.3067 ms]
+//! medium/32               time:   [2.4295 ms 2.4461 ms 2.4623 ms]
+//! medium/64               time:   [4.3973 ms 4.4458 ms 4.4875 ms]
+//! medium/128              time:   [7.5955 ms 7.7847 ms 7.9481 ms]
 //! ```
 
 use bytes::{Buf, Bytes};
 use criterion::{BenchmarkId, Criterion};
-use pageserver::{
-    config::PageServerConf,
-    walrecord::NeonWalRecord,
-    walredo::{PostgresRedoManager, ProcessKind},
-};
+use pageserver::{config::PageServerConf, walrecord::NeonWalRecord, walredo::PostgresRedoManager};
 use pageserver_api::{key::Key, shard::TenantShardId};
 use std::{
     sync::Arc,
@@ -80,39 +60,33 @@ use tokio::{sync::Barrier, task::JoinSet};
 use utils::{id::TenantId, lsn::Lsn};
 
 fn bench(c: &mut Criterion) {
-    for process_kind in &[ProcessKind::Async, ProcessKind::Sync] {
-        {
-            let nclients = [1, 2, 4, 8, 16, 32, 64, 128];
-            for nclients in nclients {
-                let mut group = c.benchmark_group(format!("{process_kind}-short"));
-                group.bench_with_input(
-                    BenchmarkId::from_parameter(nclients),
-                    &nclients,
-                    |b, nclients| {
-                        let redo_work = Arc::new(Request::short_input());
-                        b.iter_custom(|iters| {
-                            bench_impl(*process_kind, Arc::clone(&redo_work), iters, *nclients)
-                        });
-                    },
-                );
-            }
+    {
+        let nclients = [1, 2, 4, 8, 16, 32, 64, 128];
+        for nclients in nclients {
+            let mut group = c.benchmark_group("short");
+            group.bench_with_input(
+                BenchmarkId::from_parameter(nclients),
+                &nclients,
+                |b, nclients| {
+                    let redo_work = Arc::new(Request::short_input());
+                    b.iter_custom(|iters| bench_impl(Arc::clone(&redo_work), iters, *nclients));
+                },
+            );
         }
+    }
 
-        {
-            let nclients = [1, 2, 4, 8, 16, 32, 64, 128];
-            for nclients in nclients {
-                let mut group = c.benchmark_group(format!("{process_kind}-medium"));
-                group.bench_with_input(
-                    BenchmarkId::from_parameter(nclients),
-                    &nclients,
-                    |b, nclients| {
-                        let redo_work = Arc::new(Request::medium_input());
-                        b.iter_custom(|iters| {
-                            bench_impl(*process_kind, Arc::clone(&redo_work), iters, *nclients)
-                        });
-                    },
-                );
-            }
+    {
+        let nclients = [1, 2, 4, 8, 16, 32, 64, 128];
+        for nclients in nclients {
+            let mut group = c.benchmark_group("medium");
+            group.bench_with_input(
+                BenchmarkId::from_parameter(nclients),
+                &nclients,
+                |b, nclients| {
+                    let redo_work = Arc::new(Request::medium_input());
+                    b.iter_custom(|iters| bench_impl(Arc::clone(&redo_work), iters, *nclients));
+                },
+            );
         }
     }
 }
@@ -120,16 +94,10 @@ criterion::criterion_group!(benches, bench);
 criterion::criterion_main!(benches);
 
 // Returns the sum of each client's wall-clock time spent executing their share of the n_redos.
-fn bench_impl(
-    process_kind: ProcessKind,
-    redo_work: Arc<Request>,
-    n_redos: u64,
-    nclients: u64,
-) -> Duration {
+fn bench_impl(redo_work: Arc<Request>, n_redos: u64, nclients: u64) -> Duration {
     let repo_dir = camino_tempfile::tempdir_in(env!("CARGO_TARGET_TMPDIR")).unwrap();
 
-    let mut conf = PageServerConf::dummy_conf(repo_dir.path().to_path_buf());
-    conf.walredo_process_kind = process_kind;
+    let conf = PageServerConf::dummy_conf(repo_dir.path().to_path_buf());
     let conf = Box::leak(Box::new(conf));
     let tenant_shard_id = TenantShardId::unsharded(TenantId::generate());
 
@@ -145,40 +113,25 @@ fn bench_impl(
     let manager = PostgresRedoManager::new(conf, tenant_shard_id);
     let manager = Arc::new(manager);
 
-    // divide the amount of work equally among the clients.
-    let nredos_per_client = n_redos / nclients;
     for _ in 0..nclients {
         rt.block_on(async {
             tasks.spawn(client(
                 Arc::clone(&manager),
                 Arc::clone(&start),
                 Arc::clone(&redo_work),
-                nredos_per_client,
+                // divide the amount of work equally among the clients
+                n_redos / nclients,
             ))
         });
     }
 
