Compare commits

..

2 Commits

Author SHA1 Message Date
Conrad Ludgate
6d2bbffdab only for console 2023-12-15 12:28:50 +00:00
Conrad Ludgate
7151bcc175 proxy console force http2 2023-12-15 12:26:51 +00:00
230 changed files with 4697 additions and 13380 deletions

View File

@@ -1,2 +0,0 @@
[profile.default]
slow-timeout = "1m"

View File

@@ -11,7 +11,7 @@ on:
# │ │ ┌───────────── day of the month (1 - 31)
# │ │ │ ┌───────────── month (1 - 12 or JAN-DEC)
# │ │ │ │ ┌───────────── day of the week (0 - 6 or SUN-SAT)
- cron: '0 3 * * *' # run once a day, timezone is utc
- cron: '0 3 * * *' # run once a day, timezone is utc
workflow_dispatch: # adds ability to run this manually
inputs:
@@ -23,21 +23,6 @@ on:
type: boolean
description: 'Publish perf report. If not set, the report will be published only for the main branch'
required: false
collect_olap_explain:
type: boolean
description: 'Collect EXPLAIN ANALYZE for OLAP queries. If not set, EXPLAIN ANALYZE will not be collected'
required: false
default: false
collect_pg_stat_statements:
type: boolean
description: 'Collect pg_stat_statements for OLAP queries. If not set, pg_stat_statements will not be collected'
required: false
default: false
run_AWS_RDS_AND_AURORA:
type: boolean
description: 'AWS-RDS and AWS-AURORA normally only run on Saturday. Set this to true to run them on every workflow_dispatch'
required: false
default: false
defaults:
run:
@@ -128,8 +113,6 @@ jobs:
# - neon-captest-reuse: Reusing existing project
# - rds-aurora: Aurora Postgres Serverless v2 with autoscaling from 0.5 to 2 ACUs
# - rds-postgres: RDS Postgres db.m5.large instance (2 vCPU, 8 GiB) with gp3 EBS storage
env:
RUN_AWS_RDS_AND_AURORA: ${{ github.event.inputs.run_AWS_RDS_AND_AURORA || 'false' }}
runs-on: ubuntu-latest
outputs:
pgbench-compare-matrix: ${{ steps.pgbench-compare-matrix.outputs.matrix }}
@@ -169,7 +152,7 @@ jobs:
]
}'
if [ "$(date +%A)" = "Saturday" ] || [ ${RUN_AWS_RDS_AND_AURORA} = "true" ]; then
if [ "$(date +%A)" = "Saturday" ]; then
matrix=$(echo "$matrix" | jq '.include += [{ "platform": "rds-postgres" },
{ "platform": "rds-aurora" }]')
fi
@@ -188,9 +171,9 @@ jobs:
]
}'
if [ "$(date +%A)" = "Saturday" ] || [ ${RUN_AWS_RDS_AND_AURORA} = "true" ]; then
if [ "$(date +%A)" = "Saturday" ]; 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
@@ -354,8 +337,6 @@ jobs:
POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
DEFAULT_PG_VERSION: 14
TEST_OUTPUT: /tmp/test_output
TEST_OLAP_COLLECT_EXPLAIN: ${{ github.event.inputs.collect_olap_explain }}
TEST_OLAP_COLLECT_PG_STAT_STATEMENTS: ${{ github.event.inputs.collect_pg_stat_statements }}
BUILD_TYPE: remote
SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
PLATFORM: ${{ matrix.platform }}
@@ -418,8 +399,6 @@ jobs:
env:
VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
TEST_OLAP_COLLECT_EXPLAIN: ${{ github.event.inputs.collect_olap_explain || 'false' }}
TEST_OLAP_COLLECT_PG_STAT_STATEMENTS: ${{ github.event.inputs.collect_pg_stat_statements || 'false' }}
BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
TEST_OLAP_SCALE: 10

View File

@@ -1,105 +0,0 @@
name: Build and Push Docker Image
on:
workflow_call:
inputs:
dockerfile-path:
required: true
type: string
image-name:
required: true
type: string
outputs:
build-tools-tag:
description: "tag generated for build tools"
value: ${{ jobs.tag.outputs.build-tools-tag }}
jobs:
check-if-build-tools-dockerfile-changed:
runs-on: ubuntu-latest
outputs:
docker_file_changed: ${{ steps.dockerfile.outputs.docker_file_changed }}
steps:
- name: Check if Dockerfile.buildtools has changed
id: dockerfile
run: |
if [[ "$GITHUB_EVENT_NAME" != "pull_request" ]]; then
echo "docker_file_changed=false" >> $GITHUB_OUTPUT
exit
fi
updated_files=$(gh pr --repo neondatabase/neon diff ${{ github.event.pull_request.number }} --name-only)
if [[ $updated_files == *"Dockerfile.buildtools"* ]]; then
echo "docker_file_changed=true" >> $GITHUB_OUTPUT
fi
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
tag:
runs-on: ubuntu-latest
needs: [ check-if-build-tools-dockerfile-changed ]
outputs:
build-tools-tag: ${{steps.buildtools-tag.outputs.image_tag}}
steps:
- name: Get buildtools tag
env:
DOCKERFILE_CHANGED: ${{ needs.check-if-build-tools-dockerfile-changed.outputs.docker_file_changed }}
run: |
if [[ "$GITHUB_EVENT_NAME" == "pull_request" ]] && [[ "${DOCKERFILE_CHANGED}" == "true" ]]; then
IMAGE_TAG=$GITHUB_RUN_ID
else
IMAGE_TAG=pinned
fi
echo "image_tag=${IMAGE_TAG}" >> $GITHUB_OUTPUT
shell: bash
id: buildtools-tag
kaniko:
if: needs.check-if-build-tools-dockerfile-changed.outputs.docker_file_changed == 'true'
needs: [ tag, check-if-build-tools-dockerfile-changed ]
runs-on: [ self-hosted, dev, x64 ]
container: gcr.io/kaniko-project/executor:v1.7.0-debug
steps:
- name: Checkout
uses: actions/checkout@v1
- name: Configure ECR login
run: echo "{\"credsStore\":\"ecr-login\"}" > /kaniko/.docker/config.json
- name: Kaniko build
run: /kaniko/executor --reproducible --snapshotMode=redo --skip-unused-stages --dockerfile ${{ inputs.dockerfile-path }} --cache=true --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-amd64
kaniko-arm:
if: needs.check-if-build-tools-dockerfile-changed.outputs.docker_file_changed == 'true'
needs: [ tag, check-if-build-tools-dockerfile-changed ]
runs-on: [ self-hosted, dev, arm64 ]
container: gcr.io/kaniko-project/executor:v1.7.0-debug
steps:
- name: Checkout
uses: actions/checkout@v1
- name: Configure ECR login
run: echo "{\"credsStore\":\"ecr-login\"}" > /kaniko/.docker/config.json
- name: Kaniko build
run: /kaniko/executor --reproducible --snapshotMode=redo --skip-unused-stages --dockerfile ${{ inputs.dockerfile-path }} --cache=true --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-arm64
manifest:
if: needs.check-if-build-tools-dockerfile-changed.outputs.docker_file_changed == 'true'
name: 'manifest'
runs-on: [ self-hosted, dev, x64 ]
needs:
- tag
- kaniko
- kaniko-arm
- check-if-build-tools-dockerfile-changed
steps:
- name: Create manifest
run: docker manifest create 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }} --amend 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-amd64 --amend 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-arm64
- name: Push manifest
run: docker manifest push 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}

View File

@@ -44,6 +44,7 @@ jobs:
exit 1
tag:
needs: [ check-permissions ]
runs-on: [ self-hosted, gen3, small ]
@@ -73,19 +74,11 @@ jobs:
shell: bash
id: build-tag
build-buildtools-image:
needs: [ check-permissions ]
uses: ./.github/workflows/build_and_push_docker_image.yml
with:
dockerfile-path: Dockerfile.buildtools
image-name: build-tools
secrets: inherit
check-codestyle-python:
needs: [ check-permissions, build-buildtools-image ]
needs: [ check-permissions ]
runs-on: [ self-hosted, gen3, small ]
container:
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
options: --init
steps:
@@ -105,20 +98,20 @@ jobs:
- name: Install Python deps
run: ./scripts/pysync
- name: Run `ruff check` to ensure code format
run: poetry run ruff check .
- name: Run ruff to ensure code format
run: poetry run ruff .
- name: Run `ruff format` to ensure code format
run: poetry run ruff format --check .
- name: Run black to ensure code format
run: poetry run black --diff --check .
- name: Run mypy to check types
run: poetry run mypy .
check-codestyle-rust:
needs: [ check-permissions, build-buildtools-image ]
needs: [ check-permissions ]
runs-on: [ self-hosted, gen3, large ]
container:
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
options: --init
steps:
@@ -182,10 +175,10 @@ jobs:
run: cargo deny check --hide-inclusion-graph
build-neon:
needs: [ check-permissions, tag, build-buildtools-image ]
needs: [ check-permissions, tag ]
runs-on: [ self-hosted, gen3, large ]
container:
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
options: --init
strategy:
fail-fast: false
@@ -339,16 +332,16 @@ jobs:
run: |
${cov_prefix} mold -run cargo build $CARGO_FLAGS $CARGO_FEATURES --bins --tests
- name: Run rust tests
- name: Run cargo test
run: |
${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES
${cov_prefix} cargo test $CARGO_FLAGS $CARGO_FEATURES
# Run separate tests for real S3
export ENABLE_REAL_S3_REMOTE_STORAGE=nonempty
export REMOTE_STORAGE_S3_BUCKET=neon-github-ci-tests
export REMOTE_STORAGE_S3_BUCKET=neon-github-public-dev
export REMOTE_STORAGE_S3_REGION=eu-central-1
# Avoid `$CARGO_FEATURES` since there's no `testing` feature in the e2e tests now
${cov_prefix} cargo nextest run $CARGO_FLAGS -E 'package(remote_storage)' -E 'test(test_real_s3)'
${cov_prefix} cargo test $CARGO_FLAGS --package remote_storage --test test_real_s3
# Run separate tests for real Azure Blob Storage
# XXX: replace region with `eu-central-1`-like region
@@ -358,7 +351,7 @@ jobs:
export REMOTE_STORAGE_AZURE_CONTAINER="${{ vars.REMOTE_STORAGE_AZURE_CONTAINER }}"
export REMOTE_STORAGE_AZURE_REGION="${{ vars.REMOTE_STORAGE_AZURE_REGION }}"
# Avoid `$CARGO_FEATURES` since there's no `testing` feature in the e2e tests now
${cov_prefix} cargo nextest run $CARGO_FLAGS -E 'package(remote_storage)' -E 'test(test_real_azure)'
${cov_prefix} cargo test $CARGO_FLAGS --package remote_storage --test test_real_azure
- name: Install rust binaries
run: |
@@ -415,10 +408,10 @@ jobs:
uses: ./.github/actions/save-coverage-data
regress-tests:
needs: [ check-permissions, build-neon, build-buildtools-image, tag ]
needs: [ check-permissions, build-neon, tag ]
runs-on: [ self-hosted, gen3, large ]
container:
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
# Default shared memory is 64mb
options: --init --shm-size=512mb
strategy:
@@ -454,10 +447,10 @@ jobs:
uses: ./.github/actions/save-coverage-data
benchmarks:
needs: [ check-permissions, build-neon, build-buildtools-image ]
needs: [ check-permissions, build-neon ]
runs-on: [ self-hosted, gen3, small ]
container:
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
# Default shared memory is 64mb
options: --init --shm-size=512mb
if: github.ref_name == 'main' || contains(github.event.pull_request.labels.*.name, 'run-benchmarks')
@@ -486,12 +479,12 @@ jobs:
# while coverage is currently collected for the debug ones
create-test-report:
needs: [ check-permissions, regress-tests, coverage-report, benchmarks, build-buildtools-image ]
needs: [ check-permissions, regress-tests, coverage-report, benchmarks ]
if: ${{ !cancelled() && contains(fromJSON('["skipped", "success"]'), needs.check-permissions.result) }}
runs-on: [ self-hosted, gen3, small ]
container:
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
options: --init
steps:
@@ -533,10 +526,11 @@ jobs:
})
coverage-report:
needs: [ check-permissions, regress-tests, build-buildtools-image ]
needs: [ check-permissions, regress-tests ]
runs-on: [ self-hosted, gen3, small ]
container:
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
options: --init
strategy:
fail-fast: false
@@ -700,7 +694,7 @@ jobs:
}"
neon-image:
needs: [ check-permissions, build-buildtools-image, tag ]
needs: [ check-permissions, tag ]
runs-on: [ self-hosted, gen3, large ]
container: gcr.io/kaniko-project/executor:v1.9.2-debug
defaults:
@@ -739,7 +733,6 @@ jobs:
--context .
--build-arg GIT_VERSION=${{ github.event.pull_request.head.sha || github.sha }}
--build-arg BUILD_TAG=${{ needs.tag.outputs.build-tag }}
--build-arg TAG=${{ needs.build-buildtools-image.outputs.build-tools-tag }}
--build-arg REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com
--destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:${{needs.tag.outputs.build-tag}}
--destination neondatabase/neon:${{needs.tag.outputs.build-tag}}
@@ -750,7 +743,7 @@ jobs:
compute-tools-image:
runs-on: [ self-hosted, gen3, large ]
needs: [ check-permissions, build-buildtools-image, tag ]
needs: [ check-permissions, tag ]
container: gcr.io/kaniko-project/executor:v1.9.2-debug
defaults:
run:
@@ -785,7 +778,6 @@ jobs:
--context .
--build-arg GIT_VERSION=${{ github.event.pull_request.head.sha || github.sha }}
--build-arg BUILD_TAG=${{needs.tag.outputs.build-tag}}
--build-arg TAG=${{needs.build-buildtools-image.outputs.build-tools-tag}}
--build-arg REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com
--dockerfile Dockerfile.compute-tools
--destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{needs.tag.outputs.build-tag}}
@@ -796,7 +788,7 @@ jobs:
run: rm -rf ~/.ecr
compute-node-image:
needs: [ check-permissions, build-buildtools-image, tag ]
needs: [ check-permissions, tag ]
runs-on: [ self-hosted, gen3, large ]
container:
image: gcr.io/kaniko-project/executor:v1.9.2-debug
@@ -844,7 +836,6 @@ jobs:
--build-arg GIT_VERSION=${{ github.event.pull_request.head.sha || github.sha }}
--build-arg PG_VERSION=${{ matrix.version }}
--build-arg BUILD_TAG=${{needs.tag.outputs.build-tag}}
--build-arg TAG=${{needs.build-buildtools-image.outputs.build-tools-tag}}
--build-arg REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com
--dockerfile Dockerfile.compute-node
--destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
@@ -866,7 +857,7 @@ jobs:
run:
shell: sh -eu {0}
env:
VM_BUILDER_VERSION: v0.21.0
VM_BUILDER_VERSION: v0.19.0
steps:
- name: Checkout

View File

@@ -218,7 +218,7 @@ jobs:
# Run separate tests for real S3
export ENABLE_REAL_S3_REMOTE_STORAGE=nonempty
export REMOTE_STORAGE_S3_BUCKET=neon-github-ci-tests
export REMOTE_STORAGE_S3_BUCKET=neon-github-public-dev
export REMOTE_STORAGE_S3_REGION=eu-central-1
# Avoid `$CARGO_FEATURES` since there's no `testing` feature in the e2e tests now
cargo test $CARGO_FLAGS --package remote_storage --test test_real_s3

View File

@@ -1,130 +0,0 @@
name: 'Update build tools image tag'
# This workflow it used to update tag of build tools in ECR.
# The most common use case is adding/moving `pinned` tag to `${GITHUB_RUN_IT}` image.
on:
workflow_dispatch:
inputs:
from-tag:
description: 'Source tag'
required: true
type: string
to-tag:
description: 'Destination tag'
required: true
type: string
default: 'pinned'
defaults:
run:
shell: bash -euo pipefail {0}
env:
AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
permissions: {}
jobs:
tag-image:
runs-on: [ self-hosted, gen3, small ]
container: golang:1.19-bullseye
env:
IMAGE: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools
FROM_TAG: ${{ inputs.from-tag }}
TO_TAG: ${{ inputs.to-tag }}
outputs:
next-digest-buildtools: ${{ steps.next-digest.outputs.next-digest-buildtools }}
prev-digest-buildtools: ${{ steps.prev-digest.outputs.prev-digest-buildtools }}
steps:
- name: Install Crane & ECR helper
run: |
go install github.com/google/go-containerregistry/cmd/crane@a54d64203cffcbf94146e04069aae4a97f228ee2 # v0.16.1
go install github.com/awslabs/amazon-ecr-credential-helper/ecr-login/cli/docker-credential-ecr-login@adf1bafd791ae7d4ff098108b1e91f36a4da5404 # v0.7.1
- name: Configure ECR login
run: |
mkdir /github/home/.docker/
echo "{\"credsStore\":\"ecr-login\"}" > /github/home/.docker/config.json
- name: Get source image digest
id: next-digest
run: |
NEXT_DIGEST=$(crane digest ${IMAGE}:${FROM_TAG} || true)
if [ -z "${NEXT_DIGEST}" ]; then
echo >&2 "Image ${IMAGE}:${FROM_TAG} does not exist"
exit 1
fi
echo "Current ${IMAGE}@${FROM_TAG} image is ${IMAGE}@${NEXT_DIGEST}"
echo "next-digest-buildtools=$NEXT_DIGEST" >> $GITHUB_OUTPUT
- name: Get destination image digest (if already exists)
id: prev-digest
run: |
PREV_DIGEST=$(crane digest ${IMAGE}:${TO_TAG} || true)
if [ -z "${PREV_DIGEST}" ]; then
echo >&2 "Image ${IMAGE}:${TO_TAG} does not exist (it's ok)"
else
echo >&2 "Current ${IMAGE}@${TO_TAG} image is ${IMAGE}@${PREV_DIGEST}"
echo "prev-digest-buildtools=$PREV_DIGEST" >> $GITHUB_OUTPUT
fi
- name: Tag image
run: |
crane tag "${IMAGE}:${FROM_TAG}" "${TO_TAG}"
rollback-tag-image:
needs: tag-image
if: ${{ !success() }}
runs-on: [ self-hosted, gen3, small ]
container: golang:1.19-bullseye
env:
IMAGE: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools
FROM_TAG: ${{ inputs.from-tag }}
TO_TAG: ${{ inputs.to-tag }}
steps:
- name: Install Crane & ECR helper
run: |
go install github.com/google/go-containerregistry/cmd/crane@a54d64203cffcbf94146e04069aae4a97f228ee2 # v0.16.1
go install github.com/awslabs/amazon-ecr-credential-helper/ecr-login/cli/docker-credential-ecr-login@adf1bafd791ae7d4ff098108b1e91f36a4da5404 # v0.7.1
- name: Configure ECR login
run: |
mkdir /github/home/.docker/
echo "{\"credsStore\":\"ecr-login\"}" > /github/home/.docker/config.json
- name: Restore previous tag if needed
run: |
NEXT_DIGEST="${{ needs.tag-image.outputs.next-digest-buildtools }}"
PREV_DIGEST="${{ needs.tag-image.outputs.prev-digest-buildtools }}"
if [ -z "${NEXT_DIGEST}" ]; then
echo >&2 "Image ${IMAGE}:${FROM_TAG} does not exist, nothing to rollback"
exit 0
fi
if [ -z "${PREV_DIGEST}" ]; then
# I guess we should delete the tag here/untag the image, but crane does not support it
# - https://github.com/google/go-containerregistry/issues/999
echo >&2 "Image ${IMAGE}:${TO_TAG} did not exist, but it was created by the job, no need to rollback"
exit 0
fi
CURRENT_DIGEST=$(crane digest "${IMAGE}:${TO_TAG}")
if [ "${CURRENT_DIGEST}" == "${NEXT_DIGEST}" ]; then
crane tag "${IMAGE}@${PREV_DIGEST}" "${TO_TAG}"
echo >&2 "Successfully restored ${TO_TAG} tag from ${IMAGE}@${CURRENT_DIGEST} to ${IMAGE}@${PREV_DIGEST}"
else
echo >&2 "Image ${IMAGE}:${TO_TAG}@${CURRENT_DIGEST} is not required to be restored"
fi

1
.gitignore vendored
View File

@@ -6,7 +6,6 @@ __pycache__/
test_output/
.vscode
.idea
neon.iml
/.neon
/integration_tests/.neon

View File

@@ -70,17 +70,3 @@ We're using the following approach to make it work:
- The label gets removed automatically, so to run CI again with new changes, the label should be added again (after the review)
For details see [`approved-for-ci-run.yml`](.github/workflows/approved-for-ci-run.yml)
## How do I add the "pinned" tag to an buildtools image?
We use the `pinned` tag for `Dockerfile.buildtools` build images in our CI/CD setup, currently adding the `pinned` tag is a manual operation.
You can call it from GitHub UI: https://github.com/neondatabase/neon/actions/workflows/update_build_tools_image.yml,
or using GitHub CLI:
```bash
gh workflow -R neondatabase/neon run update_build_tools_image.yml \
-f from-tag=6254913013 \
-f to-tag=pinned \
# Default `-f to-tag` is `pinned`, so the parameter can be omitted.
```

