Fix documentation format issues

## Problem

See https://neondb.slack.com/archives/C03QLRH7PPD/p1712529369520409

In case of statements CREATE TABLE AS SELECT... or INSERT FROM SELECT...
we are fetching data from source table and storing it in destination
table. It cause problems with prefetch last-written-lsn is known for the
pages of source table
(which for example happens after compute restart). In this case we get
get global value of last-written-lsn which is changed frequently as far
as we are writing pages of destination table. As a result request-isn
for the prefetch and request-let when this page is actually needed are
different and we got exported prefetch request. So it actually disarms
prefetch.


## Summary of changes

Proposed simple patch stores last-written LSN for the page when it is
not found. So next time we will request last-written LSN for this page,
we will get the same value (certainly if the page was not changed).

## Checklist before requesting a review

- [ ] I have performed a self-review of my code.
- [ ] If it is a core feature, I have added thorough tests.
- [ ] Do we need to implement analytics? if so did you add the relevant
metrics to the dashboard?
- [ ] If this PR requires public announcement, mark it with
/release-notes label and add several sentences in this section.

## Checklist before merging

- [ ] Do not forget to reformat commit message to not include the above
checklist

---------

Co-authored-by: Konstantin Knizhnik <knizhnik@neon.tech>

.dockerignore

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

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

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

.github/workflows/benchmarking.yml

@@ -147,15 +147,16 @@ 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" }]
+          "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" }]
         }'
 
         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
@@ -171,7 +172,7 @@ 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
@@ -190,7 +191,7 @@ 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 '.')" >> $GITHUB_OUTPUT
@@ -253,6 +254,9 @@ jobs:
           neon-captest-reuse)
             CONNSTR=${{ 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 }}
             ;;
@@ -270,11 +274,15 @@ jobs:
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERY="SELECT version();"
+        QUERIES=("SELECT version()")
         if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERY="${QUERY} SHOW neon.tenant_id; SHOW neon.timeline_id;"
+          QUERIES+=("SHOW neon.tenant_id")
+          QUERIES+=("SHOW neon.timeline_id")
         fi
-        psql ${CONNSTR} -c "${QUERY}"
+
+        for q in "${QUERIES[@]}"; do
+          psql ${CONNSTR} -c "${q}"
+        done
 
     - name: Benchmark init
       uses: ./.github/actions/run-python-test-set
@@ -401,11 +409,15 @@ jobs:
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERY="SELECT version();"
+        QUERIES=("SELECT version()")
         if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERY="${QUERY} SHOW neon.tenant_id; SHOW neon.timeline_id;"
+          QUERIES+=("SHOW neon.tenant_id")
+          QUERIES+=("SHOW neon.timeline_id")
         fi
-        psql ${CONNSTR} -c "${QUERY}"
+
+        for q in "${QUERIES[@]}"; do
+          psql ${CONNSTR} -c "${q}"
+        done
 
     - name: ClickBench benchmark
       uses: ./.github/actions/run-python-test-set
@@ -507,11 +519,15 @@ jobs:
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERY="SELECT version();"
+        QUERIES=("SELECT version()")
         if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERY="${QUERY} SHOW neon.tenant_id; SHOW neon.timeline_id;"
+          QUERIES+=("SHOW neon.tenant_id")
+          QUERIES+=("SHOW neon.timeline_id")
         fi
-        psql ${CONNSTR} -c "${QUERY}"
+
+        for q in "${QUERIES[@]}"; do
+          psql ${CONNSTR} -c "${q}"
+        done
 
     - name: Run TPC-H benchmark
       uses: ./.github/actions/run-python-test-set
@@ -597,11 +613,15 @@ jobs:
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERY="SELECT version();"
+        QUERIES=("SELECT version()")
         if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERY="${QUERY} SHOW neon.tenant_id; SHOW neon.timeline_id;"
+          QUERIES+=("SHOW neon.tenant_id")
+          QUERIES+=("SHOW neon.timeline_id")
         fi
-        psql ${CONNSTR} -c "${QUERY}"
+
+        for q in "${QUERIES[@]}"; do
+          psql ${CONNSTR} -c "${q}"
+        done
 
     - name: Run user examples
       uses: ./.github/actions/run-python-test-set

.github/workflows/build_and_test.yml

@@ -461,6 +461,7 @@ jobs:
 
       - name: Pytest regression tests
         uses: ./.github/actions/run-python-test-set
+        timeout-minutes: 60
         with:
           build_type: ${{ matrix.build_type }}
           test_selection: regress
@@ -474,7 +475,7 @@ jobs:
           TEST_RESULT_CONNSTR: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}
           CHECK_ONDISK_DATA_COMPATIBILITY: nonempty
           BUILD_TAG: ${{ needs.tag.outputs.build-tag }}
-          PAGESERVER_VIRTUAL_FILE_IO_ENGINE: std-fs
+          PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
           PAGESERVER_GET_VECTORED_IMPL: vectored
 
       # Temporary disable this step until we figure out why it's so flaky
@@ -554,7 +555,7 @@ jobs:
           VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
           PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
           TEST_RESULT_CONNSTR: "${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}"
-          PAGESERVER_VIRTUAL_FILE_IO_ENGINE: std-fs
+          PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
       # XXX: no coverage data handling here, since benchmarks are run on release builds,
       # while coverage is currently collected for the debug ones
 
@@ -1120,18 +1121,36 @@ 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 [[ "$GITHUB_REF_NAME" == "release" ]]; then
+            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main \
+              -f deployPgSniRouter=false \
+              -f deployProxy=false \
+              -f deployStorage=true \
+              -f deployStorageBroker=true \
+              -f deployStorageController=true \
+              -f branch=main \
+              -f dockerTag=${{needs.tag.outputs.build-tag}} \
+              -f deployPreprodRegion=true
+
             gh workflow --repo neondatabase/aws run deploy-prod.yml --ref main \
               -f deployPgSniRouter=false \
               -f deployProxy=false \
               -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/trigger-e2e-tests.yml

@@ -62,14 +62,14 @@ jobs:
 
   trigger-e2e-tests:
     needs: [ tag ]
-    runs-on: [ self-hosted, gen3, small ]
+    runs-on: ubuntu-latest
     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)
@@ -79,41 +79,55 @@ jobs:
             fi
           done
 
-      - name: Set PR's status to pending and request a remote CI test
+      - name: Set e2e-platforms
+        id: e2e-platforms
+        env:
+          PR_NUMBER: ${{ github.event.pull_request.number }}
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         run: |
-          # For pull requests, GH Actions set "github.sha" variable to point at a fake merge commit
-          # but we need to use a real sha of a latest commit in the PR's branch for the e2e job,
-          # to place a job run status update later.
-          COMMIT_SHA=${{ 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 }}}
+          # Default set of platforms to run e2e tests on
+          platforms='["docker", "k8s"]'
 
-          REMOTE_REPO="${{ github.repository_owner }}/cloud"
+          # If the PR changes vendor/, pgxn/ or libs/vm_monitor/ directories, or Dockerfile.compute-node, add k8s-neonvm to the list of platforms.
+          # If the workflow run is not a pull request, add k8s-neonvm to the list.
+          if [ "$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
 
-          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\"
-            }"
+          echo "e2e-platforms=${platforms}" | tee -a $GITHUB_OUTPUT
 
-          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 }}\"
-              }
-            }"
+      - 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"
+
+          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"
+
+          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}"

CODEOWNERS

@@ -1,12 +1,13 @@
 /compute_tools/ @neondatabase/control-plane @neondatabase/compute
-/control_plane/attachment_service @neondatabase/storage
+/storage_controller @neondatabase/storage
 /libs/pageserver_api/ @neondatabase/storage
-/libs/postgres_ffi/ @neondatabase/compute
+/libs/postgres_ffi/ @neondatabase/compute @neondatabase/safekeepers
 /libs/remote_storage/ @neondatabase/storage
 /libs/safekeeper_api/ @neondatabase/safekeepers
 /libs/vm_monitor/ @neondatabase/autoscaling
 /pageserver/ @neondatabase/storage
 /pgxn/ @neondatabase/compute
+/pgxn/neon/ @neondatabase/compute @neondatabase/safekeepers
 /proxy/ @neondatabase/proxy
 /safekeeper/ @neondatabase/safekeepers
 /vendor/ @neondatabase/compute

Cargo.toml

@@ -3,7 +3,7 @@ resolver = "2"
 members = [
     "compute_tools",
     "control_plane",
-    "control_plane/attachment_service",
+    "control_plane/storcon_cli",
     "pageserver",
     "pageserver/compaction",
     "pageserver/ctl",
@@ -12,6 +12,7 @@ members = [
     "proxy",
     "safekeeper",
     "storage_broker",
+    "storage_controller",
     "s3_scrubber",
     "workspace_hack",
     "trace",
@@ -43,6 +44,7 @@ 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"
@@ -52,10 +54,12 @@ 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"
 aws-credential-types = "1.1.4"
+aws-sigv4 = { version = "1.2.0", features = ["sign-http"] }
+aws-types = "1.1.7"
 axum = { version = "0.6.20", features = ["ws"] }
 base64 = "0.13.0"
 bincode = "1.3"
@@ -76,6 +80,7 @@ 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"
@@ -88,11 +93,12 @@ hex = "0.4"
 hex-literal = "0.4"
 hmac = "0.12.1"
 hostname = "0.3.1"
+http = {version = "1.1.0", features = ["std"]}
 http-types = { version = "2", default-features = false }
 humantime = "2.1"
 humantime-serde = "1.1.1"
 hyper = "0.14"
-hyper-tungstenite = "0.11"
+hyper-tungstenite = "0.13.0"
 inotify = "0.10.2"
 ipnet = "2.9.0"
 itertools = "0.10"
@@ -101,6 +107,8 @@ lasso = "0.7"
 leaky-bucket = "1.0.1"
 libc = "0.2"
 md5 = "0.7.0"
+measured = { version = "0.0.20", features=["lasso"] }
+measured-process = { version = "0.0.20" }
 memoffset = "0.8"
 native-tls = "0.2"
 nix = { version = "0.27", features = ["fs", "process", "socket", "signal", "poll"] }
@@ -120,7 +128,7 @@ procfs = "0.14"
 prometheus = {version = "0.13", default_features=false, features = ["process"]} # removes protobuf dependency
 prost = "0.11"
 rand = "0.8"
-redis = { version = "0.24.0", features = ["tokio-rustls-comp", "keep-alive"] }
+redis = { version = "0.25.2", features = ["tokio-rustls-comp", "keep-alive"] }
 regex = "1.10.2"
 reqwest = { version = "0.11", default-features = false, features = ["rustls-tls"] }
 reqwest-tracing = { version = "0.4.7", features = ["opentelemetry_0_20"] }
@@ -148,11 +156,12 @@ smol_str = { version = "0.2.0", features = ["serde"] }
 socket2 = "0.5"
 strum = "0.24"
 strum_macros = "0.24"
+"subtle"  = "2.5.0"
 svg_fmt = "0.4.1"
 sync_wrapper = "0.1.2"
 tar = "0.4"
 task-local-extensions = "0.1.4"
-test-context = "0.1"
+test-context = "0.3"
 thiserror = "1.0"
 tikv-jemallocator = "0.5"
 tikv-jemalloc-ctl = "0.5"

Dockerfile.build-tools

@@ -135,7 +135,7 @@ WORKDIR /home/nonroot
 
 # Rust
 # Please keep the version of llvm (installed above) in sync with rust llvm (`rustc --version --verbose | grep LLVM`)
-ENV RUSTC_VERSION=1.76.0
+ENV RUSTC_VERSION=1.77.0
 ENV RUSTUP_HOME="/home/nonroot/.rustup"
 ENV PATH="/home/nonroot/.cargo/bin:${PATH}"
 RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && whoami && \
@@ -149,7 +149,7 @@ RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux
     cargo install --git https://github.com/paritytech/cachepot && \
     cargo install rustfilt && \
     cargo install cargo-hakari && \
-    cargo install cargo-deny && \
+    cargo install cargo-deny --locked && \
     cargo install cargo-hack && \
     cargo install cargo-nextest && \
     rm -rf /home/nonroot/.cargo/registry && \

Dockerfile.compute-node

@@ -944,6 +944,9 @@ 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)

Makefile

@@ -51,7 +51,7 @@ CARGO_BUILD_FLAGS += $(filter -j1,$(MAKEFLAGS))
 CARGO_CMD_PREFIX += $(if $(filter n,$(MAKEFLAGS)),,+)
 # Force cargo not to print progress bar
 CARGO_CMD_PREFIX += CARGO_TERM_PROGRESS_WHEN=never CI=1
-# Set PQ_LIB_DIR to make sure `attachment_service` get linked with bundled libpq (through diesel)
+# Set PQ_LIB_DIR to make sure `storage_controller` get linked with bundled libpq (through diesel)
 CARGO_CMD_PREFIX += PQ_LIB_DIR=$(POSTGRES_INSTALL_DIR)/v16/lib
 
 #

README.md

@@ -238,6 +238,14 @@ If you encounter errors during setting up the initial tenant, it's best to stop
 
 ## Running tests
 
+### Rust unit tests
+
+We are using [`cargo-nextest`](https://nexte.st/) to run the tests in Github Workflows.
+Some crates do not support running plain `cargo test` anymore, prefer `cargo nextest run` instead.
+You can install `cargo-nextest` with `cargo install cargo-nextest`.
+
+### Integration tests
+
 Ensure your dependencies are installed as described [here](https://github.com/neondatabase/neon#dependency-installation-notes).
 
 ```sh

clippy.toml

@@ -2,6 +2,8 @@ disallowed-methods = [
     "tokio::task::block_in_place",
     # Allow this for now, to deny it later once we stop using Handle::block_on completely
     # "tokio::runtime::Handle::block_on",
+    # use tokio_epoll_uring_ext instead
+    "tokio_epoll_uring::thread_local_system",
 ]
 
 disallowed-macros = [

compute_tools/README.md

@@ -32,6 +32,29 @@ compute_ctl -D /var/db/postgres/compute \
             -b /usr/local/bin/postgres
 ```
 
+## State Diagram
+
+Computes can be in various states. Below is a diagram that details how a
+compute moves between states.
+
+```mermaid
+%% https://mermaid.js.org/syntax/stateDiagram.html
+stateDiagram-v2
+  [*] --> Empty : Compute spawned
+  Empty --> ConfigurationPending : Waiting for compute spec
+  ConfigurationPending --> Configuration : Received compute spec
+  Configuration --> Failed : Failed to configure the compute
+  Configuration --> Running : Compute has been configured
+  Empty --> Init : Compute spec is immediately available
+  Empty --> TerminationPending : Requested termination
+  Init --> Failed : Failed to start Postgres
+  Init --> Running : Started Postgres
+  Running --> TerminationPending : Requested termination
+  TerminationPending --> Terminated : Terminated compute
+  Failed --> [*] : Compute exited
+  Terminated --> [*] : Compute exited
+```
+
 ## Tests
 
 Cargo formatter:

compute_tools/src/compute.rs

@@ -17,6 +17,7 @@ use chrono::{DateTime, Utc};
 use futures::future::join_all;
 use futures::stream::FuturesUnordered;
 use futures::StreamExt;
+use nix::unistd::Pid;
 use postgres::error::SqlState;
 use postgres::{Client, NoTls};
 use tracing::{debug, error, info, instrument, warn};
@@ -722,8 +723,12 @@ impl ComputeNode {
         // Stop it when it's ready
         info!("waiting for postgres");
         wait_for_postgres(&mut pg, Path::new(pgdata))?;
-        pg.kill()?;
-        info!("sent kill signal");
+        // SIGQUIT orders postgres to exit immediately. We don't want to SIGKILL
+        // it to avoid orphaned processes prowling around while datadir is
+        // wiped.
+        let pm_pid = Pid::from_raw(pg.id() as i32);
+        kill(pm_pid, Signal::SIGQUIT)?;
+        info!("sent SIGQUIT signal");
         pg.wait()?;
         info!("done prewarming");
 
@@ -1257,10 +1262,12 @@ LIMIT 100",
         .await
         .map_err(DownloadError::Other);
 
-        self.ext_download_progress
-            .write()
-            .expect("bad lock")
-            .insert(ext_archive_name.to_string(), (download_start, true));
+        if download_size.is_ok() {
+            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::PgOptionsSerialize;
-use compute_api::spec::{ComputeMode, ComputeSpec};
+use crate::pg_helpers::{GenericOptionExt, PgOptionsSerialize};
+use compute_api::spec::{ComputeMode, ComputeSpec, GenericOption};
 
 /// Check that `line` is inside a text file and put it there if it is not.
 /// Create file if it doesn't exist.
@@ -17,6 +17,7 @@ pub fn line_in_file(path: &Path, line: &str) -> Result<bool> {
         .write(true)
         .create(true)
         .append(false)
+        .truncate(false)
         .open(path)?;
     let buf = io::BufReader::new(&file);
     let mut count: usize = 0;
@@ -91,6 +92,27 @@ 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)
 }
 
-trait GenericOptionExt {
+pub trait GenericOptionExt {
     fn to_pg_option(&self) -> String;
     fn to_pg_setting(&self) -> String;
 }

compute_tools/src/spec.rs

@@ -745,7 +745,12 @@ pub fn handle_extension_neon(client: &mut Client) -> Result<()> {
     // - extension was already installed and is up to date
     let query = "ALTER EXTENSION neon UPDATE";
     info!("update neon extension version with query: {}", query);
-    client.simple_query(query)?;
+    if let Err(e) = client.simple_query(query) {
+        error!(
+            "failed to upgrade neon extension during `handle_extension_neon`: {}",
+            e
+        );
+    }
 
     Ok(())
 }
@@ -804,6 +809,7 @@ $$;"#,
         "",
         "",
         "",
+        "",
         // Add new migrations below.
     ];
 

control_plane/Cargo.toml

@@ -12,6 +12,7 @@ clap.workspace = true
 comfy-table.workspace = true
 futures.workspace = true
 git-version.workspace = true
+humantime.workspace = true
 nix.workspace = true
 once_cell.workspace = true
 postgres.workspace = true

control_plane/attachment_service/src/metrics.rs

@@ -1,32 +0,0 @@
-use metrics::{register_int_counter, register_int_counter_vec, IntCounter, IntCounterVec};
-use once_cell::sync::Lazy;
-
-pub(crate) struct ReconcilerMetrics {
-    pub(crate) spawned: IntCounter,
-    pub(crate) complete: IntCounterVec,
-}
-
-impl ReconcilerMetrics {
-    // Labels used on [`Self::complete`]
-    pub(crate) const SUCCESS: &'static str = "ok";
-    pub(crate) const ERROR: &'static str = "success";
-    pub(crate) const CANCEL: &'static str = "cancel";
-}
-
-pub(crate) static RECONCILER: Lazy<ReconcilerMetrics> = Lazy::new(|| ReconcilerMetrics {
-    spawned: register_int_counter!(
-        "storage_controller_reconcile_spawn",
-        "Count of how many times we spawn a reconcile task",
-    )
-    .expect("failed to define a metric"),
-    complete: register_int_counter_vec!(
-        "storage_controller_reconcile_complete",
-        "Reconciler tasks completed, broken down by success/failure/cancelled",
-        &["status"],
-    )
-    .expect("failed to define a metric"),
-});
-
-pub fn preinitialize_metrics() {
-    Lazy::force(&RECONCILER);
-}

control_plane/src/background_process.rs

@@ -86,7 +86,10 @@ where
         .stdout(process_log_file)
         .stderr(same_file_for_stderr)
         .args(args);
-    let filled_cmd = fill_remote_storage_secrets_vars(fill_rust_env_vars(background_command));
+
+    let filled_cmd = fill_env_vars_prefixed_neon(fill_remote_storage_secrets_vars(
+        fill_rust_env_vars(background_command),
+    ));
     filled_cmd.envs(envs);
 
     let pid_file_to_check = match &initial_pid_file {
@@ -268,6 +271,15 @@ 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)
@@ -294,7 +306,7 @@ 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 the fork, even though
+    //      parent now got implicitly copied by 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

@@ -8,15 +8,13 @@
 use anyhow::{anyhow, bail, Context, Result};
 use clap::{value_parser, Arg, ArgAction, ArgMatches, Command, ValueEnum};
 use compute_api::spec::ComputeMode;
-use control_plane::attachment_service::AttachmentService;
 use control_plane::endpoint::ComputeControlPlane;
 use control_plane::local_env::{InitForceMode, LocalEnv};
 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::{
-    NodeAvailability, NodeConfigureRequest, NodeSchedulingPolicy, PlacementPolicy,
-};
+use pageserver_api::controller_api::PlacementPolicy;
 use pageserver_api::models::{
     ShardParameters, TenantCreateRequest, TimelineCreateRequest, TimelineInfo,
 };
@@ -138,7 +136,7 @@ fn main() -> Result<()> {
             "start" => rt.block_on(handle_start_all(sub_args, &env)),
             "stop" => rt.block_on(handle_stop_all(sub_args, &env)),
             "pageserver" => rt.block_on(handle_pageserver(sub_args, &env)),
-            "attachment_service" => rt.block_on(handle_attachment_service(sub_args, &env)),
+            "storage_controller" => rt.block_on(handle_storage_controller(sub_args, &env)),
             "safekeeper" => rt.block_on(handle_safekeeper(sub_args, &env)),
             "endpoint" => rt.block_on(handle_endpoint(sub_args, &env)),
             "mappings" => handle_mappings(sub_args, &mut env),
@@ -437,7 +435,7 @@ async fn handle_tenant(
 
             let placement_policy = match create_match.get_one::<String>("placement-policy") {
                 Some(s) if !s.is_empty() => serde_json::from_str::<PlacementPolicy>(s)?,
-                _ => PlacementPolicy::Single,
+                _ => PlacementPolicy::Attached(0),
             };
 
             let tenant_conf = PageServerNode::parse_config(tenant_conf)?;
@@ -445,14 +443,14 @@ async fn handle_tenant(
             // If tenant ID was not specified, generate one
             let tenant_id = parse_tenant_id(create_match)?.unwrap_or_else(TenantId::generate);
 
-            // We must register the tenant with the attachment service, so
+            // We must register the tenant with the storage controller, so
             // that when the pageserver restarts, it will be re-attached.
-            let attachment_service = AttachmentService::from_env(env);
-            attachment_service
+            let storage_controller = StorageController::from_env(env);
+            storage_controller
                 .tenant_create(TenantCreateRequest {
                     // Note that ::unsharded here isn't actually because the tenant is unsharded, its because the
-                    // attachment service expecfs a shard-naive tenant_id in this attribute, and the TenantCreateRequest
-                    // type is used both in attachment service (for creating tenants) and in pageserver (for creating shards)
+                    // storage controller expecfs a shard-naive tenant_id in this attribute, and the TenantCreateRequest
+                    // type is used both in storage controller (for creating tenants) and in pageserver (for creating shards)
                     new_tenant_id: TenantShardId::unsharded(tenant_id),
                     generation: None,
                     shard_parameters: ShardParameters {
@@ -476,9 +474,9 @@ async fn handle_tenant(
                 .context("Failed to parse postgres version from the argument string")?;
 
             // FIXME: passing None for ancestor_start_lsn is not kosher in a sharded world: we can't have
-            // different shards picking different start lsns.  Maybe we have to teach attachment service
+            // different shards picking different start lsns.  Maybe we have to teach storage controller
             // to let shard 0 branch first and then propagate the chosen LSN to other shards.
-            attachment_service
+            storage_controller
                 .tenant_timeline_create(
                     tenant_id,
                     TimelineCreateRequest {
@@ -523,84 +521,6 @@ async fn handle_tenant(
                 .with_context(|| format!("Tenant config failed for tenant with id {tenant_id}"))?;
             println!("tenant {tenant_id} successfully configured on the pageserver");
         }
-        Some(("migrate", matches)) => {
-            let tenant_shard_id = get_tenant_shard_id(matches, env)?;
-            let new_pageserver = get_pageserver(env, matches)?;
-            let new_pageserver_id = new_pageserver.conf.id;
-
-            let attachment_service = AttachmentService::from_env(env);
-            attachment_service
-                .tenant_migrate(tenant_shard_id, new_pageserver_id)
-                .await?;
-
-            println!("tenant {tenant_shard_id} migrated to {}", new_pageserver_id);
-        }
-        Some(("status", matches)) => {
-            let tenant_id = get_tenant_id(matches, env)?;
-
-            let mut shard_table = comfy_table::Table::new();
-            shard_table.set_header(["Shard", "Pageserver", "Physical Size"]);
-
-            let mut tenant_synthetic_size = None;
-
-            let attachment_service = AttachmentService::from_env(env);
-            for shard in attachment_service.tenant_locate(tenant_id).await?.shards {
-                let pageserver =
-                    PageServerNode::from_env(env, env.get_pageserver_conf(shard.node_id)?);
-
-                let size = pageserver
-                    .http_client
-                    .tenant_details(shard.shard_id)
-                    .await?
-                    .tenant_info
-                    .current_physical_size
-                    .unwrap();
-
-                shard_table.add_row([
-                    format!("{}", shard.shard_id.shard_slug()),
-                    format!("{}", shard.node_id.0),
-                    format!("{} MiB", size / (1024 * 1024)),
-                ]);
-
-                if shard.shard_id.is_zero() {
-                    tenant_synthetic_size =
-                        Some(pageserver.tenant_synthetic_size(shard.shard_id).await?);
-                }
-            }
-
-            let Some(synthetic_size) = tenant_synthetic_size else {
-                bail!("Shard 0 not found")
-            };
-
-            let mut tenant_table = comfy_table::Table::new();
-            tenant_table.add_row(["Tenant ID".to_string(), tenant_id.to_string()]);
-            tenant_table.add_row([
-                "Synthetic size".to_string(),
-                format!("{} MiB", synthetic_size.size.unwrap_or(0) / (1024 * 1024)),
-            ]);
-
-            println!("{tenant_table}");
-            println!("{shard_table}");
-        }
-        Some(("shard-split", matches)) => {
-            let tenant_id = get_tenant_id(matches, env)?;
-            let shard_count: u8 = matches.get_one::<u8>("shard-count").cloned().unwrap_or(0);
-
-            let attachment_service = AttachmentService::from_env(env);
-            let result = attachment_service
-                .tenant_split(tenant_id, shard_count)
-                .await?;
-            println!(
-                "Split tenant {} into shards {}",
-                tenant_id,
-                result
-                    .new_shards
-                    .iter()
-                    .map(|s| format!("{:?}", s))
-                    .collect::<Vec<_>>()
-                    .join(",")
-            );
-        }
 
         Some((sub_name, _)) => bail!("Unexpected tenant subcommand '{}'", sub_name),
         None => bail!("no tenant subcommand provided"),
@@ -613,7 +533,7 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
 
     match timeline_match.subcommand() {
         Some(("list", list_match)) => {
-            // TODO(sharding): this command shouldn't have to specify a shard ID: we should ask the attachment service
+            // TODO(sharding): this command shouldn't have to specify a shard ID: we should ask the storage controller
             // where shard 0 is attached, and query there.
             let tenant_shard_id = get_tenant_shard_id(list_match, env)?;
             let timelines = pageserver.timeline_list(&tenant_shard_id).await?;
@@ -633,7 +553,7 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
             let new_timeline_id_opt = parse_timeline_id(create_match)?;
             let new_timeline_id = new_timeline_id_opt.unwrap_or(TimelineId::generate());
 
-            let attachment_service = AttachmentService::from_env(env);
+            let storage_controller = StorageController::from_env(env);
             let create_req = TimelineCreateRequest {
                 new_timeline_id,
                 ancestor_timeline_id: None,
@@ -641,7 +561,7 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
                 ancestor_start_lsn: None,
                 pg_version: Some(pg_version),
             };
-            let timeline_info = attachment_service
+            let timeline_info = storage_controller
                 .tenant_timeline_create(tenant_id, create_req)
                 .await?;
 
@@ -730,7 +650,7 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
                 .transpose()
                 .context("Failed to parse ancestor start Lsn from the request")?;
             let new_timeline_id = TimelineId::generate();
-            let attachment_service = AttachmentService::from_env(env);
+            let storage_controller = StorageController::from_env(env);
             let create_req = TimelineCreateRequest {
                 new_timeline_id,
                 ancestor_timeline_id: Some(ancestor_timeline_id),
@@ -738,7 +658,7 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
                 ancestor_start_lsn: start_lsn,
                 pg_version: None,
             };
-            let timeline_info = attachment_service
+            let timeline_info = storage_controller
                 .tenant_timeline_create(tenant_id, create_req)
                 .await?;
 
@@ -767,7 +687,7 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
 
     match sub_name {
         "list" => {
-            // TODO(sharding): this command shouldn't have to specify a shard ID: we should ask the attachment service
+            // TODO(sharding): this command shouldn't have to specify a shard ID: we should ask the storage controller
             // where shard 0 is attached, and query there.
             let tenant_shard_id = get_tenant_shard_id(sub_args, env)?;
             let timeline_infos = get_timeline_infos(env, &tenant_shard_id)
@@ -952,21 +872,21 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
                 (
                     vec![(parsed.0, parsed.1.unwrap_or(5432))],
                     // If caller is telling us what pageserver to use, this is not a tenant which is
-                    // full managed by attachment service, therefore not sharded.
+                    // full managed by storage controller, therefore not sharded.
                     ShardParameters::DEFAULT_STRIPE_SIZE,
                 )
             } else {
                 // Look up the currently attached location of the tenant, and its striping metadata,
                 // to pass these on to postgres.
-                let attachment_service = AttachmentService::from_env(env);
-                let locate_result = attachment_service.tenant_locate(endpoint.tenant_id).await?;
+                let storage_controller = StorageController::from_env(env);
+                let locate_result = storage_controller.tenant_locate(endpoint.tenant_id).await?;
                 let pageservers = locate_result
                     .shards
                     .into_iter()
                     .map(|shard| {
                         (
                             Host::parse(&shard.listen_pg_addr)
-                                .expect("Attachment service reported bad hostname"),
+                                .expect("Storage controller reported bad hostname"),
                             shard.listen_pg_port,
                         )
                     })
@@ -1015,8 +935,8 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
                         pageserver.pg_connection_config.port(),
                     )]
                 } else {
-                    let attachment_service = AttachmentService::from_env(env);
-                    attachment_service
+                    let storage_controller = StorageController::from_env(env);
+                    storage_controller
                         .tenant_locate(endpoint.tenant_id)
                         .await?
                         .shards
@@ -1024,7 +944,7 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
                         .map(|shard| {
                             (
                                 Host::parse(&shard.listen_pg_addr)
-                                    .expect("Attachment service reported malformed host"),
+                                    .expect("Storage controller reported malformed host"),
                                 shard.listen_pg_port,
                             )
                         })
@@ -1100,9 +1020,8 @@ fn get_pageserver(env: &local_env::LocalEnv, args: &ArgMatches) -> Result<PageSe
 async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
     match sub_match.subcommand() {
         Some(("start", subcommand_args)) => {
-            let register = subcommand_args.get_one::<bool>("register").unwrap_or(&true);
             if let Err(e) = get_pageserver(env, subcommand_args)?
-                .start(&pageserver_config_overrides(subcommand_args), *register)
+                .start(&pageserver_config_overrides(subcommand_args))
                 .await
             {
                 eprintln!("pageserver start failed: {e}");
@@ -1131,7 +1050,7 @@ async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
             }
 
             if let Err(e) = pageserver
-                .start(&pageserver_config_overrides(subcommand_args), false)
+                .start(&pageserver_config_overrides(subcommand_args))
                 .await
             {
                 eprintln!("pageserver start failed: {e}");
@@ -1139,21 +1058,6 @@ 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 attachment_service = AttachmentService::from_env(env);
-            attachment_service
-                .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"),
@@ -1170,11 +1074,11 @@ async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
     Ok(())
 }
 
-async fn handle_attachment_service(
+async fn handle_storage_controller(
     sub_match: &ArgMatches,
     env: &local_env::LocalEnv,
 ) -> Result<()> {
-    let svc = AttachmentService::from_env(env);
+    let svc = StorageController::from_env(env);
     match sub_match.subcommand() {
         Some(("start", _start_match)) => {
             if let Err(e) = svc.start().await {
@@ -1194,8 +1098,8 @@ async fn handle_attachment_service(
                 exit(1);
             }
         }
-        Some((sub_name, _)) => bail!("Unexpected attachment_service subcommand '{}'", sub_name),
-        None => bail!("no attachment_service subcommand provided"),
+        Some((sub_name, _)) => bail!("Unexpected storage_controller subcommand '{}'", sub_name),
+        None => bail!("no storage_controller subcommand provided"),
     }
     Ok(())
 }
@@ -1280,11 +1184,11 @@ async fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
 
     broker::start_broker_process(env).await?;
 
-    // Only start the attachment service if the pageserver is configured to need it
+    // Only start the storage controller if the pageserver is configured to need it
     if env.control_plane_api.is_some() {
-        let attachment_service = AttachmentService::from_env(env);
-        if let Err(e) = attachment_service.start().await {
-            eprintln!("attachment_service start failed: {:#}", e);
+        let storage_controller = StorageController::from_env(env);
+        if let Err(e) = storage_controller.start().await {
+            eprintln!("storage_controller start failed: {:#}", e);
             try_stop_all(env, true).await;
             exit(1);
         }
@@ -1293,7 +1197,7 @@ async fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
     for ps_conf in &env.pageservers {
         let pageserver = PageServerNode::from_env(env, ps_conf);
         if let Err(e) = pageserver
-            .start(&pageserver_config_overrides(sub_match), true)
+            .start(&pageserver_config_overrides(sub_match))
             .await
         {
             eprintln!("pageserver {} start failed: {:#}", ps_conf.id, e);
@@ -1356,9 +1260,9 @@ async fn try_stop_all(env: &local_env::LocalEnv, immediate: bool) {
     }
 
     if env.control_plane_api.is_some() {
-        let attachment_service = AttachmentService::from_env(env);
-        if let Err(e) = attachment_service.stop(immediate).await {
-            eprintln!("attachment service stop failed: {e:#}");
+        let storage_controller = StorageController::from_env(env);
+        if let Err(e) = storage_controller.stop(immediate).await {
+            eprintln!("storage controller stop failed: {e:#}");
         }
     }
 }
@@ -1575,18 +1479,6 @@ fn cli() -> Command {
             .subcommand(Command::new("config")
                 .arg(tenant_id_arg.clone())
                 .arg(Arg::new("config").short('c').num_args(1).action(ArgAction::Append).required(false)))
-            .subcommand(Command::new("migrate")
-                .about("Migrate a tenant from one pageserver to another")
-                .arg(tenant_id_arg.clone())
-                .arg(pageserver_id_arg.clone()))
-            .subcommand(Command::new("status")
-                .about("Human readable summary of the tenant's shards and attachment locations")
-                .arg(tenant_id_arg.clone()))
-            .subcommand(Command::new("shard-split")
-                .about("Increase the number of shards in the tenant")
-                .arg(tenant_id_arg.clone())
-                .arg(Arg::new("shard-count").value_parser(value_parser!(u8)).long("shard-count").action(ArgAction::Set).help("Number of shards in the new tenant (default 1)"))
-                )
         )
         .subcommand(
             Command::new("pageserver")
@@ -1596,11 +1488,7 @@ fn cli() -> Command {
                 .subcommand(Command::new("status"))
                 .subcommand(Command::new("start")
                     .about("Start local pageserver")
-                    .arg(pageserver_config_args.clone()).arg(Arg::new("register")
-                    .long("register")
-                    .default_value("true").required(false)
-                    .value_parser(value_parser!(bool))
-                    .value_name("register"))
+                    .arg(pageserver_config_args.clone())
                 )
                 .subcommand(Command::new("stop")
                     .about("Stop local pageserver")
@@ -1610,17 +1498,11 @@ 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("attachment_service")
+            Command::new("storage_controller")
                 .arg_required_else_help(true)
-                .about("Manage attachment_service")
+                .about("Manage storage_controller")
                 .subcommand(Command::new("start").about("Start local pageserver").arg(pageserver_config_args.clone()))
                 .subcommand(Command::new("stop").about("Stop local pageserver")
                             .arg(stop_mode_arg.clone()))

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 the data directory, and
+//! the basebackup from the pageserver to initialize the data directory, and
 //! finally launches the PostgreSQL process. It watches the PostgreSQL process
 //! until it exits.
 //!
@@ -57,9 +57,9 @@ use serde::{Deserialize, Serialize};
 use url::Host;
 use utils::id::{NodeId, TenantId, TimelineId};
 
-use crate::attachment_service::AttachmentService;
 use crate::local_env::LocalEnv;
 use crate::postgresql_conf::PostgresConf;
+use crate::storage_controller::StorageController;
 
 use compute_api::responses::{ComputeState, ComputeStatus};
 use compute_api::spec::{Cluster, ComputeFeature, ComputeMode, ComputeSpec};
@@ -750,17 +750,17 @@ impl Endpoint {
         let postgresql_conf = self.read_postgresql_conf()?;
         spec.cluster.postgresql_conf = Some(postgresql_conf);
 
-        // If we weren't given explicit pageservers, query the attachment service
+        // If we weren't given explicit pageservers, query the storage controller
         if pageservers.is_empty() {
-            let attachment_service = AttachmentService::from_env(&self.env);
-            let locate_result = attachment_service.tenant_locate(self.tenant_id).await?;
+            let storage_controller = StorageController::from_env(&self.env);
+            let locate_result = storage_controller.tenant_locate(self.tenant_id).await?;
             pageservers = locate_result
                 .shards
                 .into_iter()
                 .map(|shard| {
                     (
                         Host::parse(&shard.listen_pg_addr)
-                            .expect("Attachment service reported bad hostname"),
+                            .expect("Storage controller reported bad hostname"),
                         shard.listen_pg_port,
                     )
                 })
@@ -774,7 +774,10 @@ impl Endpoint {
             spec.shard_stripe_size = stripe_size.map(|s| s.0 as usize);
         }
 