-    let elapsed = rt.block_on(async move {
-        let mut total_wallclock_time = Duration::ZERO;
+    rt.block_on(async move {
+        let mut total_wallclock_time = std::time::Duration::from_millis(0);
         while let Some(res) = tasks.join_next().await {
             total_wallclock_time += res.unwrap();
         }
         total_wallclock_time
-    });
-
-    // consistency check to ensure process kind setting worked
-    if nredos_per_client > 0 {
-        assert_eq!(
-            manager
-                .status()
-                .process
-                .map(|p| p.kind)
-                .expect("the benchmark work causes a walredo process to be spawned"),
-            std::borrow::Cow::Borrowed(process_kind.into())
-        );
-    }
-
-    elapsed
+    })
 }
 
 async fn client(

pageserver/client/src/mgmt_api.rs

@@ -128,12 +128,12 @@ impl Client {
 
     pub async fn timeline_info(
         &self,
-        tenant_shard_id: TenantShardId,
+        tenant_id: TenantId,
         timeline_id: TimelineId,
         force_await_logical_size: ForceAwaitLogicalSize,
     ) -> Result<pageserver_api::models::TimelineInfo> {
         let uri = format!(
-            "{}/v1/tenant/{tenant_shard_id}/timeline/{timeline_id}",
+            "{}/v1/tenant/{tenant_id}/timeline/{timeline_id}",
             self.mgmt_api_endpoint
         );
 
@@ -151,11 +151,11 @@ impl Client {
 
     pub async fn keyspace(
         &self,
-        tenant_shard_id: TenantShardId,
+        tenant_id: TenantId,
         timeline_id: TimelineId,
     ) -> Result<pageserver_api::models::partitioning::Partitioning> {
         let uri = format!(
-            "{}/v1/tenant/{tenant_shard_id}/timeline/{timeline_id}/keyspace",
+            "{}/v1/tenant/{tenant_id}/timeline/{timeline_id}/keyspace",
             self.mgmt_api_endpoint
         );
         self.get(&uri)
@@ -279,7 +279,7 @@ impl Client {
         lazy: bool,
     ) -> Result<()> {
         let req_body = TenantLocationConfigRequest {
-            tenant_id: None,
+            tenant_id: Some(tenant_shard_id),
             config,
         };
 

pageserver/compaction/Cargo.toml

@@ -11,6 +11,7 @@ default = []
 anyhow.workspace = true
 async-compression.workspace = true
 async-stream.workspace = true
+async-trait.workspace = true
 byteorder.workspace = true
 bytes.workspace = true
 chrono = { workspace = true, features = ["serde"] }

pageserver/compaction/src/compact_tiered.rs

@@ -43,8 +43,7 @@ pub async fn compact_tiered<E: CompactionJobExecutor>(
     fanout: u64,
     ctx: &E::RequestContext,
 ) -> anyhow::Result<()> {
-    assert!(fanout >= 1, "fanout needs to be at least 1 but is {fanout}");
-    let exp_base = fanout.max(2);
+    assert!(fanout >= 2);
     // Start at L0
     let mut current_level_no = 0;
     let mut current_level_target_height = target_file_size;
@@ -107,7 +106,7 @@ pub async fn compact_tiered<E: CompactionJobExecutor>(
             break;
         }
         current_level_no += 1;
-        current_level_target_height = current_level_target_height.saturating_mul(exp_base);
+        current_level_target_height = current_level_target_height.saturating_mul(fanout);
     }
     Ok(())
 }

pageserver/compaction/src/helpers.rs

@@ -180,7 +180,7 @@ where
                 match top.deref_mut() {
                     LazyLoadLayer::Unloaded(ref mut l) => {
                         let fut = l.load_keys(this.ctx);
-                        this.load_future.set(Some(Box::pin(fut)));
+                        this.load_future.set(Some(fut));
                         continue;
                     }
                     LazyLoadLayer::Loaded(ref mut entries) => {

pageserver/compaction/src/interface.rs

@@ -3,6 +3,7 @@
 //!
 //! All the heavy lifting is done by the create_image and create_delta
 //! functions that the implementor provides.
+use async_trait::async_trait;
 use futures::Future;
 use pageserver_api::{key::Key, keyspace::key_range_size};
 use std::ops::Range;
@@ -140,16 +141,18 @@ pub trait CompactionLayer<K: CompactionKey + ?Sized> {
 
     fn is_delta(&self) -> bool;
 }
+
+#[async_trait]
 pub trait CompactionDeltaLayer<E: CompactionJobExecutor + ?Sized>: CompactionLayer<E::Key> {
     type DeltaEntry<'a>: CompactionDeltaEntry<'a, E::Key>
     where
         Self: 'a;
 
     /// Return all keys in this delta layer.
-    fn load_keys<'a>(
+    async fn load_keys<'a>(
         &self,
         ctx: &E::RequestContext,
-    ) -> impl Future<Output = anyhow::Result<Vec<Self::DeltaEntry<'_>>>> + Send;
+    ) -> anyhow::Result<Vec<Self::DeltaEntry<'_>>>;
 }
 
 pub trait CompactionImageLayer<E: CompactionJobExecutor + ?Sized>: CompactionLayer<E::Key> {}

pageserver/compaction/src/simulator.rs

@@ -2,6 +2,7 @@ mod draw;
 
 use draw::{LayerTraceEvent, LayerTraceFile, LayerTraceOp};
 
+use async_trait::async_trait;
 use futures::StreamExt;
 use rand::Rng;
 use tracing::info;
@@ -138,6 +139,7 @@ impl interface::CompactionLayer<Key> for Arc<MockDeltaLayer> {
     }
 }
 
+#[async_trait]
 impl interface::CompactionDeltaLayer<MockTimeline> for Arc<MockDeltaLayer> {
     type DeltaEntry<'a> = MockRecord;