241
Cargo.lock generated
View File

@@ -190,9 +190,9 @@ dependencies = [
[[package]]
name = "async-compression"
version = "0.4.5"
version = "0.4.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "bc2d0cfb2a7388d34f590e76686704c494ed7aaceed62ee1ba35cbf363abc2a5"
checksum = "5b0122885821398cc923ece939e24d1056a2384ee719432397fa9db87230ff11"
dependencies = [
"flate2",
"futures-core",
@@ -233,7 +233,7 @@ checksum = "16e62a023e7c117e27523144c5d2459f4397fcc3cab0085af8e2224f643a0193"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -244,7 +244,7 @@ checksum = "b9ccdd8f2a161be9bd5c023df56f1b2a0bd1d83872ae53b71a84a12c9bf6e842"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -881,7 +881,7 @@ dependencies = [
"regex",
"rustc-hash",
"shlex",
"syn 2.0.32",
"syn 2.0.28",
"which",
]
@@ -1095,7 +1095,7 @@ dependencies = [
"heck",
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -1161,7 +1161,6 @@ dependencies = [
"flate2",
"futures",
"hyper",
"nix 0.26.2",
"notify",
"num_cpus",
"opentelemetry",
@@ -1169,10 +1168,8 @@ dependencies = [
"regex",
"remote_storage",
"reqwest",
"rust-ini",
"serde",
"serde_json",
"signal-hook",
"tar",
"tokio",
"tokio-postgres",
@@ -1204,26 +1201,6 @@ version = "0.9.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "28c122c3980598d243d63d9a704629a2d748d101f278052ff068be5a4423ab6f"
[[package]]
name = "const-random"
version = "0.1.17"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "5aaf16c9c2c612020bcfd042e170f6e32de9b9d75adb5277cdbbd2e2c8c8299a"
dependencies = [
"const-random-macro",
]
[[package]]
name = "const-random-macro"
version = "0.1.16"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f9d839f2a20b0aee515dc581a6172f2321f96cab76c1a38a4c584a194955390e"
dependencies = [
"getrandom 0.2.11",
"once_cell",
"tiny-keccak",
]
[[package]]
name = "const_fn"
version = "0.4.9"
@@ -1268,19 +1245,16 @@ name = "control_plane"
version = "0.1.0"
dependencies = [
"anyhow",
"async-trait",
"camino",
"clap",
"comfy-table",
"compute_api",
"futures",
"git-version",
"hex",
"hyper",
"nix 0.26.2",
"once_cell",
"pageserver_api",
"pageserver_client",
"postgres",
"postgres_backend",
"postgres_connection",
@@ -1294,8 +1268,6 @@ dependencies = [
"tar",
"thiserror",
"tokio",
"tokio-postgres",
"tokio-util",
"toml",
"tracing",
"url",
@@ -1456,12 +1428,6 @@ dependencies = [
"winapi",
]
[[package]]
name = "crunchy"
version = "0.2.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "7a81dae078cea95a014a339291cec439d2f232ebe854a9d672b796c6afafa9b7"
[[package]]
name = "crypto-bigint"
version = "0.4.9"
@@ -1515,7 +1481,7 @@ dependencies = [
"proc-macro2",
"quote",
"strsim",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -1526,7 +1492,7 @@ checksum = "29a358ff9f12ec09c3e61fef9b5a9902623a695a46a917b07f269bff1445611a"
dependencies = [
"darling_core",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -1601,16 +1567,7 @@ checksum = "487585f4d0c6655fe74905e2504d8ad6908e4db67f744eb140876906c2f3175d"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
]
[[package]]
name = "dlv-list"
version = "0.5.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "442039f5147480ba31067cb00ada1adae6892028e40e45fc5de7b7df6dcc1b5f"
dependencies = [
"const-random",
"syn 2.0.28",
]
[[package]]
@@ -1704,7 +1661,7 @@ dependencies = [
"darling",
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -1958,7 +1915,7 @@ checksum = "89ca545a94061b6365f2c7355b4b32bd20df3ff95f02da9329b34ccc3bd6ee72"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -2144,20 +2101,6 @@ dependencies = [
"hashbrown 0.13.2",
]
[[package]]
name = "hdrhistogram"
version = "7.5.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "765c9198f173dd59ce26ff9f95ef0aafd0a0fe01fb9d72841bc5066a4c06511d"
dependencies = [
"base64 0.21.1",
"byteorder",
"crossbeam-channel",
"flate2",
"nom",
"num-traits",
]
[[package]]
name = "heapless"
version = "0.8.0"
@@ -2539,14 +2482,13 @@ dependencies = [
[[package]]
name = "jsonwebtoken"
version = "9.2.0"
version = "8.3.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "5c7ea04a7c5c055c175f189b6dc6ba036fd62306b58c66c9f6389036c503a3f4"
checksum = "6971da4d9c3aa03c3d8f3ff0f4155b534aad021292003895a469716b2a230378"
dependencies = [
"base64 0.21.1",
"js-sys",
"pem 3.0.3",
"ring 0.17.6",
"pem 1.1.1",
"ring 0.16.20",
"serde",
"serde_json",
"simple_asn1",
@@ -2959,7 +2901,7 @@ checksum = "a948666b637a0f465e8564c73e89d4dde00d72d4d473cc972f390fc3dcee7d9c"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -3081,16 +3023,6 @@ dependencies = [
"tokio-stream",
]
[[package]]
name = "ordered-multimap"
version = "0.7.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "a4d6a8c22fc714f0c2373e6091bf6f5e9b37b1bc0b1184874b7e0a4e303d318f"
dependencies = [
"dlv-list",
"hashbrown 0.14.0",
]
[[package]]
name = "os_info"
version = "3.7.0"
@@ -3119,28 +3051,6 @@ dependencies = [
"sha2",
]
[[package]]
name = "pagebench"
version = "0.1.0"
dependencies = [
"anyhow",
"clap",
"futures",
"hdrhistogram",
"humantime",
"humantime-serde",
"pageserver",
"pageserver_api",
"pageserver_client",
"rand 0.8.5",
"serde",
"serde_json",
"tokio",
"tracing",
"utils",
"workspace_hack",
]
[[package]]
name = "pagectl"
version = "0.1.0"
@@ -3230,7 +3140,6 @@ dependencies = [
"tokio",
"tokio-io-timeout",
"tokio-postgres",
"tokio-stream",
"tokio-tar",
"tokio-util",
"toml_edit",
@@ -3253,7 +3162,6 @@ dependencies = [
"enum-map",
"hex",
"postgres_ffi",
"rand 0.8.5",
"serde",
"serde_json",
"serde_with",
@@ -3264,27 +3172,6 @@ dependencies = [
"workspace_hack",
]
[[package]]
name = "pageserver_client"
version = "0.1.0"
dependencies = [
"anyhow",
"async-trait",
"bytes",
"futures",
"pageserver_api",
"postgres",
"reqwest",
"serde",
"thiserror",
"tokio",
"tokio-postgres",
"tokio-stream",
"tokio-util",
"utils",
"workspace_hack",
]
[[package]]
name = "parking"
version = "2.1.1"
@@ -3376,19 +3263,18 @@ checksum = "19b17cddbe7ec3f8bc800887bab5e717348c95ea2ca0b1bf0837fb964dc67099"
[[package]]
name = "pem"
version = "2.0.1"
version = "1.1.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "6b13fe415cdf3c8e44518e18a7c95a13431d9bdf6d15367d82b23c377fdd441a"
checksum = "a8835c273a76a90455d7344889b0964598e3316e2a79ede8e36f16bdcf2228b8"
dependencies = [
"base64 0.21.1",
"serde",
"base64 0.13.1",
]
[[package]]
name = "pem"
version = "3.0.3"
version = "2.0.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "1b8fcc794035347fb64beda2d3b462595dd2753e3f268d89c5aae77e8cf2c310"
checksum = "6b13fe415cdf3c8e44518e18a7c95a13431d9bdf6d15367d82b23c377fdd441a"
dependencies = [
"base64 0.21.1",
"serde",
@@ -3445,7 +3331,7 @@ checksum = "39407670928234ebc5e6e580247dd567ad73a3578460c5990f9503df207e8f07"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -3652,7 +3538,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "3b69d39aab54d069e7f2fe8cb970493e7834601ca2d8c65fd7bbd183578080d1"
dependencies = [
"proc-macro2",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -4260,20 +4146,10 @@ dependencies = [
"regex",
"relative-path",
"rustc_version",
"syn 2.0.32",
"syn 2.0.28",
"unicode-ident",
]
[[package]]
name = "rust-ini"
version = "0.20.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "3e0698206bcb8882bf2a9ecb4c1e7785db57ff052297085a6efd4fe42302068a"
dependencies = [
"cfg-if",
"ordered-multimap",
]
[[package]]
name = "rustc-demangle"
version = "0.1.23"
@@ -4405,20 +4281,17 @@ dependencies = [
"async-stream",
"aws-config",
"aws-sdk-s3",
"aws-smithy-async",
"bincode",
"bytes",
"chrono",
"clap",
"crc32c",
"either",
"futures",
"futures-util",
"hex",
"histogram",
"itertools",
"pageserver",
"pageserver_api",
"rand 0.8.5",
"remote_storage",
"reqwest",
@@ -4451,7 +4324,6 @@ dependencies = [
"clap",
"const_format",
"crc32c",
"fail",
"fs2",
"futures",
"git-version",
@@ -4475,7 +4347,6 @@ dependencies = [
"serde",
"serde_json",
"serde_with",
"sha2",
"signal-hook",
"storage_broker",
"thiserror",
@@ -4528,12 +4399,12 @@ checksum = "d29ab0c6d3fc0ee92fe66e2d99f700eab17a8d57d1c1d3b748380fb20baa78cd"
[[package]]
name = "sct"
version = "0.7.1"
version = "0.7.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "da046153aa2352493d6cb7da4b6e5c0c057d8a1d0a9aa8560baffdd945acd414"
checksum = "d53dcdb7c9f8158937a7981b48accfd39a43af418591a5d008c7b22b5e1b7ca4"
dependencies = [
"ring 0.17.6",
"untrusted 0.9.0",
"ring 0.16.20",
"untrusted 0.7.1",
]
[[package]]
@@ -4709,7 +4580,7 @@ checksum = "aafe972d60b0b9bee71a91b92fee2d4fb3c9d7e8f6b179aa99f27203d99a4816"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -4790,7 +4661,7 @@ dependencies = [
"darling",
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -5057,9 +4928,9 @@ dependencies = [
[[package]]
name = "syn"
version = "2.0.32"
version = "2.0.28"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "239814284fd6f1a4ffe4ca893952cdd93c224b6a1571c9a9eadd670295c0c9e2"
checksum = "04361975b3f5e348b2189d8dc55bc942f278b2d482a6a0365de5bdd62d351567"
dependencies = [
"proc-macro2",
"quote",
@@ -5189,7 +5060,7 @@ checksum = "f9456a42c5b0d803c8cd86e73dd7cc9edd429499f37a3550d286d5e86720569f"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -5232,15 +5103,6 @@ dependencies = [
"time-core",
]
[[package]]
name = "tiny-keccak"
version = "2.0.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "2c9d3793400a45f954c52e73d068316d76b6f4e36977e3fcebb13a2721e80237"
dependencies = [
"crunchy",
]
[[package]]
name = "tinytemplate"
version = "1.2.1"
@@ -5316,7 +5178,7 @@ checksum = "5b8a1e28f2deaa14e508979454cb3a223b10b938b45af148bc0986de36f1923b"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -5617,7 +5479,7 @@ checksum = "0f57e3ca2a01450b1a921183a9c9cbfda207fd822cef4ccb00a65402cbba7a74"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -5884,7 +5746,6 @@ dependencies = [
"chrono",
"const_format",
"criterion",
"fail",
"futures",
"heapless",
"hex",
@@ -6063,7 +5924,7 @@ dependencies = [
"once_cell",
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
"wasm-bindgen-shared",
]
@@ -6097,7 +5958,7 @@ checksum = "e128beba882dd1eb6200e1dc92ae6c5dbaa4311aa7bb211ca035779e5efc39f8"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
"wasm-bindgen-backend",
"wasm-bindgen-shared",
]
@@ -6409,7 +6270,6 @@ dependencies = [
"futures-io",
"futures-sink",
"futures-util",
"getrandom 0.2.11",
"hex",
"hmac",
"hyper",
@@ -6421,7 +6281,6 @@ dependencies = [
"num-bigint",
"num-integer",
"num-traits",
"once_cell",
"prost",
"rand 0.8.5",
"regex",
@@ -6436,7 +6295,7 @@ dependencies = [
"smallvec",
"subtle",
"syn 1.0.109",
"syn 2.0.32",
"syn 2.0.28",
"time",
"time-macros",
"tokio",
@@ -6498,22 +6357,22 @@ dependencies = [
[[package]]
name = "zerocopy"
version = "0.7.31"
version = "0.7.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "1c4061bedbb353041c12f413700357bec76df2c7e2ca8e4df8bac24c6bf68e3d"
checksum = "7a7af71d8643341260a65f89fa60c0eeaa907f34544d8f6d9b0df72f069b5e74"
dependencies = [
"zerocopy-derive",
]
[[package]]
name = "zerocopy-derive"
version = "0.7.31"
version = "0.7.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "b3c129550b3e6de3fd0ba67ba5c81818f9805e58b8d7fee80a3a59d2c9fc601a"
checksum = "9731702e2f0617ad526794ae28fbc6f6ca8849b5ba729666c2a5bc4b6ddee2cd"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.32",
"syn 2.0.28",
]
[[package]]
@@ -6524,28 +6383,30 @@ checksum = "2a0956f1ba7c7909bfb66c2e9e4124ab6f6482560f6628b5aaeba39207c9aad9"
[[package]]
name = "zstd"
version = "0.13.0"
version = "0.12.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "bffb3309596d527cfcba7dfc6ed6052f1d39dfbd7c867aa2e865e4a449c10110"
checksum = "1a27595e173641171fc74a1232b7b1c7a7cb6e18222c11e9dfb9888fa424c53c"
dependencies = [
"zstd-safe",
]
[[package]]
name = "zstd-safe"
version = "7.0.0"
version = "6.0.6"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "43747c7422e2924c11144d5229878b98180ef8b06cca4ab5af37afc8a8d8ea3e"
checksum = "ee98ffd0b48ee95e6c5168188e44a54550b1564d9d530ee21d5f0eaed1069581"
dependencies = [
"libc",
"zstd-sys",
]
[[package]]
name = "zstd-sys"
version = "2.0.9+zstd.1.5.5"
version = "2.0.8+zstd.1.5.5"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9e16efa8a874a0481a574084d34cc26fdb3b99627480f785888deb6386506656"
checksum = "5556e6ee25d32df2586c098bbfa278803692a20d0ab9565e049480d52707ec8c"
dependencies = [
"cc",
"libc",
"pkg-config",
]

View File

@@ -5,8 +5,6 @@ members = [
"control_plane",
"pageserver",
"pageserver/ctl",
"pageserver/client",
"pageserver/pagebench",
"proxy",
"safekeeper",
"storage_broker",
@@ -80,7 +78,6 @@ futures-util = "0.3"
git-version = "0.3"
hashbrown = "0.13"
hashlink = "0.8.1"
hdrhistogram = "7.5.2"
hex = "0.4"
hex-literal = "0.4"
hmac = "0.12.1"
@@ -93,7 +90,7 @@ hyper-tungstenite = "0.11"
inotify = "0.10.2"
ipnet = "2.9.0"
itertools = "0.10"
jsonwebtoken = "9"
jsonwebtoken = "8"
libc = "0.2"
md5 = "0.7.0"
memoffset = "0.8"
@@ -185,7 +182,6 @@ compute_api = { version = "0.1", path = "./libs/compute_api/" }
consumption_metrics = { version = "0.1", path = "./libs/consumption_metrics/" }
metrics = { version = "0.1", path = "./libs/metrics/" }
pageserver_api = { version = "0.1", path = "./libs/pageserver_api/" }
pageserver_client = { path = "./pageserver/client" }
postgres_backend = { version = "0.1", path = "./libs/postgres_backend/" }
postgres_connection = { version = "0.1", path = "./libs/postgres_connection/" }
postgres_ffi = { version = "0.1", path = "./libs/postgres_ffi/" }

View File

@@ -3,7 +3,7 @@
### By default, the binaries inside the image have some mock parameters and can start, but are not intended to be used
### inside this image in the real deployments.
ARG REPOSITORY=neondatabase
ARG IMAGE=build-tools
ARG IMAGE=rust
ARG TAG=pinned
# Build Postgres

View File

@@ -1,166 +0,0 @@
FROM debian:bullseye-slim
# Add nonroot user
RUN useradd -ms /bin/bash nonroot -b /home
SHELL ["/bin/bash", "-c"]
# System deps
RUN set -e \
&& apt update \
&& apt install -y \
autoconf \
automake \
bison \
build-essential \
ca-certificates \
cmake \
curl \
flex \
git \
gnupg \
gzip \
jq \
libcurl4-openssl-dev \
libbz2-dev \
libffi-dev \
liblzma-dev \
libncurses5-dev \
libncursesw5-dev \
libpq-dev \
libreadline-dev \
libseccomp-dev \
libsqlite3-dev \
libssl-dev \
libstdc++-10-dev \
libtool \
libxml2-dev \
libxmlsec1-dev \
libxxhash-dev \
lsof \
make \
netcat \
net-tools \
openssh-client \
parallel \
pkg-config \
unzip \
wget \
xz-utils \
zlib1g-dev \
zstd \
&& rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
# protobuf-compiler (protoc)
ENV PROTOC_VERSION 25.1
RUN curl -fsSL "https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOC_VERSION}/protoc-${PROTOC_VERSION}-linux-$(uname -m | sed 's/aarch64/aarch_64/g').zip" -o "protoc.zip" \
&& unzip -q protoc.zip -d protoc \
&& mv protoc/bin/protoc /usr/local/bin/protoc \
&& mv protoc/include/google /usr/local/include/google \
&& rm -rf protoc.zip protoc
# LLVM
ENV LLVM_VERSION=17
RUN curl -fsSL 'https://apt.llvm.org/llvm-snapshot.gpg.key' | apt-key add - \
&& echo "deb http://apt.llvm.org/bullseye/ llvm-toolchain-bullseye-${LLVM_VERSION} main" > /etc/apt/sources.list.d/llvm.stable.list \
&& apt update \
&& apt install -y clang-${LLVM_VERSION} llvm-${LLVM_VERSION} \
&& bash -c 'for f in /usr/bin/clang*-${LLVM_VERSION} /usr/bin/llvm*-${LLVM_VERSION}; do ln -s "${f}" "${f%-${LLVM_VERSION}}"; done' \
&& rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
# PostgreSQL 14
RUN curl -fsSL 'https://www.postgresql.org/media/keys/ACCC4CF8.asc' | apt-key add - \
&& echo 'deb http://apt.postgresql.org/pub/repos/apt bullseye-pgdg main' > /etc/apt/sources.list.d/pgdg.list \
&& apt update \
&& apt install -y postgresql-client-14 \
&& rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
# AWS CLI
RUN curl "https://awscli.amazonaws.com/awscli-exe-linux-$(uname -m).zip" -o "awscliv2.zip" \
&& unzip -q awscliv2.zip \
&& ./aws/install \
&& rm awscliv2.zip
# Mold: A Modern Linker
ENV MOLD_VERSION v2.4.0
RUN set -e \
&& git clone https://github.com/rui314/mold.git \
&& mkdir mold/build \
&& cd mold/build \
&& git checkout ${MOLD_VERSION} \
&& cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_COMPILER=clang++ .. \
&& cmake --build . -j $(nproc) \
&& cmake --install . \
&& cd .. \
&& rm -rf mold
# LCOV
# Build lcov from a fork:
# It includes several bug fixes on top on v2.0 release (https://github.com/linux-test-project/lcov/compare/v2.0...master)
# And patches from us:
# - Generates json file with code coverage summary (https://github.com/neondatabase/lcov/commit/426e7e7a22f669da54278e9b55e6d8caabd00af0.tar.gz)
RUN for package in Capture::Tiny DateTime Devel::Cover Digest::MD5 File::Spec JSON::XS Memory::Process Time::HiRes JSON; do yes | perl -MCPAN -e "CPAN::Shell->notest('install', '$package')"; done \
&& wget https://github.com/neondatabase/lcov/archive/426e7e7a22f669da54278e9b55e6d8caabd00af0.tar.gz -O lcov.tar.gz \
&& echo "61a22a62e20908b8b9e27d890bd0ea31f567a7b9668065589266371dcbca0992 lcov.tar.gz" | sha256sum --check \
&& mkdir -p lcov && tar -xzf lcov.tar.gz -C lcov --strip-components=1 \
&& cd lcov \
&& make install \
&& rm -rf ../lcov.tar.gz
# Switch to nonroot user
USER nonroot:nonroot
WORKDIR /home/nonroot
# Python
ENV PYTHON_VERSION=3.9.2 \
PYENV_ROOT=/home/nonroot/.pyenv \
PATH=/home/nonroot/.pyenv/shims:/home/nonroot/.pyenv/bin:/home/nonroot/.poetry/bin:$PATH
RUN set -e \
&& cd $HOME \
&& curl -sSO https://raw.githubusercontent.com/pyenv/pyenv-installer/master/bin/pyenv-installer \
&& chmod +x pyenv-installer \
&& ./pyenv-installer \
&& export PYENV_ROOT=/home/nonroot/.pyenv \
&& export PATH="$PYENV_ROOT/bin:$PATH" \
&& export PATH="$PYENV_ROOT/shims:$PATH" \
&& pyenv install ${PYTHON_VERSION} \
&& pyenv global ${PYTHON_VERSION} \
&& python --version \
&& pip install --upgrade pip \
&& pip --version \
&& pip install pipenv wheel poetry
# Switch to nonroot user (again)
USER nonroot:nonroot
WORKDIR /home/nonroot
# Rust
# Please keep the version of llvm (installed above) in sync with rust llvm (`rustc --version --verbose | grep LLVM`)
ENV RUSTC_VERSION=1.74.0
ENV RUSTUP_HOME="/home/nonroot/.rustup"
ENV PATH="/home/nonroot/.cargo/bin:${PATH}"
RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && whoami && \
chmod +x rustup-init && \
./rustup-init -y --default-toolchain ${RUSTC_VERSION} && \
rm rustup-init && \
export PATH="$HOME/.cargo/bin:$PATH" && \
. "$HOME/.cargo/env" && \
cargo --version && rustup --version && \
rustup component add llvm-tools-preview rustfmt clippy && \
cargo install --git https://github.com/paritytech/cachepot && \
cargo install rustfilt && \
cargo install cargo-hakari && \
cargo install cargo-deny && \
cargo install cargo-hack && \
cargo install cargo-nextest && \
rm -rf /home/nonroot/.cargo/registry && \
rm -rf /home/nonroot/.cargo/git
ENV RUSTC_WRAPPER=cachepot
# Show versions
RUN whoami \
&& python --version \
&& pip --version \
&& cargo --version --verbose \
&& rustup --version --verbose \
&& rustc --version --verbose \
&& clang --version

View File

@@ -1,6 +1,6 @@
ARG PG_VERSION
ARG REPOSITORY=neondatabase
ARG IMAGE=build-tools
ARG IMAGE=rust
ARG TAG=pinned
ARG BUILD_TAG
@@ -48,29 +48,7 @@ RUN cd postgres && \
echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgrowlocks.control && \
echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgstattuple.control && \
echo 'trusted = true' >> /usr/local/pgsql/share/extension/refint.control && \
echo 'trusted = true' >> /usr/local/pgsql/share/extension/xml2.control && \
# We need to grant EXECUTE on pg_stat_statements_reset() to neon_superuser.
# In vanilla postgres this function is limited to Postgres role superuser.
# In neon we have neon_superuser role that is not a superuser but replaces superuser in some cases.
# We could add the additional grant statements to the postgres repository but it would be hard to maintain,
# whenever we need to pick up a new postgres version and we want to limit the changes in our postgres fork,
# so we do it here.
old_list="pg_stat_statements--1.0--1.1.sql pg_stat_statements--1.1--1.2.sql pg_stat_statements--1.2--1.3.sql pg_stat_statements--1.3--1.4.sql pg_stat_statements--1.4--1.5.sql pg_stat_statements--1.4.sql pg_stat_statements--1.5--1.6.sql"; \
# the first loop is for pg_stat_statement extension version <= 1.6
for file in /usr/local/pgsql/share/extension/pg_stat_statements--*.sql; do \
filename=$(basename "$file"); \
if echo "$old_list" | grep -q -F "$filename"; then \
echo 'GRANT EXECUTE ON FUNCTION pg_stat_statements_reset() TO neon_superuser;' >> $file; \
fi; \
done; \
# the second loop is for pg_stat_statement extension versions >= 1.7,
# where pg_stat_statement_reset() got 3 additional arguments
for file in /usr/local/pgsql/share/extension/pg_stat_statements--*.sql; do \
filename=$(basename "$file"); \
if ! echo "$old_list" | grep -q -F "$filename"; then \
echo 'GRANT EXECUTE ON FUNCTION pg_stat_statements_reset(Oid, Oid, bigint) TO neon_superuser;' >> $file; \
fi; \
done
echo 'trusted = true' >> /usr/local/pgsql/share/extension/xml2.control
#########################################################################################
#
@@ -591,23 +569,6 @@ RUN wget https://github.com/ChenHuajun/pg_roaringbitmap/archive/refs/tags/v0.5.4
make -j $(getconf _NPROCESSORS_ONLN) install && \
echo 'trusted = true' >> /usr/local/pgsql/share/extension/roaringbitmap.control
#########################################################################################
#
# Layer "pg-semver-pg-build"
# compile pg_semver extension
#
#########################################################################################
FROM build-deps AS pg-semver-pg-build
COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
ENV PATH "/usr/local/pgsql/bin/:$PATH"
RUN wget https://github.com/theory/pg-semver/archive/refs/tags/v0.32.1.tar.gz -O pg_semver.tar.gz && \
echo "fbdaf7512026d62eec03fad8687c15ed509b6ba395bff140acd63d2e4fbe25d7 pg_semver.tar.gz" | sha256sum --check && \
mkdir pg_semver-src && cd pg_semver-src && tar xvzf ../pg_semver.tar.gz --strip-components=1 -C . && \
make -j $(getconf _NPROCESSORS_ONLN) && \
make -j $(getconf _NPROCESSORS_ONLN) install && \
echo 'trusted = true' >> /usr/local/pgsql/share/extension/semver.control
#########################################################################################
#
# Layer "pg-embedding-pg-build"
@@ -807,7 +768,6 @@ COPY --from=pg-pgx-ulid-build /usr/local/pgsql/ /usr/local/pgsql/
COPY --from=rdkit-pg-build /usr/local/pgsql/ /usr/local/pgsql/
COPY --from=pg-uuidv7-pg-build /usr/local/pgsql/ /usr/local/pgsql/
COPY --from=pg-roaringbitmap-pg-build /usr/local/pgsql/ /usr/local/pgsql/
COPY --from=pg-semver-pg-build /usr/local/pgsql/ /usr/local/pgsql/
COPY --from=pg-embedding-pg-build /usr/local/pgsql/ /usr/local/pgsql/
COPY --from=wal2json-pg-build /usr/local/pgsql /usr/local/pgsql
COPY pgxn/ pgxn/