-        let client = reqwest::Client::new();
+        let client = reqwest::Client::builder()
+            .timeout(Duration::from_secs(30))
+            .build()
+            .unwrap();
         let response = client
             .post(format!(
                 "http://{}:{}/configure",

control_plane/src/lib.rs

@@ -6,7 +6,6 @@
 //! local installations.
 #![deny(clippy::undocumented_unsafe_blocks)]
 
-pub mod attachment_service;
 mod background_process;
 pub mod broker;
 pub mod endpoint;
@@ -14,3 +13,4 @@ pub mod local_env;
 pub mod pageserver;
 pub mod postgresql_conf;
 pub mod safekeeper;
+pub mod storage_controller;

control_plane/src/local_env.rs

@@ -72,13 +72,13 @@ pub struct LocalEnv {
     #[serde(default)]
     pub safekeepers: Vec<SafekeeperConf>,
 
-    // Control plane upcall API for pageserver: if None, we will not run attachment_service.  If set, this will
+    // Control plane upcall API for pageserver: if None, we will not run storage_controller  If set, this will
     // be propagated into each pageserver's configuration.
     #[serde(default)]
     pub control_plane_api: Option<Url>,
 
-    // Control plane upcall API for attachment service.  If set, this will be propagated into the
-    // attachment service's configuration.
+    // Control plane upcall API for storage controller.  If set, this will be propagated into the
+    // storage controller's configuration.
     #[serde(default)]
     pub control_plane_compute_hook_api: Option<Url>,
 
@@ -114,7 +114,7 @@ impl NeonBroker {
 }
 
 #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
-#[serde(default)]
+#[serde(default, deny_unknown_fields)]
 pub struct PageServerConf {
     // node id
     pub id: NodeId,
@@ -126,6 +126,9 @@ pub struct PageServerConf {
     // auth type used for the PG and HTTP ports
     pub pg_auth_type: AuthType,
     pub http_auth_type: AuthType,
+
+    pub(crate) virtual_file_io_engine: Option<String>,
+    pub(crate) get_vectored_impl: Option<String>,
 }
 
 impl Default for PageServerConf {
@@ -136,6 +139,8 @@ impl Default for PageServerConf {
             listen_http_addr: String::new(),
             pg_auth_type: AuthType::Trust,
             http_auth_type: AuthType::Trust,
+            virtual_file_io_engine: None,
+            get_vectored_impl: None,
         }
     }
 }
@@ -227,10 +232,10 @@ impl LocalEnv {
         self.neon_distrib_dir.join("pageserver")
     }
 
-    pub fn attachment_service_bin(&self) -> PathBuf {
-        // Irrespective of configuration, attachment service binary is always
+    pub fn storage_controller_bin(&self) -> PathBuf {
+        // Irrespective of configuration, storage controller binary is always
         // run from the same location as neon_local.  This means that for compatibility
-        // tests that run old pageserver/safekeeper, they still run latest attachment service.
+        // tests that run old pageserver/safekeeper, they still run latest storage controller.
         let neon_local_bin_dir = env::current_exe().unwrap().parent().unwrap().to_owned();
         neon_local_bin_dir.join("storage_controller")
     }

control_plane/src/pageserver.rs

@@ -17,7 +17,6 @@ use std::time::Duration;
 use anyhow::{bail, Context};
 use camino::Utf8PathBuf;
 use futures::SinkExt;
-use pageserver_api::controller_api::NodeRegisterRequest;
 use pageserver_api::models::{
     self, LocationConfig, ShardParameters, TenantHistorySize, TenantInfo, TimelineInfo,
 };
@@ -31,7 +30,6 @@ use utils::{
     lsn::Lsn,
 };
 
-use crate::attachment_service::AttachmentService;
 use crate::local_env::PageServerConf;
 use crate::{background_process, local_env::LocalEnv};
 
@@ -80,18 +78,39 @@ impl PageServerNode {
     ///
     /// These all end up on the command line of the `pageserver` binary.
     fn neon_local_overrides(&self, cli_overrides: &[&str]) -> Vec<String> {
-        let id = format!("id={}", self.conf.id);
         // FIXME: the paths should be shell-escaped to handle paths with spaces, quotas etc.
         let pg_distrib_dir_param = format!(
             "pg_distrib_dir='{}'",
             self.env.pg_distrib_dir_raw().display()
         );
 
-        let http_auth_type_param = format!("http_auth_type='{}'", self.conf.http_auth_type);
-        let listen_http_addr_param = format!("listen_http_addr='{}'", self.conf.listen_http_addr);
+        let PageServerConf {
+            id,
+            listen_pg_addr,
+            listen_http_addr,
+            pg_auth_type,
+            http_auth_type,
+            virtual_file_io_engine,
+            get_vectored_impl,
+        } = &self.conf;
 
-        let pg_auth_type_param = format!("pg_auth_type='{}'", self.conf.pg_auth_type);
-        let listen_pg_addr_param = format!("listen_pg_addr='{}'", self.conf.listen_pg_addr);
+        let id = format!("id={}", id);
+
+        let http_auth_type_param = format!("http_auth_type='{}'", http_auth_type);
+        let listen_http_addr_param = format!("listen_http_addr='{}'", listen_http_addr);
+
+        let pg_auth_type_param = format!("pg_auth_type='{}'", pg_auth_type);
+        let listen_pg_addr_param = format!("listen_pg_addr='{}'", listen_pg_addr);
+        let virtual_file_io_engine = if let Some(virtual_file_io_engine) = virtual_file_io_engine {
+            format!("virtual_file_io_engine='{virtual_file_io_engine}'")
+        } else {
+            String::new()
+        };
+        let get_vectored_impl = if let Some(get_vectored_impl) = get_vectored_impl {
+            format!("get_vectored_impl='{get_vectored_impl}'")
+        } else {
+            String::new()
+        };
 
         let broker_endpoint_param = format!("broker_endpoint='{}'", self.env.broker.client_url());
 
@@ -103,6 +122,8 @@ impl PageServerNode {
             listen_http_addr_param,
             listen_pg_addr_param,
             broker_endpoint_param,
+            virtual_file_io_engine,
+            get_vectored_impl,
         ];
 
         if let Some(control_plane_api) = &self.env.control_plane_api {
@@ -111,9 +132,9 @@ impl PageServerNode {
                 control_plane_api.as_str()
             ));
 
-            // Attachment service uses the same auth as pageserver: if JWT is enabled
+            // Storage controller uses the same auth as pageserver: if JWT is enabled
             // for us, we will also need it to talk to them.
-            if matches!(self.conf.http_auth_type, AuthType::NeonJWT) {
+            if matches!(http_auth_type, AuthType::NeonJWT) {
                 let jwt_token = self
                     .env
                     .generate_auth_token(&Claims::new(None, Scope::GenerationsApi))
@@ -131,8 +152,7 @@ impl PageServerNode {
             ));
         }
 
-        if self.conf.http_auth_type != AuthType::Trust || self.conf.pg_auth_type != AuthType::Trust
-        {
+        if *http_auth_type != AuthType::Trust || *pg_auth_type != AuthType::Trust {
             // Keys are generated in the toplevel repo dir, pageservers' workdirs
             // are one level below that, so refer to keys with ../
             overrides.push("auth_validation_public_key_path='../auth_public_key.pem'".to_owned());
@@ -163,8 +183,8 @@ impl PageServerNode {
             .expect("non-Unicode path")
     }
 
-    pub async fn start(&self, config_overrides: &[&str], register: bool) -> anyhow::Result<()> {
-        self.start_node(config_overrides, false, register).await
+    pub async fn start(&self, config_overrides: &[&str]) -> anyhow::Result<()> {
+        self.start_node(config_overrides, false).await
     }
 
     fn pageserver_init(&self, config_overrides: &[&str]) -> anyhow::Result<()> {
@@ -202,6 +222,28 @@ impl PageServerNode {
             String::from_utf8_lossy(&init_output.stderr),
         );
 
+        // Write metadata file, used by pageserver on startup to register itself with
+        // the storage controller
+        let metadata_path = datadir.join("metadata.json");
+
+        let (_http_host, http_port) =
+            parse_host_port(&self.conf.listen_http_addr).expect("Unable to parse listen_http_addr");
+        let http_port = http_port.unwrap_or(9898);
+        // Intentionally hand-craft JSON: this acts as an implicit format compat test
+        // in case the pageserver-side structure is edited, and reflects the real life
+        // situation: the metadata is written by some other script.
+        std::fs::write(
+            metadata_path,
+            serde_json::to_vec(&serde_json::json!({
+                "host": "localhost",
+                "port": self.pg_connection_config.port(),
+                "http_host": "localhost",
+                "http_port": http_port,
+            }))
+            .unwrap(),
+        )
+        .expect("Failed to write metadata file");
+
         Ok(())
     }
 
@@ -209,27 +251,7 @@ impl PageServerNode {
         &self,
         config_overrides: &[&str],
         update_config: bool,
-        register: bool,
     ) -> anyhow::Result<()> {
-        // Register the node with the storage controller before starting pageserver: pageserver must be registered to
-        // successfully call /re-attach and finish starting up.
-        if register {
-            let attachment_service = AttachmentService::from_env(&self.env);
-            let (pg_host, pg_port) =
-                parse_host_port(&self.conf.listen_pg_addr).expect("Unable to parse listen_pg_addr");
-            let (http_host, http_port) = parse_host_port(&self.conf.listen_http_addr)
-                .expect("Unable to parse listen_http_addr");
-            attachment_service
-                .node_register(NodeRegisterRequest {
-                    node_id: self.conf.id,
-                    listen_pg_addr: pg_host.to_string(),
-                    listen_pg_port: pg_port.unwrap_or(5432),
-                    listen_http_addr: http_host.to_string(),
-                    listen_http_port: http_port.unwrap_or(80),
-                })
-                .await?;
-        }
-
         // TODO: using a thread here because start_process() is not async but we need to call check_status()
         let datadir = self.repo_path();
         print!(
@@ -367,6 +389,10 @@ 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")
@@ -479,6 +505,12 @@ 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")
@@ -554,13 +586,6 @@ impl PageServerNode {
         Ok(self.http_client.list_timelines(*tenant_shard_id).await?)
     }
 
-    pub async fn tenant_secondary_download(&self, tenant_id: &TenantShardId) -> anyhow::Result<()> {
-        Ok(self
-            .http_client
-            .tenant_secondary_download(*tenant_id)
-            .await?)
-    }
-
     pub async fn timeline_create(
         &self,
         tenant_shard_id: TenantShardId,

control_plane/src/storage_controller.rs

@@ -10,7 +10,7 @@ use pageserver_api::{
         TenantCreateRequest, TenantShardSplitRequest, TenantShardSplitResponse,
         TimelineCreateRequest, TimelineInfo,
     },
-    shard::TenantShardId,
+    shard::{ShardStripeSize, TenantShardId},
 };
 use pageserver_client::mgmt_api::ResponseErrorMessageExt;
 use postgres_backend::AuthType;
@@ -24,7 +24,7 @@ use utils::{
     id::{NodeId, TenantId},
 };
 
-pub struct AttachmentService {
+pub struct StorageController {
     env: LocalEnv,
     listen: String,
     path: Utf8PathBuf,
@@ -36,7 +36,10 @@ pub struct AttachmentService {
 
 const COMMAND: &str = "storage_controller";
 
-const ATTACHMENT_SERVICE_POSTGRES_VERSION: u32 = 16;
+const STORAGE_CONTROLLER_POSTGRES_VERSION: u32 = 16;
+
+// Use a shorter pageserver unavailability interval than the default to speed up tests.
+const NEON_LOCAL_MAX_UNAVAILABLE_INTERVAL: std::time::Duration = std::time::Duration::from_secs(10);
 
 #[derive(Serialize, Deserialize)]
 pub struct AttachHookRequest {
@@ -59,7 +62,7 @@ pub struct InspectResponse {
     pub attachment: Option<(u32, NodeId)>,
 }
 
-impl AttachmentService {
+impl StorageController {
     pub fn from_env(env: &LocalEnv) -> Self {
         let path = Utf8PathBuf::from_path_buf(env.base_data_dir.clone())
             .unwrap()
@@ -136,27 +139,27 @@ impl AttachmentService {
     }
 
     fn pid_file(&self) -> Utf8PathBuf {
-        Utf8PathBuf::from_path_buf(self.env.base_data_dir.join("attachment_service.pid"))
+        Utf8PathBuf::from_path_buf(self.env.base_data_dir.join("storage_controller.pid"))
             .expect("non-Unicode path")
     }
 
-    /// PIDFile for the postgres instance used to store attachment service state
+    /// PIDFile for the postgres instance used to store storage controller state
     fn postgres_pid_file(&self) -> Utf8PathBuf {
         Utf8PathBuf::from_path_buf(
             self.env
                 .base_data_dir
-                .join("attachment_service_postgres.pid"),
+                .join("storage_controller_postgres.pid"),
         )
         .expect("non-Unicode path")
     }
 
     /// Find the directory containing postgres binaries, such as `initdb` and `pg_ctl`
     ///
-    /// This usually uses ATTACHMENT_SERVICE_POSTGRES_VERSION of postgres, but will fall back
+    /// This usually uses STORAGE_CONTROLLER_POSTGRES_VERSION of postgres, but will fall back
     /// to other versions if that one isn't found.  Some automated tests create circumstances
     /// where only one version is available in pg_distrib_dir, such as `test_remote_extensions`.
     pub async fn get_pg_bin_dir(&self) -> anyhow::Result<Utf8PathBuf> {
-        let prefer_versions = [ATTACHMENT_SERVICE_POSTGRES_VERSION, 15, 14];
+        let prefer_versions = [STORAGE_CONTROLLER_POSTGRES_VERSION, 15, 14];
 
         for v in prefer_versions {
             let path = Utf8PathBuf::from_path_buf(self.env.pg_bin_dir(v)?).unwrap();
@@ -189,7 +192,7 @@ impl AttachmentService {
     ///
     /// Returns the database url
     pub async fn setup_database(&self) -> anyhow::Result<String> {
-        const DB_NAME: &str = "attachment_service";
+        const DB_NAME: &str = "storage_controller";
         let database_url = format!("postgresql://localhost:{}/{DB_NAME}", self.postgres_port);
 
         let pg_bin_dir = self.get_pg_bin_dir().await?;
@@ -219,10 +222,10 @@ impl AttachmentService {
     }
 
     pub async fn start(&self) -> anyhow::Result<()> {
-        // Start a vanilla Postgres process used by the attachment service for persistence.
+        // Start a vanilla Postgres process used by the storage controller for persistence.
         let pg_data_path = Utf8PathBuf::from_path_buf(self.env.base_data_dir.clone())
             .unwrap()
-            .join("attachment_service_db");
+            .join("storage_controller_db");
         let pg_bin_dir = self.get_pg_bin_dir().await?;
         let pg_log_path = pg_data_path.join("postgres.log");
 
@@ -245,7 +248,7 @@ impl AttachmentService {
             .await?;
         };
 
-        println!("Starting attachment service database...");
+        println!("Starting storage controller database...");
         let db_start_args = [
             "-w",
             "-D",
@@ -256,7 +259,7 @@ impl AttachmentService {
         ];
 
         background_process::start_process(
-            "attachment_service_db",
+            "storage_controller_db",
             &self.env.base_data_dir,
             pg_bin_dir.join("pg_ctl").as_std_path(),
             db_start_args,
@@ -269,13 +272,18 @@ impl AttachmentService {
         // Run migrations on every startup, in case something changed.
         let database_url = self.setup_database().await?;
 
+        let max_unavailable: humantime::Duration = NEON_LOCAL_MAX_UNAVAILABLE_INTERVAL.into();
+
         let mut args = vec![
             "-l",
             &self.listen,
             "-p",
             self.path.as_ref(),
+            "--dev",
             "--database-url",
             &database_url,
+            "--max-unavailable-interval",
+            &max_unavailable.to_string(),
         ]
         .into_iter()
         .map(|s| s.to_string())
@@ -300,7 +308,7 @@ impl AttachmentService {
         background_process::start_process(
             COMMAND,
             &self.env.base_data_dir,
-            &self.env.attachment_service_bin(),
+            &self.env.storage_controller_bin(),
             args,
             [(
                 "NEON_REPO_DIR".to_string(),
@@ -322,10 +330,10 @@ impl AttachmentService {
     pub async fn stop(&self, immediate: bool) -> anyhow::Result<()> {
         background_process::stop_process(immediate, COMMAND, &self.pid_file())?;
 
-        let pg_data_path = self.env.base_data_dir.join("attachment_service_db");
+        let pg_data_path = self.env.base_data_dir.join("storage_controller_db");
         let pg_bin_dir = self.get_pg_bin_dir().await?;
 
-        println!("Stopping attachment service database...");
+        println!("Stopping storage controller database...");
         let pg_stop_args = ["-D", &pg_data_path.to_string_lossy(), "stop"];
         let stop_status = Command::new(pg_bin_dir.join("pg_ctl"))
             .args(pg_stop_args)
@@ -344,10 +352,10 @@ impl AttachmentService {
             // fine that stop failed.  Otherwise it is an error that stop failed.
             const PG_STATUS_NOT_RUNNING: i32 = 3;
             if Some(PG_STATUS_NOT_RUNNING) == status_exitcode.code() {
-                println!("Attachment service data base is already stopped");
+                println!("Storage controller database is already stopped");
                 return Ok(());
             } else {
-                anyhow::bail!("Failed to stop attachment service database: {stop_status}")
+                anyhow::bail!("Failed to stop storage controller database: {stop_status}")
             }
         }
 
@@ -368,7 +376,7 @@ impl AttachmentService {
         }
     }
 
-    /// Simple HTTP request wrapper for calling into attachment service
+    /// Simple HTTP request wrapper for calling into storage controller
     async fn dispatch<RQ, RS>(
         &self,
         method: hyper::Method,
@@ -468,7 +476,7 @@ impl AttachmentService {
     pub async fn tenant_locate(&self, tenant_id: TenantId) -> anyhow::Result<TenantLocateResponse> {
         self.dispatch::<(), _>(
             Method::GET,
-            format!("control/v1/tenant/{tenant_id}/locate"),
+            format!("debug/v1/tenant/{tenant_id}/locate"),
             None,
         )
         .await
@@ -496,11 +504,15 @@ impl AttachmentService {
         &self,
         tenant_id: TenantId,
         new_shard_count: u8,
+        new_stripe_size: Option<ShardStripeSize>,
     ) -> anyhow::Result<TenantShardSplitResponse> {
         self.dispatch(
             Method::PUT,
             format!("control/v1/tenant/{tenant_id}/shard_split"),
-            Some(TenantShardSplitRequest { new_shard_count }),
+            Some(TenantShardSplitRequest {
+                new_shard_count,
+                new_stripe_size,
+            }),
         )
         .await
     }

control_plane/storcon_cli/Cargo.toml

@@ -0,0 +1,23 @@
+[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

@@ -0,0 +1,587 @@
+use std::{collections::HashMap, str::FromStr};
+
+use clap::{Parser, Subcommand};
+use hyper::Method;
+use pageserver_api::{
+    controller_api::{
+        NodeAvailabilityWrapper, NodeDescribeResponse, ShardSchedulingPolicy,
+        TenantDescribeResponse, TenantPolicyRequest,
+    },
+    models::{
+        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,
+    },
+}
+
+#[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}");
+        }
+    }
+
+    Ok(())
+}

diesel.toml

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

docs/authentication.md

@@ -70,9 +70,9 @@ Should only be used e.g. for status check/tenant creation/list.
 Should only be used e.g. for status check.
 Currently also used for connection from any pageserver to any safekeeper.
 
-"generations_api": Provides access to the upcall APIs served by the attachment service or the control plane.
+"generations_api": Provides access to the upcall APIs served by the storage controller or the control plane.
 
-"admin": Provides access to the control plane and admin APIs of the attachment service.
+"admin": Provides access to the control plane and admin APIs of the storage controller.
 
 ### CLI
 CLI generates a key pair during call to `neon_local init` with the following commands:

docs/rfcs/031-sharding-static.md

@@ -0,0 +1,408 @@
+# Sharding Phase 1: Static Key-space Sharding
+
+## Summary
+
+To enable databases with sizes approaching the capacity of a pageserver's disk,
+it is necessary to break up the storage for the database, or _shard_ it.
+
+Sharding in general is a complex area. This RFC aims to define an initial
+capability that will permit creating large-capacity databases using a static configuration
+defined at time of Tenant creation.
+
+## Motivation
+
+Currently, all data for a Tenant, including all its timelines, is stored on a single
+pageserver. The local storage required may be several times larger than the actual
+database size, due to LSM write inflation.
+
+If a database is larger than what one pageserver can hold, then it becomes impossible
+for the pageserver to hold it in local storage, as it must do to provide service to
+clients.
+
+### Prior art
+
+In Neon:
+
+- Layer File Spreading: https://www.notion.so/neondatabase/One-Pager-Layer-File-Spreading-Konstantin-21fd9b11b618475da5f39c61dd8ab7a4
+- Layer File SPreading: https://www.notion.so/neondatabase/One-Pager-Layer-File-Spreading-Christian-eb6b64182a214e11b3fceceee688d843
+- Key Space partitioning: https://www.notion.so/neondatabase/One-Pager-Key-Space-Partitioning-Stas-8e3a28a600a04a25a68523f42a170677
+
+Prior art in other distributed systems is too broad to capture here: pretty much
+any scale out storage system does something like this.
+
+## Requirements
+
+- Enable creating a large (for example, 16TiB) database without requiring dedicated
+  pageserver nodes.
+- Share read/write bandwidth costs for large databases across pageservers, as well
+  as storage capacity, in order to avoid large capacity databases acting as I/O hotspots
+  that disrupt service to other tenants.
+- Our data distribution scheme should handle sparse/nonuniform keys well, since postgres
+  does not write out a single contiguous ranges of page numbers.
+
+_Note: the definition of 'large database' is arbitrary, but the lower bound is to ensure that a database
+that a user might create on a current-gen enterprise SSD should also work well on
+Neon. The upper bound is whatever postgres can handle: i.e. we must make sure that the
+pageserver backend is not the limiting factor in the database size_.
+
+## Non Goals
+
+- Independently distributing timelines within the same tenant. If a tenant has many
+  timelines, then sharding may be a less efficient mechanism for distributing load than
+  sharing out timelines between pageservers.
+- Distributing work in the LSN dimension: this RFC focuses on the Key dimension only,
+  based on the idea that separate mechanisms will make sense for each dimension.
+
+## Impacted Components
+
+pageserver, control plane, postgres/smgr
+
+## Terminology
+
+**Key**: a postgres page number, qualified by relation. In the sense that the pageserver is a versioned key-value store,
+the page number is the key in that store. `Key` is a literal data type in existing code.
+
+**LSN dimension**: this just means the range of LSNs (history), when talking about the range
+of keys and LSNs as a two dimensional space.
+
+## Implementation
+
+### Key sharding vs. LSN sharding
+
+When we think of sharding across the two dimensional key/lsn space, this is an
+opportunity to think about how the two dimensions differ:
+
+- Sharding the key space distributes the _write_ workload of ingesting data
+  and compacting. This work must be carefully managed so that exactly one
+  node owns a given key.
+- Sharding the LSN space distributes the _historical read_ workload. This work
+  can be done by anyone without any special coordination, as long as they can
+  see the remote index and layers.
+
+The key sharding is the harder part, and also the more urgent one, to support larger
+capacity databases. Because distributing historical LSN read work is a relatively
+simpler problem that most users don't have, we defer it to future work. It is anticipated
+that some quite simple P2P offload model will enable distributing work for historical
+reads: a node which is low on space can call out to peer to ask it to download and
+serve reads from a historical layer.
+
+### Key mapping scheme
+
+Having decided to focus on key sharding, we must next decide how we will map
+keys to shards. It is proposed to use a "wide striping" approach, to obtain a good compromise
+between data locality and avoiding entire large relations mapping to the same shard.
+
+We will define two spaces:
+
+- Key space: unsigned integer
+- Shard space: integer from 0 to N-1, where we have N shards.
+
+### Key -> Shard mapping
+
+Keys are currently defined in the pageserver's getpage@lsn interface as follows:
+
+```
+pub struct Key {
+    pub field1: u8,
+    pub field2: u32,
+    pub field3: u32,
+    pub field4: u32,
+    pub field5: u8,
+    pub field6: u32,
+}
+
+
+fn rel_block_to_key(rel: RelTag, blknum: BlockNumber) -> Key {
+    Key {
+        field1: 0x00,
+        field2: rel.spcnode,
+        field3: rel.dbnode,
+        field4: rel.relnode,
+        field5: rel.forknum,
+        field6: blknum,
+    }
+}
+```
+
+_Note: keys for relation metadata are ignored here, as this data will be mirrored to all
+shards. For distribution purposes, we only care about user data keys_
+
+The properties we want from our Key->Shard mapping are:
+
+- Locality in `blknum`, such that adjacent `blknum` will usually map to
+  the same stripe and consequently land on the same shard, even though the overall
+  collection of blocks in a relation will be spread over many stripes and therefore
+  many shards.
+- Avoid the same blknum on different relations landing on the same stripe, so that
+  with many small relations we do not end up aliasing data to the same stripe/shard.
+- Avoid vulnerability to aliasing in the values of relation identity fields, such that
+  if there are patterns in the value of `relnode`, these do not manifest as patterns
+  in data placement.
+
+To accomplish this, the blknum is used to select a stripe, and stripes are
+assigned to shards in a pseudorandom order via a hash. The motivation for
+pseudo-random distribution (rather than sequential mapping of stripe to shard)
+is to avoid I/O hotspots when sequentially reading multiple relations: we don't want
+all relations' stripes to touch pageservers in the same order.
+
+To map a `Key` to a shard:
+
+- Hash the `Key` field 4 (relNode).
+- Divide field 6 (`blknum`) field by the stripe size in pages, and combine the
+  hash of this with the hash from the previous step.
+- The total hash modulo the shard count gives the shard holding this key.
+
+Why don't we use the other fields in the Key?
+
+- We ignore `forknum` for key mapping, because it distinguishes different classes of data
+  in the same relation, and we would like to keep the data in a relation together.
+- We would like to use spcNode and dbNode, but cannot. Postgres database creation operations can refer to an existing database as a template, such that the created
+  database's blocks differ only by spcNode and dbNode from the original. To enable running
+  this type of creation without cross-pageserver communication, we must ensure that these
+  blocks map to the same shard -- we do this by excluding spcNode and dbNode from the hash.
+
+### Data placement examples
+
+For example, consider the extreme large databases cases of postgres data layout in a system with 8 shards
+and a stripe size of 32k pages:
+
+- A single large relation: `blknum` division will break the data up into 4096
+  stripes, which will be scattered across the shards.
+- 4096 relations of of 32k pages each: each relation will map to exactly one stripe,
+  and that stripe will be placed according to the hash of the key fields 4. The
+  data placement will be statistically uniform across shards.
+
+Data placement will be more uneven on smaller databases:
+
+- A tenant with 2 shards and 2 relations of one stripe size each: there is a 50% chance
+  that both relations land on the same shard and no data lands on the other shard.
+- A tenant with 8 shards and one relation of size 12 stripes: 4 shards will have double
+  the data of the other four shards.
+
+These uneven cases for small amounts of data do not matter, as long as the stripe size
+is an order of magnitude smaller than the amount of data we are comfortable holding
+in a single shard: if our system handles shard sizes up to 10-100GB, then it is not an issue if
+a tenant has some shards with 256MB size and some shards with 512MB size, even though
+the standard deviation of shard size within the tenant is very high. Our key mapping
+scheme provides a statistical guarantee that as the tenant's overall data size increases,
+uniformity of placement will improve.
+
+### Important Types
+
+#### `ShardIdentity`
+
+Provides the information needed to know whether a particular key belongs
+to a particular shard:
+
+- Layout version
+- Stripe size
+- Shard count
+- Shard index
+
+This structure's size is constant. Note that if we had used a differnet key
+mapping scheme such as consistent hashing with explicit hash ranges assigned
+to each shard, then the ShardIdentity's size would grow with the shard count: the simpler
+key mapping scheme used here enables a small fixed size ShardIdentity.
+
+### Pageserver changes
+
+#### Structural
+
+Everywhere the Pageserver currently deals with Tenants, it will move to dealing with
+`TenantShard`s, which are just a `Tenant` plus a `ShardIdentity` telling it which part
+of the keyspace it owns. An un-sharded tenant is just a `TenantShard` whose `ShardIdentity`
+covers the whole keyspace.
+
+When the pageserver writes layers and index_part.json to remote storage, it must
+include the shard index & count in the name, to avoid collisions (the count is
+necessary for future-proofing: the count will vary in time). These keys
+will also include a generation number: the [generation numbers](025-generation-numbers.md) system will work
+exactly the same for TenantShards as it does for Tenants today: each shard will have
+its own generation number.
+
+#### Storage Format: Keys
+
+For tenants with >1 shard, layer files implicitly become sparse: within the key
+range described in the layer name, the layer file for a shard will only hold the
+content relevant to stripes assigned to the shard.
+
+For this reason, the LayerFileName within a tenant is no longer unique: different shards
+may use the same LayerFileName to refer to different data. We may solve this simply
+by including the shard number in the keys used for layers.
+
+The shard number will be included as a prefix (as part of tenant ID), like this:
+
+`pageserver/v1/tenants/<tenant_id>-<shard_number><shard_count>/timelines/<timeline id>/<layer file name>-<generation>`
+
+`pageserver/v1/tenants/<tenant_id>-<shard_number><shard_count>/timelines/<timeline id>/index_part.json-<generation>`
+
+Reasons for this particular format:
+
+- Use of a prefix is convenient for implementation (no need to carry the shard ID everywhere
+  we construct a layer file name), and enables efficient listing of index_parts within
+  a particular shard-timeline prefix.
+- Including the shard _count_ as well as shard number means that in future when we implement
+  shard splitting, it will be possible for a parent shard and one of its children to write
+  the same layer file without a name collision. For example, a parent shard 0_1 might split
+  into two (0_2, 1_2), and in the process of splitting shard 0_2 could write a layer or index_part
+  that is distinct from what shard 0_1 would have written at the same place.
+
+In practice, we expect shard counts to be relatively small, so a `u8` will be sufficient,
+and therefore the shard part of the path can be a fixed-length hex string like `{:02X}{:02X}`,
+for example a single-shard tenant's prefix will be `0001`.
+
+For backward compatibility, we may define a special `ShardIdentity` that has shard_count==0,
+and use this as a cue to construct paths with no prefix at all.
+
+#### Storage Format: Indices
+
+In the phase 1 described in this RFC, shards only reference layers they write themselves. However,
+when we implement shard splitting in future, it will be useful to enable shards to reference layers
+written by other shards (specifically the parent shard during a split), so that shards don't
+have to exhaustively copy all data into their own shard-prefixed keys.
+
+To enable this, the `IndexPart` structure will be extended to store the (shard number, shard count)
+tuple on each layer, such that it can construct paths for layers written by other shards. This
+naturally raises the question of who "owns" such layers written by ancestral shards: this problem
+will be addressed in phase 2.
+
+For backward compatibility, any index entry without shard information will be assumed to be
+in the legacy shardidentity.
+
+#### WAL Ingest
+
+In Phase 1, all shards will subscribe to the safekeeper to download WAL content. They will filter
+it down to the pages relevant to their shard:
+
+- For ordinary user data writes, only retain a write if it matches the ShardIdentity
+- For metadata describing relations etc, all shards retain these writes.
+
+The pageservers must somehow give the safekeeper correct feedback on remote_consistent_lsn:
+one solution here is for the 0th shard to periodically peek at the IndexParts for all the other shards,
+and have only the 0th shard populate remote_consistent_lsn. However, this is relatively
+expensive: if the safekeeper can be made shard-aware then it could be taught to use
+the max() of all shards' remote_consistent_lsns to decide when to trim the WAL.
+
+#### Compaction/GC
+
+No changes needed.
+
+The pageserver doesn't have to do anything special during compaction
+or GC. It is implicitly operating on the subset of keys that map to its ShardIdentity.
+This will result in sparse layer files, containing keys only in the stripes that this
+shard owns. Where optimizations currently exist in compaction for spotting "gaps" in
+the key range, these should be updated to ignore gaps that are due to sharding, to
+avoid spuriously splitting up layers ito stripe-sized pieces.
+
+### Compute Endpoints
+
+Compute endpoints will need to:
+
+- Accept a vector of connection strings as part of their configuration from the control plane
+- Route pageserver requests according to mapping the hash of key to the correct
+  entry in the vector of connection strings.
+
+Doing this in compute rather than routing requests via a single pageserver is
+necessary to enable sharding tenants without adding latency from extra hops.
+
+### Control Plane
+
+Tenants, or _Projects_ in the control plane, will each own a set of TenantShards (this will
+be 1 for small tenants). Logic for placement of tenant shards is just the same as the current logic for placing
+tenants.
+
+Tenant lifecycle operations like deletion will require fanning-out to all the shards
+in the tenant. The same goes for timeline creation and deletion: a timeline should
+not be considered created until it has been created in all shards.
+
+#### Selectively enabling sharding for large tenants
+
+Initially, we will explicitly enable sharding for large tenants only.
+
+In future, this hint mechanism will become optional when we implement automatic
+re-sharding of tenants.
+
+## Future Phases
+
+This section exists to indicate what will likely come next after this phase.
+
+Phases 2a and 2b are amenable to execution in parallel.
+
+### Phase 2a: WAL fan-out
+
+**Problem**: when all shards consume the whole WAL, the network bandwidth used
+for transmitting the WAL from safekeeper to pageservers is multiplied by a factor
+of the shard count.
+
+Network bandwidth is not our most pressing bottleneck, but it is likely to become
+a problem if we set a modest shard count (~8) on a significant number of tenants,
+especially as those larger tenants which we shard are also likely to have higher
+write bandwidth than average.
+
+### Phase 2b: Shard Splitting
+
+**Problem**: the number of shards in a tenant is defined at creation time and cannot
+be changed. This causes excessive sharding for most small tenants, and an upper
+bound on scale for very large tenants.
+
+To address this, a _splitting_ feature will later be added. One shard can split its
+data into a number of children by doing a special compaction operation to generate
+image layers broken up child-shard-wise, and then writing out an `index_part.json` for
+each child. This will then require external coordination (by the control plane) to
+safely attach these new child shards and then move them around to distribute work.
+The opposite _merging_ operation can also be imagined, but is unlikely to be implemented:
+once a Tenant has been sharded, the marginal efficiency benefit of merging is unlikely to justify
+the risk/complexity of implementing such a rarely-encountered scenario.
+
+### Phase N (future): distributed historical reads
+
+**Problem**: while sharding based on key is good for handling changes in overall
+database size, it is less suitable for spiky/unpredictable changes in the read
+workload to historical layers. Sudden increases in historical reads could result
+in sudden increases in local disk capacity required for a TenantShard.
+
+Example: the extreme case of this would be to run a tenant for a year, then create branches
+with ancestors at monthly intervals. This could lead to a sudden 12x inflation in
+the on-disk capacity footprint of a TenantShard, since it would be serving reads
+from all those disparate historical layers.
+
+If we can respond fast enough, then key-sharding a tenant more finely can help with
+this, but splitting may be a relatively expensive operation and the increased historical
+read load may be transient.
+
+A separate mechanism for handling heavy historical reads could be something like
+a gossip mechanism for pageservers to communicate
+about their workload, and then a getpageatlsn offload mechanism where one pageserver can
+ask another to go read the necessary layers from remote storage to serve the read. This
+requires relativly little coordination because it is read-only: any node can service any
+read. All reads to a particular shard would still flow through one node, but the
+disk capactity & I/O impact of servicing the read would be distributed.
+
+## FAQ/Alternatives
+
+### Why stripe the data, rather than using contiguous ranges of keyspace for each shard?
+
+When a database is growing under a write workload, writes may predominantly hit the
+end of the keyspace, creating a bandwidth hotspot on that shard. Similarly, if the user
+is intensively re-writing a particular relation, if that relation lived in a particular
+shard then it would not achieve our goal of distributing the write work across shards.
+
+### Why not proxy read requests through one pageserver, so that endpoints don't have to change?
+
+1. This would not achieve scale-out of network bandwidth: a busy tenant with a large
+   database would still cause a load hotspot on the pageserver routing its read requests.
+2. The additional hop through the "proxy" pageserver would add latency and overall
+   resource cost (CPU, network bandwidth)
+
+### Layer File Spreading: use one pageserver as the owner of a tenant, and have it spread out work on a per-layer basis to peers
+
+In this model, there would be no explicit sharding of work, but the pageserver to which
+a tenant is attached would not hold all layers on its disk: instead, it would call out
+to peers to have them store some layers, and call out to those peers to request reads
+in those layers.
+
+This mechanism will work well for distributing work in the LSN dimension, but in the key
+space dimension it has the major limitation of requiring one node to handle all
+incoming writes, and compactions. Even if the write workload for a large database
+fits in one pageserver, it will still be a hotspot and such tenants may still
+de-facto require their own pageserver.