View File

@@ -1,7 +1,7 @@
# First transient image to build compute_tools binaries
# NB: keep in sync with rust image version in .github/workflows/build_and_test.yml
ARG REPOSITORY=neondatabase
ARG IMAGE=build-tools
ARG IMAGE=rust
ARG TAG=pinned
ARG BUILD_TAG

View File

@@ -29,14 +29,13 @@ See developer documentation in [SUMMARY.md](/docs/SUMMARY.md) for more informati
```bash
apt install build-essential libtool libreadline-dev zlib1g-dev flex bison libseccomp-dev \
libssl-dev clang pkg-config libpq-dev cmake postgresql-client protobuf-compiler \
libcurl4-openssl-dev openssl python3-poetry lsof libicu-dev
libcurl4-openssl-dev openssl python-poetry lsof libicu-dev
```
* On Fedora, these packages are needed:
```bash
dnf install flex bison readline-devel zlib-devel openssl-devel \
libseccomp-devel perl clang cmake postgresql postgresql-contrib protobuf-compiler \
protobuf-devel libcurl-devel openssl poetry lsof libicu-devel libpq-devel python3-devel \
libffi-devel
protobuf-devel libcurl-devel openssl poetry lsof libicu-devel
```
* On Arch based systems, these packages are needed:
```bash

View File

@@ -13,7 +13,6 @@ clap.workspace = true
flate2.workspace = true
futures.workspace = true
hyper = { workspace = true, features = ["full"] }
nix.workspace = true
notify.workspace = true
num_cpus.workspace = true
opentelemetry.workspace = true
@@ -21,7 +20,6 @@ postgres.workspace = true
regex.workspace = true
serde.workspace = true
serde_json.workspace = true
signal-hook.workspace = true
tar.workspace = true
reqwest = { workspace = true, features = ["json"] }
tokio = { workspace = true, features = ["rt", "rt-multi-thread"] }
@@ -39,6 +37,5 @@ workspace_hack.workspace = true
toml_edit.workspace = true
remote_storage = { version = "0.1", path = "../libs/remote_storage/" }
vm_monitor = { version = "0.1", path = "../libs/vm_monitor/" }
zstd = "0.13"
zstd = "0.12.4"
bytes = "1.0"
rust-ini = "0.20.0"

View File

@@ -31,31 +31,25 @@
//! -C 'postgresql://cloud_admin@localhost/postgres' \
//! -S /var/db/postgres/specs/current.json \
//! -b /usr/local/bin/postgres \
//! -r http://pg-ext-s3-gateway \
//! --pgbouncer-connstr 'host=localhost port=6432 dbname=pgbouncer user=cloud_admin sslmode=disable'
//! --pgbouncer-ini-path /etc/pgbouncer.ini \
//! -r http://pg-ext-s3-gateway
//! ```
//!
use std::collections::HashMap;
use std::fs::File;
use std::path::Path;
use std::process::exit;
use std::sync::atomic::Ordering;
use std::sync::{mpsc, Arc, Condvar, Mutex, RwLock};
use std::{thread, time::Duration};
use anyhow::{Context, Result};
use chrono::Utc;
use clap::Arg;
use nix::sys::signal::{kill, Signal};
use signal_hook::consts::{SIGQUIT, SIGTERM};
use signal_hook::{consts::SIGINT, iterator::Signals};
use tracing::{error, info};
use url::Url;
use compute_api::responses::ComputeStatus;
use compute_tools::compute::{ComputeNode, ComputeState, ParsedSpec, PG_PID, SYNC_SAFEKEEPERS_PID};
use compute_tools::compute::{ComputeNode, ComputeState, ParsedSpec};
use compute_tools::configurator::launch_configurator;
use compute_tools::extension_server::get_pg_version;
use compute_tools::http::api::launch_http_server;
@@ -71,13 +65,6 @@ const BUILD_TAG_DEFAULT: &str = "latest";
fn main() -> Result<()> {
init_tracing_and_logging(DEFAULT_LOG_LEVEL)?;
let mut signals = Signals::new([SIGINT, SIGTERM, SIGQUIT])?;
thread::spawn(move || {
for sig in signals.forever() {
handle_exit_signal(sig);
}
});
let build_tag = option_env!("BUILD_TAG")
.unwrap_or(BUILD_TAG_DEFAULT)
.to_string();
@@ -112,9 +99,6 @@ fn main() -> Result<()> {
let spec_json = matches.get_one::<String>("spec");
let spec_path = matches.get_one::<String>("spec-path");
let pgbouncer_connstr = matches.get_one::<String>("pgbouncer-connstr");
let pgbouncer_ini_path = matches.get_one::<String>("pgbouncer-ini-path");
// Extract OpenTelemetry context for the startup actions from the
// TRACEPARENT and TRACESTATE env variables, and attach it to the current
// tracing context.
@@ -225,8 +209,6 @@ fn main() -> Result<()> {
ext_remote_storage: ext_remote_storage.map(|s| s.to_string()),
ext_download_progress: RwLock::new(HashMap::new()),
build_tag,
pgbouncer_connstr: pgbouncer_connstr.map(|s| s.to_string()),
pgbouncer_ini_path: pgbouncer_ini_path.map(|s| s.to_string()),
};
let compute = Arc::new(compute_node);
@@ -357,7 +339,6 @@ fn main() -> Result<()> {
let ecode = pg
.wait()
.expect("failed to start waiting on Postgres process");
PG_PID.store(0, Ordering::SeqCst);
info!("Postgres exited with code {}, shutting down", ecode);
exit_code = ecode.code()
}
@@ -512,41 +493,6 @@ fn cli() -> clap::Command {
)
.value_name("FILECACHE_CONNSTR"),
)
.arg(
Arg::new("pgbouncer-connstr")
.long("pgbouncer-connstr")
.default_value(
"host=localhost port=6432 dbname=pgbouncer user=cloud_admin sslmode=disable",
)
.value_name("PGBOUNCER_CONNSTR"),
)
.arg(
Arg::new("pgbouncer-ini-path")
.long("pgbouncer-ini-path")
// Note: this doesn't match current path for pgbouncer.ini.
// Until we fix it, we need to pass the path explicitly
// or this will be effectively no-op.
.default_value("/etc/pgbouncer.ini")
.value_name("PGBOUNCER_INI_PATH"),
)
}
/// When compute_ctl is killed, send also termination signal to sync-safekeepers
/// to prevent leakage. TODO: it is better to convert compute_ctl to async and
/// wait for termination which would be easy then.
fn handle_exit_signal(sig: i32) {
info!("received {sig} termination signal");
let ss_pid = SYNC_SAFEKEEPERS_PID.load(Ordering::SeqCst);
if ss_pid != 0 {
let ss_pid = nix::unistd::Pid::from_raw(ss_pid as i32);
kill(ss_pid, Signal::SIGTERM).ok();
}
let pg_pid = PG_PID.load(Ordering::SeqCst);
if pg_pid != 0 {
let pg_pid = nix::unistd::Pid::from_raw(pg_pid as i32);
kill(pg_pid, Signal::SIGTERM).ok();
}
exit(1);
}
#[test]

View File

@@ -6,10 +6,7 @@ use std::os::unix::fs::PermissionsExt;
use std::path::Path;
use std::process::{Command, Stdio};
use std::str::FromStr;
use std::sync::atomic::AtomicU32;
use std::sync::atomic::Ordering;
use std::sync::{Condvar, Mutex, RwLock};
use std::thread;
use std::time::Instant;
use anyhow::{Context, Result};
@@ -36,9 +33,6 @@ use crate::spec::*;
use crate::sync_sk::{check_if_synced, ping_safekeeper};
use crate::{config, extension_server};
pub static SYNC_SAFEKEEPERS_PID: AtomicU32 = AtomicU32::new(0);
pub static PG_PID: AtomicU32 = AtomicU32::new(0);
/// Compute node info shared across several `compute_ctl` threads.
pub struct ComputeNode {
// Url type maintains proper escaping
@@ -70,10 +64,6 @@ pub struct ComputeNode {
// key: ext_archive_name, value: started download time, download_completed?
pub ext_download_progress: RwLock<HashMap<String, (DateTime<Utc>, bool)>>,
pub build_tag: String,
// connection string to pgbouncer to change settings
pub pgbouncer_connstr: Option<String>,
// path to pgbouncer.ini to change settings
pub pgbouncer_ini_path: Option<String>,
}
// store some metrics about download size that might impact startup time
@@ -506,7 +496,6 @@ impl ComputeNode {
.stdout(Stdio::piped())
.spawn()
.expect("postgres --sync-safekeepers failed to start");
SYNC_SAFEKEEPERS_PID.store(sync_handle.id(), Ordering::SeqCst);
// `postgres --sync-safekeepers` will print all log output to stderr and
// final LSN to stdout. So we pipe only stdout, while stderr will be automatically
@@ -514,7 +503,6 @@ impl ComputeNode {
let sync_output = sync_handle
.wait_with_output()
.expect("postgres --sync-safekeepers failed");
SYNC_SAFEKEEPERS_PID.store(0, Ordering::SeqCst);
if !sync_output.status.success() {
anyhow::bail!(
@@ -669,7 +657,6 @@ impl ComputeNode {
})
.spawn()
.expect("cannot start postgres process");
PG_PID.store(pg.id(), Ordering::SeqCst);
wait_for_postgres(&mut pg, pgdata_path)?;
@@ -750,31 +737,6 @@ impl ComputeNode {
pub fn reconfigure(&self) -> Result<()> {
let spec = self.state.lock().unwrap().pspec.clone().unwrap().spec;
if let Some(connstr) = &self.pgbouncer_connstr {
info!("tuning pgbouncer with connstr: {:?}", connstr);
let rt = tokio::runtime::Builder::new_current_thread()
.enable_all()
.build()
.expect("failed to create rt");
// Spawn a thread to do the tuning,
// so that we don't block the main thread that starts Postgres.
let pgbouncer_settings = spec.pgbouncer_settings.clone();
let connstr_clone = connstr.clone();
let pgbouncer_ini_path = self.pgbouncer_ini_path.clone();
let _handle = thread::spawn(move || {
let res = rt.block_on(tune_pgbouncer(
pgbouncer_settings,
&connstr_clone,
pgbouncer_ini_path,
));
if let Err(err) = res {
error!("error while tuning pgbouncer: {err:?}");
}
});
}
// Write new config
let pgdata_path = Path::new(&self.pgdata);
let postgresql_conf_path = pgdata_path.join("postgresql.conf");
@@ -829,32 +791,6 @@ impl ComputeNode {
pspec.timeline_id,
);
// tune pgbouncer
if let Some(connstr) = &self.pgbouncer_connstr {
info!("tuning pgbouncer with connstr: {:?}", connstr);
let rt = tokio::runtime::Builder::new_current_thread()
.enable_all()
.build()
.expect("failed to create rt");
// Spawn a thread to do the tuning,
// so that we don't block the main thread that starts Postgres.
let pgbouncer_settings = pspec.spec.pgbouncer_settings.clone();
let connstr_clone = connstr.clone();
let pgbouncer_ini_path = self.pgbouncer_ini_path.clone();
let _handle = thread::spawn(move || {
let res = rt.block_on(tune_pgbouncer(
pgbouncer_settings,
&connstr_clone,
pgbouncer_ini_path,
));
if let Err(err) = res {
error!("error while tuning pgbouncer: {err:?}");
}
});
}
info!(
"start_compute spec.remote_extensions {:?}",
pspec.spec.remote_extensions

View File

@@ -3,7 +3,7 @@ use std::{thread, time::Duration};
use chrono::{DateTime, Utc};
use postgres::{Client, NoTls};
use tracing::{debug, info, warn};
use tracing::{debug, info};
use crate::compute::ComputeNode;
@@ -84,29 +84,6 @@ fn watch_compute_activity(compute: &ComputeNode) {
}
}
// If there are existing (logical) walsenders, do not suspend.
//
// walproposer doesn't currently show up in pg_stat_replication,
// but protect if it will be
let ws_count_query = "select count(*) from pg_stat_replication where application_name != 'walproposer';";
match cli.query_one(ws_count_query, &[]) {
Ok(r) => match r.try_get::<&str, i64>("count") {
Ok(num_ws) => {
if num_ws > 0 {
last_active = Some(Utc::now());
}
}
Err(e) => {
warn!("failed to parse ws count: {:?}", e);
continue;
}
},
Err(e) => {
warn!("failed to get list of walsenders: {:?}", e);
continue;
}
}
// Update the last activity in the shared state if we got a more recent one.
let mut state = compute.state.lock().unwrap();
// NB: `Some(<DateTime>)` is always greater than `None`.

View File

@@ -9,11 +9,9 @@ use std::process::Child;
use std::time::{Duration, Instant};
use anyhow::{bail, Result};
use ini::Ini;
use notify::{RecursiveMode, Watcher};
use postgres::{Client, Transaction};
use tokio_postgres::NoTls;
use tracing::{debug, error, info, instrument};
use tracing::{debug, instrument};
use compute_api::spec::{Database, GenericOption, GenericOptions, PgIdent, Role};
@@ -361,68 +359,3 @@ pub fn create_pgdata(pgdata: &str) -> Result<()> {
Ok(())
}
/// Update pgbouncer.ini with provided options
pub fn update_pgbouncer_ini(
pgbouncer_config: HashMap<String, String>,
pgbouncer_ini_path: &str,
) -> Result<()> {
let mut conf = Ini::load_from_file(pgbouncer_ini_path)?;
let section = conf.section_mut(Some("pgbouncer")).unwrap();
for (option_name, value) in pgbouncer_config.iter() {
section.insert(option_name, value);
}
conf.write_to_file(pgbouncer_ini_path)?;
Ok(())
}
/// Tune pgbouncer.
/// 1. Apply new config using pgbouncer admin console
/// 2. Add new values to pgbouncer.ini to preserve them after restart
pub async fn tune_pgbouncer(
pgbouncer_settings: Option<HashMap<String, String>>,
pgbouncer_connstr: &str,
pgbouncer_ini_path: Option<String>,
) -> Result<()> {
if let Some(pgbouncer_config) = pgbouncer_settings {
// Apply new config
let connect_result = tokio_postgres::connect(pgbouncer_connstr, NoTls).await;
let (client, connection) = connect_result.unwrap();
tokio::spawn(async move {
if let Err(e) = connection.await {
eprintln!("connection error: {}", e);
}
});
for (option_name, value) in pgbouncer_config.iter() {
info!(
"Applying pgbouncer setting change: {} = {}",
option_name, value
);
let query = format!("SET {} = {}", option_name, value);
let result = client.simple_query(&query).await;
info!("Applying pgbouncer setting change: {}", query);
info!("pgbouncer setting change result: {:?}", result);
if let Err(err) = result {
// Don't fail on error, just print it into log
error!(
"Failed to apply pgbouncer setting change: {}, {}",
query, err
);
};
}
// save values to pgbouncer.ini
// so that they are preserved after pgbouncer restart
if let Some(pgbouncer_ini_path) = pgbouncer_ini_path {
update_pgbouncer_ini(pgbouncer_config, &pgbouncer_ini_path)?;
}
}
Ok(())
}

View File

@@ -298,7 +298,7 @@ pub fn handle_roles(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
// safe to add more permissions here. BYPASSRLS and REPLICATION are inherited
// from neon_superuser.
let mut query: String = format!(
"CREATE ROLE {} INHERIT CREATEROLE CREATEDB BYPASSRLS REPLICATION IN ROLE neon_superuser",
"CREATE ROLE {} INHERIT CREATEROLE CREATEDB IN ROLE neon_superuser",
name.pg_quote()
);
info!("role create query: '{}'", &query);
@@ -370,49 +370,33 @@ pub fn handle_role_deletions(spec: &ComputeSpec, connstr: &str, client: &mut Cli
Ok(())
}
fn reassign_owned_objects_in_one_db(
conf: Config,
role_name: &PgIdent,
db_owner: &PgIdent,
) -> Result<()> {
let mut client = conf.connect(NoTls)?;
// This will reassign all dependent objects to the db owner
let reassign_query = format!(
"REASSIGN OWNED BY {} TO {}",
role_name.pg_quote(),
db_owner.pg_quote()
);
info!(
"reassigning objects owned by '{}' in db '{}' to '{}'",
role_name,
conf.get_dbname().unwrap_or(""),
db_owner
);
client.simple_query(&reassign_query)?;
// This now will only drop privileges of the role
let drop_query = format!("DROP OWNED BY {}", role_name.pg_quote());
client.simple_query(&drop_query)?;
Ok(())
}
// Reassign all owned objects in all databases to the owner of the database.
fn reassign_owned_objects(spec: &ComputeSpec, connstr: &str, role_name: &PgIdent) -> Result<()> {
for db in &spec.cluster.databases {
if db.owner != *role_name {
let mut conf = Config::from_str(connstr)?;
conf.dbname(&db.name);
reassign_owned_objects_in_one_db(conf, role_name, &db.owner)?;
let mut client = conf.connect(NoTls)?;
// This will reassign all dependent objects to the db owner
let reassign_query = format!(
"REASSIGN OWNED BY {} TO {}",
role_name.pg_quote(),
db.owner.pg_quote()
);
info!(
"reassigning objects owned by '{}' in db '{}' to '{}'",
role_name, &db.name, &db.owner
);
client.simple_query(&reassign_query)?;
// This now will only drop privileges of the role
let drop_query = format!("DROP OWNED BY {}", role_name.pg_quote());
client.simple_query(&drop_query)?;
}
}
// Also handle case when there are no databases in the spec.
// In this case we need to reassign objects in the default database.
let conf = Config::from_str(connstr)?;
let db_owner = PgIdent::from_str("cloud_admin")?;
reassign_owned_objects_in_one_db(conf, role_name, &db_owner)?;
Ok(())
}

View File

@@ -6,11 +6,9 @@ license.workspace = true
[dependencies]
anyhow.workspace = true
async-trait.workspace = true
camino.workspace = true
clap.workspace = true
comfy-table.workspace = true
futures.workspace = true
git-version.workspace = true
nix.workspace = true
once_cell.workspace = true
@@ -26,11 +24,10 @@ tar.workspace = true
thiserror.workspace = true
toml.workspace = true
tokio.workspace = true
tokio-postgres.workspace = true
tokio-util.workspace = true
url.workspace = true
# Note: Do not directly depend on pageserver or safekeeper; use pageserver_api or safekeeper_api
# instead, so that recompile times are better.
pageserver_api.workspace = true
pageserver_client.workspace = true
postgres_backend.workspace = true
safekeeper_api.workspace = true
postgres_connection.workspace = true

View File

@@ -9,7 +9,7 @@ pub struct AttachmentService {
env: LocalEnv,
listen: String,
path: PathBuf,
client: reqwest::Client,
client: reqwest::blocking::Client,
}
const COMMAND: &str = "attachment_service";
@@ -53,7 +53,7 @@ impl AttachmentService {
env: env.clone(),
path,
listen,
client: reqwest::ClientBuilder::new()
client: reqwest::blocking::ClientBuilder::new()
.build()
.expect("Failed to construct http client"),
}
@@ -64,7 +64,7 @@ impl AttachmentService {
.expect("non-Unicode path")
}
pub async fn start(&self) -> anyhow::Result<Child> {
pub fn start(&self) -> anyhow::Result<Child> {
let path_str = self.path.to_string_lossy();
background_process::start_process(
@@ -73,11 +73,10 @@ impl AttachmentService {
&self.env.attachment_service_bin(),
["-l", &self.listen, "-p", &path_str],
[],
background_process::InitialPidFile::Create(self.pid_file()),
background_process::InitialPidFile::Create(&self.pid_file()),
// TODO: a real status check
|| async move { anyhow::Ok(true) },
|| Ok(true),
)
.await
}
pub fn stop(&self, immediate: bool) -> anyhow::Result<()> {
@@ -85,7 +84,7 @@ impl AttachmentService {
}
/// Call into the attach_hook API, for use before handing out attachments to pageservers
pub async fn attach_hook(
pub fn attach_hook(
&self,
tenant_id: TenantId,
pageserver_id: NodeId,
@@ -105,16 +104,16 @@ impl AttachmentService {
node_id: Some(pageserver_id),
};
let response = self.client.post(url).json(&request).send().await?;
let response = self.client.post(url).json(&request).send()?;
if response.status() != StatusCode::OK {
return Err(anyhow!("Unexpected status {}", response.status()));
}
let response = response.json::<AttachHookResponse>().await?;
let response = response.json::<AttachHookResponse>()?;
Ok(response.gen)
}
pub async fn inspect(&self, tenant_id: TenantId) -> anyhow::Result<Option<(u32, NodeId)>> {
pub fn inspect(&self, tenant_id: TenantId) -> anyhow::Result<Option<(u32, NodeId)>> {
use hyper::StatusCode;
let url = self
@@ -127,12 +126,12 @@ impl AttachmentService {
let request = InspectRequest { tenant_id };
let response = self.client.post(url).json(&request).send().await?;
let response = self.client.post(url).json(&request).send()?;
if response.status() != StatusCode::OK {
return Err(anyhow!("Unexpected status {}", response.status()));
}
let response = response.json::<InspectResponse>().await?;
let response = response.json::<InspectResponse>()?;
Ok(response.attachment)
}
}

View File

@@ -44,15 +44,15 @@ const NOTICE_AFTER_RETRIES: u64 = 50;
/// Argument to `start_process`, to indicate whether it should create pidfile or if the process creates
/// it itself.
pub enum InitialPidFile {
pub enum InitialPidFile<'t> {
/// Create a pidfile, to allow future CLI invocations to manipulate the process.
Create(Utf8PathBuf),
Create(&'t Utf8Path),
/// The process will create the pidfile itself, need to wait for that event.
Expect(Utf8PathBuf),
Expect(&'t Utf8Path),
}
/// Start a background child process using the parameters given.
pub async fn start_process<F, Fut, AI, A, EI>(
pub fn start_process<F, AI, A, EI>(
process_name: &str,
datadir: &Path,
command: &Path,
@@ -62,8 +62,7 @@ pub async fn start_process<F, Fut, AI, A, EI>(
process_status_check: F,
) -> anyhow::Result<Child>
where
F: Fn() -> Fut,
Fut: std::future::Future<Output = anyhow::Result<bool>>,
F: Fn() -> anyhow::Result<bool>,
AI: IntoIterator<Item = A>,
A: AsRef<OsStr>,
// Not generic AsRef<OsStr>, otherwise empty `envs` prevents type inference
@@ -90,7 +89,7 @@ where
let filled_cmd = fill_remote_storage_secrets_vars(fill_rust_env_vars(background_command));
filled_cmd.envs(envs);
let pid_file_to_check = match &initial_pid_file {
let pid_file_to_check = match initial_pid_file {
InitialPidFile::Create(path) => {
pre_exec_create_pidfile(filled_cmd, path);
path
@@ -108,7 +107,7 @@ where
);
for retries in 0..RETRIES {
match process_started(pid, pid_file_to_check, &process_status_check).await {
match process_started(pid, Some(pid_file_to_check), &process_status_check) {
Ok(true) => {
println!("\n{process_name} started, pid: {pid}");
return Ok(spawned_process);
@@ -317,20 +316,22 @@ where
cmd
}
async fn process_started<F, Fut>(
fn process_started<F>(
pid: Pid,
pid_file_to_check: &Utf8Path,
pid_file_to_check: Option<&Utf8Path>,
status_check: &F,
) -> anyhow::Result<bool>
where
F: Fn() -> Fut,
Fut: std::future::Future<Output = anyhow::Result<bool>>,
F: Fn() -> anyhow::Result<bool>,
{
match status_check().await {
Ok(true) => match pid_file::read(pid_file_to_check)? {
PidFileRead::NotExist => Ok(false),
PidFileRead::LockedByOtherProcess(pid_in_file) => Ok(pid_in_file == pid),
PidFileRead::NotHeldByAnyProcess(_) => Ok(false),
match status_check() {
Ok(true) => match pid_file_to_check {
Some(pid_file_path) => match pid_file::read(pid_file_path)? {
PidFileRead::NotExist => Ok(false),
PidFileRead::LockedByOtherProcess(pid_in_file) => Ok(pid_in_file == pid),
PidFileRead::NotHeldByAnyProcess(_) => Ok(false),
},
None => Ok(true),
},
Ok(false) => Ok(false),
Err(e) => anyhow::bail!("process failed to start: {e}"),

View File

@@ -120,20 +120,15 @@ fn main() -> Result<()> {
let mut env = LocalEnv::load_config().context("Error loading config")?;
let original_env = env.clone();
let rt = tokio::runtime::Builder::new_current_thread()
.enable_all()
.build()
.unwrap();
let subcommand_result = match sub_name {
"tenant" => rt.block_on(handle_tenant(sub_args, &mut env)),
"timeline" => rt.block_on(handle_timeline(sub_args, &mut env)),
"start" => rt.block_on(handle_start_all(sub_args, &env)),
"tenant" => handle_tenant(sub_args, &mut env),
"timeline" => handle_timeline(sub_args, &mut env),
"start" => handle_start_all(sub_args, &env),
"stop" => 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)),
"safekeeper" => rt.block_on(handle_safekeeper(sub_args, &env)),
"endpoint" => rt.block_on(handle_endpoint(sub_args, &env)),
"pageserver" => handle_pageserver(sub_args, &env),
"attachment_service" => handle_attachment_service(sub_args, &env),
"safekeeper" => handle_safekeeper(sub_args, &env),
"endpoint" => handle_endpoint(sub_args, &env),
"mappings" => handle_mappings(sub_args, &mut env),
"pg" => bail!("'pg' subcommand has been renamed to 'endpoint'"),
_ => bail!("unexpected subcommand {sub_name}"),
@@ -274,13 +269,12 @@ fn print_timeline(
/// Returns a map of timeline IDs to timeline_id@lsn strings.
/// Connects to the pageserver to query this information.
async fn get_timeline_infos(
fn get_timeline_infos(
env: &local_env::LocalEnv,
tenant_id: &TenantId,
) -> Result<HashMap<TimelineId, TimelineInfo>> {
Ok(get_default_pageserver(env)
.timeline_list(tenant_id)
.await?
.timeline_list(tenant_id)?
.into_iter()
.map(|timeline_info| (timeline_info.timeline_id, timeline_info))
.collect())
@@ -379,14 +373,11 @@ fn pageserver_config_overrides(init_match: &ArgMatches) -> Vec<&str> {
.collect()
}
async fn handle_tenant(
tenant_match: &ArgMatches,
env: &mut local_env::LocalEnv,
) -> anyhow::Result<()> {
fn handle_tenant(tenant_match: &ArgMatches, env: &mut local_env::LocalEnv) -> anyhow::Result<()> {
let pageserver = get_default_pageserver(env);
match tenant_match.subcommand() {
Some(("list", _)) => {
for t in pageserver.tenant_list().await? {
for t in pageserver.tenant_list()? {
println!("{} {:?}", t.id, t.state);
}
}
@@ -403,16 +394,12 @@ async fn handle_tenant(
// We must register the tenant with the attachment service, so
// that when the pageserver restarts, it will be re-attached.
let attachment_service = AttachmentService::from_env(env);
attachment_service
.attach_hook(tenant_id, pageserver.conf.id)
.await?
attachment_service.attach_hook(tenant_id, pageserver.conf.id)?
} else {
None
};
pageserver
.tenant_create(tenant_id, generation, tenant_conf)
.await?;
pageserver.tenant_create(tenant_id, generation, tenant_conf)?;
println!("tenant {tenant_id} successfully created on the pageserver");
// Create an initial timeline for the new tenant
@@ -422,16 +409,14 @@ async fn handle_tenant(
.copied()
.context("Failed to parse postgres version from the argument string")?;
let timeline_info = pageserver
.timeline_create(
tenant_id,
new_timeline_id,
None,
None,
Some(pg_version),
None,
)
.await?;
let timeline_info = pageserver.timeline_create(
tenant_id,
new_timeline_id,
None,
None,
Some(pg_version),
None,
)?;
let new_timeline_id = timeline_info.timeline_id;
let last_record_lsn = timeline_info.last_record_lsn;
@@ -465,7 +450,6 @@ async fn handle_tenant(
pageserver
.tenant_config(tenant_id, tenant_conf)
.await
.with_context(|| format!("Tenant config failed for tenant with id {tenant_id}"))?;
println!("tenant {tenant_id} successfully configured on the pageserver");
}
@@ -474,7 +458,7 @@ async fn handle_tenant(
let new_pageserver = get_pageserver(env, matches)?;
let new_pageserver_id = new_pageserver.conf.id;
migrate_tenant(env, tenant_id, new_pageserver).await?;
migrate_tenant(env, tenant_id, new_pageserver)?;
println!("tenant {tenant_id} migrated to {}", new_pageserver_id);
}
@@ -484,13 +468,13 @@ async fn handle_tenant(
Ok(())
}
async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::LocalEnv) -> Result<()> {
fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::LocalEnv) -> Result<()> {
let pageserver = get_default_pageserver(env);
match timeline_match.subcommand() {
Some(("list", list_match)) => {
let tenant_id = get_tenant_id(list_match, env)?;
let timelines = pageserver.timeline_list(&tenant_id).await?;
let timelines = pageserver.timeline_list(&tenant_id)?;
print_timelines_tree(timelines, env.timeline_name_mappings())?;
}
Some(("create", create_match)) => {
@@ -506,16 +490,14 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
let new_timeline_id_opt = parse_timeline_id(create_match)?;
let timeline_info = pageserver
.timeline_create(
tenant_id,
new_timeline_id_opt,
None,
None,
Some(pg_version),
None,
)
.await?;
let timeline_info = pageserver.timeline_create(
tenant_id,
new_timeline_id_opt,
None,
None,
Some(pg_version),
None,
)?;
let new_timeline_id = timeline_info.timeline_id;
let last_record_lsn = timeline_info.last_record_lsn;
@@ -560,9 +542,7 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
let mut cplane = ComputeControlPlane::load(env.clone())?;
println!("Importing timeline into pageserver ...");
pageserver
.timeline_import(tenant_id, timeline_id, base, pg_wal, pg_version)
.await?;
pageserver.timeline_import(tenant_id, timeline_id, base, pg_wal, pg_version)?;
env.register_branch_mapping(name.to_string(), tenant_id, timeline_id)?;
println!("Creating endpoint for imported timeline ...");
@@ -598,16 +578,14 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
.map(|lsn_str| Lsn::from_str(lsn_str))
.transpose()
.context("Failed to parse ancestor start Lsn from the request")?;
let timeline_info = pageserver
.timeline_create(
tenant_id,
None,
start_lsn,
Some(ancestor_timeline_id),
None,
None,
)
.await?;
let timeline_info = pageserver.timeline_create(
tenant_id,
None,
start_lsn,
Some(ancestor_timeline_id),
None,
None,
)?;
let new_timeline_id = timeline_info.timeline_id;
let last_record_lsn = timeline_info.last_record_lsn;
@@ -626,7 +604,7 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
Ok(())
}
async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
let (sub_name, sub_args) = match ep_match.subcommand() {
Some(ep_subcommand_data) => ep_subcommand_data,
None => bail!("no endpoint subcommand provided"),
@@ -636,12 +614,10 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
match sub_name {
"list" => {
let tenant_id = get_tenant_id(sub_args, env)?;
let timeline_infos = get_timeline_infos(env, &tenant_id)
.await
.unwrap_or_else(|e| {
eprintln!("Failed to load timeline info: {}", e);
HashMap::new()
});
let timeline_infos = get_timeline_infos(env, &tenant_id).unwrap_or_else(|e| {
eprintln!("Failed to load timeline info: {}", e);
HashMap::new()
});
let timeline_name_mappings = env.timeline_name_mappings();
@@ -815,9 +791,7 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
};
println!("Starting existing endpoint {endpoint_id}...");
endpoint
.start(&auth_token, safekeepers, remote_ext_config)
.await?;
endpoint.start(&auth_token, safekeepers, remote_ext_config)?;
}
"reconfigure" => {
let endpoint_id = sub_args
@@ -835,7 +809,7 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
} else {
None
};
endpoint.reconfigure(pageserver_id).await?;
endpoint.reconfigure(pageserver_id)?;
}
"stop" => {
let endpoint_id = sub_args
@@ -901,12 +875,11 @@ fn get_pageserver(env: &local_env::LocalEnv, args: &ArgMatches) -> Result<PageSe
))
}
async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
match sub_match.subcommand() {
Some(("start", subcommand_args)) => {
if let Err(e) = get_pageserver(env, subcommand_args)?
.start(&pageserver_config_overrides(subcommand_args))
.await
{
eprintln!("pageserver start failed: {e}");
exit(1);
@@ -933,10 +906,7 @@ async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
exit(1);
}
if let Err(e) = pageserver
.start(&pageserver_config_overrides(subcommand_args))
.await
{
if let Err(e) = pageserver.start(&pageserver_config_overrides(subcommand_args)) {
eprintln!("pageserver start failed: {e}");
exit(1);
}
@@ -950,17 +920,14 @@ async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
exit(1);
}
if let Err(e) = pageserver
.start(&pageserver_config_overrides(subcommand_args))
.await
{
if let Err(e) = pageserver.start(&pageserver_config_overrides(subcommand_args)) {
eprintln!("pageserver start failed: {e}");
exit(1);
}
}
Some(("status", subcommand_args)) => {
match get_pageserver(env, subcommand_args)?.check_status().await {
match get_pageserver(env, subcommand_args)?.check_status() {
Ok(_) => println!("Page server is up and running"),
Err(err) => {
eprintln!("Page server is not available: {}", err);
@@ -975,14 +942,11 @@ async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
Ok(())
}
async fn handle_attachment_service(
sub_match: &ArgMatches,
env: &local_env::LocalEnv,
) -> Result<()> {
fn handle_attachment_service(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
let svc = AttachmentService::from_env(env);
match sub_match.subcommand() {
Some(("start", _start_match)) => {
if let Err(e) = svc.start().await {
if let Err(e) = svc.start() {
eprintln!("start failed: {e}");
exit(1);
}
@@ -1023,7 +987,7 @@ fn safekeeper_extra_opts(init_match: &ArgMatches) -> Vec<String> {
.collect()
}
async fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
let (sub_name, sub_args) = match sub_match.subcommand() {
Some(safekeeper_command_data) => safekeeper_command_data,
None => bail!("no safekeeper subcommand provided"),
@@ -1041,7 +1005,7 @@ async fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
"start" => {
let extra_opts = safekeeper_extra_opts(sub_args);
if let Err(e) = safekeeper.start(extra_opts).await {
if let Err(e) = safekeeper.start(extra_opts) {
eprintln!("safekeeper start failed: {}", e);
exit(1);
}
@@ -1067,7 +1031,7 @@ async fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
}
let extra_opts = safekeeper_extra_opts(sub_args);
if let Err(e) = safekeeper.start(extra_opts).await {
if let Err(e) = safekeeper.start(extra_opts) {
eprintln!("safekeeper start failed: {}", e);
exit(1);
}
@@ -1080,15 +1044,15 @@ async fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
Ok(())
}
async fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> anyhow::Result<()> {
fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> anyhow::Result<()> {
// Endpoints are not started automatically
broker::start_broker_process(env).await?;
broker::start_broker_process(env)?;
// Only start the attachment service 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 {
if let Err(e) = attachment_service.start() {
eprintln!("attachment_service start failed: {:#}", e);
try_stop_all(env, true);
exit(1);
@@ -1097,10 +1061,7 @@ 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))
.await
{
if let Err(e) = pageserver.start(&pageserver_config_overrides(sub_match)) {
eprintln!("pageserver {} start failed: {:#}", ps_conf.id, e);
try_stop_all(env, true);
exit(1);
@@ -1109,7 +1070,7 @@ async fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
for node in env.safekeepers.iter() {
let safekeeper = SafekeeperNode::from_env(env, node);
if let Err(e) = safekeeper.start(vec![]).await {
if let Err(e) = safekeeper.start(vec![]) {
eprintln!("safekeeper {} start failed: {:#}", safekeeper.id, e);
try_stop_all(env, false);
exit(1);

View File

@@ -11,7 +11,7 @@ use camino::Utf8PathBuf;
use crate::{background_process, local_env};
pub async fn start_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<()> {
pub fn start_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<()> {
let broker = &env.broker;
let listen_addr = &broker.listen_addr;
@@ -19,15 +19,15 @@ pub async fn start_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<(
let args = [format!("--listen-addr={listen_addr}")];
let client = reqwest::Client::new();
let client = reqwest::blocking::Client::new();
background_process::start_process(
"storage_broker",
&env.base_data_dir,
&env.storage_broker_bin(),
args,
[],
background_process::InitialPidFile::Create(storage_broker_pid_file_path(env)),
|| async {
background_process::InitialPidFile::Create(&storage_broker_pid_file_path(env)),
|| {
let url = broker.client_url();
let status_url = url.join("status").with_context(|| {
format!("Failed to append /status path to broker endpoint {url}")
@@ -36,13 +36,12 @@ pub async fn start_broker_process(env: &local_env::LocalEnv) -> anyhow::Result<(
.get(status_url)
.build()
.with_context(|| format!("Failed to construct request to broker endpoint {url}"))?;
match client.execute(request).await {
match client.execute(request) {
Ok(resp) => Ok(resp.status().is_success()),
Err(_) => Ok(false),
}
},
)
.await
.context("Failed to spawn storage_broker subprocess")?;
Ok(())
}

View File

@@ -46,8 +46,6 @@ use std::time::Duration;
use anyhow::{anyhow, bail, Context, Result};
use compute_api::spec::RemoteExtSpec;
use nix::sys::signal::kill;
use nix::sys::signal::Signal;
use serde::{Deserialize, Serialize};
use utils::id::{NodeId, TenantId, TimelineId};
@@ -441,14 +439,11 @@ impl Endpoint {
Ok(())
}
fn wait_for_compute_ctl_to_exit(&self, send_sigterm: bool) -> Result<()> {
fn wait_for_compute_ctl_to_exit(&self) -> Result<()> {
// TODO use background_process::stop_process instead
let pidfile_path = self.endpoint_path().join("compute_ctl.pid");
let pid: u32 = std::fs::read_to_string(pidfile_path)?.parse()?;
let pid = nix::unistd::Pid::from_raw(pid as i32);
if send_sigterm {
kill(pid, Signal::SIGTERM).ok();
}
crate::background_process::wait_until_stopped("compute_ctl", pid)?;
Ok(())
}
@@ -469,7 +464,7 @@ impl Endpoint {
}
}
pub async fn start(
pub fn start(
&self,
auth_token: &Option<String>,
safekeepers: Vec<NodeId>,
@@ -542,7 +537,6 @@ impl Endpoint {
safekeeper_connstrings,
storage_auth_token: auth_token.clone(),
remote_extensions,
pgbouncer_settings: None,
};
let spec_path = self.endpoint_path().join("spec.json");
std::fs::write(spec_path, serde_json::to_string_pretty(&spec)?)?;
@@ -593,7 +587,7 @@ impl Endpoint {
const MAX_ATTEMPTS: u32 = 10 * 30; // Wait up to 30 s
loop {
attempt += 1;
match self.get_status().await {
match self.get_status() {
Ok(state) => {
match state.status {
ComputeStatus::Init => {
@@ -635,8 +629,8 @@ impl Endpoint {
}
// Call the /status HTTP API
pub async fn get_status(&self) -> Result<ComputeState> {
let client = reqwest::Client::new();
pub fn get_status(&self) -> Result<ComputeState> {
let client = reqwest::blocking::Client::new();
let response = client
.request(
@@ -647,17 +641,16 @@ impl Endpoint {
self.http_address.port()
),
)
.send()
.await?;
.send()?;
// Interpret the response
let status = response.status();
if !(status.is_client_error() || status.is_server_error()) {
Ok(response.json().await?)
Ok(response.json()?)
} else {
// reqwest does not export its error construction utility functions, so let's craft the message ourselves
let url = response.url().to_owned();
let msg = match response.text().await {
let msg = match response.text() {
Ok(err_body) => format!("Error: {}", err_body),
Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
};
@@ -665,7 +658,7 @@ impl Endpoint {
}
}
pub async fn reconfigure(&self, pageserver_id: Option<NodeId>) -> Result<()> {
pub fn reconfigure(&self, pageserver_id: Option<NodeId>) -> Result<()> {
let mut spec: ComputeSpec = {
let spec_path = self.endpoint_path().join("spec.json");
let file = std::fs::File::open(spec_path)?;
@@ -694,7 +687,7 @@ impl Endpoint {
spec.pageserver_connstring = Some(format!("postgresql://no_user@{host}:{port}"));
}
let client = reqwest::Client::new();
let client = reqwest::blocking::Client::new();
let response = client
.post(format!(
"http://{}:{}/configure",
@@ -705,15 +698,14 @@ impl Endpoint {
"{{\"spec\":{}}}",
serde_json::to_string_pretty(&spec)?
))
.send()
.await?;
.send()?;
let status = response.status();
if !(status.is_client_error() || status.is_server_error()) {
Ok(())
} else {
let url = response.url().to_owned();
let msg = match response.text().await {
let msg = match response.text() {
Ok(err_body) => format!("Error: {}", err_body),
Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
};
@@ -738,15 +730,10 @@ impl Endpoint {
&None,
)?;
// Also wait for the compute_ctl process to die. It might have some
// cleanup work to do after postgres stops, like syncing safekeepers,
// etc.
// Also wait for the compute_ctl process to die. It might have some cleanup
// work to do after postgres stops, like syncing safekeepers, etc.
//
// If destroying, send it SIGTERM before waiting. Sometimes we do *not*
// want this cleanup: tests intentionally do stop when majority of
// safekeepers is down, so sync-safekeepers would hang otherwise. This
// could be a separate flag though.
self.wait_for_compute_ctl_to_exit(destroy)?;
self.wait_for_compute_ctl_to_exit()?;
if destroy {
println!(
"Destroying postgres data directory '{}'",

View File

@@ -6,24 +6,28 @@
//!
use std::borrow::Cow;
use std::collections::HashMap;
use std::io;
use std::io::Write;
use std::fs::File;
use std::io::{BufReader, Write};
use std::num::NonZeroU64;
use std::path::PathBuf;
use std::process::{Child, Command};
use std::time::Duration;
use std::{io, result};
use anyhow::{bail, Context};
use camino::Utf8PathBuf;
use futures::SinkExt;
use pageserver_api::models::{self, LocationConfig, TenantInfo, TimelineInfo};
use pageserver_api::models::{
self, LocationConfig, TenantInfo, TenantLocationConfigRequest, TimelineInfo,
};
use pageserver_api::shard::TenantShardId;
use pageserver_client::mgmt_api;
use postgres_backend::AuthType;
use postgres_connection::{parse_host_port, PgConnectionConfig};
use reqwest::blocking::{Client, RequestBuilder, Response};
use reqwest::{IntoUrl, Method};
use thiserror::Error;
use utils::auth::{Claims, Scope};
use utils::{
http::error::HttpErrorBody,
id::{TenantId, TimelineId},
lsn::Lsn,
};
@@ -34,6 +38,45 @@ use crate::{background_process, local_env::LocalEnv};
/// Directory within .neon which will be used by default for LocalFs remote storage.
pub const PAGESERVER_REMOTE_STORAGE_DIR: &str = "local_fs_remote_storage/pageserver";
#[derive(Error, Debug)]
pub enum PageserverHttpError {
#[error("Reqwest error: {0}")]
Transport(#[from] reqwest::Error),
#[error("Error: {0}")]
Response(String),
}
impl From<anyhow::Error> for PageserverHttpError {
fn from(e: anyhow::Error) -> Self {
Self::Response(e.to_string())
}
}
type Result<T> = result::Result<T, PageserverHttpError>;
pub trait ResponseErrorMessageExt: Sized {
fn error_from_body(self) -> Result<Self>;
}
impl ResponseErrorMessageExt for Response {
fn error_from_body(self) -> Result<Self> {
let status = self.status();
if !(status.is_client_error() || status.is_server_error()) {
return Ok(self);
}
// reqwest does not export its error construction utility functions, so let's craft the message ourselves
let url = self.url().to_owned();
Err(PageserverHttpError::Response(
match self.json::<HttpErrorBody>() {
Ok(err_body) => format!("Error: {}", err_body.msg),
Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
},
))
}
}
//
// Control routines for pageserver.
//
@@ -44,7 +87,8 @@ pub struct PageServerNode {
pub pg_connection_config: PgConnectionConfig,
pub conf: PageServerConf,
pub env: LocalEnv,
pub http_client: mgmt_api::Client,
pub http_client: Client,
pub http_base_url: String,
}
impl PageServerNode {
@@ -56,19 +100,8 @@ impl PageServerNode {
pg_connection_config: PgConnectionConfig::new_host_port(host, port),
conf: conf.clone(),
env: env.clone(),
http_client: mgmt_api::Client::new(
format!("http://{}", conf.listen_http_addr),
{
match conf.http_auth_type {
AuthType::Trust => None,
AuthType::NeonJWT => Some(
env.generate_auth_token(&Claims::new(None, Scope::PageServerApi))
.unwrap(),
),
}
}
.as_deref(),
),
http_client: Client::new(),
http_base_url: format!("http://{}/v1", conf.listen_http_addr),
}
}
@@ -149,8 +182,8 @@ impl PageServerNode {
.expect("non-Unicode path")
}
pub async fn start(&self, config_overrides: &[&str]) -> anyhow::Result<Child> {
self.start_node(config_overrides, false).await
pub fn start(&self, config_overrides: &[&str]) -> anyhow::Result<Child> {
self.start_node(config_overrides, false)
}
fn pageserver_init(&self, config_overrides: &[&str]) -> anyhow::Result<()> {
@@ -191,12 +224,7 @@ impl PageServerNode {
Ok(())
}
async fn start_node(
&self,
config_overrides: &[&str],
update_config: bool,
) -> anyhow::Result<Child> {
// TODO: using a thread here because start_process() is not async but we need to call check_status()
fn start_node(&self, config_overrides: &[&str], update_config: bool) -> anyhow::Result<Child> {
let datadir = self.repo_path();
print!(
"Starting pageserver node {} at '{}' in {:?}",
@@ -204,7 +232,7 @@ impl PageServerNode {
self.pg_connection_config.raw_address(),
datadir
);
io::stdout().flush().context("flush stdout")?;
io::stdout().flush()?;
let datadir_path_str = datadir.to_str().with_context(|| {
format!(
@@ -216,23 +244,20 @@ impl PageServerNode {
if update_config {
args.push(Cow::Borrowed("--update-config"));
}
background_process::start_process(
"pageserver",
&datadir,
&self.env.pageserver_bin(),
args.iter().map(Cow::as_ref),
self.pageserver_env_variables()?,
background_process::InitialPidFile::Expect(self.pid_file()),
|| async {
let st = self.check_status().await;
match st {
Ok(()) => Ok(true),
Err(mgmt_api::Error::ReceiveBody(_)) => Ok(false),
Err(e) => Err(anyhow::anyhow!("Failed to check node status: {e}")),
}
background_process::InitialPidFile::Expect(&self.pid_file()),
|| match self.check_status() {
Ok(()) => Ok(true),
Err(PageserverHttpError::Transport(_)) => Ok(false),
Err(e) => Err(anyhow::anyhow!("Failed to check node status: {e}")),
},
)
.await
}
fn pageserver_basic_args<'a>(
@@ -278,12 +303,7 @@ impl PageServerNode {
background_process::stop_process(immediate, "pageserver", &self.pid_file())
}
pub async fn page_server_psql_client(
&self,
) -> anyhow::Result<(
tokio_postgres::Client,
tokio_postgres::Connection<tokio_postgres::Socket, tokio_postgres::tls::NoTlsStream>,
)> {
pub fn page_server_psql_client(&self) -> anyhow::Result<postgres::Client> {
let mut config = self.pg_connection_config.clone();
if self.conf.pg_auth_type == AuthType::NeonJWT {
let token = self
@@ -291,18 +311,36 @@ impl PageServerNode {
.generate_auth_token(&Claims::new(None, Scope::PageServerApi))?;