docs/rfcs/032-shard-splitting.md

@@ -0,0 +1,479 @@
+# Shard splitting
+
+## Summary
+
+This RFC describes a new pageserver API for splitting an existing tenant shard into
+multiple shards, and describes how to use this API to safely increase the total
+shard count of a tenant.
+
+## Motivation
+
+In the [sharding RFC](031-sharding-static.md), a mechanism was introduced to scale
+tenants beyond the capacity of a single pageserver by breaking up the key space
+into stripes, and distributing these stripes across many pageservers. However,
+the shard count was defined once at tenant creation time and not varied thereafter.
+
+In practice, the expected size of a database is rarely known at creation time, and
+it is inefficient to enable sharding for very small tenants: we need to be
+able to create a tenant with a small number of shards (such as 1), and later expand
+when it becomes clear that the tenant has grown in size to a point where sharding
+is beneficial.
+
+### Prior art
+
+Many distributed systems have the problem of choosing how many shards to create for
+tenants that do not specify an expected size up-front. There are a couple of general
+approaches:
+
+- Write to a key space in order, and start a new shard when the highest key advances
+  past some point. This doesn't work well for Neon, because we write to our key space
+  in many different contiguous ranges (per relation), rather than in one contiguous
+  range. To adapt to this kind of model, we would need a sharding scheme where each
+  relation had its own range of shards, which would be inefficient for the common
+  case of databases with many small relations.
+- Monitor the system, and automatically re-shard at some size threshold. For
+  example in Ceph, the [pg_autoscaler](https://github.com/ceph/ceph/blob/49c27499af4ee9a90f69fcc6bf3597999d6efc7b/src/pybind/mgr/pg_autoscaler/module.py)
+  component monitors the size of each RADOS Pool, and adjusts the number of Placement
+  Groups (Ceph's shard equivalent).
+
+## Requirements
+
+- A configurable capacity limit per-shard is enforced.
+- Changes in shard count do not interrupt service beyond requiring postgres
+  to reconnect (i.e. milliseconds).
+- Human being does not have to choose shard count
+
+## Non Goals
+
+- Shard splitting is always a tenant-global operation: we will not enable splitting
+  one shard while leaving others intact.
+- The inverse operation (shard merging) is not described in this RFC. This is a lower
+  priority than splitting, because databases grow more often than they shrink, and
+  a database with many shards will still work properly if the stored data shrinks, just
+  with slightly more overhead (e.g. redundant WAL replication)
+- Shard splitting is only initiated based on capacity bounds, not load. Splitting
+  a tenant based on load will make sense for some medium-capacity, high-load workloads,
+  but is more complex to reason about and likely is not desirable until we have
+  shard merging to reduce the shard count again if the database becomes less busy.
+
+## Impacted Components
+
+pageserver, storage controller
+
+(the _storage controller_ is the evolution of what was called `attachment_service` in our test environment)
+
+## Terminology
+
+**Parent** shards are the shards that exist before a split. **Child** shards are
+the new shards created during a split.
+
+**Shard** is synonymous with _tenant shard_.
+
+**Shard Index** is the 2-tuple of shard number and shard count, written in
+paths as {:02x}{:02x}, e.g. `0001`.
+
+## Background
+
+In the implementation section, a couple of existing aspects of sharding are important
+to remember:
+
+- Shard identifiers contain the shard number and count, so that "shard 0 of 1" (`0001`) is
+  a distinct shard from "shard 0 of 2" (`0002`). This is the case in key paths, local
+  storage paths, and remote index metadata.
+- Remote layer file paths contain the shard index of the shard that created them, and
+  remote indices contain the same index to enable building the layer file path. A shard's
+  index may reference layers that were created by another shard.
+- Local tenant shard directories include the shard index. All layers downloaded by
+  a tenant shard are stored in this shard-prefixed path, even if those layers were
+  initially created by another shard: tenant shards do not read and write one anothers'
+  paths.
+- The `Tenant` pageserver type represents one tenant _shard_, not the whole tenant.
+  This is for historical reasons and will be cleaned up in future, but the existing
+  name is used here to help comprehension when reading code.
+
+## Implementation
+
+Note: this section focuses on the correctness of the core split process. This will
+be fairly inefficient in a naive implementation, and several important optimizations
+are described in a later section.
+
+There are broadly two parts to the implementation:
+
+1. The pageserver split API, which splits one shard on one pageserver
+2. The overall tenant split proccess which is coordinated by the storage controller,
+   and calls into the pageserver split API as needed.
+
+### Pageserver Split API
+
+The pageserver will expose a new API endpoint at `/v1/tenant/:tenant_shard_id/shard_split`
+that takes the new total shard count in the body.
+
+The pageserver split API operates on one tenant shard, on one pageserver. External
+coordination is required to use it safely, this is described in the later
+'Split procedure' section.
+
+#### Preparation
+
+First identify the shard indices for the new child shards. These are deterministic,
+calculated from the parent shard's index, and the number of children being created (this
+is an input to the API, and validated to be a power of two). In a trivial example, splitting
+0001 in two always results in 0002 and 0102.
+
+Child shard indices are chosen such that the childrens' parts of the keyspace will
+be subsets of the parent's parts of the keyspace.
+
+#### Step 1: write new remote indices
+
+In remote storage, splitting is very simple: we may just write new index_part.json
+objects for each child shard, containing exactly the same layers as the parent shard.
+
+The children will have more data than they need, but this avoids any exhausive
+re-writing or copying of layer files.
+
+The index key path includes a generation number: the parent shard's current
+attached generation number will also be used for the child shards' indices. This
+makes the operation safely retryable: if everything crashes and restarts, we may
+call the split API again on the parent shard, and the result will be some new remote
+indices for the child shards, under a higher generation number.
+
+#### Step 2: start new `Tenant` objects
+
+A new `Tenant` object may be instantiated for each child shard, while the parent
+shard still exists. When calling the tenant_spawn function for this object,
+the remote index from step 1 will be read, and the child shard will start
+to ingest WAL to catch up from whatever was in the remote storage at step 1.
+
+We now wait for child shards' WAL ingestion to catch up with the parent shard,
+so that we can safely tear down the parent shard without risking an availability
+gap to clients reading recent LSNs.
+
+#### Step 3: tear down parent `Tenant` object
+
+Once child shards are running and have caught up with WAL ingest, we no longer
+need the parent shard. Note that clients may still be using it -- when we
+shut it down, any page_service handlers will also shut down, causing clients
+to disconnect. When the client reconnects, it will re-lookup the tenant,
+and hit the child shard instead of the parent (shard lookup from page_service
+should bias toward higher ShardCount shards).
+
+Note that at this stage the page service client has not yet been notified of
+any split. In the trivial single split example:
+
+- Shard 0001 is gone: Tenant object torn down
+- Shards 0002 and 0102 are running on the same pageserver where Shard 0001 used to live.
+- Clients will continue to connect to that server thinking that shard 0001 is there,
+  and all requests will work, because any key that was in shard 0001 is definitely
+  available in either shard 0002 or shard 0102.
+- Eventually, the storage controller (not the pageserver) will decide to migrate
+  some child shards away: at that point it will do a live migration, ensuring
+  that the client has an updated configuration before it detaches anything
+  from the original server.
+
+#### Complete
+
+When we send a 200 response to the split request, we are promising the caller:
+
+- That the child shards are persistent in remote storage
+- That the parent shard has been shut down
+
+This enables the caller to proceed with the overall shard split operation, which
+may involve other shards on other pageservers.
+
+### Storage Controller Split procedure
+
+Splitting a tenant requires calling the pageserver split API, and tracking
+enough state to ensure recovery + completion in the event of any component (pageserver
+or storage controller) crashing (or request timing out) during the split.
+
+1. call the split API on all existing shards. Ensure that the resulting
+   child shards are pinned to their pageservers until _all_ the split calls are done.
+   This pinning may be implemented as a "split bit" on the tenant shards, that
+   blocks any migrations, and also acts as a sign that if we restart, we must go
+   through some recovery steps to resume the split.
+2. Once all the split calls are done, we may unpin the child shards (clear
+   the split bit). The split is now complete: subsequent steps are just migrations,
+   not strictly part of the split.
+3. Try to schedule new pageserver locations for the child shards, using
+   a soft anti-affinity constraint to place shards from the same tenant onto different
+   pageservers.
+
+Updating computes about the new shard count is not necessary until we migrate
+any of the child shards away from the parent's location.
+
+### Recovering from failures
+
+#### Rolling back an incomplete split
+
+An incomplete shard split may be rolled back quite simply, by attaching the parent shards to pageservers,
+and detaching child shards. This will lose any WAL ingested into the children after the parents
+were detached earlier, but the parents will catch up.
+
+No special pageserver API is needed for this. From the storage controllers point of view, the
+procedure is:
+
+1. For all parent shards in the tenant, ensure they are attached
+2. For all child shards, ensure they are not attached
+3. Drop child shards from the storage controller's database, and clear the split bit on the parent shards.
+
+Any remote storage content for child shards is left behind. This is similar to other cases where
+we may leave garbage objects in S3 (e.g. when we upload a layer but crash before uploading an
+index that references it). Future online scrub/cleanup functionality can remove these objects, or
+they will be removed when the tenant is deleted, as tenant deletion lists all objects in the prefix,
+which would include any child shards that were rolled back.
+
+If any timelines had been created on child shards, they will be lost when rolling back. To mitigate
+this, we will **block timeline creation during splitting**, so that we can safely roll back until
+the split is complete, without risking losing timelines.
+
+Rolling back an incomplete split will happen automatically if a split fails due to some fatal
+reason, and will not be accessible via an API:
+
+- A pageserver fails to complete its split API request after too many retries
+- A pageserver returns a fatal unexpected error such as 400 or 500
+- The storage controller database returns a non-retryable error
+- Some internal invariant is violated in the storage controller split code
+
+#### Rolling back a complete split
+
+A complete shard split may be rolled back similarly to an incomplete split, with the following
+modifications:
+
+- The parent shards will no longer exist in the storage controller database, so these must
+  be re-synthesized somehow: the hard part of this is figuring the parent shards' generations. This
+  may be accomplished either by probing in S3, or by retaining some tombstone state for deleted
+  shards in the storage controller database.
+- Any timelines that were created after the split complete will disappear when rolling back
+  to the tenant shards. For this reason, rolling back after a complete split should only
+  be done due to serious issues where loss of recently created timelines is acceptable, or
+  in cases where we have confirmed that no timelines were created in the intervening period.
+- Parent shards' layers must not have been deleted: this property will come "for free" when
+  we first roll out sharding, by simply not implementing deletion of parent layers after
+  a split. When we do implement such deletion (see "Cleaning up parent-shard layers" in the
+  Optimizations section), it should apply a TTL to layers such that we have a
+  defined walltime window in which rollback will be possible.
+
+The storage controller will expose an API for rolling back a complete split, for use
+in the field if we encounter some critical bug with a post-split tenant.
+
+#### Retrying API calls during Pageserver Restart
+
+When a pageserver restarts during a split API call, it may witness on-disk content for both parent and
+child shards from an ongoing split. This does not intrinsically break anything, and the
+pageserver may include all these shards in its `/re-attach` request to the storage controller.
+
+In order to support such restarts, it is important that the storage controller stores
+persistent records of each child shard before it calls into a pageserver, as these child shards
+may require generation increments via a `/re-attach` request.
+
+The pageserver restart will also result in a failed API call from the storage controller's point
+of view. Recall that if _any_ pageserver fails to split, the overall split operation may not
+complete, and all shards must remain pinned to their current pageserver locations until the
+split is done.
+
+The pageserver API calls during splitting will retry on transient errors, so that
+short availability gaps do not result in a failure of the overall operation. The
+split in progress will be automatically rolled back if the threshold for API
+retries is reached (e.g. if a pageserver stays offline for longer than a typical
+restart).
+
+#### Rollback on Storage Controller Restart
+
+On startup, the storage controller will inspect the split bit for tenant shards that
+it loads from the database. If any splits are in progress:
+
+- Database content will be reverted to the parent shards
+- Child shards will be dropped from memory
+- The parent and child shards will be included in the general startup reconciliation that
+  the storage controller does: any child shards will be detached from pageservers because
+  they don't exist in the storage controller's expected set of shards, and parent shards
+  will be attached if they aren't already.
+
+#### Storage controller API request failures/retries
+
+The split request handler will implement idempotency: if the [`Tenant`] requested to split
+doesn't exist, we will check for the would-be child shards, and if they already exist,
+we consider the request complete.
+
+If a request is retried while the original request is still underway, then the split
+request handler will notice an InProgress marker in TenantManager, and return 503
+to encourage the client to backoff/retry. This is the same as the general pageserver
+API handling for calls that try to act on an InProgress shard.
+
+#### Compute start/restart during a split
+
+If a compute starts up during split, it will be configured with the old sharding
+configuration. This will work for reads irrespective of the progress of the split
+as long as no child hards have been migrated away from their original location, and
+this is guaranteed in the split procedure (see earlier section).
+
+#### Pageserver fails permanently during a split
+
+If a pageserver permanently fails (i.e. the storage controller availability state for it
+goes to Offline) while a split is in progress, the splitting operation will roll back, and
+during the roll back it will skip any API calls to the offline pageserver. If the offline
+pageserver becomes available again, any stale locations will be cleaned up via the normal reconciliation process (the `/re-attach` API).
+
+### Handling secondary locations
+
+For correctness, it is not necessary to split secondary locations. We can simply detach
+the secondary locations for parent shards, and then attach new secondary locations
+for child shards.
+
+Clearly this is not optimal, as it will result in re-downloads of layer files that
+were already present on disk. See "Splitting secondary locations"
+
+### Conditions to trigger a split
+
+The pageserver will expose a new API for reporting on shards that are candidates
+for split: this will return a top-N report of the largest tenant shards by
+physical size (remote size). This should exclude any tenants that are already
+at the maximum configured shard count.
+
+The API would look something like:
+`/v1/top_n_tenant?shard_count_lt=8&sort_by=resident_size`
+
+The storage controller will poll that API across all pageservers it manages at some appropriate interval (e.g. 60 seconds).
+
+A split operation will be started when the tenant exceeds some threshold. This threshold
+should be _less than_ how large we actually want shards to be, perhaps much less. That's to
+minimize the amount of work involved in splitting -- if we want 100GiB shards, we shouldn't
+wait for a tenant to exceed 100GiB before we split anything. Some data analysis of existing
+tenant size distribution may be useful here: if we can make a statement like "usually, if
+a tenant has exceeded 20GiB they're probably going to exceed 100GiB later", then we might
+make our policy to split a tenant at 20GiB.
+
+The finest split we can do is by factors of two, but we can do higher-cardinality splits
+too, and this will help to reduce the overhead of repeatedly re-splitting a tenant
+as it grows. An example of a very simple heuristic for early deployment of the splitting
+feature would be: "Split tenants into 8 shards when their physical size exceeds 64GiB": that
+would give us two kinds of tenant (1 shard and 8 shards), and the confidence that once we had
+split a tenant, it will not need re-splitting soon after.
+
+## Optimizations
+
+### Flush parent shard to remote storage during split
+
+Any data that is in WAL but not remote storage at time of split will need
+to be replayed by child shards when they start for the first time. To minimize
+this work, we may flush the parent shard to remote storage before writing the
+remote indices for child shards.
+
+It is important that this flush is subject to some time bounds: we may be splitting
+in response to a surge of write ingest, so it may be time-critical to split. A
+few seconds to flush latest data should be sufficient to optimize common cases without
+running the risk of holding up a split for a harmful length of time when a parent
+shard is being written heavily. If the flush doesn't complete in time, we may proceed
+to shut down the parent shard and carry on with the split.
+
+### Hard linking parent layers into child shard directories
+
+Before we start the Tenant objects for child shards, we may pre-populate their
+local storage directories with hard links to the layer files already present
+in the parent shard's local directory. When the child shard starts and downloads
+its remote index, it will find all those layer files already present on local disk.
+
+This avoids wasting download capacity and makes splitting faster, but more importantly
+it avoids taking up a factor of N more disk space when splitting 1 shard into N.
+
+This mechanism will work well in typical flows where shards are migrated away
+promptly after a split, but for the general case including what happens when
+layers are evicted and re-downloaded after a split, see the 'Proactive compaction'
+section below.
+
+### Filtering during compaction
+
+Compaction, especially image layer generation, should skip any keys that are
+present in a shard's layer files, but do not match the shard's ShardIdentity's
+is_key_local() check. This avoids carrying around data for longer than necessary
+in post-split compactions.
+
+This was already implemented in https://github.com/neondatabase/neon/pull/6246
+
+### Proactive compaction
+
+In remote storage, there is little reason to rewrite any data on a shard split:
+all the children can reference parent layers via the very cheap write of the child
+index_part.json.
+
+In local storage, things are more nuanced. During the initial split there is no
+capacity cost to duplicating parent layers, if we implement the hard linking
+optimization described above. However, as soon as any layers are evicted from
+local disk and re-downloaded, the downloaded layers will not be hard-links any more:
+they'll have real capacity footprint. That isn't a problem if we migrate child shards
+away from the parent node swiftly, but it risks a significant over-use of local disk
+space if we do not.
+
+For example, if we did an 8-way split of a shard, and then _didn't_ migrate 7 of
+the shards elsewhere, then churned all the layers in all the shards via eviction,
+then we would blow up the storage capacity used on the node by 8x. If we're splitting
+a 100GB shard, that could take the pageserver to the point of exhausting disk space.
+
+To avoid this scenario, we could implement a special compaction mode where we just
+read historic layers, drop unwanted keys, and write back the layer file. This
+is pretty expensive, but useful if we have split a large shard and are not going to
+migrate the child shards away.
+
+The heuristic conditions for triggering such a compaction are:
+
+- A) eviction plus time: if a child shard
+  has existed for more than a time threshold, and has been requested to perform at least one eviction, then it becomes urgent for this child shard to execute a proactive compaction to reduce its storage footprint, at the cost of I/O load.
+- B) resident size plus time: we may inspect the resident layers and calculate how
+  many of them include the overhead of storing pre-split keys. After some time
+  threshold (different to the one in case A) we still have such layers occupying
+  local disk space, then we should proactively compact them.
+
+### Cleaning up parent-shard layers
+
+It is functionally harmless to leave parent shard layers in remote storage indefinitely.
+They would be cleaned up in the event of the tenant's deletion.
+
+As an optimization to avoid leaking remote storage capacity (which costs money), we may
+lazily clean up parent shard layers once no child shards reference them.
+
+This may be done _very_ lazily: e.g. check every PITR interval. The cleanup procedure is:
+
+- list all the key prefixes beginning with the tenant ID, and select those shard prefixes
+  which do not belong to the most-recently-split set of shards (_ancestral shards_, i.e. `shard*count < max(shard_count) over all shards)`, and those shard prefixes which do have the latest shard count (_current shards_)
+- If there are no _ancestral shard_ prefixes found, we have nothing to clean up and
+  may drop out now.
+- find the latest-generation index for each _current shard_, read all and accumulate the set of layers belonging to ancestral shards referenced by these indices.
+- for all ancestral shards, list objects in the prefix and delete any layer which was not
+  referenced by a current shard.
+
+If this cleanup is scheduled for 1-2 PITR periods after the split, there is a good chance that child shards will have written their own image layers covering the whole keyspace, such that all parent shard layers will be deletable.
+
+The cleanup may be done by the scrubber (external process), or we may choose to have
+the zeroth shard in the latest generation do the work -- there is no obstacle to one shard
+reading the other shard's indices at runtime, and we do not require visibility of the
+latest index writes.
+
+Cleanup should be artificially delayed by some period (for example 24 hours) to ensure
+that we retain the option to roll back a split in case of bugs.
+
+### Splitting secondary locations
+
+We may implement a pageserver API similar to the main splitting API, which does a simpler
+operation for secondary locations: it would not write anything to S3, instead it would simply
+create the child shard directory on local disk, hard link in directories from the parent,
+and set up the in memory (TenantSlot) state for the children.
+
+Similar to attached locations, a subset of secondary locations will probably need re-locating
+after the split is complete, to avoid leaving multiple child shards on the same pageservers,
+where they may use excessive space for the tenant.
+
+## FAQ/Alternatives
+
+### What should the thresholds be set to?
+
+Shard size limit: the pre-sharding default capacity quota for databases was 200GiB, so this could be a starting point for the per-shard size limit.
+
+Max shard count:
+
+- The safekeeper overhead to sharding is currently O(N) network bandwidth because
+  the un-filtered WAL is sent to all shards. To avoid this growing out of control,
+  a limit of 8 shards should be temporarily imposed until WAL filtering is implemented
+  on the safekeeper.
+- there is also little benefit to increasing the shard count beyond the number
+  of pageservers in a region.
+
+### Is it worth just rewriting all the data during a split to simplify reasoning about space?

docs/sourcetree.md

@@ -7,6 +7,11 @@ 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.

libs/metrics/Cargo.toml

@@ -10,11 +10,13 @@ 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

@@ -40,7 +40,7 @@ macro_rules! register_hll {
     }};
 
     ($N:literal, $NAME:expr, $HELP:expr $(,)?) => {{
-        $crate::register_hll!($N, $crate::opts!($NAME, $HELP), $LABELS_NAMES)
+        $crate::register_hll!($N, $crate::opts!($NAME, $HELP))
     }};
 }
 

libs/metrics/src/lib.rs

@@ -4,6 +4,17 @@
 //! a default registry.
 #![deny(clippy::undocumented_unsafe_blocks)]
 
+use measured::{
+    label::{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,
@@ -11,6 +22,7 @@ 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};
@@ -23,13 +35,11 @@ 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 mod metric_vec_duration;
 pub use hll::{HyperLogLog, HyperLogLogVec};
 #[cfg(target_os = "linux")]
 pub mod more_process_metrics;
@@ -60,7 +70,7 @@ 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>) -> Result<()> {
+pub fn register_internal(c: Box<dyn Collector>) -> prometheus::Result<()> {
     INTERNAL_REGISTRY.register(c)
 }
 
@@ -97,6 +107,127 @@ 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",
@@ -106,6 +237,7 @@ 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.
@@ -118,7 +250,6 @@ pub fn set_build_info_metric(revision: &str, build_tag: &str) {
 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);
@@ -152,6 +283,7 @@ 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 {
@@ -189,7 +321,10 @@ 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]) -> Result<GenericCounterPair<P>> {
+    pub fn get_metric_with_label_values(
+        &self,
+        vals: &[&str],
+    ) -> prometheus::Result<GenericCounterPair<P>> {
         Ok(GenericCounterPair {
             inc: self.inc.get_metric_with_label_values(vals)?,
             dec: self.dec.get_metric_with_label_values(vals)?,
@@ -202,7 +337,7 @@ impl<P: Atomic> GenericCounterPairVec<P> {
         self.get_metric_with_label_values(vals).unwrap()
     }
 
-    pub fn remove_label_values(&self, res: &mut [Result<()>; 2], vals: &[&str]) {
+    pub fn remove_label_values(&self, res: &mut [prometheus::Result<()>; 2], vals: &[&str]) {
         res[0] = self.inc.remove_label_values(vals);
         res[1] = self.dec.remove_label_values(vals);
     }

libs/metrics/src/metric_vec_duration.rs

@@ -1,23 +0,0 @@
-//! Helpers for observing duration on `HistogramVec` / `CounterVec` / `GaugeVec` / `MetricVec<T>`.
-
-use std::{future::Future, time::Instant};
-
-pub trait DurationResultObserver {
-    fn observe_result<T, E>(&self, res: &Result<T, E>, duration: std::time::Duration);
-}
-
-pub async fn observe_async_block_duration_by_result<
-    T,
-    E,
-    F: Future<Output = Result<T, E>>,
-    O: DurationResultObserver,
->(
-    observer: &O,
-    block: F,
-) -> Result<T, E> {
-    let start = Instant::now();
-    let result = block.await;
-    let duration = start.elapsed();
-    observer.observe_result(&result, duration);
-    result
-}

libs/pageserver_api/src/controller_api.rs

@@ -2,11 +2,14 @@ use std::str::FromStr;
 
 /// Request/response types for the storage controller
 /// API (`/control/v1` prefix).  Implemented by the server
-/// in [`attachment_service::http`]
+/// in [`storage_controller::http`]
 use serde::{Deserialize, Serialize};
-use utils::id::NodeId;
+use utils::id::{NodeId, TenantId};
 
-use crate::{models::ShardParameters, shard::TenantShardId};
+use crate::{
+    models::{ShardParameters, TenantConfig},
+    shard::{ShardStripeSize, TenantShardId},
+};
 
 #[derive(Serialize, Deserialize)]
 pub struct TenantCreateResponseShard {
@@ -35,10 +38,16 @@ pub struct NodeRegisterRequest {
 pub struct NodeConfigureRequest {
     pub node_id: NodeId,
 
-    pub availability: Option<NodeAvailability>,
+    pub availability: Option<NodeAvailabilityWrapper>,
     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,
@@ -57,6 +66,48 @@ pub struct TenantLocateResponse {
     pub shard_params: ShardParameters,
 }
 
+#[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,
+
+    pub node_attached: Option<NodeId>,
+    pub node_secondary: Vec<NodeId>,
+
+    pub last_error: String,
+
+    /// A task is currently running to reconcile this tenant's intent state with the state on pageservers
+    pub is_reconciling: bool,
+    /// This shard failed in sending a compute notification to the cloud control plane, and a retry is pending.
+    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
 /// TODO: higher level "Reschedule tenant" operation where the request
 /// specifies some constraints, e.g. asking it to get off particular node(s)
@@ -66,31 +117,94 @@ pub struct TenantShardMigrateRequest {
     pub node_id: NodeId,
 }
 
-#[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq)]
+/// 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)]
+pub struct UtilizationScore(pub u64);
+
+impl UtilizationScore {
+    pub fn worst() -> Self {
+        UtilizationScore(u64::MAX)
+    }
+}
+
+#[derive(Serialize, Deserialize, Clone, Copy, Debug)]
+#[serde(into = "NodeAvailabilityWrapper")]
 pub enum NodeAvailability {
     // Normal, happy state
-    Active,
+    Active(UtilizationScore),
     // Offline: Tenants shouldn't try to attach here, but they may assume that their
     // secondary locations on this node still exist.  Newly added nodes are in this
     // state until we successfully contact them.
     Offline,
 }
 
-impl FromStr for NodeAvailability {
-    type Err = anyhow::Error;
+impl PartialEq for NodeAvailability {
+    fn eq(&self, other: &Self) -> bool {
+        use NodeAvailability::*;
+        matches!((self, other), (Active(_), Active(_)) | (Offline, Offline))
+    }
+}
 
-    fn from_str(s: &str) -> Result<Self, Self::Err> {
-        match s {
-            "active" => Ok(Self::Active),
-            "offline" => Ok(Self::Offline),
-            _ => Err(anyhow::anyhow!("Unknown availability state '{s}'")),
+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)]
+pub enum NodeAvailabilityWrapper {
+    Active,
+    Offline,
+}
+
+impl From<NodeAvailabilityWrapper> for NodeAvailability {
+    fn from(val: NodeAvailabilityWrapper) -> Self {
+        match val {
+            // Assume the worst utilisation score to begin with. It will later be updated by
+            // the heartbeats.
+            NodeAvailabilityWrapper::Active => NodeAvailability::Active(UtilizationScore::worst()),
+            NodeAvailabilityWrapper::Offline => NodeAvailability::Offline,
         }
     }
 }
 
-/// FIXME: this is a duplicate of the type in the attachment_service crate, because the
-/// type needs to be defined with diesel traits in there.
-#[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq)]
+impl From<NodeAvailability> for NodeAvailabilityWrapper {
+    fn from(val: NodeAvailability) -> Self {
+        match val {
+            NodeAvailability::Active(_) => NodeAvailabilityWrapper::Active,
+            NodeAvailability::Offline => NodeAvailabilityWrapper::Offline,
+        }
+    }
+}
+
+#[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,
+
+    // 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
+    }
+}
+
+#[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, Debug)]
 pub enum NodeSchedulingPolicy {
     Active,
     Filling,
@@ -129,11 +243,8 @@ impl From<NodeSchedulingPolicy> for String {
 /// to create secondary locations.
 #[derive(Clone, Serialize, Deserialize, Debug, PartialEq, Eq)]
 pub enum PlacementPolicy {
-    /// Cheapest way to attach a tenant: just one pageserver, no secondary
-    Single,
-    /// Production-ready way to attach a tenant: one attached pageserver and
-    /// some number of secondaries.
-    Double(usize),
+    /// Normal live state: one attached pageserver and zero or more secondaries.
+    Attached(usize),
     /// Create one secondary mode locations. This is useful when onboarding
     /// a tenant, or for an idle tenant that we might want to bring online quickly.
     Secondary,
@@ -155,14 +266,14 @@ mod test {
     /// Check stability of PlacementPolicy's serialization
     #[test]
     fn placement_policy_encoding() -> anyhow::Result<()> {
-        let v = PlacementPolicy::Double(1);
+        let v = PlacementPolicy::Attached(1);
         let encoded = serde_json::to_string(&v)?;
-        assert_eq!(encoded, "{\"Double\":1}");
+        assert_eq!(encoded, "{\"Attached\":1}");
         assert_eq!(serde_json::from_str::<PlacementPolicy>(&encoded)?, v);
 
-        let v = PlacementPolicy::Single;
+        let v = PlacementPolicy::Detached;
         let encoded = serde_json::to_string(&v)?;
-        assert_eq!(encoded, "\"Single\"");
+        assert_eq!(encoded, "\"Detached\"");
         assert_eq!(serde_json::from_str::<PlacementPolicy>(&encoded)?, v);
         Ok(())
     }

libs/pageserver_api/src/models.rs

@@ -4,6 +4,7 @@ pub mod utilization;
 pub use utilization::PageserverUtilization;
 
 use std::{
+    borrow::Cow,
     collections::HashMap,
     io::{BufRead, Read},
     num::{NonZeroU64, NonZeroUsize},
@@ -19,6 +20,7 @@ use utils::{
     history_buffer::HistoryBufferWithDropCounter,
     id::{NodeId, TenantId, TimelineId},
     lsn::Lsn,
+    serde_system_time,
 };
 
 use crate::controller_api::PlacementPolicy;
@@ -198,6 +200,13 @@ pub struct TimelineCreateRequest {
 #[derive(Serialize, Deserialize)]
 pub struct TenantShardSplitRequest {
     pub new_shard_count: u8,
+
+    // A tenant's stripe size is only meaningful the first time their shard count goes
+    // above 1: therefore during a split from 1->N shards, we may modify the stripe size.
+    //
+    // If this is set while the stripe count is being increased from an already >1 value,
+    // then the request will fail with 400.
+    pub new_stripe_size: Option<ShardStripeSize>,
 }
 
 #[derive(Serialize, Deserialize)]
@@ -293,6 +302,7 @@ 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)]
@@ -419,7 +429,7 @@ pub struct StatusResponse {
 #[derive(Serialize, Deserialize, Debug)]
 #[serde(deny_unknown_fields)]
 pub struct TenantLocationConfigRequest {
-    pub tenant_id: TenantShardId,
+    pub tenant_id: Option<TenantShardId>,
     #[serde(flatten)]
     pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
 }
@@ -570,7 +580,7 @@ pub struct TimelineInfo {
     pub walreceiver_status: String,
 }
 
-#[derive(Debug, Clone, Serialize)]
+#[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct LayerMapInfo {
     pub in_memory_layers: Vec<InMemoryLayerInfo>,
     pub historic_layers: Vec<HistoricLayerInfo>,
@@ -588,7 +598,7 @@ pub enum LayerAccessKind {
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct LayerAccessStatFullDetails {
     pub when_millis_since_epoch: u64,
-    pub task_kind: &'static str,
+    pub task_kind: Cow<'static, str>,
     pub access_kind: LayerAccessKind,
 }
 
@@ -647,23 +657,23 @@ impl LayerResidenceEvent {
     }
 }
 
-#[derive(Debug, Clone, Serialize)]
+#[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct LayerAccessStats {
     pub access_count_by_access_kind: HashMap<LayerAccessKind, u64>,
-    pub task_kind_access_flag: Vec<&'static str>,
+    pub task_kind_access_flag: Vec<Cow<'static, str>>,
     pub first: Option<LayerAccessStatFullDetails>,
     pub accesses_history: HistoryBufferWithDropCounter<LayerAccessStatFullDetails, 16>,
     pub residence_events_history: HistoryBufferWithDropCounter<LayerResidenceEvent, 16>,
 }
 
-#[derive(Debug, Clone, Serialize)]
+#[derive(Debug, Clone, Serialize, Deserialize)]
 #[serde(tag = "kind")]
 pub enum InMemoryLayerInfo {
     Open { lsn_start: Lsn },
     Frozen { lsn_start: Lsn, lsn_end: Lsn },
 }
 
-#[derive(Debug, Clone, Serialize)]
+#[derive(Debug, Clone, Serialize, Deserialize)]
 #[serde(tag = "kind")]
 pub enum HistoricLayerInfo {
     Delta {
@@ -685,6 +695,32 @@ pub enum HistoricLayerInfo {
     },
 }
 
+impl HistoricLayerInfo {
+    pub fn layer_file_name(&self) -> &str {
+        match self {
+            HistoricLayerInfo::Delta {
+                layer_file_name, ..
+            } => layer_file_name,
+            HistoricLayerInfo::Image {
+                layer_file_name, ..
+            } => layer_file_name,
+        }
+    }
+    pub fn is_remote(&self) -> bool {
+        match self {
+            HistoricLayerInfo::Delta { remote, .. } => *remote,
+            HistoricLayerInfo::Image { remote, .. } => *remote,
+        }
+    }
+    pub fn set_remote(&mut self, value: bool) {
+        let field = match self {
+            HistoricLayerInfo::Delta { remote, .. } => remote,
+            HistoricLayerInfo::Image { remote, .. } => remote,
+        };
+        *field = value;
+    }
+}
+
 #[derive(Debug, Serialize, Deserialize)]
 pub struct DownloadRemoteLayersTaskSpawnRequest {
     pub max_concurrent_downloads: NonZeroUsize,
@@ -717,6 +753,25 @@ pub struct WalRedoManagerStatus {
     pub pid: Option<u32>,
 }
 
+/// The progress of a secondary tenant is mostly useful when doing a long running download: e.g. initiating
+/// a download job, timing out while waiting for it to run, and then inspecting this status to understand
+/// what's happening.
+#[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>,
+
+    /// The number of layers currently on-disk
+    pub layers_downloaded: usize,
+    /// The number of layers in the most recently seen heatmap
+    pub layers_total: usize,
+
+    /// The number of layer bytes currently on-disk
+    pub bytes_downloaded: u64,
+    /// The number of layer bytes in the most recently seen heatmap
+    pub bytes_total: u64,
+}
+
 pub mod virtual_file {
     #[derive(
         Copy,
@@ -786,21 +841,21 @@ impl TryFrom<u8> for PagestreamBeMessageTag {
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamExistsRequest {
-    pub latest: bool,
+    pub horizon: Lsn,
     pub lsn: Lsn,
     pub rel: RelTag,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamNblocksRequest {
-    pub latest: bool,
+    pub horizon: Lsn,
     pub lsn: Lsn,
     pub rel: RelTag,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamGetPageRequest {
-    pub latest: bool,
+    pub horizon: Lsn,
     pub lsn: Lsn,
     pub rel: RelTag,
     pub blkno: u32,
@@ -808,14 +863,14 @@ pub struct PagestreamGetPageRequest {
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamDbSizeRequest {
-    pub latest: bool,
+    pub horizon: Lsn,
     pub lsn: Lsn,
     pub dbnode: u32,
 }
 
 #[derive(Debug, PartialEq, Eq)]
 pub struct PagestreamGetSlruSegmentRequest {
-    pub latest: bool,
+    pub horizon: Lsn,
     pub lsn: Lsn,
     pub kind: u8,
     pub segno: u32,
@@ -868,8 +923,8 @@ impl PagestreamFeMessage {
 
         match self {
             Self::Exists(req) => {
-                bytes.put_u8(0);
-                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u8(10);
+                bytes.put_u64(req.horizon.0);
                 bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
@@ -878,8 +933,8 @@ impl PagestreamFeMessage {
             }
 
             Self::Nblocks(req) => {
-                bytes.put_u8(1);
-                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u8(11);
+                bytes.put_u64(req.horizon.0);
                 bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
@@ -888,8 +943,8 @@ impl PagestreamFeMessage {
             }
 
             Self::GetPage(req) => {
-                bytes.put_u8(2);
-                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u8(12);
+                bytes.put_u64(req.horizon.0);
                 bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
@@ -899,15 +954,15 @@ impl PagestreamFeMessage {
             }
 
             Self::DbSize(req) => {
-                bytes.put_u8(3);
-                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u8(13);
+                bytes.put_u64(req.horizon.0);
                 bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.dbnode);
             }
 
             Self::GetSlruSegment(req) => {
-                bytes.put_u8(4);
-                bytes.put_u8(u8::from(req.latest));
+                bytes.put_u8(14);
+                bytes.put_u64(req.horizon.0);
                 bytes.put_u64(req.lsn.0);
                 bytes.put_u8(req.kind);
                 bytes.put_u32(req.segno);
@@ -924,11 +979,32 @@ impl PagestreamFeMessage {
         //
         // TODO: consider using protobuf or serde bincode for less error prone
         // serialization.
-        let msg_tag = body.read_u8()?;
+        let mut msg_tag = body.read_u8()?;
+        //
+        // Old version of protocol use commands with tags started with 0 and containing `latest` flag.
+        // New version of protocol shift command tags by 10 and pass LSN range instead of `latest` flag.
+        // Server should be able to handle both protocol version. As far as we are not passing no=w,
+        // protocol version from client to server, we make a decision based on tag range.
+        // So this code actually provides backward compatibility.
+        //
+        let horizon = if msg_tag >= 10 {
+            // new protocol
+            msg_tag -= 10; // commands tags in new protocol starts with 10
+            Lsn::from(body.read_u64::<BigEndian>()?)
+        } else {
+            // old_protocol
+            let latest = body.read_u8()? != 0;
+            if latest {
+                Lsn::MAX // get latest version
+            } else {
+                Lsn::INVALID // get version on specified LSN
+            }
+        };
+        let lsn = Lsn::from(body.read_u64::<BigEndian>()?);
         match msg_tag {
             0 => Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
-                latest: body.read_u8()? != 0,
-                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
+                horizon,
+                lsn,
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -937,8 +1013,8 @@ impl PagestreamFeMessage {
                 },
             })),
             1 => Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
-                latest: body.read_u8()? != 0,
-                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
+                horizon,
+                lsn,
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -947,8 +1023,8 @@ impl PagestreamFeMessage {
                 },
             })),
             2 => Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
-                latest: body.read_u8()? != 0,
-                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
+                horizon,
+                lsn,
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -958,14 +1034,14 @@ impl PagestreamFeMessage {
                 blkno: body.read_u32::<BigEndian>()?,
             })),
             3 => Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
-                latest: body.read_u8()? != 0,
-                lsn: Lsn::from(body.read_u64::<BigEndian>()?),
+                horizon,
+                lsn,
                 dbnode: body.read_u32::<BigEndian>()?,
             })),
             4 => Ok(PagestreamFeMessage::GetSlruSegment(
                 PagestreamGetSlruSegmentRequest {
-                    latest: body.read_u8()? != 0,
-                    lsn: Lsn::from(body.read_u64::<BigEndian>()?),
+                    horizon,
+                    lsn,
                     kind: body.read_u8()?,
                     segno: body.read_u32::<BigEndian>()?,
                 },
@@ -1093,7 +1169,7 @@ mod tests {
         // Test serialization/deserialization of PagestreamFeMessage
         let messages = vec![
             PagestreamFeMessage::Exists(PagestreamExistsRequest {
-                latest: true,
+                horizon: Lsn::MAX,
                 lsn: Lsn(4),
                 rel: RelTag {
                     forknum: 1,
@@ -1103,7 +1179,7 @@ mod tests {
                 },
             }),
             PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
-                latest: false,
+                horizon: Lsn::INVALID,
                 lsn: Lsn(4),
                 rel: RelTag {
                     forknum: 1,
@@ -1113,8 +1189,8 @@ mod tests {
                 },
             }),
             PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
-                latest: true,
-                lsn: Lsn(4),
+                horizon: Lsn::MAX,
+                lsn: Lsn::INVALID,
                 rel: RelTag {
                     forknum: 1,
                     spcnode: 2,
@@ -1124,7 +1200,7 @@ mod tests {
                 blkno: 7,
             }),
             PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
-                latest: true,
+                horizon: Lsn::MAX,
                 lsn: Lsn(4),
                 dbnode: 7,
             }),

libs/pageserver_api/src/models/utilization.rs

@@ -1,4 +1,4 @@
-use std::time::SystemTime;
+use utils::serde_system_time::SystemTime;
 
 /// Pageserver current utilization and scoring for how good candidate the pageserver would be for
 /// the next tenant.
@@ -7,7 +7,7 @@ use std::time::SystemTime;
 ///
 /// `format: int64` fields must use `ser_saturating_u63` because openapi generated clients might
 /// not handle full u64 values properly.
-#[derive(serde::Serialize, Debug)]
+#[derive(serde::Serialize, serde::Deserialize, Debug, Clone)]
 pub struct PageserverUtilization {
     /// Used disk space
     #[serde(serialize_with = "ser_saturating_u63")]
@@ -21,17 +21,9 @@ 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")]
     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))
-}
-
 /// 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
@@ -58,7 +50,9 @@ mod tests {
             disk_usage_bytes: u64::MAX,
             free_space_bytes: 0,
             utilization_score: u64::MAX,
-            captured_at: SystemTime::UNIX_EPOCH + Duration::from_secs(1708509779),
+            captured_at: SystemTime(
+                std::time::SystemTime::UNIX_EPOCH + Duration::from_secs(1708509779),
+            ),
         };
 
         let s = serde_json::to_string(&doc).unwrap();

libs/pageserver_api/src/upcall_api.rs

@@ -6,19 +6,36 @@
 use serde::{Deserialize, Serialize};
 use utils::id::NodeId;
 
-use crate::shard::TenantShardId;
+use crate::{
+    controller_api::NodeRegisterRequest, models::LocationConfigMode, shard::TenantShardId,
+};
 
+/// Upcall message sent by the pageserver to the configured `control_plane_api` on
+/// startup.
 #[derive(Serialize, Deserialize)]
 pub struct ReAttachRequest {
     pub node_id: NodeId,
+
+    /// Optional inline self-registration: this is useful with the storage controller,
+    /// if the node already has a node_id set.
+    #[serde(skip_serializing_if = "Option::is_none", default)]
+    pub register: Option<NodeRegisterRequest>,
 }
 
-#[derive(Serialize, Deserialize)]
+fn default_mode() -> LocationConfigMode {
+    LocationConfigMode::AttachedSingle
+}
+
+#[derive(Serialize, Deserialize, Debug)]
 pub struct ReAttachResponseTenant {
     pub id: TenantShardId,
-    pub gen: u32,
-}
+    /// Mandatory if LocationConfigMode is None or set to an Attached* mode
+    pub gen: Option<u32>,
 
+    /// Default value only for backward compat: this field should be set
+    #[serde(default = "default_mode")]
+    pub mode: LocationConfigMode,
+}
 #[derive(Serialize, Deserialize)]
 pub struct ReAttachResponse {
     pub tenants: Vec<ReAttachResponseTenant>,

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

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

@@ -5,7 +5,6 @@ use postgres::types::PgLsn;
 use postgres::Client;
 use postgres_ffi::{WAL_SEGMENT_SIZE, XLOG_BLCKSZ};
 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};
@@ -232,59 +231,52 @@ 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 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)>;
+    /// 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>>;
 }
 
+/// 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>, Option<PgLsn>)>,
-) -> anyhow::Result<(Vec<PgLsn>, PgLsn)> {
+    f: impl Fn(&mut C, PgLsn) -> anyhow::Result<Vec<PgLsn>>,
+) -> anyhow::Result<Vec<PgLsn>> {
     ensure_server_config(client)?;
 
     let initial_lsn = client.pg_current_wal_insert_lsn()?;
     info!("LSN initial = {}", 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"),
-            }
-        }
-    };
+    let mut intermediate_lsns = f(client, initial_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.
+    //
+    // Note: this is broken if pg_current_wal_insert_lsn is at page boundary
+    // because pg_current_wal_insert_lsn skips page headers.
     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"),
-    }
-    Ok((intermediate_lsns, last_lsn))
+    Ok(intermediate_lsns)
 }
 
 pub struct Simple;
 impl Crafter for Simple {
     const NAME: &'static str = "simple";
-    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<(Vec<PgLsn>, PgLsn)> {
+    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<Vec<PgLsn>> {
         craft_internal(client, |client, _| {
             client.execute("CREATE table t(x int)", &[])?;
-            Ok((Vec::new(), None))
+            Ok(Vec::new())
         })
     }
 }
@@ -292,29 +284,36 @@ 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>, PgLsn)> {
-        // Do not use generate_internal because here we end up with flush_lsn exactly on
+    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
         // 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()?;
-        let after_xlog_switch: PgLsn = client.query_one("SELECT pg_switch_wal()", &[])?.get(0);
-        let next_segment = PgLsn::from(0x0200_0000);
+        // 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,
+        );
         ensure!(
-            after_xlog_switch <= next_segment,
-            "XLOG_SWITCH message ended after the expected segment boundary: {} > {}",
-            after_xlog_switch,
+            xlog_switch_record_end <= next_segment,
+            "XLOG_SWITCH record ended after the expected segment boundary: {} > {}",
+            xlog_switch_record_end,
             next_segment
         );
-        Ok((vec![before_xlog_switch, after_xlog_switch], next_segment))
+        Ok(vec![before_xlog_switch, xlog_switch_record_end])
     }
 }
 
 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>, PgLsn)> {
+    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<Vec<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)?;
@@ -361,28 +360,29 @@ impl Crafter for LastWalRecordXlogSwitchEndsOnPageBoundary {
 
         // Emit the XLOG_SWITCH
         let before_xlog_switch = client.pg_current_wal_insert_lsn()?;
-        let after_xlog_switch: PgLsn = client.query_one("SELECT pg_switch_wal()", &[])?.get(0);
+        let xlog_switch_record_end: PgLsn = client.query_one("SELECT pg_switch_wal()", &[])?.get(0);
         let next_segment = PgLsn::from(0x0200_0000);
         ensure!(
-            after_xlog_switch < next_segment,
-            "XLOG_SWITCH message ended on or after the expected segment boundary: {} > {}",
-            after_xlog_switch,
+            xlog_switch_record_end < next_segment,
+            "XLOG_SWITCH record ended on or after the expected segment boundary: {} > {}",
+            xlog_switch_record_end,
             next_segment
         );
         ensure!(
-            u64::from(after_xlog_switch) as usize % XLOG_BLCKSZ == XLOG_SIZE_OF_XLOG_SHORT_PHD,
+            u64::from(xlog_switch_record_end) as usize % XLOG_BLCKSZ == XLOG_SIZE_OF_XLOG_SHORT_PHD,
             "XLOG_SWITCH message ended not on page boundary: {}, offset = {}",
-            after_xlog_switch,
-            u64::from(after_xlog_switch) as usize % XLOG_BLCKSZ
+            xlog_switch_record_end,
+            u64::from(xlog_switch_record_end) as usize % XLOG_BLCKSZ
         );
-        Ok((vec![before_xlog_switch, after_xlog_switch], next_segment))
+        Ok(vec![before_xlog_switch, xlog_switch_record_end])
     }
 }
 
-fn craft_single_logical_message(
+/// Write ~16MB logical message; it should cross WAL segment.
+fn craft_seg_size_logical_message(
     client: &mut impl postgres::GenericClient,
     transactional: bool,
-) -> anyhow::Result<(Vec<PgLsn>, PgLsn)> {
+) -> anyhow::Result<Vec<PgLsn>> {
     craft_internal(client, |client, initial_lsn| {
         ensure!(
             initial_lsn < PgLsn::from(0x0200_0000 - 1024 * 1024),
@@ -405,34 +405,24 @@ fn craft_single_logical_message(
             "Logical message crossed two segments"
         );
 
-        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)))
-        }
+        Ok(vec![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>, PgLsn)> {
-        craft_single_logical_message(client, true)
+    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)
     }
 }
 
 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>, PgLsn)> {
-        craft_single_logical_message(client, false)
+    fn craft(client: &mut impl postgres::GenericClient) -> anyhow::Result<Vec<PgLsn>> {
+        craft_seg_size_logical_message(client, false)
     }
 }

libs/postgres_ffi/wal_craft/src/xlog_utils_test.rs

@@ -11,13 +11,15 @@ 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::*;
 
@@ -38,13 +40,13 @@ fn test_end_of_wal<C: crate::Crafter>(test_name: &str) {
     }
     cfg.initdb().unwrap();
     let srv = cfg.start_server().unwrap();
-    let (intermediate_lsns, expected_end_of_wal_partial) =
-        C::craft(&mut srv.connect_with_timeout().unwrap()).unwrap();
+    let intermediate_lsns = C::craft(&mut srv.connect_with_timeout().unwrap()).unwrap();
     let intermediate_lsns: Vec<Lsn> = intermediate_lsns
         .iter()
         .map(|&lsn| u64::from(lsn).into())
         .collect();
-    let expected_end_of_wal: Lsn = u64::from(expected_end_of_wal_partial).into();
+    // Kill postgres. Note that it might have inserted to WAL something after
+    // 'craft' did its job.
     srv.kill();
 
     // Check find_end_of_wal on the initial WAL
@@ -56,7 +58,7 @@ fn test_end_of_wal<C: crate::Crafter>(test_name: &str) {
         .filter(|fname| IsXLogFileName(fname))
         .max()
         .unwrap();
-    check_pg_waldump_end_of_wal(&cfg, &last_segment, expected_end_of_wal);
+    let expected_end_of_wal = find_pg_waldump_end_of_wal(&cfg, &last_segment);
     for start_lsn in intermediate_lsns
         .iter()
         .chain(std::iter::once(&expected_end_of_wal))
@@ -91,11 +93,7 @@ fn test_end_of_wal<C: crate::Crafter>(test_name: &str) {
     }
 }
 
-fn check_pg_waldump_end_of_wal(
-    cfg: &crate::Conf,
-    last_segment: &str,
-    expected_end_of_wal: Lsn,
-) {
+fn find_pg_waldump_end_of_wal(cfg: &crate::Conf, last_segment: &str) -> Lsn {
     // Get the actual end of WAL by pg_waldump
     let waldump_output = cfg
         .pg_waldump("000000010000000000000001", last_segment)
@@ -113,11 +111,8 @@ fn check_pg_waldump_end_of_wal(
         }
     };
     let waldump_wal_end = Lsn::from_str(caps.get(1).unwrap().as_str()).unwrap();
-    info!(
-        "waldump erred on {}, expected wal end at {}",
-        waldump_wal_end, expected_end_of_wal
-    );
-    assert_eq!(waldump_wal_end, expected_end_of_wal);
+    info!("waldump erred on {}", waldump_wal_end);
+    waldump_wal_end
 }
 
 fn check_end_of_wal(
@@ -210,9 +205,9 @@ 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/Cargo.toml

@@ -18,6 +18,7 @@ camino.workspace = true
 humantime.workspace = true
 hyper = { workspace = true, features = ["stream"] }
 futures.workspace = true
+rand.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 tokio = { workspace = true, features = ["sync", "fs", "io-util"] }

libs/remote_storage/src/azure_blob.rs

@@ -157,9 +157,8 @@ impl AzureBlobStorage {
             let mut bufs = Vec::new();
             while let Some(part) = response.next().await {
                 let part = part?;
-                let etag_str: &str = part.blob.properties.etag.as_ref();
                 if etag.is_none() {
-                    etag = Some(etag.unwrap_or_else(|| etag_str.to_owned()));
+                    etag = Some(part.blob.properties.etag);
                 }
                 if last_modified.is_none() {
                     last_modified = Some(part.blob.properties.last_modified.into());
@@ -174,6 +173,16 @@ impl AzureBlobStorage {
                     .map_err(|e| DownloadError::Other(e.into()))?;
                 bufs.push(data);
             }
+
+            if bufs.is_empty() {
+                return Err(DownloadError::Other(anyhow::anyhow!(
+                    "Azure GET response contained no buffers"
+                )));
+            }
+            // unwrap safety: if these were None, bufs would be empty and we would have returned an error already
+            let etag = etag.unwrap();
+            let last_modified = last_modified.unwrap();
+
             Ok(Download {
                 download_stream: Box::pin(futures::stream::iter(bufs.into_iter().map(Ok))),
                 etag,

libs/remote_storage/src/lib.rs

@@ -42,6 +42,9 @@ pub use self::{
 };
 use s3_bucket::RequestKind;
 
+/// Azure SDK's ETag type is a simple String wrapper: we use this internally instead of repeating it here.
+pub use azure_core::Etag;
+
 pub use error::{DownloadError, TimeTravelError, TimeoutOrCancel};
 
 /// Currently, sync happens with AWS S3, that has two limits on requests per second:
@@ -291,9 +294,9 @@ pub type DownloadStream =
 pub struct Download {
     pub download_stream: DownloadStream,
     /// The last time the file was modified (`last-modified` HTTP header)
-    pub last_modified: Option<SystemTime>,
+    pub last_modified: SystemTime,
     /// A way to identify this specific version of the resource (`etag` HTTP header)
-    pub etag: Option<String>,
+    pub etag: Etag,
     /// Extra key-value data, associated with the current remote file.
     pub metadata: Option<StorageMetadata>,
 }
@@ -562,6 +565,16 @@ 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 {

libs/remote_storage/src/local_fs.rs

@@ -10,7 +10,7 @@ use std::{
     io::ErrorKind,
     num::NonZeroU32,
     pin::Pin,
-    time::{Duration, SystemTime},
+    time::{Duration, SystemTime, UNIX_EPOCH},
 };
 
 use anyhow::{bail, ensure, Context};
@@ -30,6 +30,7 @@ use crate::{
 };
 
 use super::{RemoteStorage, StorageMetadata};
+use crate::Etag;
 
 const LOCAL_FS_TEMP_FILE_SUFFIX: &str = "___temp";
 
@@ -197,6 +198,7 @@ impl LocalFs {
             fs::OpenOptions::new()
                 .write(true)
                 .create(true)
+                .truncate(true)
                 .open(&temp_file_path)
                 .await
                 .with_context(|| {
@@ -406,35 +408,37 @@ impl RemoteStorage for LocalFs {
         cancel: &CancellationToken,
     ) -> Result<Download, DownloadError> {
         let target_path = from.with_base(&self.storage_root);
-        if file_exists(&target_path).map_err(DownloadError::BadInput)? {
-            let source = ReaderStream::new(
-                fs::OpenOptions::new()
-                    .read(true)
-                    .open(&target_path)
-                    .await
-                    .with_context(|| {
-                        format!("Failed to open source file {target_path:?} to use in the download")
-                    })
-                    .map_err(DownloadError::Other)?,
-            );
 
-            let metadata = self
-                .read_storage_metadata(&target_path)
+        let file_metadata = file_metadata(&target_path).await?;
+
+        let source = ReaderStream::new(
+            fs::OpenOptions::new()
+                .read(true)
+                .open(&target_path)
                 .await
-                .map_err(DownloadError::Other)?;
+                .with_context(|| {
+                    format!("Failed to open source file {target_path:?} to use in the download")
+                })
+                .map_err(DownloadError::Other)?,
+        );
 
-            let cancel_or_timeout = crate::support::cancel_or_timeout(self.timeout, cancel.clone());
-            let source = crate::support::DownloadStream::new(cancel_or_timeout, source);
+        let metadata = self
+            .read_storage_metadata(&target_path)
+            .await
+            .map_err(DownloadError::Other)?;
 
-            Ok(Download {
-                metadata,
-                last_modified: None,
-                etag: None,
-                download_stream: Box::pin(source),
-            })
-        } else {
-            Err(DownloadError::NotFound)
-        }
+        let cancel_or_timeout = crate::support::cancel_or_timeout(self.timeout, cancel.clone());
+        let source = crate::support::DownloadStream::new(cancel_or_timeout, source);
+
+        let etag = mock_etag(&file_metadata);
+        Ok(Download {
+            metadata,
+            last_modified: file_metadata
+                .modified()
+                .map_err(|e| DownloadError::Other(anyhow::anyhow!(e).context("Reading mtime")))?,
+            etag,
+            download_stream: Box::pin(source),
+        })
     }
 
     async fn download_byte_range(
@@ -452,50 +456,51 @@ impl RemoteStorage for LocalFs {
                 return Err(DownloadError::Other(anyhow::anyhow!("Invalid range, start ({start_inclusive}) and end_exclusive ({end_exclusive:?}) difference is zero bytes")));
             }
         }
+
         let target_path = from.with_base(&self.storage_root);
-        if file_exists(&target_path).map_err(DownloadError::BadInput)? {
-            let mut source = tokio::fs::OpenOptions::new()
-                .read(true)
-                .open(&target_path)
-                .await
-                .with_context(|| {
-                    format!("Failed to open source file {target_path:?} to use in the download")
-                })
-                .map_err(DownloadError::Other)?;
-
-            let len = source
-                .metadata()
-                .await
-                .context("query file length")
-                .map_err(DownloadError::Other)?
-                .len();
-
-            source
-                .seek(io::SeekFrom::Start(start_inclusive))
-                .await
-                .context("Failed to seek to the range start in a local storage file")
-                .map_err(DownloadError::Other)?;
-
-            let metadata = self
-                .read_storage_metadata(&target_path)
-                .await
-                .map_err(DownloadError::Other)?;
-
-            let source = source.take(end_exclusive.unwrap_or(len) - start_inclusive);
-            let source = ReaderStream::new(source);
-
-            let cancel_or_timeout = crate::support::cancel_or_timeout(self.timeout, cancel.clone());
-            let source = crate::support::DownloadStream::new(cancel_or_timeout, source);
-
-            Ok(Download {
-                metadata,
-                last_modified: None,
-                etag: None,
-                download_stream: Box::pin(source),
+        let file_metadata = file_metadata(&target_path).await?;
+        let mut source = tokio::fs::OpenOptions::new()
+            .read(true)
+            .open(&target_path)
+            .await
+            .with_context(|| {
+                format!("Failed to open source file {target_path:?} to use in the download")
             })
-        } else {
-            Err(DownloadError::NotFound)
-        }
+            .map_err(DownloadError::Other)?;
+
+        let len = source
+            .metadata()
+            .await
+            .context("query file length")
+            .map_err(DownloadError::Other)?
+            .len();
+
+        source
+            .seek(io::SeekFrom::Start(start_inclusive))
+            .await
+            .context("Failed to seek to the range start in a local storage file")
+            .map_err(DownloadError::Other)?;
+
+        let metadata = self
+            .read_storage_metadata(&target_path)
+            .await
+            .map_err(DownloadError::Other)?;
+
+        let source = source.take(end_exclusive.unwrap_or(len) - start_inclusive);
+        let source = ReaderStream::new(source);
+
+        let cancel_or_timeout = crate::support::cancel_or_timeout(self.timeout, cancel.clone());
+        let source = crate::support::DownloadStream::new(cancel_or_timeout, source);
+
+        let etag = mock_etag(&file_metadata);
+        Ok(Download {
+            metadata,
+            last_modified: file_metadata
+                .modified()
+                .map_err(|e| DownloadError::Other(anyhow::anyhow!(e).context("Reading mtime")))?,
+            etag,
+            download_stream: Box::pin(source),
+        })
     }
 
     async fn delete(&self, path: &RemotePath, _cancel: &CancellationToken) -> anyhow::Result<()> {
@@ -610,13 +615,22 @@ async fn create_target_directory(target_file_path: &Utf8Path) -> anyhow::Result<
     Ok(())
 }
 
-fn file_exists(file_path: &Utf8Path) -> anyhow::Result<bool> {
-    if file_path.exists() {
-        ensure!(file_path.is_file(), "file path '{file_path}' is not a file");
-        Ok(true)
-    } else {
-        Ok(false)
-    }
+async fn file_metadata(file_path: &Utf8Path) -> Result<std::fs::Metadata, DownloadError> {
+    tokio::fs::metadata(&file_path).await.map_err(|e| {
+        if e.kind() == ErrorKind::NotFound {
+            DownloadError::NotFound
+        } else {
+            DownloadError::BadInput(e.into())
+        }
+    })
+}
+
+// Use mtime as stand-in for ETag.  We could calculate a meaningful one by md5'ing the contents of files we
+// read, but that's expensive and the local_fs test helper's whole reason for existence is to run small tests
+// quickly, with less overhead than using a mock S3 server.
+fn mock_etag(meta: &std::fs::Metadata) -> Etag {
+    let mtime = meta.modified().expect("Filesystem mtime missing");
+    format!("{}", mtime.duration_since(UNIX_EPOCH).unwrap().as_millis()).into()
 }
 
 #[cfg(test)]

libs/remote_storage/src/s3_bucket.rs

@@ -35,8 +35,8 @@ use aws_sdk_s3::{
 };
 use aws_smithy_async::rt::sleep::TokioSleep;
 
-use aws_smithy_types::byte_stream::ByteStream;
 use aws_smithy_types::{body::SdkBody, DateTime};
+use aws_smithy_types::{byte_stream::ByteStream, date_time::ConversionError};
 use bytes::Bytes;
 use futures::stream::Stream;
 use hyper::Body;
@@ -287,8 +287,17 @@ impl S3Bucket {
         let remaining = self.timeout.saturating_sub(started_at.elapsed());
 
         let metadata = object_output.metadata().cloned().map(StorageMetadata);
-        let etag = object_output.e_tag;
-        let last_modified = object_output.last_modified.and_then(|t| t.try_into().ok());
+        let etag = object_output
+            .e_tag
+            .ok_or(DownloadError::Other(anyhow::anyhow!("Missing ETag header")))?
+            .into();
+        let last_modified = object_output
+            .last_modified
+            .ok_or(DownloadError::Other(anyhow::anyhow!(
+                "Missing LastModified header"
+            )))?
+            .try_into()
+            .map_err(|e: ConversionError| DownloadError::Other(e.into()))?;
 
         let body = object_output.body;
         let body = ByteStreamAsStream::from(body);

libs/remote_storage/tests/test_real_azure.rs

@@ -57,7 +57,6 @@ enum MaybeEnabledStorage {
     Disabled,
 }
 
-#[async_trait::async_trait]
 impl AsyncTestContext for MaybeEnabledStorage {
     async fn setup() -> Self {
         ensure_logging_ready();
@@ -86,7 +85,6 @@ struct AzureWithTestBlobs {
     remote_blobs: HashSet<RemotePath>,
 }
 
-#[async_trait::async_trait]
 impl AsyncTestContext for MaybeEnabledStorageWithTestBlobs {
     async fn setup() -> Self {
         ensure_logging_ready();
@@ -148,7 +146,6 @@ 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

@@ -17,6 +17,7 @@ use remote_storage::{
 };
 use test_context::test_context;
 use test_context::AsyncTestContext;
+use tokio::io::AsyncBufReadExt;
 use tokio_util::sync::CancellationToken;
 use tracing::info;
 
@@ -117,7 +118,7 @@ async fn s3_time_travel_recovery_works(ctx: &mut MaybeEnabledStorage) -> anyhow:
     // A little check to ensure that our clock is not too far off from the S3 clock
     {
         let dl = retry(|| ctx.client.download(&path2, &cancel)).await?;
-        let last_modified = dl.last_modified.unwrap();
+        let last_modified = dl.last_modified;
         let half_wt = WAIT_TIME.mul_f32(0.5);
         let t0_hwt = t0 + half_wt;
         let t1_hwt = t1 - half_wt;
@@ -218,7 +219,6 @@ enum MaybeEnabledStorage {
     Disabled,
 }
 
-#[async_trait::async_trait]
 impl AsyncTestContext for MaybeEnabledStorage {
     async fn setup() -> Self {
         ensure_logging_ready();
@@ -247,7 +247,6 @@ struct S3WithTestBlobs {
     remote_blobs: HashSet<RemotePath>,
 }
 
-#[async_trait::async_trait]
 impl AsyncTestContext for MaybeEnabledStorageWithTestBlobs {
     async fn setup() -> Self {
         ensure_logging_ready();
@@ -309,7 +308,6 @@ struct S3WithSimpleTestBlobs {
     remote_blobs: HashSet<RemotePath>,
 }
 
-#[async_trait::async_trait]
 impl AsyncTestContext for MaybeEnabledStorageWithSimpleTestBlobs {
     async fn setup() -> Self {
         ensure_logging_ready();
@@ -484,32 +482,33 @@ async fn download_is_cancelled(ctx: &mut MaybeEnabledStorage) {
     ))
     .unwrap();
 
-    let len = upload_large_enough_file(&ctx.client, &path, &cancel).await;
+    let file_len = upload_large_enough_file(&ctx.client, &path, &cancel).await;
 
     {
-        let mut stream = ctx
+        let stream = ctx
             .client
             .download(&path, &cancel)
             .await
             .expect("download succeeds")
             .download_stream;
 
-        let first = stream
-            .next()
-            .await
-            .expect("should have the first blob")
-            .expect("should have succeeded");
+        let mut reader = std::pin::pin!(tokio_util::io::StreamReader::new(stream));
 
-        tracing::info!(len = first.len(), "downloaded first chunk");
+        let first = reader.fill_buf().await.expect("should have the first blob");
+
+        let len = first.len();
+        tracing::info!(len, "downloaded first chunk");
 
         assert!(
-            first.len() < len,
+            first.len() < file_len,
             "uploaded file is too small, we downloaded all on first chunk"
         );
 
+        reader.consume(len);
+
         cancel.cancel();
 
-        let next = stream.next().await.expect("stream should have more");
+        let next = reader.fill_buf().await;
 
         let e = next.expect_err("expected an error, but got a chunk?");
 
@@ -520,6 +519,10 @@ async fn download_is_cancelled(ctx: &mut MaybeEnabledStorage) {
                 .is_some_and(|e| matches!(e, DownloadError::Cancelled)),
             "{inner:?}"
         );
+
+        let e = DownloadError::from(e);
+
+        assert!(matches!(e, DownloadError::Cancelled), "{e:?}");
     }
 
     let cancel = CancellationToken::new();

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 the method from the previous scenario, and
+    // (If we used 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

@@ -13,6 +13,7 @@ testing = ["fail/failpoints"]
 [dependencies]
 arc-swap.workspace = true
 sentry.workspace = true
+async-compression.workspace = true
 async-trait.workspace = true
 anyhow.workspace = true
 bincode.workspace = true
@@ -21,6 +22,7 @@ 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}
@@ -36,6 +38,7 @@ serde_json.workspace = true
 signal-hook.workspace = true
 thiserror.workspace = true
 tokio.workspace = true
+tokio-tar.workspace = true
 tokio-util.workspace = true
 tracing.workspace = true
 tracing-error.workspace = true
@@ -46,6 +49,7 @@ strum.workspace = true
 strum_macros.workspace = true
 url.workspace = true
 uuid.workspace = true
+walkdir.workspace = true
 
 pq_proto.workspace = true
 postgres_connection.workspace = true

libs/utils/src/env.rs

@@ -0,0 +1,21 @@
+//! 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/history_buffer.rs

@@ -47,9 +47,10 @@ impl<T, const L: usize> ops::Deref for HistoryBufferWithDropCounter<T, L> {
     }
 }
 
-#[derive(serde::Serialize)]
+#[derive(serde::Serialize, serde::Deserialize)]
 struct SerdeRepr<T> {
     buffer: Vec<T>,
+    buffer_size: usize,
     drop_count: u64,
 }
 
@@ -61,6 +62,7 @@ where
         let HistoryBufferWithDropCounter { buffer, drop_count } = value;
         SerdeRepr {
             buffer: buffer.iter().cloned().collect(),
+            buffer_size: L,
             drop_count: *drop_count,
         }
     }
@@ -78,19 +80,52 @@ where
     }
 }
 
+impl<'de, T, const L: usize> serde::de::Deserialize<'de> for HistoryBufferWithDropCounter<T, L>
+where
+    T: Clone + serde::Deserialize<'de>,
+{
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: serde::Deserializer<'de>,
+    {
+        let SerdeRepr {
+            buffer: des_buffer,
+            drop_count,
+            buffer_size,
+        } = SerdeRepr::<T>::deserialize(deserializer)?;
+        if buffer_size != L {
+            use serde::de::Error;
+            return Err(D::Error::custom(format!(
+                "invalid buffer_size, expecting {L} got {buffer_size}"
+            )));
+        }
+        let mut buffer = HistoryBuffer::new();
+        buffer.extend(des_buffer);
+        Ok(HistoryBufferWithDropCounter { buffer, drop_count })
+    }
+}
+
 #[cfg(test)]
 mod test {
     use super::HistoryBufferWithDropCounter;
 
     #[test]
     fn test_basics() {
-        let mut b = HistoryBufferWithDropCounter::<_, 2>::default();
+        let mut b = HistoryBufferWithDropCounter::<usize, 2>::default();
         b.write(1);
         b.write(2);
         b.write(3);
         assert!(b.iter().any(|e| *e == 2));
         assert!(b.iter().any(|e| *e == 3));
         assert!(!b.iter().any(|e| *e == 1));
+
+        // round-trip serde
+        let round_tripped: HistoryBufferWithDropCounter<usize, 2> =
+            serde_json::from_str(&serde_json::to_string(&b).unwrap()).unwrap();
+        assert_eq!(
+            round_tripped.iter().cloned().collect::<Vec<_>>(),
+            b.iter().cloned().collect::<Vec<_>>()
+        );
     }
 
     #[test]

libs/utils/src/http/endpoint.rs

@@ -245,7 +245,7 @@ impl std::io::Write for ChannelWriter {
     }
 }
 
-async fn prometheus_metrics_handler(_req: Request<Body>) -> Result<Response<Body>, ApiError> {
+pub async fn prometheus_metrics_handler(_req: Request<Body>) -> Result<Response<Body>, ApiError> {
     SERVE_METRICS_COUNT.inc();
 
     let started_at = std::time::Instant::now();
@@ -367,7 +367,6 @@ pub fn make_router() -> RouterBuilder<hyper::Body, ApiError> {
         .middleware(Middleware::post_with_info(
             add_request_id_header_to_response,
         ))
-        .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
         .err_handler(route_error_handler)
 }
 

libs/utils/src/lib.rs

@@ -63,6 +63,7 @@ pub mod measured_stream;
 
 pub mod serde_percent;
 pub mod serde_regex;
+pub mod serde_system_time;
 
 pub mod pageserver_feedback;
 
@@ -87,6 +88,10 @@ pub mod failpoint_support;
 
 pub mod yielding_loop;
 
+pub mod zstd;
+
+pub mod env;
+
 /// 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/lock_file.rs

@@ -63,6 +63,7 @@ impl UnwrittenLockFile {
 pub fn create_exclusive(lock_file_path: &Utf8Path) -> anyhow::Result<UnwrittenLockFile> {
     let lock_file = fs::OpenOptions::new()
         .create(true) // O_CREAT
+        .truncate(true)
         .write(true)
         .open(lock_file_path)
         .context("open lock file")?;

libs/utils/src/pageserver_feedback.rs

@@ -29,12 +29,10 @@ pub struct PageserverFeedback {
     // Serialize with RFC3339 format.
     #[serde(with = "serde_systemtime")]
     pub replytime: SystemTime,
+    /// Used to track feedbacks from different shards. Always zero for unsharded tenants.
+    pub shard_number: u32,
 }
 
-// NOTE: Do not forget to increment this number when adding new fields to PageserverFeedback.
-// Do not remove previously available fields because this might be backwards incompatible.
-pub const PAGESERVER_FEEDBACK_FIELDS_NUMBER: u8 = 5;
-
 impl PageserverFeedback {
     pub fn empty() -> PageserverFeedback {
         PageserverFeedback {
@@ -43,6 +41,7 @@ impl PageserverFeedback {
             remote_consistent_lsn: Lsn::INVALID,
             disk_consistent_lsn: Lsn::INVALID,
             replytime: *PG_EPOCH,
+            shard_number: 0,
         }
     }
 
@@ -59,17 +58,26 @@ impl PageserverFeedback {
     //
     // TODO: change serialized fields names once all computes migrate to rename.
     pub fn serialize(&self, buf: &mut BytesMut) {
-        buf.put_u8(PAGESERVER_FEEDBACK_FIELDS_NUMBER); // # of keys
+        let buf_ptr = buf.len();
+        buf.put_u8(0); // # of keys, will be filled later
+        let mut nkeys = 0;
+
+        nkeys += 1;
         buf.put_slice(b"current_timeline_size\0");
         buf.put_i32(8);
         buf.put_u64(self.current_timeline_size);
 
+        nkeys += 1;
         buf.put_slice(b"ps_writelsn\0");
         buf.put_i32(8);
         buf.put_u64(self.last_received_lsn.0);
+
+        nkeys += 1;
         buf.put_slice(b"ps_flushlsn\0");
         buf.put_i32(8);
         buf.put_u64(self.disk_consistent_lsn.0);
+
+        nkeys += 1;
         buf.put_slice(b"ps_applylsn\0");
         buf.put_i32(8);
         buf.put_u64(self.remote_consistent_lsn.0);
@@ -80,9 +88,19 @@ impl PageserverFeedback {
             .expect("failed to serialize pg_replytime earlier than PG_EPOCH")
             .as_micros() as i64;
 
+        nkeys += 1;
         buf.put_slice(b"ps_replytime\0");
         buf.put_i32(8);
         buf.put_i64(timestamp);
+
+        if self.shard_number > 0 {
+            nkeys += 1;
+            buf.put_slice(b"shard_number\0");
+            buf.put_i32(4);
+            buf.put_u32(self.shard_number);
+        }
+
+        buf[buf_ptr] = nkeys;
     }
 
     // Deserialize PageserverFeedback message
@@ -123,6 +141,11 @@ impl PageserverFeedback {
                         rf.replytime = *PG_EPOCH - Duration::from_micros(-raw_time as u64);
                     }
                 }
+                b"shard_number" => {
+                    let len = buf.get_i32();
+                    assert_eq!(len, 4);
+                    rf.shard_number = buf.get_u32();
+                }
                 _ => {
                     let len = buf.get_i32();
                     warn!(
@@ -194,10 +217,7 @@ mod tests {
         rf.serialize(&mut data);
 