config = config.set_password(Some(token));
}
Ok(config.connect_no_tls().await?)
Ok(config.connect_no_tls()?)
}
pub async fn check_status(&self) -> mgmt_api::Result<()> {
self.http_client.status().await
fn http_request<U: IntoUrl>(&self, method: Method, url: U) -> anyhow::Result<RequestBuilder> {
let mut builder = self.http_client.request(method, url);
if self.conf.http_auth_type == AuthType::NeonJWT {
let token = self
.env
.generate_auth_token(&Claims::new(None, Scope::PageServerApi))?;
builder = builder.bearer_auth(token)
}
Ok(builder)
}
pub async fn tenant_list(&self) -> mgmt_api::Result<Vec<TenantInfo>> {
self.http_client.list_tenants().await
pub fn check_status(&self) -> Result<()> {
self.http_request(Method::GET, format!("{}/status", self.http_base_url))?
.send()?
.error_from_body()?;
Ok(())
}
pub async fn tenant_create(
pub fn tenant_list(&self) -> Result<Vec<TenantInfo>> {
Ok(self
.http_request(Method::GET, format!("{}/tenant", self.http_base_url))?
.send()?
.error_from_body()?
.json()?)
}
pub fn tenant_create(
&self,
new_tenant_id: TenantId,
generation: Option<u32>,
@@ -380,10 +418,23 @@ impl PageServerNode {
if !settings.is_empty() {
bail!("Unrecognized tenant settings: {settings:?}")
}
Ok(self.http_client.tenant_create(&request).await?)
self.http_request(Method::POST, format!("{}/tenant", self.http_base_url))?
.json(&request)
.send()?
.error_from_body()?
.json::<Option<String>>()
.with_context(|| {
format!("Failed to parse tenant creation response for tenant id: {new_tenant_id:?}")
})?
.context("No tenant id was found in the tenant creation response")
.and_then(|tenant_id_string| {
tenant_id_string.parse().with_context(|| {
format!("Failed to parse response string as tenant id: '{tenant_id_string}'")
})
})
}
pub async fn tenant_config(
pub fn tenant_config(
&self,
tenant_id: TenantId,
mut settings: HashMap<&str, &str>,
@@ -462,37 +513,54 @@ impl PageServerNode {
bail!("Unrecognized tenant settings: {settings:?}")
}
self.http_client
.tenant_config(&models::TenantConfigRequest { tenant_id, config })
.await?;
self.http_request(Method::PUT, format!("{}/tenant/config", self.http_base_url))?
.json(&models::TenantConfigRequest { tenant_id, config })
.send()?
.error_from_body()?;
Ok(())
}
pub async fn location_config(
pub fn location_config(
&self,
tenant_id: TenantId,
config: LocationConfig,
flush_ms: Option<Duration>,
) -> anyhow::Result<()> {
Ok(self
.http_client
.location_config(tenant_id, config, flush_ms)
.await?)
let req_body = TenantLocationConfigRequest { tenant_id, config };
let path = format!(
"{}/tenant/{}/location_config",
self.http_base_url, tenant_id
);
let path = if let Some(flush_ms) = flush_ms {
format!("{}?flush_ms={}", path, flush_ms.as_millis())
} else {
path
};
self.http_request(Method::PUT, path)?
.json(&req_body)
.send()?
.error_from_body()?;
Ok(())
}
pub async fn timeline_list(&self, tenant_id: &TenantId) -> anyhow::Result<Vec<TimelineInfo>> {
Ok(self.http_client.list_timelines(*tenant_id).await?)
pub fn timeline_list(&self, tenant_id: &TenantId) -> anyhow::Result<Vec<TimelineInfo>> {
let timeline_infos: Vec<TimelineInfo> = self
.http_request(
Method::GET,
format!("{}/tenant/{}/timeline", self.http_base_url, tenant_id),
)?
.send()?
.error_from_body()?
.json()?;
Ok(timeline_infos)
}
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(
pub fn timeline_create(
&self,
tenant_id: TenantId,
new_timeline_id: Option<TimelineId>,
@@ -503,14 +571,29 @@ impl PageServerNode {
) -> anyhow::Result<TimelineInfo> {
// If timeline ID was not specified, generate one
let new_timeline_id = new_timeline_id.unwrap_or(TimelineId::generate());
let req = models::TimelineCreateRequest {
self.http_request(
Method::POST,
format!("{}/tenant/{}/timeline", self.http_base_url, tenant_id),
)?
.json(&models::TimelineCreateRequest {
new_timeline_id,
ancestor_start_lsn,
ancestor_timeline_id,
pg_version,
existing_initdb_timeline_id,
};
Ok(self.http_client.timeline_create(tenant_id, &req).await?)
})
.send()?
.error_from_body()?
.json::<Option<TimelineInfo>>()
.with_context(|| {
format!("Failed to parse timeline creation response for tenant id: {tenant_id}")
})?
.with_context(|| {
format!(
"No timeline id was found in the timeline creation response for tenant {tenant_id}"
)
})
}
/// Import a basebackup prepared using either:
@@ -522,7 +605,7 @@ impl PageServerNode {
/// * `timeline_id` - id to assign to imported timeline
/// * `base` - (start lsn of basebackup, path to `base.tar` file)
/// * `pg_wal` - if there's any wal to import: (end lsn, path to `pg_wal.tar`)
pub async fn timeline_import(
pub fn timeline_import(
&self,
tenant_id: TenantId,
timeline_id: TimelineId,
@@ -530,60 +613,36 @@ impl PageServerNode {
pg_wal: Option<(Lsn, PathBuf)>,
pg_version: u32,
) -> anyhow::Result<()> {
let (client, conn) = self.page_server_psql_client().await?;
// The connection object performs the actual communication with the database,
// so spawn it off to run on its own.
tokio::spawn(async move {
if let Err(e) = conn.await {
eprintln!("connection error: {}", e);
}
});
tokio::pin!(client);
let mut client = self.page_server_psql_client()?;
// Init base reader
let (start_lsn, base_tarfile_path) = base;
let base_tarfile = tokio::fs::File::open(base_tarfile_path).await?;
let base_tarfile = tokio_util::io::ReaderStream::new(base_tarfile);
let base_tarfile = File::open(base_tarfile_path)?;
let mut base_reader = BufReader::new(base_tarfile);
// Init wal reader if necessary
let (end_lsn, wal_reader) = if let Some((end_lsn, wal_tarfile_path)) = pg_wal {
let wal_tarfile = tokio::fs::File::open(wal_tarfile_path).await?;
let wal_reader = tokio_util::io::ReaderStream::new(wal_tarfile);
let wal_tarfile = File::open(wal_tarfile_path)?;
let wal_reader = BufReader::new(wal_tarfile);
(end_lsn, Some(wal_reader))
} else {
(start_lsn, None)
};
let copy_in = |reader, cmd| {
let client = &client;
async move {
let writer = client.copy_in(&cmd).await?;
let writer = std::pin::pin!(writer);
let mut writer = writer.sink_map_err(|e| {
std::io::Error::new(std::io::ErrorKind::Other, format!("{e}"))
});
let mut reader = std::pin::pin!(reader);
writer.send_all(&mut reader).await?;
writer.into_inner().finish().await?;
anyhow::Ok(())
}
};
// Import base
copy_in(
base_tarfile,
format!(
"import basebackup {tenant_id} {timeline_id} {start_lsn} {end_lsn} {pg_version}"
),
)
.await?;
let import_cmd = format!(
"import basebackup {tenant_id} {timeline_id} {start_lsn} {end_lsn} {pg_version}"
);
let mut writer = client.copy_in(&import_cmd)?;
io::copy(&mut base_reader, &mut writer)?;
writer.finish()?;
// Import wal if necessary
if let Some(wal_reader) = wal_reader {
copy_in(
wal_reader,
format!("import wal {tenant_id} {timeline_id} {start_lsn} {end_lsn}"),
)
.await?;
if let Some(mut wal_reader) = wal_reader {
let import_cmd = format!("import wal {tenant_id} {timeline_id} {start_lsn} {end_lsn}");
let mut writer = client.copy_in(&import_cmd)?;
io::copy(&mut wal_reader, &mut writer)?;
writer.finish()?;
}
Ok(())

View File

@@ -13,6 +13,7 @@ use std::{io, result};
use anyhow::Context;
use camino::Utf8PathBuf;
use postgres_connection::PgConnectionConfig;
use reqwest::blocking::{Client, RequestBuilder, Response};
use reqwest::{IntoUrl, Method};
use thiserror::Error;
use utils::{http::error::HttpErrorBody, id::NodeId};
@@ -33,14 +34,12 @@ pub enum SafekeeperHttpError {
type Result<T> = result::Result<T, SafekeeperHttpError>;
#[async_trait::async_trait]
pub trait ResponseErrorMessageExt: Sized {
async fn error_from_body(self) -> Result<Self>;
fn error_from_body(self) -> Result<Self>;
}
#[async_trait::async_trait]
impl ResponseErrorMessageExt for reqwest::Response {
async fn error_from_body(self) -> Result<Self> {
impl ResponseErrorMessageExt for Response {
fn error_from_body(self) -> Result<Self> {
let status = self.status();
if !(status.is_client_error() || status.is_server_error()) {
return Ok(self);
@@ -49,7 +48,7 @@ impl ResponseErrorMessageExt for reqwest::Response {
// reqwest does not export its error construction utility functions, so let's craft the message ourselves
let url = self.url().to_owned();
Err(SafekeeperHttpError::Response(
match self.json::<HttpErrorBody>().await {
match self.json::<HttpErrorBody>() {
Ok(err_body) => format!("Error: {}", err_body.msg),
Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
},
@@ -70,7 +69,7 @@ pub struct SafekeeperNode {
pub pg_connection_config: PgConnectionConfig,
pub env: LocalEnv,
pub http_client: reqwest::Client,
pub http_client: Client,
pub http_base_url: String,
}
@@ -81,7 +80,7 @@ impl SafekeeperNode {
conf: conf.clone(),
pg_connection_config: Self::safekeeper_connection_config(conf.pg_port),
env: env.clone(),
http_client: reqwest::Client::new(),
http_client: Client::new(),
http_base_url: format!("http://127.0.0.1:{}/v1", conf.http_port),
}
}
@@ -104,7 +103,7 @@ impl SafekeeperNode {
.expect("non-Unicode path")
}
pub async fn start(&self, extra_opts: Vec<String>) -> anyhow::Result<Child> {
pub fn start(&self, extra_opts: Vec<String>) -> anyhow::Result<Child> {
print!(
"Starting safekeeper at '{}' in '{}'",
self.pg_connection_config.raw_address(),
@@ -192,16 +191,13 @@ impl SafekeeperNode {
&self.env.safekeeper_bin(),
&args,
[],
background_process::InitialPidFile::Expect(self.pid_file()),
|| async {
match self.check_status().await {
Ok(()) => Ok(true),
Err(SafekeeperHttpError::Transport(_)) => Ok(false),
Err(e) => Err(anyhow::anyhow!("Failed to check node status: {e}")),
}
background_process::InitialPidFile::Expect(&self.pid_file()),
|| match self.check_status() {
Ok(()) => Ok(true),
Err(SafekeeperHttpError::Transport(_)) => Ok(false),
Err(e) => Err(anyhow::anyhow!("Failed to check node status: {e}")),
},
)
.await
}
///
@@ -220,7 +216,7 @@ impl SafekeeperNode {
)
}
fn http_request<U: IntoUrl>(&self, method: Method, url: U) -> reqwest::RequestBuilder {
fn http_request<U: IntoUrl>(&self, method: Method, url: U) -> RequestBuilder {
// TODO: authentication
//if self.env.auth_type == AuthType::NeonJWT {
// builder = builder.bearer_auth(&self.env.safekeeper_auth_token)
@@ -228,12 +224,10 @@ impl SafekeeperNode {
self.http_client.request(method, url)
}
pub async fn check_status(&self) -> Result<()> {
pub fn check_status(&self) -> Result<()> {
self.http_request(Method::GET, format!("{}/{}", self.http_base_url, "status"))
.send()
.await?
.error_from_body()
.await?;
.send()?
.error_from_body()?;
Ok(())
}
}

View File

@@ -11,7 +11,6 @@ use crate::{
use pageserver_api::models::{
LocationConfig, LocationConfigMode, LocationConfigSecondary, TenantConfig,
};
use pageserver_api::shard::TenantShardId;
use std::collections::HashMap;
use std::time::Duration;
use utils::{
@@ -20,11 +19,11 @@ use utils::{
};
/// Given an attached pageserver, retrieve the LSN for all timelines
async fn get_lsns(
fn get_lsns(
tenant_id: TenantId,
pageserver: &PageServerNode,
) -> anyhow::Result<HashMap<TimelineId, Lsn>> {
let timelines = pageserver.timeline_list(&tenant_id).await?;
let timelines = pageserver.timeline_list(&tenant_id)?;
Ok(timelines
.into_iter()
.map(|t| (t.timeline_id, t.last_record_lsn))
@@ -33,17 +32,17 @@ async fn get_lsns(
/// Wait for the timeline LSNs on `pageserver` to catch up with or overtake
/// `baseline`.
async fn await_lsn(
fn await_lsn(
tenant_id: TenantId,
pageserver: &PageServerNode,
baseline: HashMap<TimelineId, Lsn>,
) -> anyhow::Result<()> {
loop {
let latest = match get_lsns(tenant_id, pageserver).await {
let latest = match get_lsns(tenant_id, pageserver) {
Ok(l) => l,
Err(_e) => {
Err(e) => {
println!(
"🕑 Waiting for pageserver {} to activate...",
"🕑 Can't get LSNs on pageserver {} yet, waiting ({e})",
pageserver.conf.id
);
std::thread::sleep(Duration::from_millis(500));
@@ -85,12 +84,12 @@ async fn await_lsn(
/// - Coordinate attach/secondary/detach on pageservers
/// - call into attachment_service for generations
/// - reconfigure compute endpoints to point to new attached pageserver
pub async fn migrate_tenant(
pub fn migrate_tenant(
env: &LocalEnv,
tenant_id: TenantId,
dest_ps: PageServerNode,
) -> anyhow::Result<()> {
println!("🤔 Checking existing status...");
// Get a new generation
let attachment_service = AttachmentService::from_env(env);
fn build_location_config(
@@ -109,18 +108,16 @@ pub async fn migrate_tenant(
}
}
let previous = attachment_service.inspect(tenant_id).await?;
let previous = attachment_service.inspect(tenant_id)?;
let mut baseline_lsns = None;
if let Some((generation, origin_ps_id)) = &previous {
let origin_ps = PageServerNode::from_env(env, env.get_pageserver_conf(*origin_ps_id)?);
if origin_ps_id == &dest_ps.conf.id {
println!("🔁 Already attached to {origin_ps_id}, freshening...");
let gen = attachment_service
.attach_hook(tenant_id, dest_ps.conf.id)
.await?;
let gen = attachment_service.attach_hook(tenant_id, dest_ps.conf.id)?;
let dest_conf = build_location_config(LocationConfigMode::AttachedSingle, gen, None);
dest_ps.location_config(tenant_id, dest_conf, None).await?;
dest_ps.location_config(tenant_id, dest_conf, None)?;
println!("✅ Migration complete");
return Ok(());
}
@@ -129,38 +126,20 @@ pub async fn migrate_tenant(
let stale_conf =
build_location_config(LocationConfigMode::AttachedStale, Some(*generation), None);
origin_ps
.location_config(tenant_id, stale_conf, Some(Duration::from_secs(10)))
.await?;
origin_ps.location_config(tenant_id, stale_conf, Some(Duration::from_secs(10)))?;
baseline_lsns = Some(get_lsns(tenant_id, &origin_ps).await?);
baseline_lsns = Some(get_lsns(tenant_id, &origin_ps)?);
}
println!(
"🔁 Downloading latest layers to destination pageserver {}",
dest_ps.conf.id
);
match dest_ps
.tenant_secondary_download(&TenantShardId::unsharded(tenant_id))
.await
{
Ok(()) => {}
Err(_) => {
println!(" (skipping, destination wasn't in secondary mode)")
}
}
let gen = attachment_service
.attach_hook(tenant_id, dest_ps.conf.id)
.await?;
let gen = attachment_service.attach_hook(tenant_id, dest_ps.conf.id)?;
let dest_conf = build_location_config(LocationConfigMode::AttachedMulti, gen, None);
println!("🔁 Attaching to pageserver {}", dest_ps.conf.id);
dest_ps.location_config(tenant_id, dest_conf, None).await?;
dest_ps.location_config(tenant_id, dest_conf, None)?;
if let Some(baseline) = baseline_lsns {
println!("🕑 Waiting for LSN to catch up...");
await_lsn(tenant_id, &dest_ps, baseline).await?;
await_lsn(tenant_id, &dest_ps, baseline)?;
}
let cplane = ComputeControlPlane::load(env.clone())?;
@@ -170,7 +149,7 @@ pub async fn migrate_tenant(
"🔁 Reconfiguring endpoint {} to use pageserver {}",
endpoint_name, dest_ps.conf.id
);
endpoint.reconfigure(Some(dest_ps.conf.id)).await?;
endpoint.reconfigure(Some(dest_ps.conf.id))?;
}
}
@@ -180,7 +159,7 @@ pub async fn migrate_tenant(
}
let other_ps = PageServerNode::from_env(env, other_ps_conf);
let other_ps_tenants = other_ps.tenant_list().await?;
let other_ps_tenants = other_ps.tenant_list()?;
// Check if this tenant is attached
let found = other_ps_tenants
@@ -202,9 +181,7 @@ pub async fn migrate_tenant(
"💤 Switching to secondary mode on pageserver {}",
other_ps.conf.id
);
other_ps
.location_config(tenant_id, secondary_conf, None)
.await?;
other_ps.location_config(tenant_id, secondary_conf, None)?;
}
println!(
@@ -212,7 +189,7 @@ pub async fn migrate_tenant(
dest_ps.conf.id
);
let dest_conf = build_location_config(LocationConfigMode::AttachedSingle, gen, None);
dest_ps.location_config(tenant_id, dest_conf, None).await?;
dest_ps.location_config(tenant_id, dest_conf, None)?;
println!("✅ Migration complete");

View File

@@ -35,7 +35,6 @@ allow = [
"Artistic-2.0",
"BSD-2-Clause",
"BSD-3-Clause",
"CC0-1.0",
"ISC",
"MIT",
"MPL-2.0",

View File

@@ -1,197 +0,0 @@
# Per-Tenant GetPage@LSN Throttling
Author: Christian Schwarz
Date: Oct 24, 2023
## Summary
This RFC proposes per-tenant throttling of GetPage@LSN requests inside Pageserver
and the interactions with its client, i.e., the neon_smgr component in Compute.
The result of implementing & executing this RFC will be a fleet-wide upper limit for
**"the highest GetPage/second that Pageserver can support for a single tenant/shard"**.
## Background
### GetPage@LSN Request Flow
Pageserver exposes its `page_service.rs` as a libpq listener.
The Computes' `neon_smgr` module connects to that libpq listener.
Once a connection is established, the protocol allows Compute to request page images at a given LSN.
We call these requests GetPage@LSN requests, or GetPage requests for short.
Other request types can be sent, but these are low traffic compared to GetPage requests
and are not the concern of this RFC.
Pageserver associates one libpq connection with one tokio task.
Per connection/task, the pq protocol is handled by the common `postgres_backend` crate.
Its `run_message_loop` function invokes the `page_service` specific `impl<IO> postgres_backend::Handler<IO> for PageServerHandler`.
Requests are processed in the order in which they arrive via the TCP-based pq protocol.
So, there is no concurrent request processing within one connection/task.
There is a degree of natural pipelining:
Compute can "fill the pipe" by sending more than one GetPage request into the libpq TCP stream.
And Pageserver can fill the pipe with responses in the other direction.
Both directions are subject to the limit of tx/rx buffers, nodelay, TCP flow control, etc.
### GetPage@LSN Access Pattern
The Compute has its own hierarchy of caches, specifically `shared_buffers` and the `local file cache` (LFC).
Compute only issues GetPage requests to Pageserver if it encounters a miss in these caches.
If the working set stops fitting into Compute's caches, requests to Pageserver increase sharply -- the Compute starts *thrashing*.
## Motivation
In INC-69, a tenant issued 155k GetPage/second for a period of 10 minutes and 60k GetPage/second for a period of 3h,
then dropping to ca 18k GetPage/second for a period of 9h.
We noticed this because of an internal GetPage latency SLO burn rate alert, i.e.,
the request latency profile during this period significantly exceeded what was acceptable according to the internal SLO.
Sadly, we do not have the observability data to determine the impact of this tenant on other tenants on the same tenants.
However, here are some illustrative data points for the 155k period:
The tenant was responsible for >= 99% of the GetPage traffic and, frankly, the overall activity on this Pageserver instance.
We were serving pages at 10 Gb/s (`155k x 8 kbyte (PAGE_SZ) per second is 1.12GiB/s = 9.4Gb/s.`)
The CPU utilization of the instance was 75% user+system.
Pageserver page cache served 1.75M accesses/second at a hit rate of ca 90%.
The hit rate for materialized pages was ca. 40%.
Curiously, IOPS to the Instance Store NVMe were very low, rarely exceeding 100.
The fact that the IOPS were so low / the materialized page cache hit rate was so high suggests that **this tenant's compute's caches were thrashing**.
The compute was of type `k8s-pod`; hence, auto-scaling could/would not have helped remediate the thrashing by provisioning more RAM.
The consequence was that the **thrashing translated into excessive GetPage requests against Pageserver**.
My claim is that it was **unhealthy to serve this workload at the pace we did**:
* it is likely that other tenants were/would have experienced high latencies (again, we sadly don't have per-tenant latency data to confirm this)
* more importantly, it was **unsustainable** to serve traffic at this pace for multiple reasons:
* **predictability of performance**: when the working set grows, the pageserver materialized page cache hit rate drops.
At some point, we're bound by the EC2 Instance Store NVMe drive's IOPS limit.
The result is an **uneven** performance profile from the Compute perspective.
* **economics**: Neon currently does not charge for IOPS, only capacity.
**We cannot afford to undercut the market in IOPS/$ this drastically; it leads to adverse selection and perverse incentives.**
For example, the 155k IOPS, which we served for 10min, would cost ca. 6.5k$/month when provisioned as an io2 EBS volume.
Even the 18k IOPS, which we served for 9h, would cost ca. 1.1k$/month when provisioned as an io2 EBS volume.
We charge 0$.
It could be economically advantageous to keep using a low-DRAM compute because Pageserver IOPS are fast enough and free.
Note: It is helpful to think of Pageserver as a disk, because it's precisely where `neon_smgr` sits:
vanilla Postgres gets its pages from disk, Neon Postgres gets them from Pageserver.
So, regarding the above performance & economic arguments, it is fair to say that we currently provide an "as-fast-as-possible-IOPS" disk that we charge for only by capacity.
## Solution: Throttling GetPage Requests
**The consequence of the above analysis must be that Pageserver throttles GetPage@LSN requests**.
That is, unless we want to start charging for provisioned GetPage@LSN/second.
Throttling sets the correct incentive for a thrashing Compute to scale up its DRAM to the working set size.
Neon Autoscaling will make this easy, [eventually](https://github.com/neondatabase/neon/pull/3913).
## The Design Space
What that remains is the question about *policy* and *mechanism*:
**Policy** concerns itself with the question of what limit applies to a given connection|timeline|tenant.
Candidates are:
* hard limit, same limit value per connection|timeline|tenant
* Per-tenant will provide an upper bound for the impact of a tenant on a given Pageserver instance.
This is a major operational pain point / risk right now.
* hard limit, configurable per connection|timeline|tenant
* This outsources policy to console/control plane, with obvious advantages for flexible structuring of what service we offer to customers.