         // Add an extra field to the buffer and adjust number of keys
-        if let Some(first) = data.first_mut() {
-            *first = PAGESERVER_FEEDBACK_FIELDS_NUMBER + 1;
-        }
-
+        data[0] += 1;
         data.put_slice(b"new_field_one\0");
         data.put_i32(8);
         data.put_u64(42);

libs/utils/src/seqwait.rs

@@ -182,6 +182,18 @@ 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

@@ -0,0 +1,55 @@
+//! 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

@@ -110,6 +110,49 @@ impl<T> OnceCell<T> {
         }
     }
 
+    /// Returns a guard to an existing initialized value, or returns an unique initialization
+    /// permit which can be used to initialize this `OnceCell` using `OnceCell::set`.
+    pub async fn get_or_init_detached(&self) -> Result<Guard<'_, T>, InitPermit> {
+        // It looks like OnceCell::get_or_init could be implemented using this method instead of
+        // duplication. However, that makes the future be !Send due to possibly holding on to the
+        // MutexGuard over an await point.
+        loop {
+            let sem = {
+                let guard = self.inner.lock().unwrap();
+                if guard.value.is_some() {
+                    return Ok(Guard(guard));
+                }
+                guard.init_semaphore.clone()
+            };
+
+            {
+                let permit = {
+                    // increment the count for the duration of queued
+                    let _guard = CountWaitingInitializers::start(self);
+                    sem.acquire().await
+                };
+
+                let Ok(permit) = permit else {
+                    let guard = self.inner.lock().unwrap();
+                    if !Arc::ptr_eq(&sem, &guard.init_semaphore) {
+                        // there was a take_and_deinit in between
+                        continue;
+                    }
+                    assert!(
+                        guard.value.is_some(),
+                        "semaphore got closed, must be initialized"
+                    );
+                    return Ok(Guard(guard));
+                };
+
+                permit.forget();
+            }
+
+            let permit = InitPermit(sem);
+            return Err(permit);
+        }
+    }
+
     /// Assuming a permit is held after previous call to [`Guard::take_and_deinit`], it can be used
     /// to complete initializing the inner value.
     ///
@@ -202,7 +245,7 @@ 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) {
+    pub fn take_and_deinit(mut self) -> (T, InitPermit) {
         let mut swapped = Inner::default();
         let sem = swapped.init_semaphore.clone();
         // acquire and forget right away, moving the control over to InitPermit
@@ -481,4 +524,39 @@ mod tests {
 
         assert_eq!("t1", *cell.get().unwrap());
     }
+
+    #[tokio::test(start_paused = true)]
+    async fn detached_init_smoke() {
+        let target = OnceCell::default();
+
+        let Err(permit) = target.get_or_init_detached().await else {
+            unreachable!("it is not initialized")
+        };
+
+        tokio::time::timeout(
+            std::time::Duration::from_secs(3600 * 24 * 7 * 365),
+            target.get_or_init(|permit2| async { Ok::<_, Infallible>((11, permit2)) }),
+        )
+        .await
+        .expect_err("should timeout since we are already holding the permit");
+
+        target.set(42, permit);
+
+        let (_answer, permit) = {
+            let guard = target
+                .get_or_init(|permit| async { Ok::<_, Infallible>((11, permit)) })
+                .await
+                .unwrap();
+
+            assert_eq!(*guard, 42);
+
+            guard.take_and_deinit()
+        };
+
+        assert!(target.get().is_none());
+
+        target.set(11, permit);
+
+        assert_eq!(*target.get().unwrap(), 11);
+    }
 }

libs/utils/src/vec_map.rs

@@ -1,27 +1,60 @@
 use std::{alloc::Layout, cmp::Ordering, ops::RangeBounds};
 
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum VecMapOrdering {
+    Greater,
+    GreaterOrEqual,
+}
+
 /// Ordered map datastructure implemented in a Vec.
 /// Append only - can only add keys that are larger than the
 /// current max key.
+/// Ordering can be adjusted using [`VecMapOrdering`]
+/// during `VecMap` construction.
 #[derive(Clone, Debug)]
-pub struct VecMap<K, V>(Vec<(K, V)>);
+pub struct VecMap<K, V> {
+    data: Vec<(K, V)>,
+    ordering: VecMapOrdering,
+}
 
 impl<K, V> Default for VecMap<K, V> {
     fn default() -> Self {
-        VecMap(Default::default())
+        VecMap {
+            data: Default::default(),
+            ordering: VecMapOrdering::Greater,
+        }
     }
 }
 
-#[derive(Debug)]
-pub struct InvalidKey;
+#[derive(thiserror::Error, Debug)]
+pub enum VecMapError {
+    #[error("Key violates ordering constraint")]
+    InvalidKey,
+    #[error("Mismatched ordering constraints")]
+    ExtendOrderingError,
+}
 
 impl<K: Ord, V> VecMap<K, V> {
+    pub fn new(ordering: VecMapOrdering) -> Self {
+        Self {
+            data: Vec::new(),
+            ordering,
+        }
+    }
+
+    pub fn with_capacity(capacity: usize, ordering: VecMapOrdering) -> Self {
+        Self {
+            data: Vec::with_capacity(capacity),
+            ordering,
+        }
+    }
+
     pub fn is_empty(&self) -> bool {
-        self.0.is_empty()
+        self.data.is_empty()
     }
 
     pub fn as_slice(&self) -> &[(K, V)] {
-        self.0.as_slice()
+        self.data.as_slice()
     }
 
     /// This function may panic if given a range where the lower bound is
@@ -29,7 +62,7 @@ impl<K: Ord, V> VecMap<K, V> {
     pub fn slice_range<R: RangeBounds<K>>(&self, range: R) -> &[(K, V)] {
         use std::ops::Bound::*;
 
-        let binary_search = |k: &K| self.0.binary_search_by_key(&k, extract_key);
+        let binary_search = |k: &K| self.data.binary_search_by_key(&k, extract_key);
 
         let start_idx = match range.start_bound() {
             Unbounded => 0,
@@ -41,7 +74,7 @@ impl<K: Ord, V> VecMap<K, V> {
         };
 
         let end_idx = match range.end_bound() {
-            Unbounded => self.0.len(),
+            Unbounded => self.data.len(),
             Included(k) => match binary_search(k) {
                 Ok(idx) => idx + 1,
                 Err(idx) => idx,
@@ -49,34 +82,30 @@ impl<K: Ord, V> VecMap<K, V> {
             Excluded(k) => binary_search(k).unwrap_or_else(std::convert::identity),
         };
 
-        &self.0[start_idx..end_idx]
+        &self.data[start_idx..end_idx]
     }
 
     /// Add a key value pair to the map.
-    /// If `key` is less than or equal to the current maximum key
-    /// the pair will not be added and InvalidKey error will be returned.
-    pub fn append(&mut self, key: K, value: V) -> Result<usize, InvalidKey> {
-        if let Some((last_key, _last_value)) = self.0.last() {
-            if &key <= last_key {
-                return Err(InvalidKey);
-            }
-        }
+    /// If `key` is not respective of the `self` ordering the
+    /// pair will not be added and `InvalidKey` error will be returned.
+    pub fn append(&mut self, key: K, value: V) -> Result<usize, VecMapError> {
+        self.validate_key_order(&key)?;
 
         let delta_size = self.instrument_vec_op(|vec| vec.push((key, value)));
         Ok(delta_size)
     }
 
     /// Update the maximum key value pair or add a new key value pair to the map.
-    /// If `key` is less than the current maximum key no updates or additions
-    /// will occur and InvalidKey error will be returned.
+    /// If `key` is not respective of the `self` ordering no updates or additions
+    /// will occur and `InvalidKey` error will be returned.
     pub fn append_or_update_last(
         &mut self,
         key: K,
         mut value: V,
-    ) -> Result<(Option<V>, usize), InvalidKey> {
-        if let Some((last_key, last_value)) = self.0.last_mut() {
+    ) -> Result<(Option<V>, usize), VecMapError> {
+        if let Some((last_key, last_value)) = self.data.last_mut() {
             match key.cmp(last_key) {
-                Ordering::Less => return Err(InvalidKey),
+                Ordering::Less => return Err(VecMapError::InvalidKey),
                 Ordering::Equal => {
                     std::mem::swap(last_value, &mut value);
                     const DELTA_SIZE: usize = 0;
@@ -100,40 +129,67 @@ impl<K: Ord, V> VecMap<K, V> {
         V: Clone,
     {
         let split_idx = self
-            .0
+            .data
             .binary_search_by_key(&cutoff, extract_key)
             .unwrap_or_else(std::convert::identity);
 
         (
-            VecMap(self.0[..split_idx].to_vec()),
-            VecMap(self.0[split_idx..].to_vec()),
+            VecMap {
+                data: self.data[..split_idx].to_vec(),
+                ordering: self.ordering,
+            },
+            VecMap {
+                data: self.data[split_idx..].to_vec(),
+                ordering: self.ordering,
+            },
         )
     }
 
     /// Move items from `other` to the end of `self`, leaving `other` empty.
-    /// If any keys in `other` is less than or equal to any key in `self`,
-    /// `InvalidKey` error will be returned and no mutation will occur.
-    pub fn extend(&mut self, other: &mut Self) -> Result<usize, InvalidKey> {
-        let self_last_opt = self.0.last().map(extract_key);
-        let other_first_opt = other.0.last().map(extract_key);
+    /// If the `other` ordering is different from `self` ordering
+    /// `ExtendOrderingError` error will be returned.
+    /// If any keys in `other` is not respective of the ordering defined in
+    /// `self`, `InvalidKey` error will be returned and no mutation will occur.
+    pub fn extend(&mut self, other: &mut Self) -> Result<usize, VecMapError> {
+        if self.ordering != other.ordering {
+            return Err(VecMapError::ExtendOrderingError);
+        }
 
-        if let (Some(self_last), Some(other_first)) = (self_last_opt, other_first_opt) {
-            if self_last >= other_first {
-                return Err(InvalidKey);
+        let other_first_opt = other.data.last().map(extract_key);
+        if let Some(other_first) = other_first_opt {
+            self.validate_key_order(other_first)?;
+        }
+
+        let delta_size = self.instrument_vec_op(|vec| vec.append(&mut other.data));
+        Ok(delta_size)
+    }
+
+    /// Validate the current last key in `self` and key being
+    /// inserted against the order defined in `self`.
+    fn validate_key_order(&self, key: &K) -> Result<(), VecMapError> {
+        if let Some(last_key) = self.data.last().map(extract_key) {
+            match (&self.ordering, &key.cmp(last_key)) {
+                (VecMapOrdering::Greater, Ordering::Less | Ordering::Equal) => {
+                    return Err(VecMapError::InvalidKey);
+                }
+                (VecMapOrdering::Greater, Ordering::Greater) => {}
+                (VecMapOrdering::GreaterOrEqual, Ordering::Less) => {
+                    return Err(VecMapError::InvalidKey);
+                }
+                (VecMapOrdering::GreaterOrEqual, Ordering::Equal | Ordering::Greater) => {}
             }
         }
 
-        let delta_size = self.instrument_vec_op(|vec| vec.append(&mut other.0));
-        Ok(delta_size)
+        Ok(())
     }
 
     /// Instrument an operation on the underlying [`Vec`].
     /// Will panic if the operation decreases capacity.
     /// Returns the increase in memory usage caused by the op.
     fn instrument_vec_op(&mut self, op: impl FnOnce(&mut Vec<(K, V)>)) -> usize {
-        let old_cap = self.0.capacity();
-        op(&mut self.0);
-        let new_cap = self.0.capacity();
+        let old_cap = self.data.capacity();
+        op(&mut self.data);
+        let new_cap = self.data.capacity();
 
         match old_cap.cmp(&new_cap) {
             Ordering::Less => {
@@ -145,6 +201,36 @@ impl<K: Ord, V> VecMap<K, V> {
             Ordering::Greater => panic!("VecMap capacity shouldn't ever decrease"),
         }
     }
+
+    /// Similar to `from_iter` defined in `FromIter` trait except
+    /// that it accepts an [`VecMapOrdering`]
+    pub fn from_iter<I: IntoIterator<Item = (K, V)>>(iter: I, ordering: VecMapOrdering) -> Self {
+        let iter = iter.into_iter();
+        let initial_capacity = {
+            match iter.size_hint() {
+                (lower_bound, None) => lower_bound,
+                (_, Some(upper_bound)) => upper_bound,
+            }
+        };
+
+        let mut vec_map = VecMap::with_capacity(initial_capacity, ordering);
+        for (key, value) in iter {
+            vec_map
+                .append(key, value)
+                .expect("The passed collection needs to be sorted!");
+        }
+
+        vec_map
+    }
+}
+
+impl<K: Ord, V> IntoIterator for VecMap<K, V> {
+    type Item = (K, V);
+    type IntoIter = std::vec::IntoIter<(K, V)>;
+
+    fn into_iter(self) -> Self::IntoIter {
+        self.data.into_iter()
+    }
 }
 
 fn extract_key<K, V>(entry: &(K, V)) -> &K {
@@ -155,7 +241,7 @@ fn extract_key<K, V>(entry: &(K, V)) -> &K {
 mod tests {
     use std::{collections::BTreeMap, ops::Bound};
 
-    use super::VecMap;
+    use super::{VecMap, VecMapOrdering};
 
     #[test]
     fn unbounded_range() {
@@ -310,5 +396,59 @@ mod tests {
         left.extend(&mut one_map).unwrap_err();
         assert_eq!(left.as_slice(), &[(0, ()), (1, ())]);
         assert_eq!(one_map.as_slice(), &[(1, ())]);
+
+        let mut map_greater_or_equal = VecMap::new(VecMapOrdering::GreaterOrEqual);
+        map_greater_or_equal.append(2, ()).unwrap();
+        map_greater_or_equal.append(2, ()).unwrap();
+
+        left.extend(&mut map_greater_or_equal).unwrap_err();
+        assert_eq!(left.as_slice(), &[(0, ()), (1, ())]);
+        assert_eq!(map_greater_or_equal.as_slice(), &[(2, ()), (2, ())]);
+    }
+
+    #[test]
+    fn extend_with_ordering() {
+        let mut left = VecMap::new(VecMapOrdering::GreaterOrEqual);
+        left.append(0, ()).unwrap();
+        assert_eq!(left.as_slice(), &[(0, ())]);
+
+        let mut greater_right = VecMap::new(VecMapOrdering::Greater);
+        greater_right.append(0, ()).unwrap();
+        left.extend(&mut greater_right).unwrap_err();
+        assert_eq!(left.as_slice(), &[(0, ())]);
+
+        let mut greater_or_equal_right = VecMap::new(VecMapOrdering::GreaterOrEqual);
+        greater_or_equal_right.append(2, ()).unwrap();
+        greater_or_equal_right.append(2, ()).unwrap();
+        left.extend(&mut greater_or_equal_right).unwrap();
+        assert_eq!(left.as_slice(), &[(0, ()), (2, ()), (2, ())]);
+    }
+
+    #[test]
+    fn vec_map_from_sorted() {
+        let vec = vec![(1, ()), (2, ()), (3, ()), (6, ())];
+        let vec_map = VecMap::from_iter(vec, VecMapOrdering::Greater);
+        assert_eq!(vec_map.as_slice(), &[(1, ()), (2, ()), (3, ()), (6, ())]);
+
+        let vec = vec![(1, ()), (2, ()), (3, ()), (3, ()), (6, ()), (6, ())];
+        let vec_map = VecMap::from_iter(vec, VecMapOrdering::GreaterOrEqual);
+        assert_eq!(
+            vec_map.as_slice(),
+            &[(1, ()), (2, ()), (3, ()), (3, ()), (6, ()), (6, ())]
+        );
+    }
+
+    #[test]
+    #[should_panic]
+    fn vec_map_from_unsorted_greater() {
+        let vec = vec![(1, ()), (2, ()), (2, ()), (3, ()), (6, ())];
+        let _ = VecMap::from_iter(vec, VecMapOrdering::Greater);
+    }
+
+    #[test]
+    #[should_panic]
+    fn vec_map_from_unsorted_greater_or_equal() {
+        let vec = vec![(1, ()), (2, ()), (3, ()), (6, ()), (5, ())];
+        let _ = VecMap::from_iter(vec, VecMapOrdering::GreaterOrEqual);
     }
 }

libs/utils/src/zstd.rs

@@ -0,0 +1,78 @@
+use std::io::SeekFrom;
+
+use anyhow::{Context, Result};
+use async_compression::{
+    tokio::{bufread::ZstdDecoder, write::ZstdEncoder},
+    zstd::CParameter,
+    Level,
+};
+use camino::Utf8Path;
+use nix::NixPath;
+use tokio::{
+    fs::{File, OpenOptions},
+    io::AsyncBufRead,
+    io::AsyncSeekExt,
+    io::AsyncWriteExt,
+};
+use tokio_tar::{Archive, Builder, HeaderMode};
+use walkdir::WalkDir;
+
+/// Creates a Zstandard tarball.
+pub async fn create_zst_tarball(path: &Utf8Path, tarball: &Utf8Path) -> Result<(File, u64)> {
+    let file = OpenOptions::new()
+        .create(true)
+        .truncate(true)
+        .read(true)
+        .write(true)
+        .open(&tarball)
+        .await
+        .with_context(|| format!("tempfile creation {tarball}"))?;
+
+    let mut paths = Vec::new();
+    for entry in WalkDir::new(path) {
+        let entry = entry?;
+        let metadata = entry.metadata().expect("error getting dir entry metadata");
+        // Also allow directories so that we also get empty directories
+        if !(metadata.is_file() || metadata.is_dir()) {
+            continue;
+        }
+        let path = entry.into_path();
+        paths.push(path);
+    }
+    // Do a sort to get a more consistent listing
+    paths.sort_unstable();
+    let zstd = ZstdEncoder::with_quality_and_params(
+        file,
+        Level::Default,
+        &[CParameter::enable_long_distance_matching(true)],
+    );
+    let mut builder = Builder::new(zstd);
+    // Use reproducible header mode
+    builder.mode(HeaderMode::Deterministic);
+    for p in paths {
+        let rel_path = p.strip_prefix(path)?;
+        if rel_path.is_empty() {
+            // The top directory should not be compressed,
+            // the tar crate doesn't like that
+            continue;
+        }
+        builder.append_path_with_name(&p, rel_path).await?;
+    }
+    let mut zstd = builder.into_inner().await?;
+    zstd.shutdown().await?;
+    let mut compressed = zstd.into_inner();
+    let compressed_len = compressed.metadata().await?.len();
+    compressed.seek(SeekFrom::Start(0)).await?;
+    Ok((compressed, compressed_len))
+}
+
+/// Creates a Zstandard tarball.
+pub async fn extract_zst_tarball(
+    path: &Utf8Path,
+    tarball: impl AsyncBufRead + Unpin,
+) -> Result<()> {
+    let decoder = Box::pin(ZstdDecoder::new(tarball));
+    let mut archive = Archive::new(decoder);
+    archive.unpack(path).await?;
+    Ok(())
+}

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 the cgroup threshold; in
+    /// Minimum fraction of total system memory reserved *before* 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.

libs/walproposer/src/api_bindings.rs

@@ -324,11 +324,11 @@ extern "C" fn finish_sync_safekeepers(wp: *mut WalProposer, lsn: XLogRecPtr) {
     }
 }
 
-extern "C" fn process_safekeeper_feedback(wp: *mut WalProposer, commit_lsn: XLogRecPtr) {
+extern "C" fn process_safekeeper_feedback(wp: *mut WalProposer, sk: *mut Safekeeper) {
     unsafe {
         let callback_data = (*(*wp).config).callback_data;
         let api = callback_data as *mut Box<dyn ApiImpl>;
-        (*api).process_safekeeper_feedback(&mut (*wp), commit_lsn)
+        (*api).process_safekeeper_feedback(&mut (*wp), &mut (*sk));
     }
 }
 

libs/walproposer/src/walproposer.rs

@@ -142,7 +142,7 @@ pub trait ApiImpl {
         todo!()
     }
 
-    fn process_safekeeper_feedback(&self, _wp: &mut WalProposer, _commit_lsn: u64) {
+    fn process_safekeeper_feedback(&mut self, _wp: &mut WalProposer, _sk: &mut Safekeeper) {
         todo!()
     }
 

pageserver/Cargo.toml

@@ -59,6 +59,7 @@ 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"] }
@@ -89,6 +90,9 @@ enumset = { workspace = true, features = ["serde"]}
 strum.workspace = true
 strum_macros.workspace = true
 
+[target.'cfg(target_os = "linux")'.dependencies]
+procfs.workspace = true
+
 [dev-dependencies]
 criterion.workspace = true
 hex-literal.workspace = true

pageserver/benches/bench_walredo.rs

@@ -1,160 +1,156 @@
-//! Simple benchmarking around walredo.
+//! Quantify a single walredo manager's throughput under N concurrent callers.
 //!
-//! Right now they hope to just set a baseline. Later we can try to expand into latency and
-//! throughput after figuring out the coordinated omission problems below.
+//! The benchmark implementation ([`bench_impl`]) is parametrized by
+//! - `redo_work` => [`Request::short_request`] or [`Request::medium_request`]
+//! - `n_redos` => number of times the benchmark shell execute the `redo_work`
+//! - `nclients` => number of clients (more on this shortly).
 //!
-//! There are two sets of inputs; `short` and `medium`. They were collected on postgres v14 by
-//! logging what happens when a sequential scan is requested on a small table, then picking out two
-//! suitable from logs.
+//! The benchmark impl sets up a multi-threaded tokio runtime with default parameters.
+//! It spawns `nclients` times [`client`] tokio tasks.
+//! Each task executes the `redo_work` `n_redos/nclients` times.
 //!
+//! We exercise the following combinations:
+//! - `redo_work = short / medium``
+//! - `nclients = [1, 2, 4, 8, 16, 32, 64, 128]`
 //!
-//! Reference data (git blame to see commit) on an i3en.3xlarge
-// ```text
-//! short/short/1           time:   [39.175 µs 39.348 µs 39.536 µs]
-//! short/short/2           time:   [51.227 µs 51.487 µs 51.755 µs]
-//! short/short/4           time:   [76.048 µs 76.362 µs 76.674 µs]
-//! short/short/8           time:   [128.94 µs 129.82 µs 130.74 µs]
-//! short/short/16          time:   [227.84 µs 229.00 µs 230.28 µs]
-//! short/short/32          time:   [455.97 µs 457.81 µs 459.90 µs]
-//! short/short/64          time:   [902.46 µs 904.84 µs 907.32 µs]
-//! short/short/128         time:   [1.7416 ms 1.7487 ms 1.7561 ms]
-//! ``
-
-use std::sync::Arc;
+//! We let `criterion` determine the `n_redos` using `iter_custom`.
+//! The idea is that for each `(redo_work, nclients)` combination,
+//! criterion will run the `bench_impl` multiple times with different `n_redos`.
+//! The `bench_impl` reports the aggregate wall clock time from the clients' perspective.
+//! Criterion will divide that by `n_redos` to compute the "time per iteration".
+//! In our case, "time per iteration" means "time per redo_work execution".
+//!
+//! NB: the way by which `iter_custom` determines the "number of iterations"
+//! is called sampling. Apparently the idea here is to detect outliers.
+//! We're not sure whether the current choice of sampling method makes sense.
+//! See https://bheisler.github.io/criterion.rs/book/user_guide/command_line_output.html#collecting-samples
+//!
+//! # Reference Numbers
+//!
+//! 2024-04-04 on i3en.3xlarge
+//!
+//! ```text
+//! short/1                 time:   [25.925 µs 26.060 µs 26.209 µs]
+//! short/2                 time:   [31.277 µs 31.483 µs 31.722 µs]
+//! short/4                 time:   [45.496 µs 45.831 µs 46.182 µs]
+//! short/8                 time:   [84.298 µs 84.920 µs 85.566 µs]
+//! short/16                time:   [185.04 µs 186.41 µs 187.88 µs]
+//! short/32                time:   [385.01 µs 386.77 µs 388.70 µs]
+//! short/64                time:   [770.24 µs 773.04 µs 776.04 µs]
+//! short/128               time:   [1.5017 ms 1.5064 ms 1.5113 ms]
+//! medium/1                time:   [106.65 µs 107.20 µs 107.85 µs]
+//! medium/2                time:   [153.28 µs 154.24 µs 155.56 µs]
+//! medium/4                time:   [325.67 µs 327.01 µs 328.71 µs]
+//! medium/8                time:   [646.82 µs 650.17 µs 653.91 µs]
+//! medium/16               time:   [1.2645 ms 1.2701 ms 1.2762 ms]
+//! medium/32               time:   [2.4409 ms 2.4550 ms 2.4692 ms]
+//! medium/64               time:   [4.6814 ms 4.7114 ms 4.7408 ms]
+//! medium/128              time:   [8.7790 ms 8.9037 ms 9.0282 ms]
+//! ```
 
 use bytes::{Buf, Bytes};
-use pageserver::{
-    config::PageServerConf, repository::Key, walrecord::NeonWalRecord, walredo::PostgresRedoManager,
+use criterion::{BenchmarkId, Criterion};
+use pageserver::{config::PageServerConf, walrecord::NeonWalRecord, walredo::PostgresRedoManager};
+use pageserver_api::{key::Key, shard::TenantShardId};
+use std::{
+    sync::Arc,
+    time::{Duration, Instant},
 };
-use pageserver_api::shard::TenantShardId;
-use tokio::task::JoinSet;
+use tokio::{sync::Barrier, task::JoinSet};
 use utils::{id::TenantId, lsn::Lsn};
 
-use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion};
+fn bench(c: &mut Criterion) {
+    {
+        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));
+                },
+            );
+        }
+    }
 
-fn redo_scenarios(c: &mut Criterion) {
-    // logging should be enabled when adding more inputs, since walredo will only report malformed
-    // input to the stderr.
-    // utils::logging::init(utils::logging::LogFormat::Plain).unwrap();
+    {
+        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));
+                },
+            );
+        }
+    }
+}
+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(redo_work: Arc<Request>, n_redos: u64, nclients: u64) -> Duration {
     let repo_dir = camino_tempfile::tempdir_in(env!("CARGO_TARGET_TMPDIR")).unwrap();
 
     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());
 
-    let manager = PostgresRedoManager::new(conf, tenant_shard_id);
-
-    let manager = Arc::new(manager);
-
-    {
-        let rt = tokio::runtime::Builder::new_current_thread()
-            .enable_all()
-            .build()
-            .unwrap();
-        tracing::info!("executing first");
-        rt.block_on(short().execute(&manager)).unwrap();
-        tracing::info!("first executed");
-    }
-
-    let thread_counts = [1, 2, 4, 8, 16, 32, 64, 128];
-
-    let mut group = c.benchmark_group("short");
-    group.sampling_mode(criterion::SamplingMode::Flat);
-
-    for thread_count in thread_counts {
-        group.bench_with_input(
-            BenchmarkId::new("short", thread_count),
-            &thread_count,
-            |b, thread_count| {
-                add_multithreaded_walredo_requesters(b, *thread_count, &manager, short);
-            },
-        );
-    }
-    drop(group);
-
-    let mut group = c.benchmark_group("medium");
-    group.sampling_mode(criterion::SamplingMode::Flat);
-
-    for thread_count in thread_counts {
-        group.bench_with_input(
-            BenchmarkId::new("medium", thread_count),
-            &thread_count,
-            |b, thread_count| {
-                add_multithreaded_walredo_requesters(b, *thread_count, &manager, medium);
-            },
-        );
-    }
-    drop(group);
-}
-
-/// Sets up a multi-threaded tokio runtime with default worker thread count,
-/// then, spawn `requesters` tasks that repeatedly:
-/// - get input from `input_factor()`
-/// - call `manager.request_redo()` with their input
-///
-/// This stress-tests the scalability of a single walredo manager at high tokio-level concurrency.
-///
-/// Using tokio's default worker thread count means the results will differ on machines
-/// with different core countrs. We don't care about that, the performance will always
-/// be different on different hardware. To compare performance of different software versions,
-/// use the same hardware.
-fn add_multithreaded_walredo_requesters(
-    b: &mut criterion::Bencher,
-    nrequesters: usize,
-    manager: &Arc<PostgresRedoManager>,
-    input_factory: fn() -> Request,
-) {
-    assert_ne!(nrequesters, 0);
-
     let rt = tokio::runtime::Builder::new_multi_thread()
         .enable_all()
         .build()
         .unwrap();
 
-    let barrier = Arc::new(tokio::sync::Barrier::new(nrequesters + 1));
+    let start = Arc::new(Barrier::new(nclients as usize));
 
-    let mut requesters = JoinSet::new();
-    for _ in 0..nrequesters {
-        let _entered = rt.enter();
-        let manager = manager.clone();
-        let barrier = barrier.clone();
-        requesters.spawn(async move {
-            loop {
-                let input = input_factory();
-                barrier.wait().await;
-                let page = input.execute(&manager).await.unwrap();
-                assert_eq!(page.remaining(), 8192);
-                barrier.wait().await;
-            }
+    let mut tasks = JoinSet::new();
+
+    let manager = PostgresRedoManager::new(conf, tenant_shard_id);
+    let manager = Arc::new(manager);
+
+    for _ in 0..nclients {
+        rt.block_on(async {
+            tasks.spawn(client(
+                Arc::clone(&manager),
+                Arc::clone(&start),
+                Arc::clone(&redo_work),
+                // divide the amount of work equally among the clients
+                n_redos / nclients,
+            ))
         });
     }
 
-    let do_one_iteration = || {
-        rt.block_on(async {
-            barrier.wait().await;
-            // wait for work to complete
-            barrier.wait().await;
-        })
-    };
-
-    b.iter_batched(
-        || {
-            // warmup
-            do_one_iteration();
-        },
-        |()| {
-            // work loop
-            do_one_iteration();
-        },
-        criterion::BatchSize::PerIteration,
-    );
-
-    rt.block_on(requesters.shutdown());
+    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
+    })
 }
 
-criterion_group!(benches, redo_scenarios);
-criterion_main!(benches);
+async fn client(
+    mgr: Arc<PostgresRedoManager>,
+    start: Arc<Barrier>,
+    redo_work: Arc<Request>,
+    n_redos: u64,
+) -> Duration {
+    start.wait().await;
+    let start = Instant::now();
+    for _ in 0..n_redos {
+        let page = redo_work.execute(&mgr).await.unwrap();
+        assert_eq!(page.remaining(), 8192);
+        // The real pageserver will rarely if ever do 2 walredos in a row without
+        // yielding to the executor.
+        tokio::task::yield_now().await;
+    }
+    start.elapsed()
+}
 
 macro_rules! lsn {
     ($input:expr) => {{
@@ -166,12 +162,46 @@ macro_rules! lsn {
     }};
 }
 
-/// Short payload, 1132 bytes.
-// pg_records are copypasted from log, where they are put with Debug impl of Bytes, which uses \0
-// for null bytes.
-#[allow(clippy::octal_escapes)]
-fn short() -> Request {
-    Request {
+/// Simple wrapper around `WalRedoManager::request_redo`.
+///
+/// In benchmarks this is cloned around.
+#[derive(Clone)]
+struct Request {
+    key: Key,
+    lsn: Lsn,
+    base_img: Option<(Lsn, Bytes)>,
+    records: Vec<(Lsn, NeonWalRecord)>,
+    pg_version: u32,
+}
+
+impl Request {
+    async fn execute(&self, manager: &PostgresRedoManager) -> anyhow::Result<Bytes> {
+        let Request {
+            key,
+            lsn,
+            base_img,
+            records,
+            pg_version,
+        } = self;
+
+        // TODO: avoid these clones
+        manager
+            .request_redo(*key, *lsn, base_img.clone(), records.clone(), *pg_version)
+            .await
+    }
+
+    fn pg_record(will_init: bool, bytes: &'static [u8]) -> NeonWalRecord {
+        let rec = Bytes::from_static(bytes);
+        NeonWalRecord::Postgres { will_init, rec }
+    }
+
+    /// Short payload, 1132 bytes.
+    // pg_records are copypasted from log, where they are put with Debug impl of Bytes, which uses \0
+    // for null bytes.
+    #[allow(clippy::octal_escapes)]
+    pub fn short_input() -> Request {
+        let pg_record = Self::pg_record;
+        Request {
         key: Key {
             field1: 0,
             field2: 1663,
@@ -194,13 +224,14 @@ fn short() -> Request {
         ],
         pg_version: 14,
     }
-}
+    }
 
-/// Medium sized payload, serializes as 26393 bytes.
-// see [`short`]
-#[allow(clippy::octal_escapes)]
-fn medium() -> Request {
-    Request {
+    /// Medium sized payload, serializes as 26393 bytes.
+    // see [`short`]
+    #[allow(clippy::octal_escapes)]
+    pub fn medium_input() -> Request {
+        let pg_record = Self::pg_record;
+        Request {
         key: Key {
             field1: 0,
             field2: 1663,
@@ -442,37 +473,5 @@ fn medium() -> Request {
         ],
         pg_version: 14,
     }
-}
-
-fn pg_record(will_init: bool, bytes: &'static [u8]) -> NeonWalRecord {
-    let rec = Bytes::from_static(bytes);
-    NeonWalRecord::Postgres { will_init, rec }
-}
-
-/// Simple wrapper around `WalRedoManager::request_redo`.
-///
-/// In benchmarks this is cloned around.
-#[derive(Clone)]
-struct Request {
-    key: Key,
-    lsn: Lsn,
-    base_img: Option<(Lsn, Bytes)>,
-    records: Vec<(Lsn, NeonWalRecord)>,
-    pg_version: u32,
-}
-
-impl Request {
-    async fn execute(self, manager: &PostgresRedoManager) -> anyhow::Result<Bytes> {
-        let Request {
-            key,
-            lsn,
-            base_img,
-            records,
-            pg_version,
-        } = self;
-
-        manager
-            .request_redo(key, lsn, base_img, records, pg_version)
-            .await
     }
 }

pageserver/client/src/mgmt_api.rs

@@ -128,12 +128,12 @@ impl Client {
 
     pub async fn timeline_info(
         &self,
-        tenant_id: TenantId,
+        tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
         force_await_logical_size: ForceAwaitLogicalSize,
     ) -> Result<pageserver_api::models::TimelineInfo> {
         let uri = format!(
-            "{}/v1/tenant/{tenant_id}/timeline/{timeline_id}",
+            "{}/v1/tenant/{tenant_shard_id}/timeline/{timeline_id}",
             self.mgmt_api_endpoint
         );
 
@@ -151,11 +151,11 @@ impl Client {
 
     pub async fn keyspace(
         &self,
-        tenant_id: TenantId,
+        tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
     ) -> Result<pageserver_api::models::partitioning::Partitioning> {
         let uri = format!(
-            "{}/v1/tenant/{tenant_id}/timeline/{timeline_id}/keyspace",
+            "{}/v1/tenant/{tenant_shard_id}/timeline/{timeline_id}/keyspace",
             self.mgmt_api_endpoint
         );
         self.get(&uri)
@@ -169,7 +169,7 @@ impl Client {
         self.request(Method::GET, uri, ()).await
     }
 
-    async fn request<B: serde::Serialize, U: reqwest::IntoUrl>(
+    async fn request_noerror<B: serde::Serialize, U: reqwest::IntoUrl>(
         &self,
         method: Method,
         uri: U,
@@ -181,7 +181,16 @@ impl Client {
         } else {
             req
         };
-        let res = req.json(&body).send().await.map_err(Error::ReceiveBody)?;
+        req.json(&body).send().await.map_err(Error::ReceiveBody)
+    }
+
+    async fn request<B: serde::Serialize, U: reqwest::IntoUrl>(
+        &self,
+        method: Method,
+        uri: U,
+        body: B,
+    ) -> Result<reqwest::Response> {
+        let res = self.request_noerror(method, uri, body).await?;
         let response = res.error_from_body().await?;
         Ok(response)
     }
@@ -240,13 +249,26 @@ impl Client {
         Ok(())
     }
 
-    pub async fn tenant_secondary_download(&self, tenant_id: TenantShardId) -> Result<()> {
-        let uri = format!(
+    pub async fn tenant_secondary_download(
+        &self,
+        tenant_id: TenantShardId,
+        wait: Option<std::time::Duration>,
+    ) -> Result<(StatusCode, SecondaryProgress)> {
+        let mut path = reqwest::Url::parse(&format!(
             "{}/v1/tenant/{}/secondary/download",
             self.mgmt_api_endpoint, tenant_id
-        );
-        self.request(Method::POST, &uri, ()).await?;
-        Ok(())
+        ))
+        .expect("Cannot build URL");
+
+        if let Some(wait) = wait {
+            path.query_pairs_mut()
+                .append_pair("wait_ms", &format!("{}", wait.as_millis()));
+        }
+
+        let response = self.request(Method::POST, path, ()).await?;
+        let status = response.status();
+        let progress: SecondaryProgress = response.json().await.map_err(Error::ReceiveBody)?;
+        Ok((status, progress))
     }
 
     pub async fn location_config(
@@ -257,7 +279,7 @@ impl Client {
         lazy: bool,
     ) -> Result<()> {
         let req_body = TenantLocationConfigRequest {
-            tenant_id: tenant_shard_id,
+            tenant_id: Some(tenant_shard_id),
             config,
         };
 
@@ -416,4 +438,77 @@ impl Client {
             .await
             .map_err(Error::ReceiveBody)
     }
+
+    pub async fn get_utilization(&self) -> Result<PageserverUtilization> {
+        let uri = format!("{}/v1/utilization", self.mgmt_api_endpoint);
+        self.get(uri)
+            .await?
+            .json()
+            .await
+            .map_err(Error::ReceiveBody)
+    }
+
+    pub async fn layer_map_info(
+        &self,
+        tenant_shard_id: TenantShardId,
+        timeline_id: TimelineId,
+    ) -> Result<LayerMapInfo> {
+        let uri = format!(
+            "{}/v1/tenant/{}/timeline/{}/layer",
+            self.mgmt_api_endpoint, tenant_shard_id, timeline_id,
+        );
+        self.get(&uri)
+            .await?
+            .json()
+            .await
+            .map_err(Error::ReceiveBody)
+    }
+
+    pub async fn layer_evict(
+        &self,
+        tenant_shard_id: TenantShardId,
+        timeline_id: TimelineId,
+        layer_file_name: &str,
+    ) -> Result<bool> {
+        let uri = format!(
+            "{}/v1/tenant/{}/timeline/{}/layer/{}",
+            self.mgmt_api_endpoint, tenant_shard_id, timeline_id, layer_file_name
+        );
+        let resp = self.request_noerror(Method::DELETE, &uri, ()).await?;
+        match resp.status() {
+            StatusCode::OK => Ok(true),
+            StatusCode::NOT_MODIFIED => Ok(false),
+            // TODO: dedupe this pattern / introduce separate error variant?
+            status => Err(match resp.json::<HttpErrorBody>().await {
+                Ok(HttpErrorBody { msg }) => Error::ApiError(status, msg),
+                Err(_) => {
+                    Error::ReceiveErrorBody(format!("Http error ({}) at {}.", status.as_u16(), uri))
+                }
+            }),
+        }
+    }
+
+    pub async fn layer_ondemand_download(
+        &self,
+        tenant_shard_id: TenantShardId,
+        timeline_id: TimelineId,
+        layer_file_name: &str,
+    ) -> Result<bool> {
+        let uri = format!(
+            "{}/v1/tenant/{}/timeline/{}/layer/{}",
+            self.mgmt_api_endpoint, tenant_shard_id, timeline_id, layer_file_name
+        );
+        let resp = self.request_noerror(Method::GET, &uri, ()).await?;
+        match resp.status() {
+            StatusCode::OK => Ok(true),
+            StatusCode::NOT_MODIFIED => Ok(false),
+            // TODO: dedupe this pattern / introduce separate error variant?
+            status => Err(match resp.json::<HttpErrorBody>().await {
+                Ok(HttpErrorBody { msg }) => Error::ApiError(status, msg),
+                Err(_) => {
+                    Error::ReceiveErrorBody(format!("Http error ({}) at {}.", status.as_u16(), uri))
+                }
+            }),
+        }
+    }
 }

pageserver/compaction/Cargo.toml

@@ -11,7 +11,6 @@ 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

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

pageserver/compaction/src/interface.rs

@@ -3,7 +3,6 @@
 //!
 //! 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;
@@ -141,18 +140,16 @@ 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.
-    async fn load_keys<'a>(
+    fn load_keys<'a>(
         &self,
         ctx: &E::RequestContext,
-    ) -> anyhow::Result<Vec<Self::DeltaEntry<'_>>>;
+    ) -> impl Future<Output = anyhow::Result<Vec<Self::DeltaEntry<'_>>>> + Send;
 }
 
 pub trait CompactionImageLayer<E: CompactionJobExecutor + ?Sized>: CompactionLayer<E::Key> {}

pageserver/compaction/src/simulator.rs

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

pageserver/ctl/Cargo.toml

@@ -12,9 +12,14 @@ bytes.workspace = true
 camino.workspace = true
 clap = { workspace = true, features = ["string"] }
 git-version.workspace = true
+humantime.workspace = true
 pageserver = { path = ".." }
+pageserver_api.workspace = true
+remote_storage = { path = "../../libs/remote_storage" }
 postgres_ffi.workspace = true
 tokio.workspace = true
+tokio-util.workspace = true
+toml_edit.workspace = true
 utils.workspace = true
 svg_fmt.workspace = true
 workspace_hack.workspace = true

pageserver/ctl/src/main.rs

@@ -9,6 +9,11 @@ mod index_part;
 mod layer_map_analyzer;
 mod layers;
 
+use std::{
+    str::FromStr,
+    time::{Duration, SystemTime},
+};
+
 use camino::{Utf8Path, Utf8PathBuf};
 use clap::{Parser, Subcommand};
 use index_part::IndexPartCmd;
@@ -20,8 +25,16 @@ use pageserver::{
     tenant::{dump_layerfile_from_path, metadata::TimelineMetadata},
     virtual_file,
 };
+use pageserver_api::shard::TenantShardId;
 use postgres_ffi::ControlFileData;
-use utils::{lsn::Lsn, project_git_version};
+use remote_storage::{RemotePath, RemoteStorageConfig};
+use tokio_util::sync::CancellationToken;
+use utils::{
+    id::TimelineId,
+    logging::{self, LogFormat, TracingErrorLayerEnablement},
+    lsn::Lsn,
+    project_git_version,
+};
 
 project_git_version!(GIT_VERSION);
 
@@ -43,6 +56,7 @@ enum Commands {
     #[command(subcommand)]
     IndexPart(IndexPartCmd),
     PrintLayerFile(PrintLayerFileCmd),
+    TimeTravelRemotePrefix(TimeTravelRemotePrefixCmd),
     DrawTimeline {},
     AnalyzeLayerMap(AnalyzeLayerMapCmd),
     #[command(subcommand)]
@@ -68,6 +82,26 @@ struct PrintLayerFileCmd {
     path: Utf8PathBuf,
 }
 
+/// Roll back the time for the specified prefix using S3 history.
+///
+/// The command is fairly low level and powerful. Validation is only very light,
+/// so it is more powerful, and thus potentially more dangerous.
+#[derive(Parser)]
+struct TimeTravelRemotePrefixCmd {
+    /// A configuration string for the remote_storage configuration.
+    ///
+    /// Example: `remote_storage = { bucket_name = "aws-storage-bucket-name", bucket_region = "us-east-2" }`
+    config_toml_str: String,
+    /// remote prefix to time travel recover. For safety reasons, we require it to contain
+    /// a timeline or tenant ID in the prefix.
+    prefix: String,
+    /// Timestamp to travel to. Given in format like `2024-01-20T10:45:45Z`. Assumes UTC and second accuracy.
+    travel_to: String,
+    /// Timestamp of the start of the operation, must be after any changes we want to roll back and after.
+    /// You can use a few seconds before invoking the command. Same format as `travel_to`.
+    done_if_after: Option<String>,
+}
+
 #[derive(Parser)]
 struct AnalyzeLayerMapCmd {
     /// Pageserver data path
@@ -78,6 +112,14 @@ struct AnalyzeLayerMapCmd {
 
 #[tokio::main]
 async fn main() -> anyhow::Result<()> {
+    logging::init(
+        LogFormat::Plain,
+        TracingErrorLayerEnablement::EnableWithRustLogFilter,
+        logging::Output::Stdout,
+    )?;
+
+    logging::replace_panic_hook_with_tracing_panic_hook().forget();
+
     let cli = CliOpts::parse();
 
     match cli.command {
@@ -105,6 +147,42 @@ async fn main() -> anyhow::Result<()> {
                 print_layerfile(&cmd.path).await?;
             }
         }
+        Commands::TimeTravelRemotePrefix(cmd) => {
+            let timestamp = humantime::parse_rfc3339(&cmd.travel_to)
+                .map_err(|_e| anyhow::anyhow!("Invalid time for travel_to: '{}'", cmd.travel_to))?;
+
+            let done_if_after = if let Some(done_if_after) = &cmd.done_if_after {
+                humantime::parse_rfc3339(done_if_after).map_err(|_e| {
+                    anyhow::anyhow!("Invalid time for done_if_after: '{}'", done_if_after)
+                })?
+            } else {
+                const SAFETY_MARGIN: Duration = Duration::from_secs(3);
+                tokio::time::sleep(SAFETY_MARGIN).await;
+                // Convert to string representation and back to get rid of sub-second values
+                let done_if_after = SystemTime::now();
+                tokio::time::sleep(SAFETY_MARGIN).await;
+                done_if_after
+            };
+
+            let timestamp = strip_subsecond(timestamp);
+            let done_if_after = strip_subsecond(done_if_after);
+
+            let Some(prefix) = validate_prefix(&cmd.prefix) else {
+                println!("specified prefix '{}' failed validation", cmd.prefix);
+                return Ok(());
+            };
+            let toml_document = toml_edit::Document::from_str(&cmd.config_toml_str)?;
+            let toml_item = toml_document
+                .get("remote_storage")
+                .expect("need remote_storage");
+            let config = RemoteStorageConfig::from_toml(toml_item)?.expect("incomplete config");
+            let storage = remote_storage::GenericRemoteStorage::from_config(&config);
+            let cancel = CancellationToken::new();
+            storage
+                .unwrap()
+                .time_travel_recover(Some(&prefix), timestamp, done_if_after, &cancel)
+                .await?;
+        }
     };
     Ok(())
 }
@@ -185,3 +263,89 @@ fn handle_metadata(
 
     Ok(())
 }
+
+/// Ensures that the given S3 prefix is sufficiently constrained.
+/// The command is very risky already and we don't want to expose something
+/// that allows usually unintentional and quite catastrophic time travel of
+/// an entire bucket, which would be a major catastrophy and away
+/// by only one character change (similar to "rm -r /home /username/foobar").
+fn validate_prefix(prefix: &str) -> Option<RemotePath> {
+    if prefix.is_empty() {
+        // Empty prefix means we want to specify the *whole* bucket
+        return None;
+    }
+    let components = prefix.split('/').collect::<Vec<_>>();
+    let (last, components) = {
+        let last = components.last()?;
+        if last.is_empty() {
+            (
+                components.iter().nth_back(1)?,
+                &components[..(components.len() - 1)],
+            )
+        } else {
+            (last, &components[..])
+        }
+    };
+    'valid: {
+        if let Ok(_timeline_id) = TimelineId::from_str(last) {
+            // Ends in either a tenant or timeline ID
+            break 'valid;
+        }
+        if *last == "timelines" {
+            if let Some(before_last) = components.iter().nth_back(1) {
+                if let Ok(_tenant_id) = TenantShardId::from_str(before_last) {
+                    // Has a valid tenant id
+                    break 'valid;
+                }
+            }
+        }
+
+        return None;
+    }
+    RemotePath::from_string(prefix).ok()
+}
+
+fn strip_subsecond(timestamp: SystemTime) -> SystemTime {
+    let ts_str = humantime::format_rfc3339_seconds(timestamp).to_string();
+    humantime::parse_rfc3339(&ts_str).expect("can't parse just created timestamp")
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_validate_prefix() {
+        assert_eq!(validate_prefix(""), None);
+        assert_eq!(validate_prefix("/"), None);
+        #[track_caller]
+        fn assert_valid(prefix: &str) {
+            let remote_path = RemotePath::from_string(prefix).unwrap();
+            assert_eq!(validate_prefix(prefix), Some(remote_path));
+        }
+        assert_valid("wal/3aa8fcc61f6d357410b7de754b1d9001/641e5342083b2235ee3deb8066819683/");
+        // Path is not relative but absolute
+        assert_eq!(
+            validate_prefix(
+                "/wal/3aa8fcc61f6d357410b7de754b1d9001/641e5342083b2235ee3deb8066819683/"
+            ),
+            None
+        );
+        assert_valid("wal/3aa8fcc61f6d357410b7de754b1d9001/");
+        // Partial tenant IDs should be invalid, S3 will match all tenants with the specific ID prefix
+        assert_eq!(validate_prefix("wal/3aa8fcc61f6d357410b7d"), None);
+        assert_eq!(validate_prefix("wal"), None);
+        assert_eq!(validate_prefix("/wal/"), None);
+        assert_valid("pageserver/v1/tenants/3aa8fcc61f6d357410b7de754b1d9001");
+        // Partial tenant ID
+        assert_eq!(
+            validate_prefix("pageserver/v1/tenants/3aa8fcc61f6d357410b"),
+            None
+        );
+        assert_valid("pageserver/v1/tenants/3aa8fcc61f6d357410b7de754b1d9001/timelines");
+        assert_valid("pageserver/v1/tenants/3aa8fcc61f6d357410b7de754b1d9001-0004/timelines");
+        assert_valid("pageserver/v1/tenants/3aa8fcc61f6d357410b7de754b1d9001/timelines/");
+        assert_valid("pageserver/v1/tenants/3aa8fcc61f6d357410b7de754b1d9001/timelines/641e5342083b2235ee3deb8066819683");
+        assert_eq!(validate_prefix("pageserver/v1/tenants/"), None);
+    }
+}

pageserver/pagebench/src/cmd/basebackup.rs

@@ -1,4 +1,5 @@
 use anyhow::Context;
+use pageserver_api::shard::TenantShardId;
 use pageserver_client::mgmt_api::ForceAwaitLogicalSize;
 use pageserver_client::page_service::BasebackupRequest;
 
@@ -95,7 +96,7 @@ async fn main_impl(
             let timeline = *timeline;
             let info = mgmt_api_client
                 .timeline_info(
-                    timeline.tenant_id,
+                    TenantShardId::unsharded(timeline.tenant_id),
                     timeline.timeline_id,
                     ForceAwaitLogicalSize::No,
                 )

pageserver/pagebench/src/cmd/getpage_latest_lsn.rs

@@ -4,6 +4,7 @@ use pageserver_api::key::{is_rel_block_key, key_to_rel_block, Key};
 use pageserver_api::keyspace::KeySpaceAccum;
 use pageserver_api::models::PagestreamGetPageRequest;
 
+use pageserver_api::shard::TenantShardId;
 use tokio_util::sync::CancellationToken;
 use utils::id::TenantTimelineId;
 use utils::lsn::Lsn;
@@ -173,7 +174,10 @@ async fn main_impl(
                 let timeline = *timeline;
                 async move {
                     let partitioning = mgmt_api_client
-                        .keyspace(timeline.tenant_id, timeline.timeline_id)
+                        .keyspace(
+                            TenantShardId::unsharded(timeline.tenant_id),
+                            timeline.timeline_id,
+                        )
                         .await?;
                     let lsn = partitioning.at_lsn;
                     let start = Instant::now();
@@ -308,7 +312,11 @@ async fn main_impl(
                     let (rel_tag, block_no) =
                         key_to_rel_block(key).expect("we filter non-rel-block keys out above");
                     PagestreamGetPageRequest {
-                        latest: rng.gen_bool(args.req_latest_probability),
+                        horizon: if rng.gen_bool(args.req_latest_probability) {
+                            Lsn::MAX
+                        } else {
+                            r.timeline_lsn
+                        },
                         lsn: r.timeline_lsn,
                         rel: rel_tag,
                         blkno: block_no,

pageserver/pagebench/src/cmd/ondemand_download_churn.rs

@@ -0,0 +1,272 @@
+use pageserver_api::{models::HistoricLayerInfo, shard::TenantShardId};
+
+use pageserver_client::mgmt_api;
+use rand::seq::SliceRandom;
+use tracing::{debug, info};
+use utils::id::{TenantTimelineId, TimelineId};
+
+use tokio::{
+    sync::{mpsc, OwnedSemaphorePermit},
+    task::JoinSet,
+};
+
+use std::{
+    num::NonZeroUsize,
+    sync::{
+        atomic::{AtomicU64, Ordering},
+        Arc,
+    },
+    time::{Duration, Instant},
+};
+
+/// Evict & on-demand download random layers.
+#[derive(clap::Parser)]
+pub(crate) struct Args {
+    #[clap(long, default_value = "http://localhost:9898")]
+    mgmt_api_endpoint: String,
+    #[clap(long)]
+    pageserver_jwt: Option<String>,
+    #[clap(long)]
+    runtime: Option<humantime::Duration>,
+    #[clap(long, default_value = "1")]
+    tasks_per_target: NonZeroUsize,
+    #[clap(long, default_value = "1")]
+    concurrency_per_target: NonZeroUsize,
+    /// Probability for sending `latest=true` in the request (uniform distribution).
+    #[clap(long)]
+    limit_to_first_n_targets: Option<usize>,
+    /// Before starting the benchmark, live-reconfigure the pageserver to use the given
+    /// [`pageserver_api::models::virtual_file::IoEngineKind`].
+    #[clap(long)]
+    set_io_engine: Option<pageserver_api::models::virtual_file::IoEngineKind>,
+    targets: Option<Vec<TenantTimelineId>>,
+}
+
+pub(crate) fn main(args: Args) -> anyhow::Result<()> {
+    let rt = tokio::runtime::Builder::new_multi_thread()
+        .enable_all()
+        .build()?;
+    let task = rt.spawn(main_impl(args));
+    rt.block_on(task).unwrap().unwrap();
+    Ok(())
+}
+
+#[derive(Debug, Default)]
+struct LiveStats {
+    evictions: AtomicU64,
+    downloads: AtomicU64,
+    timeline_restarts: AtomicU64,
+}
+
+impl LiveStats {
+    fn eviction_done(&self) {
+        self.evictions.fetch_add(1, Ordering::Relaxed);
+    }
+    fn download_done(&self) {
+        self.downloads.fetch_add(1, Ordering::Relaxed);
+    }
+    fn timeline_restart_done(&self) {
+        self.timeline_restarts.fetch_add(1, Ordering::Relaxed);
+    }
+}
+
+async fn main_impl(args: Args) -> anyhow::Result<()> {
+    let args: &'static Args = Box::leak(Box::new(args));
+
+    let mgmt_api_client = Arc::new(pageserver_client::mgmt_api::Client::new(
+        args.mgmt_api_endpoint.clone(),
+        args.pageserver_jwt.as_deref(),
+    ));
+
+    if let Some(engine_str) = &args.set_io_engine {
+        mgmt_api_client.put_io_engine(engine_str).await?;
+    }
+
+    // discover targets
+    let timelines: Vec<TenantTimelineId> = crate::util::cli::targets::discover(
+        &mgmt_api_client,
+        crate::util::cli::targets::Spec {
+            limit_to_first_n_targets: args.limit_to_first_n_targets,
+            targets: args.targets.clone(),
+        },
+    )
+    .await?;
+
+    let mut tasks = JoinSet::new();
+
+    let live_stats = Arc::new(LiveStats::default());
+    tasks.spawn({
+        let live_stats = Arc::clone(&live_stats);
+        async move {
+            let mut last_at = Instant::now();
+            loop {
+                tokio::time::sleep_until((last_at + Duration::from_secs(1)).into()).await;
+                let now = Instant::now();
+                let delta: Duration = now - last_at;
+                last_at = now;
+
+                let LiveStats {
+                    evictions,
+                    downloads,
+                    timeline_restarts,
+                } = &*live_stats;
+                let evictions = evictions.swap(0, Ordering::Relaxed) as f64 / delta.as_secs_f64();
+                let downloads = downloads.swap(0, Ordering::Relaxed) as f64 / delta.as_secs_f64();
+                let timeline_restarts = timeline_restarts.swap(0, Ordering::Relaxed);
+                info!("evictions={evictions:.2}/s downloads={downloads:.2}/s timeline_restarts={timeline_restarts}");
+            }
+        }
+    });
+
+    for tl in timelines {
+        for _ in 0..args.tasks_per_target.get() {
+            tasks.spawn(timeline_actor(
+                args,
+                Arc::clone(&mgmt_api_client),
+                tl,
+                Arc::clone(&live_stats),
+            ));
+        }
+    }
+
+    while let Some(res) = tasks.join_next().await {
+        res.unwrap();
+    }
+    Ok(())
+}
+
+async fn timeline_actor(
+    args: &'static Args,
+    mgmt_api_client: Arc<pageserver_client::mgmt_api::Client>,
+    timeline: TenantTimelineId,
+    live_stats: Arc<LiveStats>,
+) {
+    // TODO: support sharding
+    let tenant_shard_id = TenantShardId::unsharded(timeline.tenant_id);
+
+    struct Timeline {
+        joinset: JoinSet<()>,
+        layers: Vec<mpsc::Sender<OwnedSemaphorePermit>>,
+        concurrency: Arc<tokio::sync::Semaphore>,
+    }
+    loop {
+        debug!("restarting timeline");
+        let layer_map_info = mgmt_api_client
+            .layer_map_info(tenant_shard_id, timeline.timeline_id)
+            .await
+            .unwrap();
+        let concurrency = Arc::new(tokio::sync::Semaphore::new(
+            args.concurrency_per_target.get(),
+        ));
+
+        let mut joinset = JoinSet::new();
+        let layers = layer_map_info
+            .historic_layers
+            .into_iter()
+            .map(|historic_layer| {
+                let (tx, rx) = mpsc::channel(1);
+                joinset.spawn(layer_actor(
+                    tenant_shard_id,
+                    timeline.timeline_id,
+                    historic_layer,
+                    rx,
+                    Arc::clone(&mgmt_api_client),
+                    Arc::clone(&live_stats),
+                ));
+                tx
+            })
+            .collect::<Vec<_>>();
+
+        let mut timeline = Timeline {
+            joinset,
+            layers,
+            concurrency,
+        };
+
+        live_stats.timeline_restart_done();
+
+        loop {
+            assert!(!timeline.joinset.is_empty());
+            if let Some(res) = timeline.joinset.try_join_next() {
+                debug!(?res, "a layer actor exited, should not happen");
+                timeline.joinset.shutdown().await;
+                break;
+            }
+
+            let mut permit = Some(
+                Arc::clone(&timeline.concurrency)
+                    .acquire_owned()
+                    .await
+                    .unwrap(),
+            );
+
+            loop {
+                let layer_tx = {
+                    let mut rng = rand::thread_rng();
+                    timeline.layers.choose_mut(&mut rng).expect("no layers")
+                };
+                match layer_tx.try_send(permit.take().unwrap()) {
+                    Ok(_) => break,
+                    Err(e) => match e {
+                        mpsc::error::TrySendError::Full(back) => {
+                            // TODO: retrying introduces bias away from slow downloaders
+                            permit.replace(back);
+                        }
+                        mpsc::error::TrySendError::Closed(_) => panic!(),
+                    },
+                }
+            }
+        }
+    }
+}
+
+async fn layer_actor(
+    tenant_shard_id: TenantShardId,
+    timeline_id: TimelineId,
+    mut layer: HistoricLayerInfo,
+    mut rx: mpsc::Receiver<tokio::sync::OwnedSemaphorePermit>,
+    mgmt_api_client: Arc<mgmt_api::Client>,
+    live_stats: Arc<LiveStats>,
+) {
+    #[derive(Clone, Copy)]
+    enum Action {
+        Evict,
+        OnDemandDownload,
+    }
+
+    while let Some(_permit) = rx.recv().await {
+        let action = if layer.is_remote() {
+            Action::OnDemandDownload
+        } else {
+            Action::Evict
+        };
+
+        let did_it = match action {
+            Action::Evict => {
+                let did_it = mgmt_api_client
+                    .layer_evict(tenant_shard_id, timeline_id, layer.layer_file_name())
+                    .await
+                    .unwrap();
+                live_stats.eviction_done();
+                did_it
+            }
+            Action::OnDemandDownload => {
+                let did_it = mgmt_api_client
+                    .layer_ondemand_download(tenant_shard_id, timeline_id, layer.layer_file_name())
+                    .await
+                    .unwrap();
+                live_stats.download_done();
+                did_it
+            }
+        };
+        if !did_it {
+            debug!("local copy of layer map appears out of sync, re-downloading");
+            return;
+        }
+        debug!("did it");
+        layer.set_remote(match action {
+            Action::Evict => true,
+            Action::OnDemandDownload => false,
+        });
+    }
+}

pageserver/pagebench/src/cmd/trigger_initial_size_calculation.rs

@@ -1,6 +1,7 @@
 use std::sync::Arc;
 
 use humantime::Duration;
+use pageserver_api::shard::TenantShardId;
 use tokio::task::JoinSet;
 use utils::id::TenantTimelineId;
 
@@ -59,7 +60,11 @@ async fn main_impl(args: Args) -> anyhow::Result<()> {
         let mgmt_api_client = Arc::clone(&mgmt_api_client);
         js.spawn(async move {
             let info = mgmt_api_client
-                .timeline_info(tl.tenant_id, tl.timeline_id, ForceAwaitLogicalSize::Yes)
+                .timeline_info(
+                    TenantShardId::unsharded(tl.tenant_id),
+                    tl.timeline_id,
+                    ForceAwaitLogicalSize::Yes,
+                )
                 .await
                 .unwrap();
 
@@ -74,7 +79,11 @@ async fn main_impl(args: Args) -> anyhow::Result<()> {
                 while !info.current_logical_size_is_accurate {
                     ticker.tick().await;
                     info = mgmt_api_client
-                        .timeline_info(tl.tenant_id, tl.timeline_id, ForceAwaitLogicalSize::Yes)
+                        .timeline_info(
+                            TenantShardId::unsharded(tl.tenant_id),
+                            tl.timeline_id,
+                            ForceAwaitLogicalSize::Yes,
+                        )
                         .await
                         .unwrap();
                 }

pageserver/pagebench/src/main.rs

@@ -16,6 +16,7 @@ mod util {
 mod cmd {
     pub(super) mod basebackup;
     pub(super) mod getpage_latest_lsn;
+    pub(super) mod ondemand_download_churn;
     pub(super) mod trigger_initial_size_calculation;
 }
 
@@ -25,6 +26,7 @@ enum Args {
     Basebackup(cmd::basebackup::Args),
     GetPageLatestLsn(cmd::getpage_latest_lsn::Args),
     TriggerInitialSizeCalculation(cmd::trigger_initial_size_calculation::Args),
+    OndemandDownloadChurn(cmd::ondemand_download_churn::Args),
 }
 
 fn main() {
@@ -43,6 +45,7 @@ fn main() {
         Args::TriggerInitialSizeCalculation(args) => {
             cmd::trigger_initial_size_calculation::main(args)
         }
+        Args::OndemandDownloadChurn(args) => cmd::ondemand_download_churn::main(args),
     }
     .unwrap()
 }

pageserver/src/basebackup.rs

@@ -361,9 +361,10 @@ where
 
     /// Add contents of relfilenode `src`, naming it as `dst`.
     async fn add_rel(&mut self, src: RelTag, dst: RelTag) -> anyhow::Result<()> {
+        let horizon = self.lsn; // we do not need latest version
         let nblocks = self
             .timeline
-            .get_rel_size(src, Version::Lsn(self.lsn), false, self.ctx)
+            .get_rel_size(src, Version::Lsn(self.lsn), horizon, self.ctx)
             .await?;
 
         // If the relation is empty, create an empty file
@@ -384,7 +385,7 @@ where
             for blknum in startblk..endblk {
                 let img = self
                     .timeline
-                    .get_rel_page_at_lsn(src, blknum, Version::Lsn(self.lsn), false, self.ctx)
+                    .get_rel_page_at_lsn(src, blknum, Version::Lsn(self.lsn), horizon, self.ctx)
                     .await?;
                 segment_data.extend_from_slice(&img[..]);
             }

pageserver/src/bin/pageserver.rs

@@ -1,3 +1,5 @@
+#![recursion_limit = "300"]
+
 //! Main entry point for the Page Server executable.
 
 use std::env::{var, VarError};
@@ -16,6 +18,7 @@ use pageserver::metrics::{STARTUP_DURATION, STARTUP_IS_LOADING};
 use pageserver::task_mgr::WALRECEIVER_RUNTIME;
 use pageserver::tenant::{secondary, TenantSharedResources};
 use remote_storage::GenericRemoteStorage;
+use tokio::signal::unix::SignalKind;
 use tokio::time::Instant;
 use tracing::*;
 
@@ -118,6 +121,9 @@ fn main() -> anyhow::Result<()> {
         &[("node_id", &conf.id.to_string())],
     );
 
+    // after setting up logging, log the effective IO engine choice
+    info!(?conf.virtual_file_io_engine, "starting with virtual_file IO engine");
+
     let tenants_path = conf.tenants_path();
     if !tenants_path.exists() {
         utils::crashsafe::create_dir_all(conf.tenants_path())
@@ -312,6 +318,7 @@ fn start_pageserver(
     let http_listener = tcp_listener::bind(http_addr)?;
 
     let pg_addr = &conf.listen_pg_addr;
+
     info!("Starting pageserver pg protocol handler on {pg_addr}");
     let pageserver_listener = tcp_listener::bind(pg_addr)?;
 
@@ -544,7 +551,7 @@ fn start_pageserver(
         let router_state = Arc::new(
             http::routes::State::new(
                 conf,
-                tenant_manager,
+                tenant_manager.clone(),
                 http_auth.clone(),
                 remote_storage.clone(),
                 broker_client.clone(),
@@ -594,32 +601,37 @@ fn start_pageserver(
             None,
             "consumption metrics collection",
             true,
-            async move {
-                // first wait until background jobs are cleared to launch.
-                //
-                // this is because we only process active tenants and timelines, and the
-                // Timeline::get_current_logical_size will spawn the logical size calculation,
-                // which will not be rate-limited.
-                let cancel = task_mgr::shutdown_token();
+            {
+                let tenant_manager = tenant_manager.clone();
+                async move {
+                    // first wait until background jobs are cleared to launch.
+                    //
+                    // this is because we only process active tenants and timelines, and the
+                    // Timeline::get_current_logical_size will spawn the logical size calculation,
+                    // which will not be rate-limited.
+                    let cancel = task_mgr::shutdown_token();
 
-                tokio::select! {
-                    _ = cancel.cancelled() => { return Ok(()); },
-                    _ = background_jobs_barrier.wait() => {}
-                };
+                    tokio::select! {
+                        _ = cancel.cancelled() => { return Ok(()); },
+                        _ = background_jobs_barrier.wait() => {}
+                    };
 
-                pageserver::consumption_metrics::collect_metrics(
-                    metric_collection_endpoint,
-                    conf.metric_collection_interval,
-                    conf.cached_metric_collection_interval,
-                    conf.synthetic_size_calculation_interval,
-                    conf.id,
-                    local_disk_storage,
-                    cancel,
-                    metrics_ctx,
-                )
-                .instrument(info_span!("metrics_collection"))
-                .await?;
-                Ok(())
+                    pageserver::consumption_metrics::collect_metrics(
+                        tenant_manager,
+                        metric_collection_endpoint,
+                        &conf.metric_collection_bucket,
+                        conf.metric_collection_interval,
+                        conf.cached_metric_collection_interval,
+                        conf.synthetic_size_calculation_interval,
+                        conf.id,
+                        local_disk_storage,
+                        cancel,
+                        metrics_ctx,
+                    )
+                    .instrument(info_span!("metrics_collection"))
+                    .await?;
+                    Ok(())
+                }
             },
         );
     }
@@ -660,41 +672,37 @@ fn start_pageserver(
     let mut shutdown_pageserver = Some(shutdown_pageserver.drop_guard());
 
     // All started up! Now just sit and wait for shutdown signal.
-    {
-        use signal_hook::consts::*;
-        let signal_handler = BACKGROUND_RUNTIME.spawn_blocking(move || {
-            let mut signals =
-                signal_hook::iterator::Signals::new([SIGINT, SIGTERM, SIGQUIT]).unwrap();
-            return signals
-                .forever()
-                .next()
-                .expect("forever() never returns None unless explicitly closed");
-        });
-        let signal = BACKGROUND_RUNTIME
-            .block_on(signal_handler)
-            .expect("join error");
-        match signal {
-            SIGQUIT => {
-                info!("Got signal {signal}. Terminating in immediate shutdown mode",);
-                std::process::exit(111);
-            }
-            SIGINT | SIGTERM => {
-                info!("Got signal {signal}. Terminating gracefully in fast shutdown mode",);
 
-                // This cancels the `shutdown_pageserver` cancellation tree.
-                // Right now that tree doesn't reach very far, and `task_mgr` is used instead.
-                // The plan is to change that over time.
-                shutdown_pageserver.take();
-                let bg_remote_storage = remote_storage.clone();
-                let bg_deletion_queue = deletion_queue.clone();
-                BACKGROUND_RUNTIME.block_on(pageserver::shutdown_pageserver(
-                    bg_remote_storage.map(|_| bg_deletion_queue),
-                    0,
-                ));
-                unreachable!()
-            }
-            _ => unreachable!(),
-        }
+    {
+        BACKGROUND_RUNTIME.block_on(async move {
+            let mut sigint = tokio::signal::unix::signal(SignalKind::interrupt()).unwrap();
+            let mut sigterm = tokio::signal::unix::signal(SignalKind::terminate()).unwrap();
+            let mut sigquit = tokio::signal::unix::signal(SignalKind::quit()).unwrap();
+            let signal = tokio::select! {
+                _ = sigquit.recv() => {
+                    info!("Got signal SIGQUIT. Terminating in immediate shutdown mode",);
+                    std::process::exit(111);
+                }
+                _ = sigint.recv() => { "SIGINT" },
+                _ = sigterm.recv() => { "SIGTERM" },
+            };
+
+            info!("Got signal {signal}. Terminating gracefully in fast shutdown mode",);
+
+            // This cancels the `shutdown_pageserver` cancellation tree.
+            // Right now that tree doesn't reach very far, and `task_mgr` is used instead.
+            // The plan is to change that over time.
+            shutdown_pageserver.take();
+            let bg_remote_storage = remote_storage.clone();
+            let bg_deletion_queue = deletion_queue.clone();
+            pageserver::shutdown_pageserver(
+                &tenant_manager,
+                bg_remote_storage.map(|_| bg_deletion_queue),
+                0,
+            )
+            .await;
+            unreachable!()
+        })
     }
 }
 

pageserver/src/config.rs

@@ -7,8 +7,9 @@
 use anyhow::{anyhow, bail, ensure, Context, Result};
 use pageserver_api::shard::TenantShardId;
 use remote_storage::{RemotePath, RemoteStorageConfig};
+use serde;
 use serde::de::IntoDeserializer;
-use std::env;
+use std::{collections::HashMap, env};
 use storage_broker::Uri;
 use utils::crashsafe::path_with_suffix_extension;
 use utils::id::ConnectionId;
@@ -29,18 +30,17 @@ use utils::{
     logging::LogFormat,
 };
 
-use crate::disk_usage_eviction_task::DiskUsageEvictionTaskConfig;
-use crate::tenant::config::TenantConf;
 use crate::tenant::config::TenantConfOpt;
 use crate::tenant::timeline::GetVectoredImpl;
 use crate::tenant::vectored_blob_io::MaxVectoredReadBytes;
 use crate::tenant::{
     TENANTS_SEGMENT_NAME, TENANT_DELETED_MARKER_FILE_NAME, TIMELINES_SEGMENT_NAME,
 };
-use crate::virtual_file;
+use crate::{disk_usage_eviction_task::DiskUsageEvictionTaskConfig, virtual_file::io_engine};
+use crate::{tenant::config::TenantConf, virtual_file};
 use crate::{
     IGNORED_TENANT_FILE_NAME, TENANT_CONFIG_NAME, TENANT_HEATMAP_BASENAME,
-    TENANT_LOCATION_CONFIG_NAME, TIMELINE_DELETE_MARK_SUFFIX, TIMELINE_UNINIT_MARK_SUFFIX,
+    TENANT_LOCATION_CONFIG_NAME, TIMELINE_DELETE_MARK_SUFFIX,
 };
 
 use self::defaults::DEFAULT_CONCURRENT_TENANT_WARMUP;
@@ -83,6 +83,10 @@ pub mod defaults {
 
     pub const DEFAULT_INGEST_BATCH_SIZE: u64 = 100;
 
+    #[cfg(target_os = "linux")]
+    pub const DEFAULT_VIRTUAL_FILE_IO_ENGINE: &str = "tokio-epoll-uring";
+
+    #[cfg(not(target_os = "linux"))]
     pub const DEFAULT_VIRTUAL_FILE_IO_ENGINE: &str = "std-fs";
 
     pub const DEFAULT_GET_VECTORED_IMPL: &str = "sequential";
@@ -91,6 +95,8 @@ pub mod defaults {
 
     pub const DEFAULT_VALIDATE_VECTORED_GET: bool = true;
 
+    pub const DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB: usize = 0;
+
     ///
     /// Default built-in configuration file.
     ///
@@ -152,6 +158,8 @@ pub mod defaults {
 #heatmap_upload_concurrency = {DEFAULT_HEATMAP_UPLOAD_CONCURRENCY}
 #secondary_download_concurrency = {DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY}
 
+#ephemeral_bytes_per_memory_kb = {DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB}
+
 [remote_storage]
 
 "#
@@ -230,6 +238,7 @@ pub struct PageServerConf {
     // How often to send unchanged cached metrics to the metrics endpoint.
     pub cached_metric_collection_interval: Duration,
     pub metric_collection_endpoint: Option<Url>,
+    pub metric_collection_bucket: Option<RemoteStorageConfig>,
     pub synthetic_size_calculation_interval: Duration,
 
     pub disk_usage_based_eviction: Option<DiskUsageEvictionTaskConfig>,
@@ -274,6 +283,13 @@ pub struct PageServerConf {
     pub max_vectored_read_bytes: MaxVectoredReadBytes,
 
     pub validate_vectored_get: bool,
+
+    /// How many bytes of ephemeral layer content will we allow per kilobyte of RAM.  When this
+    /// is exceeded, we start proactively closing ephemeral layers to limit the total amount
+    /// of ephemeral data.
+    ///
+    /// Setting this to zero disables limits on total ephemeral layer size.
+    pub ephemeral_bytes_per_memory_kb: usize,
 }
 
 /// We do not want to store this in a PageServerConf because the latter may be logged
@@ -286,21 +302,49 @@ pub static SAFEKEEPER_AUTH_TOKEN: OnceCell<Arc<String>> = OnceCell::new();
 
 // use dedicated enum for builder to better indicate the intention
 // and avoid possible confusion with nested options
+#[derive(Clone, Default)]
 pub enum BuilderValue<T> {
     Set(T),
+    #[default]
     NotSet,
 }
 
-impl<T> BuilderValue<T> {
-    pub fn ok_or<E>(self, err: E) -> Result<T, E> {
+impl<T: Clone> BuilderValue<T> {
+    pub fn ok_or(&self, field_name: &'static str, default: BuilderValue<T>) -> anyhow::Result<T> {
         match self {
-            Self::Set(v) => Ok(v),
-            Self::NotSet => Err(err),
+            Self::Set(v) => Ok(v.clone()),
+            Self::NotSet => match default {
+                BuilderValue::Set(v) => Ok(v.clone()),
+                BuilderValue::NotSet => {
+                    anyhow::bail!("missing config value {field_name:?}")
+                }
+            },
         }
     }
 }
 
+// Certain metadata (e.g. externally-addressable name, AZ) is delivered
+// as a separate structure.  This information is not neeed by the pageserver
+// itself, it is only used for registering the pageserver with the control
+// plane and/or storage controller.
+//
+#[derive(serde::Deserialize)]
+pub(crate) struct NodeMetadata {
+    #[serde(rename = "host")]
+    pub(crate) postgres_host: String,
+    #[serde(rename = "port")]
+    pub(crate) postgres_port: u16,
+    pub(crate) http_host: String,
+    pub(crate) http_port: u16,
+
+    // Deployment tools may write fields to the metadata file beyond what we
+    // use in this type: this type intentionally only names fields that require.
+    #[serde(flatten)]
+    pub(crate) other: HashMap<String, serde_json::Value>,
+}
+
 // needed to simplify config construction
+#[derive(Default)]
 struct PageServerConfigBuilder {
     listen_pg_addr: BuilderValue<String>,
 