* Note that this is not a mechanism to guarantee a minium provisioned rate, i.e., this is not a mechanism to guarantee a certain QoS for a tenant.
* fair share among active connections|timelines|tenants per instance
* example: each connection|timeline|tenant gets a fair fraction of the machine's GetPage/second capacity
* NB: needs definition of "active", and knowledge of available GetPage/second capacity in advance
* ...
Regarding **mechanism**, it's clear that **backpressure** is the way to go.
However, we must choose between
* **implicit** backpressure through pq/TCP and
* **explicit** rejection of requests + retries with exponential backoff
Further, there is the question of how throttling GetPage@LSN will affect the **internal GetPage latency SLO**:
where do we measure the SLI for Pageserver's internal getpage latency SLO? Before or after the throttling?
And when we eventually move the measurement point into the Computes (to avoid coordinated omission),
how do we avoid counting throttling-induced latency toward the internal getpage latency SLI/SLO?
## Scope Of This RFC
**This RFC proposes introducing a hard GetPage@LSN/second limit per tenant, with the same value applying to each tenant on a Pageserver**.
This proposal is easy to implement and significantly de-risks operating large Pageservers,
based on the assumption that extremely-high-GetPage-rate-episodes like the one from the "Motivation" section are uncorrelated between tenants.
For example, suppose we pick a limit that allows up to 10 tenants to go at limit rate.
Suppose our Pageserver can serve 100k GetPage/second total at a 100% page cache miss rate.
If each tenant gets a hard limit of 10k GetPage/second, we can serve up to 10 tenants at limit speed without latency degradation.
The mechanism for backpressure will be TCP-based implicit backpressure.
The compute team isn't concerned about prefetch queue depth.
Pageserver will implement it by delaying the reading of requests from the libpq connection(s).
The rate limit will be implemented using a per-tenant token bucket.
The bucket will be be shared among all connections to the tenant.
The bucket implementation supports starvation-preventing `await`ing.
The current candidate for the implementation is [`leaky_bucket`](https://docs.rs/leaky-bucket/).
The getpage@lsn benchmark that's being added in https://github.com/neondatabase/neon/issues/5771
can be used to evaluate the overhead of sharing the bucket among connections of a tenant.
A possible technique to mitigate the impact of sharing the bucket would be to maintain a buffer of a few tokens per connection handler.
Regarding metrics / the internal GetPage latency SLO:
we will measure the GetPage latency SLO _after_ the throttler and introduce a new metric to measure the amount of throttling, quantified by:
- histogram that records the tenants' observations of queue depth before they start waiting (one such histogram per pageserver)
- histogram that records the tenants' observations of time spent waiting (one such histogram per pageserver)
Further observability measures:
- an INFO log message at frequency 1/min if the tenant/timeline/connection was throttled in that last minute.
The message will identify the tenant/timeline/connection to allow correlation with compute logs/stats.
Rollout will happen as follows:
- deploy 1: implementation + config: disabled by default, ability to enable it per tenant through tenant_conf
- experimentation in staging and later production to study impact & interaction with auto-scaling
- determination of a sensible global default value
- the value will be chosen as high as possible ...
- ... but low enough to work towards this RFC's goal that one tenant should not be able to dominate a pageserver instance.
- deploy 2: implementation fixes if any + config: enabled by default with the aforementioned global default
- reset of the experimental per-tenant overrides
- gain experience & lower the limit over time
- we stop lowering the limit as soon as this RFC's goal is achieved, i.e.,
once we decide that in practice the chosen value sufficiently de-risks operating large pageservers
The per-tenant override will remain for emergencies and testing.
But since Console doesn't preserve it during tenant migrations, it isn't durably configurable for the tenant.
Toward the upper layers of the Neon stack, the resulting limit will be
**"the highest GetPage/second that Pageserver can support for a single tenant"**.
### Rationale
We decided against error + retry because of worries about starvation.
## Future Work
Enable per-tenant emergency override of the limit via Console.
Should be part of a more general framework to specify tenant config overrides.
**NB:** this is **not** the right mechanism to _sell_ different max GetPage/second levels to users,
or _auto-scale_ the GetPage/second levels. Such functionality will require a separate RFC that
concerns itself with GetPage/second capacity planning.
Compute-side metrics for GetPage latency.
Back-channel to inform Compute/Autoscaling/ControlPlane that the project is being throttled.
Compute-side neon_smgr improvements to avoid sending the same GetPage request multiple times if multiple backends experience a cache miss.
Dealing with read-only endpoints: users use read-only endpoints to scale reads for a single tenant.
Possibly there are also assumptions around read-only endpoints not affecting the primary read-write endpoint's performance.
With per-tenant rate limiting, we will not meet that expectation.
However, we can currently only scale per tenant.
Soon, we will have sharding (#5505), which will apply the throttling on a per-shard basis.
But, that's orthogonal to scaling reads: if many endpoints hit one shard, they share the same throttling limit.
To solve this properly, I think we'll need replicas for tenants / shard.
To performance-isolate a tenant's endpoints from each other, we'd then route them to different replicas.

View File

@@ -129,13 +129,13 @@ Run `poetry shell` to activate the virtual environment.
Alternatively, use `poetry run` to run a single command in the venv, e.g. `poetry run pytest`.
### Obligatory checks
We force code formatting via `ruff`, and type hints via `mypy`.
We force code formatting via `black`, `ruff`, and type hints via `mypy`.
Run the following commands in the repository's root (next to `pyproject.toml`):
```bash
poetry run ruff format . # All code is reformatted
poetry run ruff check . # Python linter
poetry run mypy . # Ensure there are no typing errors
poetry run black . # All code is reformatted
poetry run ruff . # Python linter
poetry run mypy . # Ensure there are no typing errors
```
**WARNING**: do not run `mypy` from a directory other than the root of the repository.

View File

@@ -73,8 +73,6 @@ pub struct ComputeSpec {
// information about available remote extensions
pub remote_extensions: Option<RemoteExtSpec>,
pub pgbouncer_settings: Option<HashMap<String, String>>,
}
/// Feature flag to signal `compute_ctl` to enable certain experimental functionality.

View File

@@ -243,9 +243,5 @@
"public_extensions": [
"postgis"
]
},
"pgbouncer_settings": {
"default_pool_size": "42",
"pool_mode": "session"
}
}

View File

@@ -24,4 +24,3 @@ workspace_hack.workspace = true
[dev-dependencies]
bincode.workspace = true
rand.workspace = true

View File

@@ -142,39 +142,5 @@ impl Key {
}
pub fn is_rel_block_key(key: &Key) -> bool {
key.field1 == 0x00 && key.field4 != 0 && key.field6 != 0xffffffff
}
impl std::str::FromStr for Key {
type Err = anyhow::Error;
fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
Self::from_hex(s)
}
}
#[cfg(test)]
mod tests {
use std::str::FromStr;
use crate::key::Key;
use rand::Rng;
use rand::SeedableRng;
#[test]
fn display_fromstr_bijection() {
let mut rng = rand::rngs::StdRng::seed_from_u64(42);
let key = Key {
field1: rng.gen(),
field2: rng.gen(),
field3: rng.gen(),
field4: rng.gen(),
field5: rng.gen(),
field6: rng.gen(),
};
assert_eq!(key, Key::from_str(&format!("{key}")).unwrap());
}
key.field1 == 0x00 && key.field4 != 0
}

View File

@@ -5,7 +5,6 @@ use const_format::formatcp;
/// Public API types
pub mod control_api;
pub mod key;
pub mod keyspace;
pub mod models;
pub mod reltag;
pub mod shard;

View File

@@ -1,8 +1,5 @@
pub mod partitioning;
use std::{
collections::HashMap,
io::Read,
num::{NonZeroU64, NonZeroUsize},
time::SystemTime,
};
@@ -20,7 +17,7 @@ use utils::{
use crate::{reltag::RelTag, shard::TenantShardId};
use anyhow::bail;
use bytes::{Buf, BufMut, Bytes, BytesMut};
use bytes::{BufMut, Bytes, BytesMut};
/// The state of a tenant in this pageserver.
///
@@ -370,14 +367,6 @@ pub struct TenantInfo {
pub attachment_status: TenantAttachmentStatus,
}
#[derive(Serialize, Deserialize, Clone)]
pub struct TenantDetails {
#[serde(flatten)]
pub tenant_info: TenantInfo,
pub timelines: Vec<TimelineId>,
}
/// This represents the output of the "timeline_detail" and "timeline_list" API calls.
#[derive(Debug, Serialize, Deserialize, Clone)]
pub struct TimelineInfo {
@@ -557,6 +546,19 @@ pub enum DownloadRemoteLayersTaskState {
ShutDown,
}
pub type ConfigureFailpointsRequest = Vec<FailpointConfig>;
/// Information for configuring a single fail point
#[derive(Debug, Serialize, Deserialize)]
pub struct FailpointConfig {
/// Name of the fail point
pub name: String,
/// List of actions to take, using the format described in `fail::cfg`
///
/// We also support `actions = "exit"` to cause the fail point to immediately exit.
pub actions: String,
}
#[derive(Debug, Serialize, Deserialize)]
pub struct TimelineGcRequest {
pub gc_horizon: Option<u64>,
@@ -572,7 +574,6 @@ pub enum PagestreamFeMessage {
}
// Wrapped in libpq CopyData
#[derive(strum_macros::EnumProperty)]
pub enum PagestreamBeMessage {
Exists(PagestreamExistsResponse),
Nblocks(PagestreamNblocksResponse),
@@ -581,29 +582,6 @@ pub enum PagestreamBeMessage {
DbSize(PagestreamDbSizeResponse),
}
// Keep in sync with `pagestore_client.h`
#[repr(u8)]
enum PagestreamBeMessageTag {
Exists = 100,
Nblocks = 101,
GetPage = 102,
Error = 103,
DbSize = 104,
}
impl TryFrom<u8> for PagestreamBeMessageTag {
type Error = u8;
fn try_from(value: u8) -> Result<Self, u8> {
match value {
100 => Ok(PagestreamBeMessageTag::Exists),
101 => Ok(PagestreamBeMessageTag::Nblocks),
102 => Ok(PagestreamBeMessageTag::GetPage),
103 => Ok(PagestreamBeMessageTag::Error),
104 => Ok(PagestreamBeMessageTag::DbSize),
_ => Err(value),
}
}
}
#[derive(Debug, PartialEq, Eq)]
pub struct PagestreamExistsRequest {
pub latest: bool,
@@ -759,91 +737,35 @@ impl PagestreamBeMessage {
pub fn serialize(&self) -> Bytes {
let mut bytes = BytesMut::new();
use PagestreamBeMessageTag as Tag;
match self {
Self::Exists(resp) => {
bytes.put_u8(Tag::Exists as u8);
bytes.put_u8(100); /* tag from pagestore_client.h */
bytes.put_u8(resp.exists as u8);
}
Self::Nblocks(resp) => {
bytes.put_u8(Tag::Nblocks as u8);
bytes.put_u8(101); /* tag from pagestore_client.h */
bytes.put_u32(resp.n_blocks);
}
Self::GetPage(resp) => {
bytes.put_u8(Tag::GetPage as u8);
bytes.put_u8(102); /* tag from pagestore_client.h */
bytes.put(&resp.page[..]);
}
Self::Error(resp) => {
bytes.put_u8(Tag::Error as u8);
bytes.put_u8(103); /* tag from pagestore_client.h */
bytes.put(resp.message.as_bytes());
bytes.put_u8(0); // null terminator
}
Self::DbSize(resp) => {
bytes.put_u8(Tag::DbSize as u8);
bytes.put_u8(104); /* tag from pagestore_client.h */
bytes.put_i64(resp.db_size);
}
}
bytes.into()
}
pub fn deserialize(buf: Bytes) -> anyhow::Result<Self> {
let mut buf = buf.reader();
let msg_tag = buf.read_u8()?;
use PagestreamBeMessageTag as Tag;
let ok =
match Tag::try_from(msg_tag).map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))? {
Tag::Exists => {
let exists = buf.read_u8()?;
Self::Exists(PagestreamExistsResponse {
exists: exists != 0,
})
}
Tag::Nblocks => {
let n_blocks = buf.read_u32::<BigEndian>()?;
Self::Nblocks(PagestreamNblocksResponse { n_blocks })
}
Tag::GetPage => {
let mut page = vec![0; 8192]; // TODO: use MaybeUninit
buf.read_exact(&mut page)?;
PagestreamBeMessage::GetPage(PagestreamGetPageResponse { page: page.into() })
}
Tag::Error => {
let buf = buf.get_ref();
let cstr = std::ffi::CStr::from_bytes_until_nul(buf)?;
let rust_str = cstr.to_str()?;
PagestreamBeMessage::Error(PagestreamErrorResponse {
message: rust_str.to_owned(),
})
}
Tag::DbSize => {
let db_size = buf.read_i64::<BigEndian>()?;
Self::DbSize(PagestreamDbSizeResponse { db_size })
}
};
let remaining = buf.into_inner();
if !remaining.is_empty() {
anyhow::bail!(
"remaining bytes in msg with tag={msg_tag}: {}",
remaining.len()
);
}
Ok(ok)
}
pub fn kind(&self) -> &'static str {
match self {
Self::Exists(_) => "Exists",
Self::Nblocks(_) => "Nblocks",
Self::GetPage(_) => "GetPage",
Self::Error(_) => "Error",
Self::DbSize(_) => "DbSize",
}
}
}
#[cfg(test)]

View File

@@ -1,151 +0,0 @@
use utils::lsn::Lsn;
#[derive(Debug, PartialEq, Eq)]
pub struct Partitioning {
pub keys: crate::keyspace::KeySpace,
pub at_lsn: Lsn,
}
impl serde::Serialize for Partitioning {
fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
pub struct KeySpace<'a>(&'a crate::keyspace::KeySpace);
impl<'a> serde::Serialize for KeySpace<'a> {
fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
use serde::ser::SerializeSeq;
let mut seq = serializer.serialize_seq(Some(self.0.ranges.len()))?;
for kr in &self.0.ranges {
seq.serialize_element(&KeyRange(kr))?;
}
seq.end()
}
}
use serde::ser::SerializeMap;
let mut map = serializer.serialize_map(Some(2))?;
map.serialize_key("keys")?;
map.serialize_value(&KeySpace(&self.keys))?;
map.serialize_key("at_lsn")?;
map.serialize_value(&WithDisplay(&self.at_lsn))?;
map.end()
}
}
pub struct WithDisplay<'a, T>(&'a T);
impl<'a, T: std::fmt::Display> serde::Serialize for WithDisplay<'a, T> {
fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
serializer.collect_str(&self.0)
}
}
pub struct KeyRange<'a>(&'a std::ops::Range<crate::key::Key>);
impl<'a> serde::Serialize for KeyRange<'a> {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
use serde::ser::SerializeTuple;
let mut t = serializer.serialize_tuple(2)?;
t.serialize_element(&WithDisplay(&self.0.start))?;
t.serialize_element(&WithDisplay(&self.0.end))?;
t.end()
}
}
impl<'a> serde::Deserialize<'a> for Partitioning {
fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
where
D: serde::Deserializer<'a>,
{
pub struct KeySpace(crate::keyspace::KeySpace);
impl<'de> serde::Deserialize<'de> for KeySpace {
fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
where
D: serde::Deserializer<'de>,
{
#[serde_with::serde_as]
#[derive(serde::Deserialize)]
#[serde(transparent)]
struct Key(#[serde_as(as = "serde_with::DisplayFromStr")] crate::key::Key);
#[serde_with::serde_as]
#[derive(serde::Deserialize)]
struct Range(Key, Key);
let ranges: Vec<Range> = serde::Deserialize::deserialize(deserializer)?;
Ok(Self(crate::keyspace::KeySpace {
ranges: ranges
.into_iter()
.map(|Range(start, end)| (start.0..end.0))
.collect(),
}))
}
}
#[serde_with::serde_as]
#[derive(serde::Deserialize)]
struct De {
keys: KeySpace,
#[serde_as(as = "serde_with::DisplayFromStr")]
at_lsn: Lsn,
}
let de: De = serde::Deserialize::deserialize(deserializer)?;
Ok(Self {
at_lsn: de.at_lsn,
keys: de.keys.0,
})
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_serialization_roundtrip() {
let reference = r#"
{
"keys": [
[
"000000000000000000000000000000000000",
"000000000000000000000000000000000001"
],
[
"000000067F00000001000000000000000000",
"000000067F00000001000000000000000002"
],
[
"030000000000000000000000000000000000",
"030000000000000000000000000000000003"
]
],
"at_lsn": "0/2240160"
}
"#;
let de: Partitioning = serde_json::from_str(reference).unwrap();
let ser = serde_json::to_string(&de).unwrap();
let ser_de: serde_json::Value = serde_json::from_str(&ser).unwrap();
assert_eq!(
ser_de,
serde_json::from_str::<'_, serde_json::Value>(reference).unwrap()
);
}
}

View File

@@ -81,10 +81,6 @@ impl TenantShardId {
pub fn is_zero(&self) -> bool {
self.shard_number == ShardNumber(0)
}
pub fn is_unsharded(&self) -> bool {
self.shard_number == ShardNumber(0) && self.shard_count == ShardCount(0)
}
}
/// Formatting helper
@@ -163,7 +159,7 @@ impl From<[u8; 18]> for TenantShardId {
/// shard we're dealing with, but do not need to know the full ShardIdentity (because
/// we won't be doing any page->shard mapping), and do not need to know the fully qualified
/// TenantShardId.
#[derive(Eq, PartialEq, PartialOrd, Ord, Clone, Copy, Hash)]
#[derive(Eq, PartialEq, PartialOrd, Ord, Clone, Copy)]
pub struct ShardIndex {
pub shard_number: ShardNumber,
pub shard_count: ShardCount,
@@ -422,21 +418,6 @@ impl ShardIdentity {
}
}
/// Return true if the key should be discarded if found in this shard's
/// data store, e.g. during compaction after a split
pub fn is_key_disposable(&self, key: &Key) -> bool {
if key_is_shard0(key) {
// Q: Why can't we dispose of shard0 content if we're not shard 0?
// A: because the WAL ingestion logic currently ingests some shard 0
// content on all shards, even though it's only read on shard 0. If we
// dropped it, then subsequent WAL ingest to these keys would encounter
// an error.
false
} else {
!self.is_key_local(key)
}
}
pub fn shard_slug(&self) -> String {
if self.count > ShardCount(0) {
format!("-{:02x}{:02x}", self.number.0, self.count.0)
@@ -530,7 +511,12 @@ fn key_is_shard0(key: &Key) -> bool {
// relation pages are distributed to shards other than shard zero. Everything else gets
// stored on shard 0. This guarantees that shard 0 can independently serve basebackup
// requests, and any request other than those for particular blocks in relations.
!is_rel_block_key(key)
//
// In this condition:
// - is_rel_block_key includes only relations, i.e. excludes SLRU data and
// all metadata.
// - field6 is set to -1 for relation size pages.
!(is_rel_block_key(key) && key.field6 != 0xffffffff)
}
/// Provide the same result as the function in postgres `hashfn.h` with the same name

View File

@@ -35,12 +35,6 @@ pub enum QueryError {
/// We were instructed to shutdown while processing the query
#[error("Shutting down")]
Shutdown,
/// Query handler indicated that client should reconnect
#[error("Server requested reconnect")]
Reconnect,
/// Query named an entity that was not found
#[error("Not found: {0}")]
NotFound(std::borrow::Cow<'static, str>),
/// Authentication failure
#[error("Unauthorized: {0}")]
Unauthorized(std::borrow::Cow<'static, str>),
@@ -60,9 +54,9 @@ impl From<io::Error> for QueryError {
impl QueryError {
pub fn pg_error_code(&self) -> &'static [u8; 5] {
match self {
Self::Disconnected(_) | Self::SimulatedConnectionError | Self::Reconnect => b"08006", // connection failure
Self::Disconnected(_) | Self::SimulatedConnectionError => b"08006", // connection failure
Self::Shutdown => SQLSTATE_ADMIN_SHUTDOWN,
Self::Unauthorized(_) | Self::NotFound(_) => SQLSTATE_INTERNAL_ERROR,
Self::Unauthorized(_) => SQLSTATE_INTERNAL_ERROR,
Self::Other(_) => SQLSTATE_INTERNAL_ERROR, // internal error
}
}
@@ -431,11 +425,6 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
info!("Stopped due to shutdown");
Ok(())
}
Err(QueryError::Reconnect) => {
// Dropping out of this loop implicitly disconnects
info!("Stopped due to handler reconnect request");
Ok(())
}
Err(QueryError::Disconnected(e)) => {
info!("Disconnected ({e:#})");
// Disconnection is not an error: we just use it that way internally to drop
@@ -985,9 +974,7 @@ impl<'a, IO: AsyncRead + AsyncWrite + Unpin> AsyncWrite for CopyDataWriter<'a, I
pub fn short_error(e: &QueryError) -> String {
match e {
QueryError::Disconnected(connection_error) => connection_error.to_string(),
QueryError::Reconnect => "reconnect".to_string(),
QueryError::Shutdown => "shutdown".to_string(),
QueryError::NotFound(_) => "not found".to_string(),
QueryError::Unauthorized(_e) => "JWT authentication error".to_string(),
QueryError::SimulatedConnectionError => "simulated connection error".to_string(),
QueryError::Other(e) => format!("{e:#}"),
@@ -1009,15 +996,9 @@ fn log_query_error(query: &str, e: &QueryError) {
QueryError::SimulatedConnectionError => {
error!("query handler for query '{query}' failed due to a simulated connection error")
}
QueryError::Reconnect => {
info!("query handler for '{query}' requested client to reconnect")
}
QueryError::Shutdown => {
info!("query handler for '{query}' cancelled during tenant shutdown")
}
QueryError::NotFound(reason) => {
info!("query handler for '{query}' entity not found: {reason}")
}
QueryError::Unauthorized(e) => {
warn!("query handler for '{query}' failed with authentication error: {e}");
}

View File

@@ -163,18 +163,8 @@ impl PgConnectionConfig {
}
/// Connect using postgres protocol with TLS disabled.
pub async fn connect_no_tls(
&self,
) -> Result<
(
tokio_postgres::Client,
tokio_postgres::Connection<tokio_postgres::Socket, tokio_postgres::tls::NoTlsStream>,
),
postgres::Error,
> {
self.to_tokio_postgres_config()
.connect(postgres::NoTls)
.await
pub fn connect_no_tls(&self) -> Result<postgres::Client, postgres::Error> {
postgres::Config::from(self.to_tokio_postgres_config()).connect(postgres::NoTls)
}
}

View File

@@ -117,8 +117,6 @@ impl AzureBlobStorage {
) -> Result<Download, DownloadError> {
let mut response = builder.into_stream();
let mut etag = None;
let mut last_modified = None;
let mut metadata = HashMap::new();
// TODO give proper streaming response instead of buffering into RAM
// https://github.com/neondatabase/neon/issues/5563
@@ -126,13 +124,6 @@ impl AzureBlobStorage {
let mut bufs = Vec::new();
while let Some(part) = response.next().await {
let part = part.map_err(to_download_error)?;