@@ -341,6 +385,7 @@ struct PageServerConfigBuilder {
     cached_metric_collection_interval: BuilderValue<Duration>,
     metric_collection_endpoint: BuilderValue<Option<Url>>,
     synthetic_size_calculation_interval: BuilderValue<Duration>,
+    metric_collection_bucket: BuilderValue<Option<RemoteStorageConfig>>,
 
     disk_usage_based_eviction: BuilderValue<Option<DiskUsageEvictionTaskConfig>>,
 
@@ -366,10 +411,13 @@ struct PageServerConfigBuilder {
     max_vectored_read_bytes: BuilderValue<MaxVectoredReadBytes>,
 
     validate_vectored_get: BuilderValue<bool>,
+
+    ephemeral_bytes_per_memory_kb: BuilderValue<usize>,
 }
 
-impl Default for PageServerConfigBuilder {
-    fn default() -> Self {
+impl PageServerConfigBuilder {
+    #[inline(always)]
+    fn default_values() -> Self {
         use self::BuilderValue::*;
         use defaults::*;
         Self {
@@ -422,6 +470,8 @@ impl Default for PageServerConfigBuilder {
             .expect("cannot parse default synthetic size calculation interval")),
             metric_collection_endpoint: Set(DEFAULT_METRIC_COLLECTION_ENDPOINT),
 
+            metric_collection_bucket: Set(None),
+
             disk_usage_based_eviction: Set(None),
 
             test_remote_failures: Set(0),
@@ -449,6 +499,7 @@ impl Default for PageServerConfigBuilder {
                 NonZeroUsize::new(DEFAULT_MAX_VECTORED_READ_BYTES).unwrap(),
             )),
             validate_vectored_get: Set(DEFAULT_VALIDATE_VECTORED_GET),
+            ephemeral_bytes_per_memory_kb: Set(DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB),
         }
     }
 }
@@ -553,6 +604,13 @@ impl PageServerConfigBuilder {
         self.metric_collection_endpoint = BuilderValue::Set(metric_collection_endpoint)
     }
 
+    pub fn metric_collection_bucket(
+        &mut self,
+        metric_collection_bucket: Option<RemoteStorageConfig>,
+    ) {
+        self.metric_collection_bucket = BuilderValue::Set(metric_collection_bucket)
+    }
+
     pub fn synthetic_size_calculation_interval(
         &mut self,
         synthetic_size_calculation_interval: Duration,
@@ -621,126 +679,103 @@ impl PageServerConfigBuilder {
         self.validate_vectored_get = BuilderValue::Set(value);
     }
 
+    pub fn get_ephemeral_bytes_per_memory_kb(&mut self, value: usize) {
+        self.ephemeral_bytes_per_memory_kb = BuilderValue::Set(value);
+    }
+
     pub fn build(self) -> anyhow::Result<PageServerConf> {
-        let concurrent_tenant_warmup = self
-            .concurrent_tenant_warmup
-            .ok_or(anyhow!("missing concurrent_tenant_warmup"))?;
-        let concurrent_tenant_size_logical_size_queries = self
-            .concurrent_tenant_size_logical_size_queries
-            .ok_or(anyhow!(
-                "missing concurrent_tenant_size_logical_size_queries"
-            ))?;
-        Ok(PageServerConf {
-            listen_pg_addr: self
-                .listen_pg_addr
-                .ok_or(anyhow!("missing listen_pg_addr"))?,
-            listen_http_addr: self
-                .listen_http_addr
-                .ok_or(anyhow!("missing listen_http_addr"))?,
-            availability_zone: self
-                .availability_zone
-                .ok_or(anyhow!("missing availability_zone"))?,
-            wait_lsn_timeout: self
-                .wait_lsn_timeout
-                .ok_or(anyhow!("missing wait_lsn_timeout"))?,
-            wal_redo_timeout: self
-                .wal_redo_timeout
-                .ok_or(anyhow!("missing wal_redo_timeout"))?,
-            superuser: self.superuser.ok_or(anyhow!("missing superuser"))?,
-            page_cache_size: self
-                .page_cache_size
-                .ok_or(anyhow!("missing page_cache_size"))?,
-            max_file_descriptors: self
-                .max_file_descriptors
-                .ok_or(anyhow!("missing max_file_descriptors"))?,
-            workdir: self.workdir.ok_or(anyhow!("missing workdir"))?,
-            pg_distrib_dir: self
-                .pg_distrib_dir
-                .ok_or(anyhow!("missing pg_distrib_dir"))?,
-            http_auth_type: self
-                .http_auth_type
-                .ok_or(anyhow!("missing http_auth_type"))?,
-            pg_auth_type: self.pg_auth_type.ok_or(anyhow!("missing pg_auth_type"))?,
-            auth_validation_public_key_path: self
-                .auth_validation_public_key_path
-                .ok_or(anyhow!("missing auth_validation_public_key_path"))?,
-            remote_storage_config: self
-                .remote_storage_config
-                .ok_or(anyhow!("missing remote_storage_config"))?,
-            id: self.id.ok_or(anyhow!("missing id"))?,
-            // TenantConf is handled separately
-            default_tenant_conf: TenantConf::default(),
-            broker_endpoint: self
-                .broker_endpoint
-                .ok_or(anyhow!("No broker endpoints provided"))?,
-            broker_keepalive_interval: self
-                .broker_keepalive_interval
-                .ok_or(anyhow!("No broker keepalive interval provided"))?,
-            log_format: self.log_format.ok_or(anyhow!("missing log_format"))?,
-            concurrent_tenant_warmup: ConfigurableSemaphore::new(concurrent_tenant_warmup),
-            concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore::new(
-                concurrent_tenant_size_logical_size_queries,
-            ),
-            eviction_task_immitated_concurrent_logical_size_queries: ConfigurableSemaphore::new(
-                concurrent_tenant_size_logical_size_queries,
-            ),
-            metric_collection_interval: self
-                .metric_collection_interval
-                .ok_or(anyhow!("missing metric_collection_interval"))?,
-            cached_metric_collection_interval: self
-                .cached_metric_collection_interval
-                .ok_or(anyhow!("missing cached_metric_collection_interval"))?,
-            metric_collection_endpoint: self
-                .metric_collection_endpoint
-                .ok_or(anyhow!("missing metric_collection_endpoint"))?,
-            synthetic_size_calculation_interval: self
-                .synthetic_size_calculation_interval
-                .ok_or(anyhow!("missing synthetic_size_calculation_interval"))?,
-            disk_usage_based_eviction: self
-                .disk_usage_based_eviction
-                .ok_or(anyhow!("missing disk_usage_based_eviction"))?,
-            test_remote_failures: self
-                .test_remote_failures
-                .ok_or(anyhow!("missing test_remote_failuers"))?,
-            ondemand_download_behavior_treat_error_as_warn: self
-                .ondemand_download_behavior_treat_error_as_warn
-                .ok_or(anyhow!(
-                    "missing ondemand_download_behavior_treat_error_as_warn"
-                ))?,
-            background_task_maximum_delay: self
-                .background_task_maximum_delay
-                .ok_or(anyhow!("missing background_task_maximum_delay"))?,
-            control_plane_api: self
-                .control_plane_api
-                .ok_or(anyhow!("missing control_plane_api"))?,
-            control_plane_api_token: self
-                .control_plane_api_token
-                .ok_or(anyhow!("missing control_plane_api_token"))?,
-            control_plane_emergency_mode: self
-                .control_plane_emergency_mode
-                .ok_or(anyhow!("missing control_plane_emergency_mode"))?,
-            heatmap_upload_concurrency: self
-                .heatmap_upload_concurrency
-                .ok_or(anyhow!("missing heatmap_upload_concurrency"))?,
-            secondary_download_concurrency: self
-                .secondary_download_concurrency
-                .ok_or(anyhow!("missing secondary_download_concurrency"))?,
-            ingest_batch_size: self
-                .ingest_batch_size
-                .ok_or(anyhow!("missing ingest_batch_size"))?,
-            virtual_file_io_engine: self
-                .virtual_file_io_engine
-                .ok_or(anyhow!("missing virtual_file_io_engine"))?,
-            get_vectored_impl: self
-                .get_vectored_impl
-                .ok_or(anyhow!("missing get_vectored_impl"))?,
-            max_vectored_read_bytes: self
-                .max_vectored_read_bytes
-                .ok_or(anyhow!("missing max_vectored_read_bytes"))?,
-            validate_vectored_get: self
-                .validate_vectored_get
-                .ok_or(anyhow!("missing validate_vectored_get"))?,
-        })
+        let default = Self::default_values();
+
+        macro_rules! conf {
+            (USING DEFAULT { $($field:ident,)* } CUSTOM LOGIC { $($custom_field:ident : $custom_value:expr,)* } ) => {
+                PageServerConf {
+                    $(
+                        $field: self.$field.ok_or(stringify!($field), default.$field)?,
+                    )*
+                    $(
+                        $custom_field: $custom_value,
+                    )*
+                }
+            };
+        }
+
+        Ok(conf!(
+            USING DEFAULT
+            {
+                listen_pg_addr,
+                listen_http_addr,
+                availability_zone,
+                wait_lsn_timeout,
+                wal_redo_timeout,
+                superuser,
+                page_cache_size,
+                max_file_descriptors,
+                workdir,
+                pg_distrib_dir,
+                http_auth_type,
+                pg_auth_type,
+                auth_validation_public_key_path,
+                remote_storage_config,
+                id,
+                broker_endpoint,
+                broker_keepalive_interval,
+                log_format,
+                metric_collection_interval,
+                cached_metric_collection_interval,
+                metric_collection_endpoint,
+                metric_collection_bucket,
+                synthetic_size_calculation_interval,
+                disk_usage_based_eviction,
+                test_remote_failures,
+                ondemand_download_behavior_treat_error_as_warn,
+                background_task_maximum_delay,
+                control_plane_api,
+                control_plane_api_token,
+                control_plane_emergency_mode,
+                heatmap_upload_concurrency,
+                secondary_download_concurrency,
+                ingest_batch_size,
+                get_vectored_impl,
+                max_vectored_read_bytes,
+                validate_vectored_get,
+                ephemeral_bytes_per_memory_kb,
+            }
+            CUSTOM LOGIC
+            {
+                // TenantConf is handled separately
+                default_tenant_conf: TenantConf::default(),
+                concurrent_tenant_warmup: ConfigurableSemaphore::new({
+                    self
+                        .concurrent_tenant_warmup
+                        .ok_or("concurrent_tenant_warmpup",
+                               default.concurrent_tenant_warmup)?
+                }),
+                concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore::new(
+                    self
+                        .concurrent_tenant_size_logical_size_queries
+                        .ok_or("concurrent_tenant_size_logical_size_queries",
+                               default.concurrent_tenant_size_logical_size_queries.clone())?
+                ),
+                eviction_task_immitated_concurrent_logical_size_queries: ConfigurableSemaphore::new(
+                    // re-use `concurrent_tenant_size_logical_size_queries`
+                    self
+                        .concurrent_tenant_size_logical_size_queries
+                        .ok_or("eviction_task_immitated_concurrent_logical_size_queries",
+                               default.concurrent_tenant_size_logical_size_queries.clone())?,
+                ),
+                virtual_file_io_engine: match self.virtual_file_io_engine {
+                    BuilderValue::Set(v) => v,
+                    BuilderValue::NotSet => match crate::virtual_file::io_engine_feature_test().context("auto-detect virtual_file_io_engine")? {
+                        io_engine::FeatureTestResult::PlatformPreferred(v) => v, // make no noise
+                        io_engine::FeatureTestResult::Worse { engine, remark } => {
+                            // TODO: bubble this up to the caller so we can tracing::warn! it.
+                            eprintln!("auto-detected IO engine is not platform-preferred: engine={engine:?} remark={remark:?}");
+                            engine
+                        }
+                    },
+                },
+            }
+        ))
     }
 }
 
@@ -757,6 +792,10 @@ impl PageServerConf {
         self.workdir.join("deletion")
     }
 
+    pub fn metadata_path(&self) -> Utf8PathBuf {
+        self.workdir.join("metadata.json")
+    }
+
     pub fn deletion_list_path(&self, sequence: u64) -> Utf8PathBuf {
         // Encode a version in the filename, so that if we ever switch away from JSON we can
         // increment this.
@@ -816,18 +855,7 @@ impl PageServerConf {
             .join(timeline_id.to_string())
     }
 
-    pub fn timeline_uninit_mark_file_path(
-        &self,
-        tenant_shard_id: TenantShardId,
-        timeline_id: TimelineId,
-    ) -> Utf8PathBuf {
-        path_with_suffix_extension(
-            self.timeline_path(&tenant_shard_id, &timeline_id),
-            TIMELINE_UNINIT_MARK_SUFFIX,
-        )
-    }
-
-    pub fn timeline_delete_mark_file_path(
+    pub(crate) fn timeline_delete_mark_file_path(
         &self,
         tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
@@ -838,7 +866,10 @@ impl PageServerConf {
         )
     }
 
-    pub fn tenant_deleted_mark_file_path(&self, tenant_shard_id: &TenantShardId) -> Utf8PathBuf {
+    pub(crate) fn tenant_deleted_mark_file_path(
+        &self,
+        tenant_shard_id: &TenantShardId,
+    ) -> Utf8PathBuf {
         self.tenant_path(tenant_shard_id)
             .join(TENANT_DELETED_MARKER_FILE_NAME)
     }
@@ -942,6 +973,9 @@ impl PageServerConf {
                     let endpoint = parse_toml_string(key, item)?.parse().context("failed to parse metric_collection_endpoint")?;
                     builder.metric_collection_endpoint(Some(endpoint));
                 },
+                "metric_collection_bucket" => {
+                    builder.metric_collection_bucket(RemoteStorageConfig::from_toml(item)?)
+                }
                 "synthetic_size_calculation_interval" =>
                     builder.synthetic_size_calculation_interval(parse_toml_duration(key, item)?),
                 "test_remote_failures" => builder.test_remote_failures(parse_toml_u64(key, item)?),
@@ -995,6 +1029,9 @@ impl PageServerConf {
                 "validate_vectored_get" => {
                     builder.get_validate_vectored_get(parse_toml_bool("validate_vectored_get", item)?)
                 }
+                "ephemeral_bytes_per_memory_kb" => {
+                    builder.get_ephemeral_bytes_per_memory_kb(parse_toml_u64("ephemeral_bytes_per_memory_kb", item)? as usize)
+                }
                 _ => bail!("unrecognized pageserver option '{key}'"),
             }
         }
@@ -1057,6 +1094,7 @@ impl PageServerConf {
             metric_collection_interval: Duration::from_secs(60),
             cached_metric_collection_interval: Duration::from_secs(60 * 60),
             metric_collection_endpoint: defaults::DEFAULT_METRIC_COLLECTION_ENDPOINT,
+            metric_collection_bucket: None,
             synthetic_size_calculation_interval: Duration::from_secs(60),
             disk_usage_based_eviction: None,
             test_remote_failures: 0,
@@ -1075,6 +1113,7 @@ impl PageServerConf {
                     .expect("Invalid default constant"),
             ),
             validate_vectored_get: defaults::DEFAULT_VALIDATE_VECTORED_GET,
+            ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB,
         }
     }
 }
@@ -1289,6 +1328,7 @@ background_task_maximum_delay = '334 s'
                     defaults::DEFAULT_CACHED_METRIC_COLLECTION_INTERVAL
                 )?,
                 metric_collection_endpoint: defaults::DEFAULT_METRIC_COLLECTION_ENDPOINT,
+                metric_collection_bucket: None,
                 synthetic_size_calculation_interval: humantime::parse_duration(
                     defaults::DEFAULT_SYNTHETIC_SIZE_CALCULATION_INTERVAL
                 )?,
@@ -1311,6 +1351,7 @@ background_task_maximum_delay = '334 s'
                         .expect("Invalid default constant")
                 ),
                 validate_vectored_get: defaults::DEFAULT_VALIDATE_VECTORED_GET,
+                ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB
             },
             "Correct defaults should be used when no config values are provided"
         );
@@ -1363,6 +1404,7 @@ background_task_maximum_delay = '334 s'
                 metric_collection_interval: Duration::from_secs(222),
                 cached_metric_collection_interval: Duration::from_secs(22200),
                 metric_collection_endpoint: Some(Url::parse("http://localhost:80/metrics")?),
+                metric_collection_bucket: None,
                 synthetic_size_calculation_interval: Duration::from_secs(333),
                 disk_usage_based_eviction: None,
                 test_remote_failures: 0,
@@ -1381,6 +1423,7 @@ background_task_maximum_delay = '334 s'
                         .expect("Invalid default constant")
                 ),
                 validate_vectored_get: defaults::DEFAULT_VALIDATE_VECTORED_GET,
+                ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB
             },
             "Should be able to parse all basic config values correctly"
         );

pageserver/src/consumption_metrics.rs

@@ -3,10 +3,13 @@
 use crate::context::{DownloadBehavior, RequestContext};
 use crate::task_mgr::{self, TaskKind, BACKGROUND_RUNTIME};
 use crate::tenant::tasks::BackgroundLoopKind;
-use crate::tenant::{mgr, LogicalSizeCalculationCause, PageReconstructError, Tenant};
+use crate::tenant::{
+    mgr::TenantManager, LogicalSizeCalculationCause, PageReconstructError, Tenant,
+};
 use camino::Utf8PathBuf;
 use consumption_metrics::EventType;
 use pageserver_api::models::TenantState;
+use remote_storage::{GenericRemoteStorage, RemoteStorageConfig};
 use reqwest::Url;
 use std::collections::HashMap;
 use std::sync::Arc;
@@ -40,7 +43,9 @@ type Cache = HashMap<MetricsKey, (EventType, u64)>;
 /// Main thread that serves metrics collection
 #[allow(clippy::too_many_arguments)]
 pub async fn collect_metrics(
+    tenant_manager: Arc<TenantManager>,
     metric_collection_endpoint: &Url,
+    metric_collection_bucket: &Option<RemoteStorageConfig>,
     metric_collection_interval: Duration,
     _cached_metric_collection_interval: Duration,
     synthetic_size_calculation_interval: Duration,
@@ -65,15 +70,19 @@ pub async fn collect_metrics(
         None,
         "synthetic size calculation",
         false,
-        async move {
-            calculate_synthetic_size_worker(
-                synthetic_size_calculation_interval,
-                &cancel,
-                &worker_ctx,
-            )
-            .instrument(info_span!("synthetic_size_worker"))
-            .await?;
-            Ok(())
+        {
+            let tenant_manager = tenant_manager.clone();
+            async move {
+                calculate_synthetic_size_worker(
+                    tenant_manager,
+                    synthetic_size_calculation_interval,
+                    &cancel,
+                    &worker_ctx,
+                )
+                .instrument(info_span!("synthetic_size_worker"))
+                .await?;
+                Ok(())
+            }
         },
     );
 
@@ -94,13 +103,27 @@ pub async fn collect_metrics(
         .build()
         .expect("Failed to create http client with timeout");
 
+    let bucket_client = if let Some(bucket_config) = metric_collection_bucket {
+        match GenericRemoteStorage::from_config(bucket_config) {
+            Ok(client) => Some(client),
+            Err(e) => {
+                // Non-fatal error: if we were given an invalid config, we will proceed
+                // with sending metrics over the network, but not to S3.
+                tracing::warn!("Invalid configuration for metric_collection_bucket: {e}");
+                None
+            }
+        }
+    } else {
+        None
+    };
+
     let node_id = node_id.to_string();
 
     loop {
         let started_at = Instant::now();
 
         // these are point in time, with variable "now"
-        let metrics = metrics::collect_all_metrics(&cached_metrics, &ctx).await;
+        let metrics = metrics::collect_all_metrics(&tenant_manager, &cached_metrics, &ctx).await;
 
         let metrics = Arc::new(metrics);
 
@@ -118,10 +141,18 @@ pub async fn collect_metrics(
                     tracing::error!("failed to persist metrics to {path:?}: {e:#}");
                 }
             }
+
+            if let Some(bucket_client) = &bucket_client {
+                let res =
+                    upload::upload_metrics_bucket(bucket_client, &cancel, &node_id, &metrics).await;
+                if let Err(e) = res {
+                    tracing::error!("failed to upload to S3: {e:#}");
+                }
+            }
         };
 
         let upload = async {
-            let res = upload::upload_metrics(
+            let res = upload::upload_metrics_http(
                 &client,
                 metric_collection_endpoint,
                 &cancel,
@@ -132,7 +163,7 @@ pub async fn collect_metrics(
             .await;
             if let Err(e) = res {
                 // serialization error which should never happen
-                tracing::error!("failed to upload due to {e:#}");
+                tracing::error!("failed to upload via HTTP due to {e:#}");
             }
         };
 
@@ -247,6 +278,7 @@ async fn reschedule(
 
 /// Caclculate synthetic size for each active tenant
 async fn calculate_synthetic_size_worker(
+    tenant_manager: Arc<TenantManager>,
     synthetic_size_calculation_interval: Duration,
     cancel: &CancellationToken,
     ctx: &RequestContext,
@@ -259,7 +291,7 @@ async fn calculate_synthetic_size_worker(
     loop {
         let started_at = Instant::now();
 
-        let tenants = match mgr::list_tenants().await {
+        let tenants = match tenant_manager.list_tenants() {
             Ok(tenants) => tenants,
             Err(e) => {
                 warn!("cannot get tenant list: {e:#}");
@@ -278,10 +310,14 @@ async fn calculate_synthetic_size_worker(
                 continue;
             }
 
-            let Ok(tenant) = mgr::get_tenant(tenant_shard_id, true) else {
+            let Ok(tenant) = tenant_manager.get_attached_tenant_shard(tenant_shard_id) else {
                 continue;
             };
 
+            if !tenant.is_active() {
+                continue;
+            }
+
             // there is never any reason to exit calculate_synthetic_size_worker following any
             // return value -- we don't need to care about shutdown because no tenant is found when
             // pageserver is shut down.
@@ -319,9 +355,7 @@ async fn calculate_and_log(tenant: &Tenant, cancel: &CancellationToken, ctx: &Re
     };
 
     // this error can be returned if timeline is shutting down, but it does not
-    // mean the synthetic size worker should terminate. we do not need any checks
-    // in this function because `mgr::get_tenant` will error out after shutdown has
-    // progressed to shutting down tenants.
+    // mean the synthetic size worker should terminate.
     let shutting_down = matches!(
         e.downcast_ref::<PageReconstructError>(),
         Some(PageReconstructError::Cancelled | PageReconstructError::AncestorStopping(_))

pageserver/src/consumption_metrics/metrics.rs

@@ -1,3 +1,4 @@
+use crate::tenant::mgr::TenantManager;
 use crate::{context::RequestContext, tenant::timeline::logical_size::CurrentLogicalSize};
 use chrono::{DateTime, Utc};
 use consumption_metrics::EventType;
@@ -181,6 +182,7 @@ impl MetricsKey {
 }
 
 pub(super) async fn collect_all_metrics(
+    tenant_manager: &Arc<TenantManager>,
     cached_metrics: &Cache,
     ctx: &RequestContext,
 ) -> Vec<RawMetric> {
@@ -188,7 +190,7 @@ pub(super) async fn collect_all_metrics(
 
     let started_at = std::time::Instant::now();
 
-    let tenants = match crate::tenant::mgr::list_tenants().await {
+    let tenants = match tenant_manager.list_tenants() {
         Ok(tenants) => tenants,
         Err(err) => {
             tracing::error!("failed to list tenants: {:?}", err);
@@ -200,7 +202,8 @@ pub(super) async fn collect_all_metrics(
         if state != TenantState::Active || !id.is_zero() {
             None
         } else {
-            crate::tenant::mgr::get_tenant(id, true)
+            tenant_manager
+                .get_attached_tenant_shard(id)
                 .ok()
                 .map(|tenant| (id.tenant_id, tenant))
         }

pageserver/src/consumption_metrics/upload.rs

@@ -1,4 +1,9 @@
+use std::time::SystemTime;
+
+use chrono::{DateTime, Utc};
 use consumption_metrics::{Event, EventChunk, IdempotencyKey, CHUNK_SIZE};
+use remote_storage::{GenericRemoteStorage, RemotePath};
+use tokio::io::AsyncWriteExt;
 use tokio_util::sync::CancellationToken;
 use tracing::Instrument;
 
@@ -13,8 +18,9 @@ struct Ids {
     pub(super) timeline_id: Option<TimelineId>,
 }
 
+/// Serialize and write metrics to an HTTP endpoint
 #[tracing::instrument(skip_all, fields(metrics_total = %metrics.len()))]
-pub(super) async fn upload_metrics(
+pub(super) async fn upload_metrics_http(
     client: &reqwest::Client,
     metric_collection_endpoint: &reqwest::Url,
     cancel: &CancellationToken,
@@ -74,6 +80,60 @@ pub(super) async fn upload_metrics(
     Ok(())
 }
 
+/// Serialize and write metrics to a remote storage object
+#[tracing::instrument(skip_all, fields(metrics_total = %metrics.len()))]
+pub(super) async fn upload_metrics_bucket(
+    client: &GenericRemoteStorage,
+    cancel: &CancellationToken,
+    node_id: &str,
+    metrics: &[RawMetric],
+) -> anyhow::Result<()> {
+    if metrics.is_empty() {
+        // Skip uploads if we have no metrics, so that readers don't have to handle the edge case
+        // of an empty object.
+        return Ok(());
+    }
+
+    // Compose object path
+    let datetime: DateTime<Utc> = SystemTime::now().into();
+    let ts_prefix = datetime.format("year=%Y/month=%m/day=%d/%H:%M:%SZ");
+    let path = RemotePath::from_string(&format!("{ts_prefix}_{node_id}.ndjson.gz"))?;
+
+    // Set up a gzip writer into a buffer
+    let mut compressed_bytes: Vec<u8> = Vec::new();
+    let compressed_writer = std::io::Cursor::new(&mut compressed_bytes);
+    let mut gzip_writer = async_compression::tokio::write::GzipEncoder::new(compressed_writer);
+
+    // Serialize and write into compressed buffer
+    let started_at = std::time::Instant::now();
+    for res in serialize_in_chunks(CHUNK_SIZE, metrics, node_id) {
+        let (_chunk, body) = res?;
+        gzip_writer.write_all(&body).await?;
+    }
+    gzip_writer.flush().await?;
+    gzip_writer.shutdown().await?;
+    let compressed_length = compressed_bytes.len();
+
+    // Write to remote storage
+    client
+        .upload_storage_object(
+            futures::stream::once(futures::future::ready(Ok(compressed_bytes.into()))),
+            compressed_length,
+            &path,
+            cancel,
+        )
+        .await?;
+    let elapsed = started_at.elapsed();
+
+    tracing::info!(
+        compressed_length,
+        elapsed_ms = elapsed.as_millis(),
+        "write metrics bucket at {path}",
+    );
+
+    Ok(())
+}
+
 // The return type is quite ugly, but we gain testability in isolation
 fn serialize_in_chunks<'a, F>(
     chunk_size: usize,

pageserver/src/control_plane_client.rs

@@ -2,17 +2,22 @@ use std::collections::HashMap;
 
 use futures::Future;
 use pageserver_api::{
+    controller_api::NodeRegisterRequest,
     shard::TenantShardId,
     upcall_api::{
-        ReAttachRequest, ReAttachResponse, ValidateRequest, ValidateRequestTenant, ValidateResponse,
+        ReAttachRequest, ReAttachResponse, ReAttachResponseTenant, ValidateRequest,
+        ValidateRequestTenant, ValidateResponse,
     },
 };
 use serde::{de::DeserializeOwned, Serialize};
 use tokio_util::sync::CancellationToken;
 use url::Url;
-use utils::{backoff, generation::Generation, id::NodeId};
+use utils::{backoff, failpoint_support, generation::Generation, id::NodeId};
 
-use crate::config::PageServerConf;
+use crate::{
+    config::{NodeMetadata, PageServerConf},
+    virtual_file::on_fatal_io_error,
+};
 
 /// The Pageserver's client for using the control plane API: this is a small subset
 /// of the overall control plane API, for dealing with generations (see docs/rfcs/025-generation-numbers.md)
@@ -32,7 +37,10 @@ pub enum RetryForeverError {
 pub trait ControlPlaneGenerationsApi {
     fn re_attach(
         &self,
-    ) -> impl Future<Output = Result<HashMap<TenantShardId, Generation>, RetryForeverError>> + Send;
+        conf: &PageServerConf,
+    ) -> impl Future<
+        Output = Result<HashMap<TenantShardId, ReAttachResponseTenant>, RetryForeverError>,
+    > + Send;
     fn validate(
         &self,
         tenants: Vec<(TenantShardId, Generation)>,
@@ -110,13 +118,59 @@ impl ControlPlaneClient {
 
 impl ControlPlaneGenerationsApi for ControlPlaneClient {
     /// Block until we get a successful response, or error out if we are shut down
-    async fn re_attach(&self) -> Result<HashMap<TenantShardId, Generation>, RetryForeverError> {
+    async fn re_attach(
+        &self,
+        conf: &PageServerConf,
+    ) -> Result<HashMap<TenantShardId, ReAttachResponseTenant>, RetryForeverError> {
         let re_attach_path = self
             .base_url
             .join("re-attach")
             .expect("Failed to build re-attach path");
+
+        // Include registration content in the re-attach request if a metadata file is readable
+        let metadata_path = conf.metadata_path();
+        let register = match tokio::fs::read_to_string(&metadata_path).await {
+            Ok(metadata_str) => match serde_json::from_str::<NodeMetadata>(&metadata_str) {
+                Ok(m) => {
+                    // Since we run one time at startup, be generous in our logging and
+                    // dump all metadata.
+                    tracing::info!(
+                        "Loaded node metadata: postgres {}:{}, http {}:{}, other fields: {:?}",
+                        m.postgres_host,
+                        m.postgres_port,
+                        m.http_host,
+                        m.http_port,
+                        m.other
+                    );
+
+                    Some(NodeRegisterRequest {
+                        node_id: conf.id,
+                        listen_pg_addr: m.postgres_host,
+                        listen_pg_port: m.postgres_port,
+                        listen_http_addr: m.http_host,
+                        listen_http_port: m.http_port,
+                    })
+                }
+                Err(e) => {
+                    tracing::error!("Unreadable metadata in {metadata_path}: {e}");
+                    None
+                }
+            },
+            Err(e) => {
+                if e.kind() == std::io::ErrorKind::NotFound {
+                    // This is legal: we may have been deployed with some external script
+                    // doing registration for us.
+                    tracing::info!("Metadata file not found at {metadata_path}");
+                } else {
+                    on_fatal_io_error(&e, &format!("Loading metadata at {metadata_path}"))
+                }
+                None
+            }
+        };
+
         let request = ReAttachRequest {
             node_id: self.node_id,
+            register,
         };
 
         fail::fail_point!("control-plane-client-re-attach");
@@ -130,7 +184,7 @@ impl ControlPlaneGenerationsApi for ControlPlaneClient {
         Ok(response
             .tenants
             .into_iter()
-            .map(|t| (t.id, Generation::new(t.gen)))
+            .map(|rart| (rart.id, rart))
             .collect::<HashMap<_, _>>())
     }
 
@@ -156,7 +210,10 @@ impl ControlPlaneGenerationsApi for ControlPlaneClient {
                 .collect(),
         };
 
-        fail::fail_point!("control-plane-client-validate");
+        failpoint_support::sleep_millis_async!("control-plane-client-validate-sleep", &self.cancel);
+        if self.cancel.is_cancelled() {
+            return Err(RetryForeverError::ShuttingDown);
+        }
 
         let response: ValidateResponse = self.retry_http_forever(&re_attach_path, request).await?;
 

pageserver/src/deletion_queue.rs

@@ -724,8 +724,8 @@ impl DeletionQueue {
 mod test {
     use camino::Utf8Path;
     use hex_literal::hex;
-    use pageserver_api::shard::ShardIndex;
-    use std::io::ErrorKind;
+    use pageserver_api::{shard::ShardIndex, upcall_api::ReAttachResponseTenant};
+    use std::{io::ErrorKind, time::Duration};
     use tracing::info;
 
     use remote_storage::{RemoteStorageConfig, RemoteStorageKind};
@@ -831,9 +831,13 @@ mod test {
     }
 
     impl ControlPlaneGenerationsApi for MockControlPlane {
-        async fn re_attach(&self) -> Result<HashMap<TenantShardId, Generation>, RetryForeverError> {
+        async fn re_attach(
+            &self,
+            _conf: &PageServerConf,
+        ) -> Result<HashMap<TenantShardId, ReAttachResponseTenant>, RetryForeverError> {
             unimplemented!()
         }
+
         async fn validate(
             &self,
             tenants: Vec<(TenantShardId, Generation)>,

pageserver/src/disk_usage_eviction_task.rs

@@ -61,7 +61,6 @@ use crate::{
     metrics::disk_usage_based_eviction::METRICS,
     task_mgr::{self, TaskKind, BACKGROUND_RUNTIME},
     tenant::{
-        self,
         mgr::TenantManager,
         remote_timeline_client::LayerFileMetadata,
         secondary::SecondaryTenant,
@@ -814,8 +813,8 @@ async fn collect_eviction_candidates(
     const LOG_DURATION_THRESHOLD: std::time::Duration = std::time::Duration::from_secs(10);
 
     // get a snapshot of the list of tenants
-    let tenants = tenant::mgr::list_tenants()
-        .await
+    let tenants = tenant_manager
+        .list_tenants()
         .context("get list of tenants")?;
 
     // TODO: avoid listing every layer in every tenant: this loop can block the executor,
@@ -827,8 +826,12 @@ async fn collect_eviction_candidates(
         if cancel.is_cancelled() {
             return Ok(EvictionCandidates::Cancelled);
         }
-        let tenant = match tenant::mgr::get_tenant(tenant_id, true) {
-            Ok(tenant) => tenant,
+        let tenant = match tenant_manager.get_attached_tenant_shard(tenant_id) {
+            Ok(tenant) if tenant.is_active() => tenant,
+            Ok(_) => {
+                debug!(tenant_id=%tenant_id.tenant_id, shard_id=%tenant_id.shard_slug(), "Tenant shard is not active");
+                continue;
+            }
             Err(e) => {
                 // this can happen if tenant has lifecycle transition after we fetched it
                 debug!("failed to get tenant: {e:#}");

pageserver/src/http/openapi_spec.yml

@@ -567,9 +567,9 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/ServiceUnavailableError"
-  /v1/tenant/{tenant_id}/location_config:
+  /v1/tenant/{tenant_shard_id}/location_config:
     parameters:
-      - name: tenant_id
+      - name: tenant_shard_id
         in: path
         required: true
         schema:
@@ -932,6 +932,75 @@ paths:
               schema:
                 $ref: "#/components/schemas/ServiceUnavailableError"
 
+  /v1/tenant/{tenant_shard_id}/heatmap_upload:
+    parameters:
+      - name: tenant_shard_id
+        in: path
+        required: true
+        schema:
+          type: string
+    post:
+      description: |
+        If the location is in an attached mode, upload the current state to the remote heatmap
+      responses:
+        "200":
+          description: Success
+        "500":
+          description: Generic operation error
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/Error"
+        "503":
+          description: Temporarily unavailable, please retry.
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/ServiceUnavailableError"
+
+  /v1/tenant/{tenant_shard_id}/secondary/download:
+    parameters:
+      - name: tenant_shard_id
+        in: path
+        required: true
+        schema:
+          type: string
+      - name: wait_ms
+        description: If set, we will wait this long for download to complete, and if it isn't complete then return 202
+        in: query
+        required: false
+        schema:
+          type: integer
+    post:
+      description: |
+        If the location is in secondary mode, download latest heatmap and layers
+      responses:
+        "200":
+          description: Success
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/SecondaryProgress"
+        "202":
+          description: Download has started but not yet finished
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/SecondaryProgress"
+        "500":
+          description: Generic operation error
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/Error"
+        "503":
+          description: Temporarily unavailable, please retry.
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/ServiceUnavailableError"
+
+
 
   /v1/tenant/{tenant_id}/timeline/:
     parameters:
@@ -969,7 +1038,7 @@ paths:
                   format: hex
       responses:
         "201":
-          description: TimelineInfo
+          description: Timeline was created, or already existed with matching parameters
           content:
             application/json:
               schema:
@@ -999,11 +1068,17 @@ paths:
               schema:
                 $ref: "#/components/schemas/Error"
         "409":
-          description: Timeline already exists, creation skipped
+          description: Timeline already exists, with different parameters.  Creation cannot proceed.
           content:
             application/json:
               schema:
                 $ref: "#/components/schemas/ConflictError"
+        "429":
+          description: A creation request was sent for the same Timeline Id while a creation was already in progress.  Back off and retry.
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/Error"
         "500":
           description: Generic operation error
           content:
@@ -1314,10 +1389,11 @@ components:
     TenantLocationConfigRequest:
       type: object
       required:
-        - tenant_id
+        - mode
       properties:
         tenant_id:
           type: string
+          description: Not used, scheduled for removal.
         mode:
           type: string
           enum: ["AttachedSingle", "AttachedMulti", "AttachedStale", "Secondary", "Detached"]
@@ -1391,7 +1467,7 @@ components:
         trace_read_requests:
           type: boolean
         heatmap_period:
-          type: integer
+          type: string
     TenantConfigResponse:
       type: object
       properties:
@@ -1553,7 +1629,7 @@ components:
           type: integer
           format: int64
           minimum: 0
-          description: The amount of disk space currently utilized by layer files.
+          description: The amount of disk space currently used.
         free_space_bytes:
           type: integer
           format: int64
@@ -1569,6 +1645,37 @@ components:
             Lower is better score for how good this pageserver would be for the next tenant.
             The default or maximum value can be returned in situations when a proper score cannot (yet) be calculated.
 
+    SecondaryProgress:
+      type: object
+      required:
+        - heatmap_mtime
+        - layers_downloaded
+        - layers_total
+        - bytes_downloaded
+        - bytes_total
+      properties:
+        heatmap_mtime:
+          type: string
+          format: date-time
+          description: Modification time of the most recently downloaded layer heatmap (RFC 3339 format)
+        layers_downloaded:
+          type: integer
+          format: int64
+          description: How many layers from the latest layer heatmap are present on disk
+        bytes_downloaded:
+          type: integer
+          format: int64
+          description: How many bytes of layer content from the latest layer heatmap are present on disk
+        layers_total:
+          type: integer
+          format: int64
+          description: How many layers were in the latest layer heatmap
+        bytes_total:
+          type: integer
+          format: int64
+          description: How many bytes of layer content were in the latest layer heatmap
+
+
     Error:
       type: object
       required:

pageserver/src/http/routes.rs

@@ -36,6 +36,7 @@ use tokio_util::sync::CancellationToken;
 use tracing::*;
 use utils::auth::JwtAuth;
 use utils::failpoint_support::failpoints_handler;
+use utils::http::endpoint::prometheus_metrics_handler;
 use utils::http::endpoint::request_span;
 use utils::http::json::json_request_or_empty_body;
 use utils::http::request::{get_request_param, must_get_query_param, parse_query_param};
@@ -48,8 +49,8 @@ use crate::task_mgr::TaskKind;
 use crate::tenant::config::{LocationConf, TenantConfOpt};
 use crate::tenant::mgr::GetActiveTenantError;
 use crate::tenant::mgr::{
-    GetTenantError, SetNewTenantConfigError, TenantManager, TenantMapError, TenantMapInsertError,
-    TenantSlotError, TenantSlotUpsertError, TenantStateError,
+    GetTenantError, TenantManager, TenantMapError, TenantMapInsertError, TenantSlotError,
+    TenantSlotUpsertError, TenantStateError,
 };
 use crate::tenant::mgr::{TenantSlot, UpsertLocationError};
 use crate::tenant::remote_timeline_client;
@@ -248,16 +249,11 @@ impl From<GetTenantError> for ApiError {
     fn from(tse: GetTenantError) -> ApiError {
         match tse {
             GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
-            GetTenantError::Broken(reason) => {
-                ApiError::InternalServerError(anyhow!("tenant is broken: {}", reason))
-            }
             GetTenantError::NotActive(_) => {
                 // Why is this not `ApiError::NotFound`?
                 // Because we must be careful to never return 404 for a tenant if it does
                 // in fact exist locally. If we did, the caller could draw the conclusion
                 // that it can attach the tenant to another PS and we'd be in split-brain.
-                //
-                // (We can produce this variant only in `mgr::get_tenant(..., active=true)` calls).
                 ApiError::ResourceUnavailable("Tenant not yet active".into())
             }
             GetTenantError::MapState(e) => ApiError::ResourceUnavailable(format!("{e}").into()),
@@ -268,6 +264,9 @@ impl From<GetTenantError> for ApiError {
 impl From<GetActiveTenantError> for ApiError {
     fn from(e: GetActiveTenantError) -> ApiError {
         match e {
+            GetActiveTenantError::Broken(reason) => {
+                ApiError::InternalServerError(anyhow!("tenant is broken: {}", reason))
+            }
             GetActiveTenantError::WillNotBecomeActive(_) => ApiError::Conflict(format!("{}", e)),
             GetActiveTenantError::Cancelled => ApiError::ShuttingDown,
             GetActiveTenantError::NotFound(gte) => gte.into(),
@@ -278,19 +277,6 @@ impl From<GetActiveTenantError> for ApiError {
     }
 }
 
-impl From<SetNewTenantConfigError> for ApiError {
-    fn from(e: SetNewTenantConfigError) -> ApiError {
-        match e {
-            SetNewTenantConfigError::GetTenant(tid) => {
-                ApiError::NotFound(anyhow!("tenant {}", tid).into())
-            }
-            e @ (SetNewTenantConfigError::Persist(_) | SetNewTenantConfigError::Other(_)) => {
-                ApiError::InternalServerError(anyhow::Error::new(e))
-            }
-        }
-    }
-}
-
 impl From<crate::tenant::DeleteTimelineError> for ApiError {
     fn from(value: crate::tenant::DeleteTimelineError) -> Self {
         use crate::tenant::DeleteTimelineError::*;
@@ -494,7 +480,7 @@ async fn timeline_create_handler(
     async {
         let tenant = state
             .tenant_manager
-            .get_attached_tenant_shard(tenant_shard_id, false)?;
+            .get_attached_tenant_shard(tenant_shard_id)?;
 
         tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
 
@@ -534,10 +520,13 @@ async fn timeline_create_handler(
                     HttpErrorBody::from_msg("Tenant shutting down".to_string()),
                 )
             }
-            Err(
-                tenant::CreateTimelineError::Conflict
-                | tenant::CreateTimelineError::AlreadyCreating,
-            ) => json_response(StatusCode::CONFLICT, ()),
+            Err(e @ tenant::CreateTimelineError::Conflict) => {
+                json_response(StatusCode::CONFLICT, HttpErrorBody::from_msg(e.to_string()))
+            }
+            Err(e @ tenant::CreateTimelineError::AlreadyCreating) => json_response(
+                StatusCode::TOO_MANY_REQUESTS,
+                HttpErrorBody::from_msg(e.to_string()),
+            ),
             Err(tenant::CreateTimelineError::AncestorLsn(err)) => json_response(
                 StatusCode::NOT_ACCEPTABLE,
                 HttpErrorBody::from_msg(format!("{err:#}")),
@@ -580,7 +569,7 @@ async fn timeline_list_handler(
     let response_data = async {
         let tenant = state
             .tenant_manager
-            .get_attached_tenant_shard(tenant_shard_id, false)?;
+            .get_attached_tenant_shard(tenant_shard_id)?;
 
         tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
 
@@ -618,6 +607,7 @@ async fn timeline_preserve_initdb_handler(
     let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
     let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
     check_permission(&request, Some(tenant_shard_id.tenant_id))?;
+    let state = get_state(&request);
 
     // Part of the process for disaster recovery from safekeeper-stored WAL:
     // If we don't recover into a new timeline but want to keep the timeline ID,
@@ -625,7 +615,9 @@ async fn timeline_preserve_initdb_handler(
     // location where timeline recreation cand find it.
 
     async {
-        let tenant = mgr::get_tenant(tenant_shard_id, false)?;
+        let tenant = state
+            .tenant_manager
+            .get_attached_tenant_shard(tenant_shard_id)?;
 
         let timeline = tenant
             .get_timeline(timeline_id, false)
@@ -667,7 +659,7 @@ async fn timeline_detail_handler(
     let timeline_info = async {
         let tenant = state
             .tenant_manager
-            .get_attached_tenant_shard(tenant_shard_id, false)?;
+            .get_attached_tenant_shard(tenant_shard_id)?;
 
         tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
 
@@ -854,7 +846,7 @@ async fn timeline_delete_handler(
 
     let tenant = state
         .tenant_manager
-        .get_attached_tenant_shard(tenant_shard_id, false)
+        .get_attached_tenant_shard(tenant_shard_id)
         .map_err(|e| {
             match e {
                 // GetTenantError has a built-in conversion to ApiError, but in this context we don't
@@ -885,14 +877,16 @@ async fn tenant_detach_handler(
 
     let state = get_state(&request);
     let conf = state.conf;
-    mgr::detach_tenant(
-        conf,
-        tenant_shard_id,
-        detach_ignored.unwrap_or(false),
-        &state.deletion_queue_client,
-    )
-    .instrument(info_span!("tenant_detach", %tenant_id, shard_id=%tenant_shard_id.shard_slug()))
-    .await?;
+    state
+        .tenant_manager
+        .detach_tenant(
+            conf,
+            tenant_shard_id,
+            detach_ignored.unwrap_or(false),
+            &state.deletion_queue_client,
+        )
+        .instrument(info_span!("tenant_detach", %tenant_id, shard_id=%tenant_shard_id.shard_slug()))
+        .await?;
 
     json_response(StatusCode::OK, ())
 }
@@ -970,10 +964,11 @@ async fn tenant_list_handler(
     _cancel: CancellationToken,
 ) -> Result<Response<Body>, ApiError> {
     check_permission(&request, None)?;
+    let state = get_state(&request);
 
-    let response_data = mgr::list_tenants()
-        .instrument(info_span!("tenant_list"))
-        .await
+    let response_data = state
+        .tenant_manager
+        .list_tenants()
         .map_err(|_| {
             ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".into())
         })?
@@ -996,9 +991,27 @@ async fn tenant_status(
 ) -> Result<Response<Body>, ApiError> {
     let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
     check_permission(&request, Some(tenant_shard_id.tenant_id))?;
+    let state = get_state(&request);
+
+    // In tests, sometimes we want to query the state of a tenant without auto-activating it if it's currently waiting.
+    let activate = true;
+    #[cfg(feature = "testing")]
+    let activate = parse_query_param(&request, "activate")?.unwrap_or(activate);
 
     let tenant_info = async {
-        let tenant = mgr::get_tenant(tenant_shard_id, false)?;
+        let tenant = state
+            .tenant_manager
+            .get_attached_tenant_shard(tenant_shard_id)?;
+
+        if activate {
+            // This is advisory: we prefer to let the tenant activate on-demand when this function is
+            // called, but it is still valid to return 200 and describe the current state of the tenant
+            // if it doesn't make it into an active state.
+            tenant
+                .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
+                .await
+                .ok();
+        }
 
         // Calculate total physical size of all timelines
         let mut current_physical_size = 0;
@@ -1071,9 +1084,7 @@ async fn tenant_size_handler(
     let inputs_only: Option<bool> = parse_query_param(&request, "inputs_only")?;
     let retention_period: Option<u64> = parse_query_param(&request, "retention_period")?;
     let headers = request.headers();
-
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
-    let tenant = mgr::get_tenant(tenant_shard_id, true)?;
+    let state = get_state(&request);
 
     if !tenant_shard_id.is_zero() {
         return Err(ApiError::BadRequest(anyhow!(
@@ -1081,6 +1092,12 @@ async fn tenant_size_handler(
         )));
     }
 
+    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let tenant = state
+        .tenant_manager
+        .get_attached_tenant_shard(tenant_shard_id)?;
+    tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
+
     // this can be long operation
     let inputs = tenant
         .gather_size_inputs(
@@ -1149,9 +1166,19 @@ async fn tenant_shard_split_handler(
     let state = get_state(&request);
     let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
 
+    let tenant = state
+        .tenant_manager
+        .get_attached_tenant_shard(tenant_shard_id)?;
+    tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
+
     let new_shards = state
         .tenant_manager
-        .shard_split(tenant_shard_id, ShardCount::new(req.new_shard_count), &ctx)
+        .shard_split(
+            tenant,
+            ShardCount::new(req.new_shard_count),
+            req.new_stripe_size,
+            &ctx,
+        )
         .await
         .map_err(ApiError::InternalServerError)?;
 
@@ -1365,8 +1392,11 @@ async fn get_tenant_config_handler(
 ) -> Result<Response<Body>, ApiError> {
     let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
     check_permission(&request, Some(tenant_shard_id.tenant_id))?;
+    let state = get_state(&request);
 
-    let tenant = mgr::get_tenant(tenant_shard_id, false)?;
+    let tenant = state
+        .tenant_manager
+        .get_attached_tenant_shard(tenant_shard_id)?;
 
     let response = HashMap::from([
         (
@@ -1394,13 +1424,31 @@ async fn update_tenant_config_handler(
     let tenant_id = request_data.tenant_id;
     check_permission(&request, Some(tenant_id))?;
 
-    let tenant_conf =
+    let new_tenant_conf =
         TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
 
     let state = get_state(&request);
-    mgr::set_new_tenant_config(state.conf, tenant_conf, tenant_id)
-        .instrument(info_span!("tenant_config", %tenant_id))
-        .await?;
+
+    let tenant_shard_id = TenantShardId::unsharded(tenant_id);
+
+    let tenant = state
+        .tenant_manager
+        .get_attached_tenant_shard(tenant_shard_id)?;
+    tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
+
+    // This is a legacy API that only operates on attached tenants: the preferred
+    // API to use is the location_config/ endpoint, which lets the caller provide
+    // the full LocationConf.
+    let location_conf = LocationConf::attached_single(
+        new_tenant_conf.clone(),
+        tenant.get_generation(),
+        &ShardParameters::default(),
+    );
+
+    crate::tenant::Tenant::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
+        .await
+        .map_err(ApiError::InternalServerError)?;
+    tenant.set_new_tenant_config(new_tenant_conf);
 
     json_response(StatusCode::OK, ())
 }
@@ -1423,13 +1471,14 @@ async fn put_tenant_location_config_handler(
     // The `Detached` state is special, it doesn't upsert a tenant, it removes
     // its local disk content and drops it from memory.
     if let LocationConfigMode::Detached = request_data.config.mode {
-        if let Err(e) =
-            mgr::detach_tenant(conf, tenant_shard_id, true, &state.deletion_queue_client)
-                .instrument(info_span!("tenant_detach",
-                    tenant_id = %tenant_shard_id.tenant_id,
-                    shard_id = %tenant_shard_id.shard_slug()
-                ))
-                .await
+        if let Err(e) = state
+            .tenant_manager
+            .detach_tenant(conf, tenant_shard_id, true, &state.deletion_queue_client)
+            .instrument(info_span!("tenant_detach",
+                tenant_id = %tenant_shard_id.tenant_id,
+                shard_id = %tenant_shard_id.shard_slug()
+            ))
+            .await
         {
             match e {
                 TenantStateError::SlotError(TenantSlotError::NotFound(_)) => {
@@ -1623,10 +1672,12 @@ async fn handle_tenant_break(
 ) -> Result<Response<Body>, ApiError> {
     let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
 
-    let tenant = crate::tenant::mgr::get_tenant(tenant_shard_id, true)
-        .map_err(|_| ApiError::Conflict(String::from("no active tenant found")))?;
-
-    tenant.set_broken("broken from test".to_owned()).await;
+    let state = get_state(&r);
+    state
+        .tenant_manager
+        .get_attached_tenant_shard(tenant_shard_id)?
+        .set_broken("broken from test".to_owned())
+        .await;
 
     json_response(StatusCode::OK, ())
 }
@@ -1643,8 +1694,7 @@ async fn timeline_gc_handler(
     let gc_req: TimelineGcRequest = json_request(&mut request).await?;
 
     let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
-    let wait_task_done =
-        mgr::immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx).await?;
+    let wait_task_done = mgr::immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx)?;
     let gc_result = wait_task_done
         .await
         .context("wait for gc task")
@@ -1871,7 +1921,7 @@ async fn active_timeline_of_active_tenant(
     tenant_shard_id: TenantShardId,
     timeline_id: TimelineId,
 ) -> Result<Arc<Timeline>, ApiError> {
-    let tenant = tenant_manager.get_attached_tenant_shard(tenant_shard_id, false)?;
+    let tenant = tenant_manager.get_attached_tenant_shard(tenant_shard_id)?;
 
     tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
 
@@ -1982,13 +2032,42 @@ async fn secondary_download_handler(
 ) -> Result<Response<Body>, ApiError> {
     let state = get_state(&request);
     let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
-    state
-        .secondary_controller
-        .download_tenant(tenant_shard_id)
-        .await
-        .map_err(ApiError::InternalServerError)?;
+    let wait = parse_query_param(&request, "wait_ms")?.map(Duration::from_millis);
 
-    json_response(StatusCode::OK, ())
+    // We don't need this to issue the download request, but:
+    // - it enables us to cleanly return 404 if we get a request for an absent shard
+    // - we will use this to provide status feedback in the response
+    let Some(secondary_tenant) = state
+        .tenant_manager
+        .get_secondary_tenant_shard(tenant_shard_id)
+    else {
+        return Err(ApiError::NotFound(
+            anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
+        ));
+    };
+
+    let timeout = wait.unwrap_or(Duration::MAX);
+
+    let status = match tokio::time::timeout(
+        timeout,
+        state.secondary_controller.download_tenant(tenant_shard_id),
+    )
+    .await
+    {
+        // Download job ran to completion.
+        Ok(Ok(())) => StatusCode::OK,
+        // Edge case: downloads aren't usually fallible: things like a missing heatmap are considered
+        // okay.  We could get an error here in the unlikely edge case that the tenant
+        // was detached between our check above and executing the download job.
+        Ok(Err(e)) => return Err(ApiError::InternalServerError(e)),
+        // A timeout is not an error: we have started the download, we're just not done
+        // yet.  The caller will get a response body indicating status.
+        Err(_) => StatusCode::ACCEPTED,
+    };
+
+    let progress = secondary_tenant.progress.lock().unwrap().clone();
+
+    json_response(status, progress)
 }
 
 async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
@@ -2048,6 +2127,10 @@ async fn get_utilization(
     r: Request<Body>,
     _cancel: CancellationToken,
 ) -> Result<Response<Body>, ApiError> {
+    fail::fail_point!("get-utilization-http-handler", |_| {
+        Err(ApiError::ResourceUnavailable("failpoint".into()))
+    });
+
     // this probably could be completely public, but lets make that change later.
     check_permission(&r, None)?;
 
@@ -2103,6 +2186,16 @@ where
     R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
     H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
 {
+    if request.uri() != &"/v1/failpoints".parse::<Uri>().unwrap() {
+        fail::fail_point!("api-503", |_| Err(ApiError::ResourceUnavailable(
+            "failpoint".into()
+        )));
+
+        fail::fail_point!("api-500", |_| Err(ApiError::InternalServerError(
+            anyhow::anyhow!("failpoint")
+        )));
+    }
+
     // Spawn a new task to handle the request, to protect the handler from unexpected
     // async cancellations. Most pageserver functions are not async cancellation safe.
     // We arm a drop-guard, so that if Hyper drops the Future, we signal the task
@@ -2214,6 +2307,7 @@ pub fn make_router(
 
     Ok(router
         .data(state)
+        .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
         .get("/v1/status", |r| api_handler(r, status_handler))
         .put("/v1/failpoints", |r| {
             testing_api_handler("manage failpoints", r, failpoints_handler)
@@ -2247,7 +2341,7 @@ pub fn make_router(
         .get("/v1/location_config", |r| {
             api_handler(r, list_location_config_handler)
         })
-        .get("/v1/location_config/:tenant_id", |r| {
+        .get("/v1/location_config/:tenant_shard_id", |r| {
             api_handler(r, get_location_config_handler)
         })
         .put(

pageserver/src/import_datadir.rs

@@ -2,28 +2,21 @@
 //! Import data and WAL from a PostgreSQL data directory and WAL segments into
 //! a neon Timeline.
 //!
-use std::io::SeekFrom;
 use std::path::{Path, PathBuf};
 
 use anyhow::{bail, ensure, Context, Result};
-use async_compression::tokio::bufread::ZstdDecoder;
-use async_compression::{tokio::write::ZstdEncoder, zstd::CParameter, Level};
 use bytes::Bytes;
 use camino::Utf8Path;
 use futures::StreamExt;
-use nix::NixPath;
-use tokio::fs::{File, OpenOptions};
-use tokio::io::{AsyncBufRead, AsyncRead, AsyncReadExt, AsyncSeekExt, AsyncWriteExt};
+use pageserver_api::key::rel_block_to_key;
+use tokio::io::{AsyncRead, AsyncReadExt};
 use tokio_tar::Archive;
-use tokio_tar::Builder;
-use tokio_tar::HeaderMode;
 use tracing::*;
 use walkdir::WalkDir;
 
 use crate::context::RequestContext;
 use crate::metrics::WAL_INGEST;
 use crate::pgdatadir_mapping::*;
-use crate::tenant::remote_timeline_client::INITDB_PATH;
 use crate::tenant::Timeline;
 use crate::walingest::WalIngest;
 use crate::walrecord::DecodedWALRecord;
@@ -178,7 +171,10 @@ async fn import_rel(
         let r = reader.read_exact(&mut buf).await;
         match r {
             Ok(_) => {
-                modification.put_rel_page_image(rel, blknum, Bytes::copy_from_slice(&buf))?;
+                let key = rel_block_to_key(rel, blknum);
+                if modification.tline.get_shard_identity().is_key_local(&key) {
+                    modification.put_rel_page_image(rel, blknum, Bytes::copy_from_slice(&buf))?;
+                }
             }
 
             // TODO: UnexpectedEof is expected
@@ -633,65 +629,3 @@ async fn read_all_bytes(reader: &mut (impl AsyncRead + Unpin)) -> Result<Bytes>
     reader.read_to_end(&mut buf).await?;
     Ok(Bytes::from(buf))
 }
-
-pub async fn create_tar_zst(pgdata_path: &Utf8Path, tmp_path: &Utf8Path) -> Result<(File, u64)> {
-    let file = OpenOptions::new()
-        .create(true)
-        .truncate(true)
-        .read(true)
-        .write(true)
-        .open(&tmp_path)
-        .await
-        .with_context(|| format!("tempfile creation {tmp_path}"))?;
-
-    let mut paths = Vec::new();
-    for entry in WalkDir::new(pgdata_path) {
-        let entry = entry?;
-        let metadata = entry.metadata().expect("error getting dir entry metadata");
-        // Also allow directories so that we also get empty directories
-        if !(metadata.is_file() || metadata.is_dir()) {
-            continue;
-        }
-        let path = entry.into_path();
-        paths.push(path);
-    }
-    // Do a sort to get a more consistent listing
-    paths.sort_unstable();
-    let zstd = ZstdEncoder::with_quality_and_params(
-        file,
-        Level::Default,
-        &[CParameter::enable_long_distance_matching(true)],
-    );
-    let mut builder = Builder::new(zstd);
-    // Use reproducible header mode
-    builder.mode(HeaderMode::Deterministic);
-    for path in paths {
-        let rel_path = path.strip_prefix(pgdata_path)?;
-        if rel_path.is_empty() {
-            // The top directory should not be compressed,
-            // the tar crate doesn't like that
-            continue;
-        }
-        builder.append_path_with_name(&path, rel_path).await?;
-    }
-    let mut zstd = builder.into_inner().await?;
-    zstd.shutdown().await?;
-    let mut compressed = zstd.into_inner();
-    let compressed_len = compressed.metadata().await?.len();
-    const INITDB_TAR_ZST_WARN_LIMIT: u64 = 2 * 1024 * 1024;
-    if compressed_len > INITDB_TAR_ZST_WARN_LIMIT {
-        warn!("compressed {INITDB_PATH} size of {compressed_len} is above limit {INITDB_TAR_ZST_WARN_LIMIT}.");
-    }
-    compressed.seek(SeekFrom::Start(0)).await?;
-    Ok((compressed, compressed_len))
-}
-
-pub async fn extract_tar_zst(
-    pgdata_path: &Utf8Path,
-    tar_zst: impl AsyncBufRead + Unpin,
-) -> Result<()> {
-    let tar = Box::pin(ZstdDecoder::new(tar_zst));
-    let mut archive = Archive::new(tar);
-    archive.unpack(pgdata_path).await?;
-    Ok(())
-}

pageserver/src/lib.rs

@@ -31,6 +31,7 @@ pub mod walredo;
 use crate::task_mgr::TaskKind;
 use camino::Utf8Path;
 use deletion_queue::DeletionQueue;
+use tenant::mgr::TenantManager;
 use tracing::info;
 
 /// Current storage format version
@@ -53,7 +54,11 @@ static ZERO_PAGE: bytes::Bytes = bytes::Bytes::from_static(&[0u8; 8192]);
 pub use crate::metrics::preinitialize_metrics;
 
 #[tracing::instrument(skip_all, fields(%exit_code))]
-pub async fn shutdown_pageserver(deletion_queue: Option<DeletionQueue>, exit_code: i32) {
+pub async fn shutdown_pageserver(
+    tenant_manager: &TenantManager,
+    deletion_queue: Option<DeletionQueue>,
+    exit_code: i32,
+) {
     use std::time::Duration;
     // Shut down the libpq endpoint task. This prevents new connections from
     // being accepted.
@@ -67,7 +72,7 @@ pub async fn shutdown_pageserver(deletion_queue: Option<DeletionQueue>, exit_cod
     // Shut down all the tenants. This flushes everything to disk and kills
     // the checkpoint and GC tasks.
     timed(
-        tenant::mgr::shutdown_all_tenants(),
+        tenant_manager.shutdown(),
         "shutdown all tenants",
         Duration::from_secs(5),
     )
@@ -114,27 +119,27 @@ pub const METADATA_FILE_NAME: &str = "metadata";
 
 /// Per-tenant configuration file.
 /// Full path: `tenants/<tenant_id>/config`.
-pub const TENANT_CONFIG_NAME: &str = "config";
+pub(crate) const TENANT_CONFIG_NAME: &str = "config";
 
 /// Per-tenant configuration file.
 /// Full path: `tenants/<tenant_id>/config`.
-pub const TENANT_LOCATION_CONFIG_NAME: &str = "config-v1";
+pub(crate) const TENANT_LOCATION_CONFIG_NAME: &str = "config-v1";
 
 /// Per-tenant copy of their remote heatmap, downloaded into the local
 /// tenant path while in secondary mode.
-pub const TENANT_HEATMAP_BASENAME: &str = "heatmap-v1.json";
+pub(crate) const TENANT_HEATMAP_BASENAME: &str = "heatmap-v1.json";
 
 /// A suffix used for various temporary files. Any temporary files found in the
 /// data directory at pageserver startup can be automatically removed.
-pub const TEMP_FILE_SUFFIX: &str = "___temp";
+pub(crate) const TEMP_FILE_SUFFIX: &str = "___temp";
 
 /// A marker file to mark that a timeline directory was not fully initialized.
 /// If a timeline directory with this marker is encountered at pageserver startup,
 /// the timeline directory and the marker file are both removed.
 /// Full path: `tenants/<tenant_id>/timelines/<timeline_id>___uninit`.
-pub const TIMELINE_UNINIT_MARK_SUFFIX: &str = "___uninit";
+pub(crate) const TIMELINE_UNINIT_MARK_SUFFIX: &str = "___uninit";
 
-pub const TIMELINE_DELETE_MARK_SUFFIX: &str = "___delete";
+pub(crate) const TIMELINE_DELETE_MARK_SUFFIX: &str = "___delete";
 
 /// A marker file to prevent pageserver from loading a certain tenant on restart.
 /// Different from [`TIMELINE_UNINIT_MARK_SUFFIX`] due to semantics of the corresponding
@@ -161,11 +166,11 @@ fn ends_with_suffix(path: &Utf8Path, suffix: &str) -> bool {
 // from the directory name. Instead create type "UninitMark(TimelineId)" and only parse it once
 // from the name.
 
-pub fn is_uninit_mark(path: &Utf8Path) -> bool {
+pub(crate) fn is_uninit_mark(path: &Utf8Path) -> bool {
     ends_with_suffix(path, TIMELINE_UNINIT_MARK_SUFFIX)
 }
 
-pub fn is_delete_mark(path: &Utf8Path) -> bool {
+pub(crate) fn is_delete_mark(path: &Utf8Path) -> bool {
     ends_with_suffix(path, TIMELINE_DELETE_MARK_SUFFIX)
 }
 

pageserver/src/metrics.rs

@@ -1,5 +1,4 @@
 use enum_map::EnumMap;
-use metrics::metric_vec_duration::DurationResultObserver;
 use metrics::{
     register_counter_vec, register_gauge_vec, register_histogram, register_histogram_vec,
     register_int_counter, register_int_counter_pair_vec, register_int_counter_vec,
@@ -168,7 +167,7 @@ impl GetVectoredLatency {
 pub(crate) static GET_VECTORED_LATENCY: Lazy<GetVectoredLatency> = Lazy::new(|| {
     let inner = register_histogram_vec!(
         "pageserver_get_vectored_seconds",
-        "Time spent in get_vectored",
+        "Time spent in get_vectored, excluding time spent in timeline_get_throttle.",
         &["task_kind"],
         CRITICAL_OP_BUCKETS.into(),
     )
@@ -436,7 +435,7 @@ pub(crate) static RESIDENT_PHYSICAL_SIZE_GLOBAL: Lazy<UIntGauge> = Lazy::new(||
 static REMOTE_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
     register_uint_gauge_vec!(
         "pageserver_remote_physical_size",
-        "The size of the layer files present in the remote storage that are listed in the the remote index_part.json.",
+        "The size of the layer files present in the remote storage that are listed in the remote index_part.json.",
         // Corollary: If any files are missing from the index part, they won't be included here.
         &["tenant_id", "shard_id", "timeline_id"]
     )
@@ -700,6 +699,14 @@ pub static STARTUP_IS_LOADING: Lazy<UIntGauge> = Lazy::new(|| {
     .expect("Failed to register pageserver_startup_is_loading")
 });
 
+pub(crate) static TIMELINE_EPHEMERAL_BYTES: Lazy<UIntGauge> = Lazy::new(|| {
+    register_uint_gauge!(
+        "pageserver_timeline_ephemeral_bytes",
+        "Total number of bytes in ephemeral layers, summed for all timelines.  Approximate, lazily updated."
+    )
+    .expect("Failed to register metric")
+});
+
 /// Metrics related to the lifecycle of a [`crate::tenant::Tenant`] object: things
 /// like how long it took to load.
 ///
@@ -1283,11 +1290,65 @@ pub(crate) static BASEBACKUP_QUERY_TIME: Lazy<BasebackupQueryTime> = Lazy::new(|
     })
 });
 
-impl DurationResultObserver for BasebackupQueryTime {
-    fn observe_result<T, E>(&self, res: &Result<T, E>, duration: std::time::Duration) {
+pub(crate) struct BasebackupQueryTimeOngoingRecording<'a, 'c> {
+    parent: &'a BasebackupQueryTime,
+    ctx: &'c RequestContext,
+    start: std::time::Instant,
+}
+
+impl BasebackupQueryTime {
+    pub(crate) fn start_recording<'c: 'a, 'a>(
+        &'a self,
+        ctx: &'c RequestContext,
+    ) -> BasebackupQueryTimeOngoingRecording<'_, '_> {
+        let start = Instant::now();
+        match ctx.micros_spent_throttled.open() {
+            Ok(()) => (),
+            Err(error) => {
+                use utils::rate_limit::RateLimit;
+                static LOGGED: Lazy<Mutex<RateLimit>> =
+                    Lazy::new(|| Mutex::new(RateLimit::new(Duration::from_secs(10))));
+                let mut rate_limit = LOGGED.lock().unwrap();
+                rate_limit.call(|| {
+                    warn!(error, "error opening micros_spent_throttled; this message is logged at a global rate limit");
+                });
+            }
+        }
+        BasebackupQueryTimeOngoingRecording {
+            parent: self,
+            ctx,
+            start,
+        }
+    }
+}
+
+impl<'a, 'c> BasebackupQueryTimeOngoingRecording<'a, 'c> {
+    pub(crate) fn observe<T, E>(self, res: &Result<T, E>) {
+        let elapsed = self.start.elapsed();
+        let ex_throttled = self
+            .ctx
+            .micros_spent_throttled
+            .close_and_checked_sub_from(elapsed);
+        let ex_throttled = match ex_throttled {
+            Ok(ex_throttled) => ex_throttled,
+            Err(error) => {
+                use utils::rate_limit::RateLimit;
+                static LOGGED: Lazy<Mutex<RateLimit>> =
+                    Lazy::new(|| Mutex::new(RateLimit::new(Duration::from_secs(10))));
+                let mut rate_limit = LOGGED.lock().unwrap();
+                rate_limit.call(|| {
+                    warn!(error, "error deducting time spent throttled; this message is logged at a global rate limit");
+                });
+                elapsed
+            }
+        };
         let label_value = if res.is_ok() { "ok" } else { "error" };
-        let metric = self.0.get_metric_with_label_values(&[label_value]).unwrap();
-        metric.observe(duration.as_secs_f64());
+        let metric = self
+            .parent
+            .0
+            .get_metric_with_label_values(&[label_value])
+            .unwrap();
+        metric.observe(ex_throttled.as_secs_f64());
     }
 }
 
@@ -1422,12 +1483,18 @@ pub(crate) static DELETION_QUEUE: Lazy<DeletionQueueMetrics> = Lazy::new(|| {
 });
 
 pub(crate) struct WalIngestMetrics {
+    pub(crate) bytes_received: IntCounter,
     pub(crate) records_received: IntCounter,
     pub(crate) records_committed: IntCounter,
     pub(crate) records_filtered: IntCounter,
 }
 
 pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMetrics {
+    bytes_received: register_int_counter!(
+        "pageserver_wal_ingest_bytes_received",
+        "Bytes of WAL ingested from safekeepers",
+    )
+    .unwrap(),
     records_received: register_int_counter!(
         "pageserver_wal_ingest_records_received",
         "Number of WAL records received from safekeepers"
@@ -1964,10 +2031,8 @@ impl TimelineMetrics {
     pub(crate) fn resident_physical_size_get(&self) -> u64 {
         self.resident_physical_size_gauge.get()
     }
-}
 
-impl Drop for TimelineMetrics {
-    fn drop(&mut self) {
+    pub(crate) fn shutdown(&self) {
         let tenant_id = &self.tenant_id;
         let timeline_id = &self.timeline_id;
         let shard_id = &self.shard_id;
@@ -2035,6 +2100,7 @@ pub(crate) fn remove_tenant_metrics(tenant_shard_id: &TenantShardId) {
 use futures::Future;
 use pin_project_lite::pin_project;
 use std::collections::HashMap;
+use std::num::NonZeroUsize;
 use std::pin::Pin;
 use std::sync::{Arc, Mutex};
 use std::task::{Context, Poll};
@@ -2414,7 +2480,8 @@ impl<F: Future<Output = Result<O, E>>, O, E> Future for MeasuredRemoteOp<F> {
 }
 
 pub mod tokio_epoll_uring {
-    use metrics::UIntGauge;
+    use metrics::{register_int_counter, UIntGauge};
+    use once_cell::sync::Lazy;
 
     pub struct Collector {
         descs: Vec<metrics::core::Desc>,
@@ -2422,15 +2489,13 @@ pub mod tokio_epoll_uring {
         systems_destroyed: UIntGauge,
     }
 
-    const NMETRICS: usize = 2;
-
     impl metrics::core::Collector for Collector {
         fn desc(&self) -> Vec<&metrics::core::Desc> {
             self.descs.iter().collect()
         }
 
         fn collect(&self) -> Vec<metrics::proto::MetricFamily> {
-            let mut mfs = Vec::with_capacity(NMETRICS);
+            let mut mfs = Vec::with_capacity(Self::NMETRICS);
             let tokio_epoll_uring::metrics::Metrics {
                 systems_created,
                 systems_destroyed,
@@ -2444,6 +2509,8 @@ pub mod tokio_epoll_uring {
     }
 
     impl Collector {
+        const NMETRICS: usize = 2;
+
         #[allow(clippy::new_without_default)]
         pub fn new() -> Self {
             let mut descs = Vec::new();
@@ -2477,6 +2544,22 @@ pub mod tokio_epoll_uring {
             }
         }
     }
+
+    pub(crate) static THREAD_LOCAL_LAUNCH_SUCCESSES: Lazy<metrics::IntCounter> = Lazy::new(|| {
+        register_int_counter!(
+            "pageserver_tokio_epoll_uring_pageserver_thread_local_launch_success_count",
+            "Number of times where thread_local_system creation spanned multiple executor threads",
+        )
+        .unwrap()
+    });
+
+    pub(crate) static THREAD_LOCAL_LAUNCH_FAILURES: Lazy<metrics::IntCounter> = Lazy::new(|| {
+        register_int_counter!(
+            "pageserver_tokio_epoll_uring_pageserver_thread_local_launch_failures_count",
+            "Number of times thread_local_system creation failed and was retried after back-off.",
+        )
+        .unwrap()
+    });
 }
 
 pub(crate) mod tenant_throttling {
@@ -2587,6 +2670,26 @@ pub(crate) mod disk_usage_based_eviction {
     pub(crate) static METRICS: Lazy<Metrics> = Lazy::new(Metrics::default);
 }
 
+static TOKIO_EXECUTOR_THREAD_COUNT: Lazy<UIntGaugeVec> = Lazy::new(|| {
+    register_uint_gauge_vec!(
+        "pageserver_tokio_executor_thread_configured_count",
+        "Total number of configued tokio executor threads in the process.
+         The `setup` label denotes whether we're running with multiple runtimes or a single runtime.",
+        &["setup"],
+    )
+    .unwrap()
+});
+
+pub(crate) fn set_tokio_runtime_setup(setup: &str, num_threads: NonZeroUsize) {
+    static SERIALIZE: std::sync::Mutex<()> = std::sync::Mutex::new(());
+    let _guard = SERIALIZE.lock().unwrap();
+    TOKIO_EXECUTOR_THREAD_COUNT.reset();
+    TOKIO_EXECUTOR_THREAD_COUNT
+        .get_metric_with_label_values(&[setup])
+        .unwrap()
+        .set(u64::try_from(num_threads.get()).unwrap());
+}
+
 pub fn preinitialize_metrics() {
     // Python tests need these and on some we do alerting.
     //
@@ -2605,6 +2708,8 @@ pub fn preinitialize_metrics() {
         &WALRECEIVER_BROKER_UPDATES,
         &WALRECEIVER_CANDIDATES_ADDED,
         &WALRECEIVER_CANDIDATES_REMOVED,
+        &tokio_epoll_uring::THREAD_LOCAL_LAUNCH_FAILURES,
+        &tokio_epoll_uring::THREAD_LOCAL_LAUNCH_SUCCESSES,
     ]
     .into_iter()
     .for_each(|c| {
@@ -2623,6 +2728,12 @@ pub fn preinitialize_metrics() {
     Lazy::force(&crate::tenant::storage_layer::layer::LAYER_IMPL_METRICS);
     Lazy::force(&disk_usage_based_eviction::METRICS);
 
+    for state_name in pageserver_api::models::TenantState::VARIANTS {
+        // initialize the metric for all gauges, otherwise the time series might seemingly show
+        // values from last restart.
+        TENANT_STATE_METRIC.with_label_values(&[state_name]).set(0);
+    }
+
     // countervecs
     [&BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT]
         .into_iter()