let etag_str: &str = part.blob.properties.etag.as_ref();
if etag.is_none() {
etag = Some(etag.unwrap_or_else(|| etag_str.to_owned()));
}
if last_modified.is_none() {
last_modified = Some(part.blob.properties.last_modified.into());
}
if let Some(blob_meta) = part.blob.metadata {
metadata.extend(blob_meta.iter().map(|(k, v)| (k.to_owned(), v.to_owned())));
}
@@ -145,8 +136,6 @@ impl AzureBlobStorage {
}
Ok(Download {
download_stream: Box::pin(futures::stream::iter(bufs.into_iter().map(Ok))),
etag,
last_modified,
metadata: Some(StorageMetadata(metadata)),
})
}
@@ -322,12 +311,6 @@ impl RemoteStorage for AzureBlobStorage {
}
Ok(())
}
async fn copy(&self, _from: &RemotePath, _to: &RemotePath) -> anyhow::Result<()> {
Err(anyhow::anyhow!(
"copy for azure blob storage is not implemented"
))
}
}
pin_project_lite::pin_project! {

View File

@@ -14,9 +14,7 @@ mod local_fs;
mod s3_bucket;
mod simulate_failures;
use std::{
collections::HashMap, fmt::Debug, num::NonZeroUsize, pin::Pin, sync::Arc, time::SystemTime,
};
use std::{collections::HashMap, fmt::Debug, num::NonZeroUsize, pin::Pin, sync::Arc};
use anyhow::{bail, Context};
use camino::{Utf8Path, Utf8PathBuf};
@@ -207,18 +205,10 @@ pub trait RemoteStorage: Send + Sync + 'static {
async fn delete(&self, path: &RemotePath) -> anyhow::Result<()>;
async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()>;
/// Copy a remote object inside a bucket from one path to another.
async fn copy(&self, from: &RemotePath, to: &RemotePath) -> anyhow::Result<()>;
}
pub type DownloadStream = Pin<Box<dyn Stream<Item = std::io::Result<Bytes>> + Unpin + Send + Sync>>;
pub struct Download {
pub download_stream: DownloadStream,
/// The last time the file was modified (`last-modified` HTTP header)
pub last_modified: Option<SystemTime>,
/// A way to identify this specific version of the resource (`etag` HTTP header)
pub etag: Option<String>,
pub download_stream: Pin<Box<dyn Stream<Item = std::io::Result<Bytes>> + Unpin + Send + Sync>>,
/// Extra key-value data, associated with the current remote file.
pub metadata: Option<StorageMetadata>,
}
@@ -377,15 +367,6 @@ impl GenericRemoteStorage {
Self::Unreliable(s) => s.delete_objects(paths).await,
}
}
pub async fn copy_object(&self, from: &RemotePath, to: &RemotePath) -> anyhow::Result<()> {
match self {
Self::LocalFs(s) => s.copy(from, to).await,
Self::AwsS3(s) => s.copy(from, to).await,
Self::AzureBlob(s) => s.copy(from, to).await,
Self::Unreliable(s) => s.copy(from, to).await,
}
}
}
impl GenericRemoteStorage {
@@ -672,7 +653,6 @@ impl ConcurrencyLimiter {
RequestKind::Put => &self.write,
RequestKind::List => &self.read,
RequestKind::Delete => &self.write,
RequestKind::Copy => &self.write,
}
}

View File

@@ -18,7 +18,7 @@ use tokio_util::io::ReaderStream;
use tracing::*;
use utils::{crashsafe::path_with_suffix_extension, fs_ext::is_directory_empty};
use crate::{Download, DownloadError, DownloadStream, Listing, ListingMode, RemotePath};
use crate::{Download, DownloadError, Listing, ListingMode, RemotePath};
use super::{RemoteStorage, StorageMetadata};
@@ -331,8 +331,6 @@ impl RemoteStorage for LocalFs {
.map_err(DownloadError::Other)?;
Ok(Download {
metadata,
last_modified: None,
etag: None,
download_stream: Box::pin(source),
})
} else {
@@ -374,17 +372,17 @@ impl RemoteStorage for LocalFs {
.await
.map_err(DownloadError::Other)?;
let download_stream: DownloadStream = match end_exclusive {
Some(end_exclusive) => Box::pin(ReaderStream::new(
source.take(end_exclusive - start_inclusive),
)),
None => Box::pin(ReaderStream::new(source)),
};
Ok(Download {
metadata,
last_modified: None,
etag: None,
download_stream,
Ok(match end_exclusive {
Some(end_exclusive) => Download {
metadata,
download_stream: Box::pin(ReaderStream::new(
source.take(end_exclusive - start_inclusive),
)),
},
None => Download {
metadata,
download_stream: Box::pin(ReaderStream::new(source)),
},
})
} else {
Err(DownloadError::NotFound)
@@ -409,20 +407,6 @@ impl RemoteStorage for LocalFs {
}
Ok(())
}
async fn copy(&self, from: &RemotePath, to: &RemotePath) -> anyhow::Result<()> {
let from_path = from.with_base(&self.storage_root);
let to_path = to.with_base(&self.storage_root);
create_target_directory(&to_path).await?;
fs::copy(&from_path, &to_path).await.with_context(|| {
format!(
"Failed to copy file from '{from_path}' to '{to_path}'",
from_path = from_path,
to_path = to_path
)
})?;
Ok(())
}
}
fn storage_metadata_path(original_path: &Utf8Path) -> Utf8PathBuf {

View File

@@ -16,7 +16,6 @@ use aws_config::{
environment::credentials::EnvironmentVariableCredentialsProvider,
imds::credentials::ImdsCredentialsProvider,
meta::credentials::CredentialsProviderChain,
profile::ProfileFileCredentialsProvider,
provider_config::ProviderConfig,
retry::{RetryConfigBuilder, RetryMode},
web_identity_token::WebIdentityTokenCredentialsProvider,
@@ -75,29 +74,20 @@ impl S3Bucket {
let region = Some(Region::new(aws_config.bucket_region.clone()));
let provider_conf = ProviderConfig::without_region().with_region(region.clone());
let credentials_provider = {
// uses "AWS_ACCESS_KEY_ID", "AWS_SECRET_ACCESS_KEY"
CredentialsProviderChain::first_try(
"env",
EnvironmentVariableCredentialsProvider::new(),
)
// uses "AWS_PROFILE" / `aws sso login --profile <profile>`
.or_else(
"profile-sso",
ProfileFileCredentialsProvider::builder()
.configure(&provider_conf)
.build(),
)
// uses "AWS_WEB_IDENTITY_TOKEN_FILE", "AWS_ROLE_ARN", "AWS_ROLE_SESSION_NAME"
// needed to access remote extensions bucket
.or_else(
"token",
.or_else("token", {
let provider_conf = ProviderConfig::without_region().with_region(region.clone());
WebIdentityTokenCredentialsProvider::builder()
.configure(&provider_conf)
.build(),
)
.build()
})
// uses imds v2
.or_else("imds", ImdsCredentialsProvider::builder().build())
};
@@ -228,11 +218,17 @@ impl S3Bucket {
let started_at = ScopeGuard::into_inner(started_at);
if get_object.is_err() {
metrics::BUCKET_METRICS.req_seconds.observe_elapsed(
kind,
AttemptOutcome::Err,
started_at,
);
}
match get_object {
Ok(object_output) => {
let metadata = object_output.metadata().cloned().map(StorageMetadata);
let etag = object_output.e_tag.clone();
let last_modified = object_output.last_modified.and_then(|t| t.try_into().ok());
let body = object_output.body;
let body = ByteStreamAsStream::from(body);
@@ -241,33 +237,15 @@ impl S3Bucket {
Ok(Download {
metadata,
etag,
last_modified,
download_stream: Box::pin(body),
})
}
Err(SdkError::ServiceError(e)) if matches!(e.err(), GetObjectError::NoSuchKey(_)) => {
// Count this in the AttemptOutcome::Ok bucket, because 404 is not
// an error: we expect to sometimes fetch an object and find it missing,
// e.g. when probing for timeline indices.
metrics::BUCKET_METRICS.req_seconds.observe_elapsed(
kind,
AttemptOutcome::Ok,
started_at,
);
Err(DownloadError::NotFound)
}
Err(e) => {
metrics::BUCKET_METRICS.req_seconds.observe_elapsed(
kind,
AttemptOutcome::Err,
started_at,
);
Err(DownloadError::Other(
anyhow::Error::new(e).context("download s3 object"),
))
}
Err(e) => Err(DownloadError::Other(
anyhow::Error::new(e).context("download s3 object"),
)),
}
}
}
@@ -493,38 +471,6 @@ impl RemoteStorage for S3Bucket {
Ok(())
}
async fn copy(&self, from: &RemotePath, to: &RemotePath) -> anyhow::Result<()> {
let kind = RequestKind::Copy;
let _guard = self.permit(kind).await;
let started_at = start_measuring_requests(kind);
// we need to specify bucket_name as a prefix
let copy_source = format!(
"{}/{}",
self.bucket_name,
self.relative_path_to_s3_object(from)
);
let res = self
.client
.copy_object()
.bucket(self.bucket_name.clone())
.key(self.relative_path_to_s3_object(to))
.copy_source(copy_source)
.send()
.await;
let started_at = ScopeGuard::into_inner(started_at);
metrics::BUCKET_METRICS
.req_seconds
.observe_elapsed(kind, &res, started_at);
res?;
Ok(())
}
async fn download(&self, from: &RemotePath) -> Result<Download, DownloadError> {
// if prefix is not none then download file `prefix/from`
// if prefix is none then download file `from`

View File

@@ -11,7 +11,6 @@ pub(crate) enum RequestKind {
Put = 1,
Delete = 2,
List = 3,
Copy = 4,
}
use RequestKind::*;
@@ -23,7 +22,6 @@ impl RequestKind {
Put => "put_object",
Delete => "delete_object",
List => "list_objects",
Copy => "copy_object",
}
}
const fn as_index(&self) -> usize {
@@ -31,7 +29,7 @@ impl RequestKind {
}
}
pub(super) struct RequestTyped<C>([C; 5]);
pub(super) struct RequestTyped<C>([C; 4]);
impl<C> RequestTyped<C> {
pub(super) fn get(&self, kind: RequestKind) -> &C {
@@ -40,8 +38,8 @@ impl<C> RequestTyped<C> {
fn build_with(mut f: impl FnMut(RequestKind) -> C) -> Self {
use RequestKind::*;
let mut it = [Get, Put, Delete, List, Copy].into_iter();
let arr = std::array::from_fn::<C, 5, _>(|index| {
let mut it = [Get, Put, Delete, List].into_iter();
let arr = std::array::from_fn::<C, 4, _>(|index| {
let next = it.next().unwrap();
assert_eq!(index, next.as_index());
f(next)

View File

@@ -162,11 +162,4 @@ impl RemoteStorage for UnreliableWrapper {
}
Ok(())
}
async fn copy(&self, from: &RemotePath, to: &RemotePath) -> anyhow::Result<()> {
// copy is equivalent to download + upload
self.attempt(RemoteOp::Download(from.clone()))?;
self.attempt(RemoteOp::Upload(to.clone()))?;
self.inner.copy_object(from, to).await
}
}

View File

@@ -1,200 +0,0 @@
use std::collections::HashSet;
use std::ops::ControlFlow;
use std::path::PathBuf;
use std::sync::Arc;
use anyhow::Context;
use bytes::Bytes;
use camino::Utf8Path;
use futures::stream::Stream;
use once_cell::sync::OnceCell;
use remote_storage::{Download, GenericRemoteStorage, RemotePath};
use tokio::task::JoinSet;
use tracing::{debug, error, info};
static LOGGING_DONE: OnceCell<()> = OnceCell::new();
pub(crate) fn upload_stream(
content: std::borrow::Cow<'static, [u8]>,
) -> (
impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
usize,
) {
use std::borrow::Cow;
let content = match content {
Cow::Borrowed(x) => Bytes::from_static(x),
Cow::Owned(vec) => Bytes::from(vec),
};
wrap_stream(content)
}
pub(crate) fn wrap_stream(
content: bytes::Bytes,
) -> (
impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
usize,
) {
let len = content.len();
let content = futures::future::ready(Ok(content));
(futures::stream::once(content), len)
}
pub(crate) async fn download_to_vec(dl: Download) -> anyhow::Result<Vec<u8>> {
let mut buf = Vec::new();
tokio::io::copy_buf(
&mut tokio_util::io::StreamReader::new(dl.download_stream),
&mut buf,
)
.await?;
Ok(buf)
}
// Uploads files `folder{j}/blob{i}.txt`. See test description for more details.
pub(crate) async fn upload_simple_remote_data(
client: &Arc<GenericRemoteStorage>,
upload_tasks_count: usize,
) -> ControlFlow<HashSet<RemotePath>, HashSet<RemotePath>> {
info!("Creating {upload_tasks_count} remote files");
let mut upload_tasks = JoinSet::new();
for i in 1..upload_tasks_count + 1 {
let task_client = Arc::clone(client);
upload_tasks.spawn(async move {
let blob_path = PathBuf::from(format!("folder{}/blob_{}.txt", i / 7, i));
let blob_path = RemotePath::new(
Utf8Path::from_path(blob_path.as_path()).expect("must be valid blob path"),
)
.with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
debug!("Creating remote item {i} at path {blob_path:?}");
let (data, len) = upload_stream(format!("remote blob data {i}").into_bytes().into());
task_client.upload(data, len, &blob_path, None).await?;
Ok::<_, anyhow::Error>(blob_path)
});
}
let mut upload_tasks_failed = false;
let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
while let Some(task_run_result) = upload_tasks.join_next().await {
match task_run_result
.context("task join failed")
.and_then(|task_result| task_result.context("upload task failed"))
{
Ok(upload_path) => {
uploaded_blobs.insert(upload_path);
}
Err(e) => {
error!("Upload task failed: {e:?}");
upload_tasks_failed = true;
}
}
}
if upload_tasks_failed {
ControlFlow::Break(uploaded_blobs)
} else {
ControlFlow::Continue(uploaded_blobs)
}
}
pub(crate) async fn cleanup(
client: &Arc<GenericRemoteStorage>,
objects_to_delete: HashSet<RemotePath>,
) {
info!(
"Removing {} objects from the remote storage during cleanup",
objects_to_delete.len()
);
let mut delete_tasks = JoinSet::new();
for object_to_delete in objects_to_delete {
let task_client = Arc::clone(client);
delete_tasks.spawn(async move {
debug!("Deleting remote item at path {object_to_delete:?}");
task_client
.delete(&object_to_delete)
.await
.with_context(|| format!("{object_to_delete:?} removal"))
});
}
while let Some(task_run_result) = delete_tasks.join_next().await {
match task_run_result {
Ok(task_result) => match task_result {
Ok(()) => {}
Err(e) => error!("Delete task failed: {e:?}"),
},
Err(join_err) => error!("Delete task did not finish correctly: {join_err}"),
}
}
}
pub(crate) struct Uploads {
pub(crate) prefixes: HashSet<RemotePath>,
pub(crate) blobs: HashSet<RemotePath>,
}
pub(crate) async fn upload_remote_data(
client: &Arc<GenericRemoteStorage>,
base_prefix_str: &'static str,
upload_tasks_count: usize,
) -> ControlFlow<Uploads, Uploads> {
info!("Creating {upload_tasks_count} remote files");
let mut upload_tasks = JoinSet::new();
for i in 1..upload_tasks_count + 1 {
let task_client = Arc::clone(client);
upload_tasks.spawn(async move {
let prefix = format!("{base_prefix_str}/sub_prefix_{i}/");
let blob_prefix = RemotePath::new(Utf8Path::new(&prefix))
.with_context(|| format!("{prefix:?} to RemotePath conversion"))?;
let blob_path = blob_prefix.join(Utf8Path::new(&format!("blob_{i}")));
debug!("Creating remote item {i} at path {blob_path:?}");
let (data, data_len) =
upload_stream(format!("remote blob data {i}").into_bytes().into());
task_client.upload(data, data_len, &blob_path, None).await?;
Ok::<_, anyhow::Error>((blob_prefix, blob_path))
});
}
let mut upload_tasks_failed = false;
let mut uploaded_prefixes = HashSet::with_capacity(upload_tasks_count);
let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
while let Some(task_run_result) = upload_tasks.join_next().await {
match task_run_result
.context("task join failed")
.and_then(|task_result| task_result.context("upload task failed"))
{
Ok((upload_prefix, upload_path)) => {
uploaded_prefixes.insert(upload_prefix);
uploaded_blobs.insert(upload_path);
}
Err(e) => {
error!("Upload task failed: {e:?}");
upload_tasks_failed = true;
}
}
}
let uploads = Uploads {
prefixes: uploaded_prefixes,
blobs: uploaded_blobs,
};
if upload_tasks_failed {
ControlFlow::Break(uploads)
} else {
ControlFlow::Continue(uploads)
}
}
pub(crate) fn ensure_logging_ready() {
LOGGING_DONE.get_or_init(|| {
utils::logging::init(
utils::logging::LogFormat::Test,
utils::logging::TracingErrorLayerEnablement::Disabled,
utils::logging::Output::Stdout,
)
.expect("logging init failed");
});
}

View File

@@ -2,23 +2,23 @@ use std::collections::HashSet;
use std::env;
use std::num::NonZeroUsize;
use std::ops::ControlFlow;
use std::path::PathBuf;
use std::sync::Arc;
use std::time::UNIX_EPOCH;
use anyhow::Context;
use bytes::Bytes;
use camino::Utf8Path;
use futures::stream::Stream;
use once_cell::sync::OnceCell;
use remote_storage::{
AzureConfig, GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind,
AzureConfig, Download, GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind,
};
use test_context::{test_context, AsyncTestContext};
use tracing::{debug, info};
use tokio::task::JoinSet;
use tracing::{debug, error, info};
mod common;
use common::{
cleanup, download_to_vec, ensure_logging_ready, upload_remote_data, upload_simple_remote_data,
upload_stream, wrap_stream,
};
static LOGGING_DONE: OnceCell<()> = OnceCell::new();
const ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME: &str = "ENABLE_REAL_AZURE_REMOTE_STORAGE";
@@ -30,7 +30,7 @@ const BASE_PREFIX: &str = "test";
/// If real Azure tests are disabled, the test passes, skipping any real test run: currently, there's no way to mark the test ignored in runtime with the
/// deafult test framework, see https://github.com/rust-lang/rust/issues/68007 for details.
///
/// First, the test creates a set of Azure blobs with keys `/${random_prefix_part}/${base_prefix_str}/sub_prefix_${i}/blob_${i}` in [`upload_remote_data`]
/// First, the test creates a set of Azure blobs with keys `/${random_prefix_part}/${base_prefix_str}/sub_prefix_${i}/blob_${i}` in [`upload_azure_data`]
/// where
/// * `random_prefix_part` is set for the entire Azure client during the Azure client creation in [`create_azure_client`], to avoid multiple test runs interference
/// * `base_prefix_str` is a common prefix to use in the client requests: we would want to ensure that the client is able to list nested prefixes inside the bucket
@@ -97,7 +97,7 @@ async fn azure_pagination_should_work(
/// Uses real Azure and requires [`ENABLE_REAL_AZURE_REMOTE_STORAGE_ENV_VAR_NAME`] and related Azure cred env vars specified. Test will skip real code and pass if env vars not set.
/// See `Azure_pagination_should_work` for more information.
///
/// First, create a set of Azure objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_remote_data`]
/// First, create a set of Azure objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_azure_data`]
/// Then performs the following queries:
/// 1. `list_files(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
/// 2. `list_files("folder1")`. This should return all files `random_prefix/folder1/blob_{i}.txt`
@@ -218,9 +218,18 @@ async fn azure_upload_download_works(ctx: &mut MaybeEnabledAzure) -> anyhow::Res
ctx.client.upload(data, len, &path, None).await?;
async fn download_and_compare(dl: Download) -> anyhow::Result<Vec<u8>> {
let mut buf = Vec::new();
tokio::io::copy_buf(
&mut tokio_util::io::StreamReader::new(dl.download_stream),
&mut buf,
)
.await?;
Ok(buf)
}
// Normal download request
let dl = ctx.client.download(&path).await?;
let buf = download_to_vec(dl).await?;
let buf = download_and_compare(dl).await?;
assert_eq!(&buf, &orig);
// Full range (end specified)
@@ -228,12 +237,12 @@ async fn azure_upload_download_works(ctx: &mut MaybeEnabledAzure) -> anyhow::Res
.client
.download_byte_range(&path, 0, Some(len as u64))
.await?;
let buf = download_to_vec(dl).await?;
let buf = download_and_compare(dl).await?;
assert_eq!(&buf, &orig);
// partial range (end specified)
let dl = ctx.client.download_byte_range(&path, 4, Some(10)).await?;
let buf = download_to_vec(dl).await?;
let buf = download_and_compare(dl).await?;
assert_eq!(&buf, &orig[4..10]);
// partial range (end beyond real end)
@@ -241,17 +250,17 @@ async fn azure_upload_download_works(ctx: &mut MaybeEnabledAzure) -> anyhow::Res
.client
.download_byte_range(&path, 8, Some(len as u64 * 100))
.await?;
let buf = download_to_vec(dl).await?;
let buf = download_and_compare(dl).await?;
assert_eq!(&buf, &orig[8..]);
// Partial range (end unspecified)
let dl = ctx.client.download_byte_range(&path, 4, None).await?;
let buf = download_to_vec(dl).await?;
let buf = download_and_compare(dl).await?;
assert_eq!(&buf, &orig[4..]);
// Full range (end unspecified)
let dl = ctx.client.download_byte_range(&path, 0, None).await?;
let buf = download_to_vec(dl).await?;
let buf = download_and_compare(dl).await?;
assert_eq!(&buf, &orig);
debug!("Cleanup: deleting file at path {path:?}");
@@ -263,6 +272,17 @@ async fn azure_upload_download_works(ctx: &mut MaybeEnabledAzure) -> anyhow::Res
Ok(())
}
fn ensure_logging_ready() {
LOGGING_DONE.get_or_init(|| {
utils::logging::init(
utils::logging::LogFormat::Test,
utils::logging::TracingErrorLayerEnablement::Disabled,
utils::logging::Output::Stdout,
)
.expect("logging init failed");
});
}
struct EnabledAzure {
client: Arc<GenericRemoteStorage>,
base_prefix: &'static str,
@@ -332,7 +352,7 @@ impl AsyncTestContext for MaybeEnabledAzureWithTestBlobs {
let enabled = EnabledAzure::setup(Some(max_keys_in_list_response)).await;
match upload_remote_data(&enabled.client, enabled.base_prefix, upload_tasks_count).await {
match upload_azure_data(&enabled.client, enabled.base_prefix, upload_tasks_count).await {
ControlFlow::Continue(uploads) => {
info!("Remote objects created successfully");
@@ -394,7 +414,7 @@ impl AsyncTestContext for MaybeEnabledAzureWithSimpleTestBlobs {
let enabled = EnabledAzure::setup(Some(max_keys_in_list_response)).await;
match upload_simple_remote_data(&enabled.client, upload_tasks_count).await {
match upload_simple_azure_data(&enabled.client, upload_tasks_count).await {
ControlFlow::Continue(uploads) => {
info!("Remote objects created successfully");
@@ -458,3 +478,166 @@ fn create_azure_client(
GenericRemoteStorage::from_config(&remote_storage_config).context("remote storage init")?,
))
}
struct Uploads {
prefixes: HashSet<RemotePath>,
blobs: HashSet<RemotePath>,
}
async fn upload_azure_data(
client: &Arc<GenericRemoteStorage>,
base_prefix_str: &'static str,
upload_tasks_count: usize,
) -> ControlFlow<Uploads, Uploads> {
info!("Creating {upload_tasks_count} Azure files");
let mut upload_tasks = JoinSet::new();
for i in 1..upload_tasks_count + 1 {
let task_client = Arc::clone(client);
upload_tasks.spawn(async move {
let prefix = format!("{base_prefix_str}/sub_prefix_{i}/");
let blob_prefix = RemotePath::new(Utf8Path::new(&prefix))
.with_context(|| format!("{prefix:?} to RemotePath conversion"))?;
let blob_path = blob_prefix.join(Utf8Path::new(&format!("blob_{i}")));
debug!("Creating remote item {i} at path {blob_path:?}");
let (data, len) = upload_stream(format!("remote blob data {i}").into_bytes().into());
task_client.upload(data, len, &blob_path, None).await?;
Ok::<_, anyhow::Error>((blob_prefix, blob_path))
});
}
let mut upload_tasks_failed = false;
let mut uploaded_prefixes = HashSet::with_capacity(upload_tasks_count);
let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
while let Some(task_run_result) = upload_tasks.join_next().await {
match task_run_result
.context("task join failed")
.and_then(|task_result| task_result.context("upload task failed"))
{
Ok((upload_prefix, upload_path)) => {
uploaded_prefixes.insert(upload_prefix);
uploaded_blobs.insert(upload_path);
}
Err(e) => {
error!("Upload task failed: {e:?}");
upload_tasks_failed = true;
}
}
}
let uploads = Uploads {
prefixes: uploaded_prefixes,
blobs: uploaded_blobs,
};
if upload_tasks_failed {
ControlFlow::Break(uploads)
} else {
ControlFlow::Continue(uploads)
}
}
async fn cleanup(client: &Arc<GenericRemoteStorage>, objects_to_delete: HashSet<RemotePath>) {
info!(
"Removing {} objects from the remote storage during cleanup",
objects_to_delete.len()
);
let mut delete_tasks = JoinSet::new();
for object_to_delete in objects_to_delete {
let task_client = Arc::clone(client);
delete_tasks.spawn(async move {
debug!("Deleting remote item at path {object_to_delete:?}");
task_client
.delete(&object_to_delete)
.await
.with_context(|| format!("{object_to_delete:?} removal"))
});
}
while let Some(task_run_result) = delete_tasks.join_next().await {
match task_run_result {
Ok(task_result) => match task_result {
Ok(()) => {}
Err(e) => error!("Delete task failed: {e:?}"),
},
Err(join_err) => error!("Delete task did not finish correctly: {join_err}"),
}
}
}
// Uploads files `folder{j}/blob{i}.txt`. See test description for more details.
async fn upload_simple_azure_data(
client: &Arc<GenericRemoteStorage>,
upload_tasks_count: usize,
) -> ControlFlow<HashSet<RemotePath>, HashSet<RemotePath>> {
info!("Creating {upload_tasks_count} Azure files");
let mut upload_tasks = JoinSet::new();
for i in 1..upload_tasks_count + 1 {
let task_client = Arc::clone(client);
upload_tasks.spawn(async move {
let blob_path = PathBuf::from(format!("folder{}/blob_{}.txt", i / 7, i));
let blob_path = RemotePath::new(
Utf8Path::from_path(blob_path.as_path()).expect("must be valid blob path"),
)
.with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
debug!("Creating remote item {i} at path {blob_path:?}");
let (data, len) = upload_stream(format!("remote blob data {i}").into_bytes().into());
task_client.upload(data, len, &blob_path, None).await?;
Ok::<_, anyhow::Error>(blob_path)
});
}
let mut upload_tasks_failed = false;
let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
while let Some(task_run_result) = upload_tasks.join_next().await {
match task_run_result
.context("task join failed")
.and_then(|task_result| task_result.context("upload task failed"))
{
Ok(upload_path) => {
uploaded_blobs.insert(upload_path);
}
Err(e) => {
error!("Upload task failed: {e:?}");
upload_tasks_failed = true;
}
}
}
if upload_tasks_failed {
ControlFlow::Break(uploaded_blobs)
} else {
ControlFlow::Continue(uploaded_blobs)
}
}
// FIXME: copypasted from test_real_s3, can't remember how to share a module which is not compiled
// to binary
fn upload_stream(
content: std::borrow::Cow<'static, [u8]>,
) -> (
impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
usize,
) {
use std::borrow::Cow;
let content = match content {
Cow::Borrowed(x) => Bytes::from_static(x),
Cow::Owned(vec) => Bytes::from(vec),
};
wrap_stream(content)
}
fn wrap_stream(
content: bytes::Bytes,
) -> (
impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
usize,
) {
let len = content.len();
let content = futures::future::ready(Ok(content));
(futures::stream::once(content), len)
}

View File

@@ -2,23 +2,23 @@ use std::collections::HashSet;
use std::env;
use std::num::NonZeroUsize;
use std::ops::ControlFlow;
use std::path::PathBuf;
use std::sync::Arc;
use std::time::UNIX_EPOCH;
use anyhow::Context;
use bytes::Bytes;
use camino::Utf8Path;
use futures::stream::Stream;
use once_cell::sync::OnceCell;
use remote_storage::{
GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind, S3Config,
};
use test_context::{test_context, AsyncTestContext};
use tracing::{debug, info};
use tokio::task::JoinSet;
use tracing::{debug, error, info};
mod common;
use common::{
cleanup, download_to_vec, ensure_logging_ready, upload_remote_data, upload_simple_remote_data,
upload_stream, wrap_stream,
};
static LOGGING_DONE: OnceCell<()> = OnceCell::new();
const ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME: &str = "ENABLE_REAL_S3_REMOTE_STORAGE";
@@ -30,7 +30,7 @@ const BASE_PREFIX: &str = "test";
/// If real S3 tests are disabled, the test passes, skipping any real test run: currently, there's no way to mark the test ignored in runtime with the
/// deafult test framework, see https://github.com/rust-lang/rust/issues/68007 for details.
///
/// First, the test creates a set of S3 objects with keys `/${random_prefix_part}/${base_prefix_str}/sub_prefix_${i}/blob_${i}` in [`upload_remote_data`]
/// First, the test creates a set of S3 objects with keys `/${random_prefix_part}/${base_prefix_str}/sub_prefix_${i}/blob_${i}` in [`upload_s3_data`]
/// where
/// * `random_prefix_part` is set for the entire S3 client during the S3 client creation in [`create_s3_client`], to avoid multiple test runs interference
/// * `base_prefix_str` is a common prefix to use in the client requests: we would want to ensure that the client is able to list nested prefixes inside the bucket
@@ -95,7 +95,7 @@ async fn s3_pagination_should_work(ctx: &mut MaybeEnabledS3WithTestBlobs) -> any
/// Uses real S3 and requires [`ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME`] and related S3 cred env vars specified. Test will skip real code and pass if env vars not set.
/// See `s3_pagination_should_work` for more information.
///
/// First, create a set of S3 objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_remote_data`]
/// First, create a set of S3 objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_s3_data`]
/// Then performs the following queries:
/// 1. `list_files(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
/// 2. `list_files("folder1")`. This should return all files `random_prefix/folder1/blob_{i}.txt`
@@ -198,65 +198,15 @@ async fn s3_delete_objects_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()>
Ok(())
}
#[test_context(MaybeEnabledS3)]
#[tokio::test]
async fn s3_upload_download_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()> {
let MaybeEnabledS3::Enabled(ctx) = ctx else {
return Ok(());
};
let path = RemotePath::new(Utf8Path::new(format!("{}/file", ctx.base_prefix).as_str()))
.with_context(|| "RemotePath conversion")?;
let orig = bytes::Bytes::from_static("remote blob data here".as_bytes());
let (data, len) = wrap_stream(orig.clone());
ctx.client.upload(data, len, &path, None).await?;
// Normal download request
let dl = ctx.client.download(&path).await?;
let buf = download_to_vec(dl).await?;
assert_eq!(&buf, &orig);
// Full range (end specified)
let dl = ctx
.client
.download_byte_range(&path, 0, Some(len as u64))
.await?;
let buf = download_to_vec(dl).await?;
assert_eq!(&buf, &orig);
// partial range (end specified)
let dl = ctx.client.download_byte_range(&path, 4, Some(10)).await?;
let buf = download_to_vec(dl).await?;
assert_eq!(&buf, &orig[4..10]);
// partial range (end beyond real end)
let dl = ctx
.client
.download_byte_range(&path, 8, Some(len as u64 * 100))
.await?;
let buf = download_to_vec(dl).await?;
assert_eq!(&buf, &orig[8..]);
// Partial range (end unspecified)
let dl = ctx.client.download_byte_range(&path, 4, None).await?;
let buf = download_to_vec(dl).await?;
assert_eq!(&buf, &orig[4..]);
// Full range (end unspecified)
let dl = ctx.client.download_byte_range(&path, 0, None).await?;
let buf = download_to_vec(dl).await?;
assert_eq!(&buf, &orig);
debug!("Cleanup: deleting file at path {path:?}");
ctx.client
.delete(&path)
.await
.with_context(|| format!("{path:?} removal"))?;
Ok(())
fn ensure_logging_ready() {
LOGGING_DONE.get_or_init(|| {
utils::logging::init(
utils::logging::LogFormat::Test,
utils::logging::TracingErrorLayerEnablement::Disabled,
utils::logging::Output::Stdout,
)
.expect("logging init failed");
});
}
struct EnabledS3 {
@@ -328,7 +278,7 @@ impl AsyncTestContext for MaybeEnabledS3WithTestBlobs {
let enabled = EnabledS3::setup(Some(max_keys_in_list_response)).await;
match upload_remote_data(&enabled.client, enabled.base_prefix, upload_tasks_count).await {
match upload_s3_data(&enabled.client, enabled.base_prefix, upload_tasks_count).await {
ControlFlow::Continue(uploads) => {
info!("Remote objects created successfully");
@@ -390,7 +340,7 @@ impl AsyncTestContext for MaybeEnabledS3WithSimpleTestBlobs {
let enabled = EnabledS3::setup(Some(max_keys_in_list_response)).await;
match upload_simple_remote_data(&enabled.client, upload_tasks_count).await {
match upload_simple_s3_data(&enabled.client, upload_tasks_count).await {
ControlFlow::Continue(uploads) => {
info!("Remote objects created successfully");
@@ -453,3 +403,166 @@ fn create_s3_client(
GenericRemoteStorage::from_config(&remote_storage_config).context("remote storage init")?,
))
}
struct Uploads {
prefixes: HashSet<RemotePath>,
blobs: HashSet<RemotePath>,
}
async fn upload_s3_data(
client: &Arc<GenericRemoteStorage>,
base_prefix_str: &'static str,
upload_tasks_count: usize,
) -> ControlFlow<Uploads, Uploads> {
info!("Creating {upload_tasks_count} S3 files");
let mut upload_tasks = JoinSet::new();
for i in 1..upload_tasks_count + 1 {
let task_client = Arc::clone(client);
upload_tasks.spawn(async move {
let prefix = format!("{base_prefix_str}/sub_prefix_{i}/");
let blob_prefix = RemotePath::new(Utf8Path::new(&prefix))
.with_context(|| format!("{prefix:?} to RemotePath conversion"))?;
let blob_path = blob_prefix.join(Utf8Path::new(&format!("blob_{i}")));
debug!("Creating remote item {i} at path {blob_path:?}");
let (data, data_len) =
upload_stream(format!("remote blob data {i}").into_bytes().into());
task_client.upload(data, data_len, &blob_path, None).await?;
Ok::<_, anyhow::Error>((blob_prefix, blob_path))
});
}
let mut upload_tasks_failed = false;
let mut uploaded_prefixes = HashSet::with_capacity(upload_tasks_count);
let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
while let Some(task_run_result) = upload_tasks.join_next().await {
match task_run_result
.context("task join failed")
.and_then(|task_result| task_result.context("upload task failed"))
{
Ok((upload_prefix, upload_path)) => {
uploaded_prefixes.insert(upload_prefix);
uploaded_blobs.insert(upload_path);
}
Err(e) => {
error!("Upload task failed: {e:?}");
upload_tasks_failed = true;
}
}
}
let uploads = Uploads {
prefixes: uploaded_prefixes,
blobs: uploaded_blobs,
};
if upload_tasks_failed {
ControlFlow::Break(uploads)
} else {
ControlFlow::Continue(uploads)
}
}
async fn cleanup(client: &Arc<GenericRemoteStorage>, objects_to_delete: HashSet<RemotePath>) {
info!(
"Removing {} objects from the remote storage during cleanup",
objects_to_delete.len()
);
let mut delete_tasks = JoinSet::new();
for object_to_delete in objects_to_delete {
let task_client = Arc::clone(client);
delete_tasks.spawn(async move {
debug!("Deleting remote item at path {object_to_delete:?}");
task_client
.delete(&object_to_delete)
.await
.with_context(|| format!("{object_to_delete:?} removal"))
});
}
while let Some(task_run_result) = delete_tasks.join_next().await {
match task_run_result {
Ok(task_result) => match task_result {
Ok(()) => {}
Err(e) => error!("Delete task failed: {e:?}"),
},
Err(join_err) => error!("Delete task did not finish correctly: {join_err}"),
}
}
}
// Uploads files `folder{j}/blob{i}.txt`. See test description for more details.
async fn upload_simple_s3_data(
client: &Arc<GenericRemoteStorage>,
upload_tasks_count: usize,
) -> ControlFlow<HashSet<RemotePath>, HashSet<RemotePath>> {
info!("Creating {upload_tasks_count} S3 files");
let mut upload_tasks = JoinSet::new();
for i in 1..upload_tasks_count + 1 {
let task_client = Arc::clone(client);
upload_tasks.spawn(async move {
let blob_path = PathBuf::from(format!("folder{}/blob_{}.txt", i / 7, i));
let blob_path = RemotePath::new(
Utf8Path::from_path(blob_path.as_path()).expect("must be valid blob path"),
)
.with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
debug!("Creating remote item {i} at path {blob_path:?}");
let (data, data_len) =
upload_stream(format!("remote blob data {i}").into_bytes().into());
task_client.upload(data, data_len, &blob_path, None).await?;
Ok::<_, anyhow::Error>(blob_path)
});
}
let mut upload_tasks_failed = false;
let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
while let Some(task_run_result) = upload_tasks.join_next().await {
match task_run_result
.context("task join failed")
.and_then(|task_result| task_result.context("upload task failed"))
{
Ok(upload_path) => {
uploaded_blobs.insert(upload_path);
}
Err(e) => {
error!("Upload task failed: {e:?}");
upload_tasks_failed = true;
}
}
}
if upload_tasks_failed {
ControlFlow::Break(uploaded_blobs)
} else {
ControlFlow::Continue(uploaded_blobs)
}
}
fn upload_stream(
content: std::borrow::Cow<'static, [u8]>,
) -> (
impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
usize,
) {
use std::borrow::Cow;
let content = match content {
Cow::Borrowed(x) => Bytes::from_static(x),
Cow::Owned(vec) => Bytes::from(vec),
};
wrap_stream(content)
}
fn wrap_stream(
content: bytes::Bytes,
) -> (
impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
usize,
) {
let len = content.len();
let content = futures::future::ready(Ok(content));
(futures::stream::once(content), len)
}

View File

@@ -51,9 +51,3 @@ pub struct SkTimelineInfo {
#[serde(default)]
pub http_connstr: Option<String>,
}
#[derive(Debug, Clone, Deserialize, Serialize)]
pub struct TimelineCopyRequest {
pub target_timeline_id: TimelineId,
pub until_lsn: Lsn,
}

View File

@@ -4,12 +4,6 @@ version = "0.1.0"
edition.workspace = true
license.workspace = true
[features]
default = []
# Enables test-only APIs, incuding failpoints. In particular, enables the `fail_point!` macro,
# which adds some runtime cost to run tests on outage conditions
testing = ["fail/failpoints"]
[dependencies]
arc-swap.workspace = true
sentry.workspace = true
@@ -22,7 +16,6 @@ chrono.workspace = true
heapless.workspace = true
hex = { workspace = true, features = ["serde"] }
hyper = { workspace = true, features = ["full"] }
fail.workspace = true
futures = { workspace = true}
jsonwebtoken.workspace = true
nix.workspace = true

View File

@@ -31,9 +31,6 @@ pub enum ApiError {
#[error("Shutting down")]
ShuttingDown,
#[error("Timeout")]
Timeout(Cow<'static, str>),
#[error(transparent)]
InternalServerError(anyhow::Error),
}
@@ -70,10 +67,6 @@ impl ApiError {
err.to_string(),
StatusCode::SERVICE_UNAVAILABLE,
),
ApiError::Timeout(err) => HttpErrorBody::response_from_msg_and_status(
err.to_string(),
StatusCode::REQUEST_TIMEOUT,
),
ApiError::InternalServerError(err) => HttpErrorBody::response_from_msg_and_status(
err.to_string(),
StatusCode::INTERNAL_SERVER_ERROR,

View File

@@ -83,10 +83,6 @@ pub mod timeout;
pub mod sync;
pub mod failpoint_support;
pub mod yielding_loop;
/// This is a shortcut to embed git sha into binaries and avoid copying the same build script to all packages
///
/// we have several cases:

View File

@@ -366,49 +366,6 @@ impl MonotonicCounter<Lsn> for RecordLsn {
}
}
/// Implements [`rand::distributions::uniform::UniformSampler`] so we can sample [`Lsn`]s.
///
/// This is used by the `pagebench` pageserver benchmarking tool.
pub struct LsnSampler(<u64 as rand::distributions::uniform::SampleUniform>::Sampler);
impl rand::distributions::uniform::SampleUniform for Lsn {
type Sampler = LsnSampler;
}
impl rand::distributions::uniform::UniformSampler for LsnSampler {
type X = Lsn;
fn new<B1, B2>(low: B1, high: B2) -> Self
where
B1: rand::distributions::uniform::SampleBorrow<Self::X> + Sized,
B2: rand::distributions::uniform::SampleBorrow<Self::X> + Sized,
{
Self(
<u64 as rand::distributions::uniform::SampleUniform>::Sampler::new(
low.borrow().0,
high.borrow().0,
),
)
}
fn new_inclusive<B1, B2>(low: B1, high: B2) -> Self
where
B1: rand::distributions::uniform::SampleBorrow<Self::X> + Sized,
B2: rand::distributions::uniform::SampleBorrow<Self::X> + Sized,
{
Self(
<u64 as rand::distributions::uniform::SampleUniform>::Sampler::new_inclusive(
low.borrow().0,
high.borrow().0,
),
)
}
fn sample<R: rand::prelude::Rng + ?Sized>(&self, rng: &mut R) -> Self::X {
Lsn(self.0.sample(rng))
}
}
#[cfg(test)]
mod tests {
use crate::bin_ser::BeSer;

View File

@@ -15,12 +15,6 @@ pub struct Gate {
name: String,
}
impl std::fmt::Debug for Gate {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
write!(f, "Gate<{}>", self.name)
}
}
/// RAII guard for a [`Gate`]: as long as this exists, calls to [`Gate::close`] will
/// not complete.
#[derive(Debug)]

View File

@@ -2,11 +2,8 @@ use std::time::Duration;
use tokio_util::sync::CancellationToken;
#[derive(thiserror::Error, Debug)]
pub enum TimeoutCancellableError {
#[error("Timed out")]
Timeout,
#[error("Cancelled")]
Cancelled,
}

View File

@@ -1,35 +0,0 @@
use tokio_util::sync::CancellationToken;
#[derive(thiserror::Error, Debug)]
pub enum YieldingLoopError {
#[error("Cancelled")]
Cancelled,
}
/// Helper for long synchronous loops, e.g. over all tenants in the system. Periodically
/// yields to avoid blocking the executor, and after resuming checks the provided
/// cancellation token to drop out promptly on shutdown.
#[inline(always)]
pub async fn yielding_loop<I, T, F>(
interval: usize,
cancel: &CancellationToken,
iter: I,
mut visitor: F,
) -> Result<(), YieldingLoopError>
where
I: Iterator<Item = T>,
F: FnMut(T),
{
for (i, item) in iter.enumerate() {
visitor(item);
if i + 1 % interval == 0 {
tokio::task::yield_now().await;
if cancel.is_cancelled() {
return Err(YieldingLoopError::Cancelled);
}
}
}
Ok(())
}

View File

@@ -1,2 +1 @@
#include "postgres.h"
#include "walproposer.h"

View File

@@ -1,6 +1,3 @@
//! Links with walproposer, pgcommon, pgport and runs bindgen on walproposer.h
//! to generate Rust bindings for it.
use std::{env, path::PathBuf, process::Command};
use anyhow::{anyhow, Context};

View File

@@ -1,6 +1,3 @@
//! A C-Rust shim: defines implementation of C walproposer API, assuming wp
//! callback_data stores Box to some Rust implementation.
#![allow(dead_code)]
use std::ffi::CStr;
@@ -8,12 +5,12 @@ use std::ffi::CString;
use crate::bindings::uint32;
use crate::bindings::walproposer_api;
use crate::bindings::NeonWALReadResult;
use crate::bindings::PGAsyncReadResult;
use crate::bindings::PGAsyncWriteResult;
use crate::bindings::Safekeeper;
use crate::bindings::Size;
use crate::bindings::StringInfoData;
use crate::bindings::TimeLineID;
use crate::bindings::TimestampTz;
use crate::bindings::WalProposer;
use crate::bindings::WalProposerConnStatusType;
@@ -178,11 +175,31 @@ extern "C" fn conn_blocking_write(
}
}
extern "C" fn recovery_download(wp: *mut WalProposer, sk: *mut Safekeeper) -> bool {
extern "C" fn recovery_download(
sk: *mut Safekeeper,
_timeline: TimeLineID,
startpos: XLogRecPtr,
endpos: XLogRecPtr,
) -> bool {
unsafe {
let callback_data = (*(*(*sk).wp).config).callback_data;
let api = callback_data as *mut Box<dyn ApiImpl>;
(*api).recovery_download(&mut (*wp), &mut (*sk))
(*api).recovery_download(&mut (*sk), startpos, endpos)
}
}
#[allow(clippy::unnecessary_cast)]
extern "C" fn wal_read(
sk: *mut Safekeeper,
buf: *mut ::std::os::raw::c_char,
startptr: XLogRecPtr,
count: Size,
) {
unsafe {
let buf = std::slice::from_raw_parts_mut(buf as *mut u8, count);
let callback_data = (*(*(*sk).wp).config).callback_data;
let api = callback_data as *mut Box<dyn ApiImpl>;
(*api).wal_read(&mut (*sk), buf, startptr)
}
}
@@ -194,28 +211,11 @@ extern "C" fn wal_reader_allocate(sk: *mut Safekeeper) {
}
}
#[allow(clippy::unnecessary_cast)]
extern "C" fn wal_read(
sk: *mut Safekeeper,
buf: *mut ::std::os::raw::c_char,
startptr: XLogRecPtr,
count: Size,
_errmsg: *mut *mut ::std::os::raw::c_char,
) -> NeonWALReadResult {
extern "C" fn free_event_set(wp: *mut WalProposer) {
unsafe {
let buf = std::slice::from_raw_parts_mut(buf as *mut u8, count);
let callback_data = (*(*(*sk).wp).config).callback_data;
let callback_data = (*(*wp).config).callback_data;
let api = callback_data as *mut Box<dyn ApiImpl>;
// TODO: errmsg is not forwarded
(*api).wal_read(&mut (*sk), buf, startptr)
}
}
extern "C" fn wal_reader_events(sk: *mut Safekeeper) -> uint32 {
unsafe {
let callback_data = (*(*(*sk).wp).config).callback_data;
let api = callback_data as *mut Box<dyn ApiImpl>;
(*api).wal_reader_events(&mut (*sk))
(*api).free_event_set(&mut (*wp));
}
}
@@ -235,14 +235,6 @@ extern "C" fn update_event_set(sk: *mut Safekeeper, events: uint32) {
}
}
extern "C" fn active_state_update_event_set(sk: *mut Safekeeper) {
unsafe {
let callback_data = (*(*(*sk).wp).config).callback_data;
let api = callback_data as *mut Box<dyn ApiImpl>;
(*api).active_state_update_event_set(&mut (*sk));
}
}
extern "C" fn add_safekeeper_event_set(sk: *mut Safekeeper, events: uint32) {
unsafe {
let callback_data = (*(*(*sk).wp).config).callback_data;
@@ -251,14 +243,6 @@ extern "C" fn add_safekeeper_event_set(sk: *mut Safekeeper, events: uint32) {
}
}
extern "C" fn rm_safekeeper_event_set(sk: *mut Safekeeper) {
unsafe {
let callback_data = (*(*(*sk).wp).config).callback_data;
let api = callback_data as *mut Box<dyn ApiImpl>;
(*api).rm_safekeeper_event_set(&mut (*sk));
}
}
extern "C" fn wait_event_set(
wp: *mut WalProposer,
timeout: ::std::os::raw::c_long,
@@ -326,6 +310,14 @@ extern "C" fn process_safekeeper_feedback(wp: *mut WalProposer, commit_lsn: XLog
}
}
extern "C" fn confirm_wal_streamed(wp: *mut WalProposer, lsn: XLogRecPtr) {
unsafe {
let callback_data = (*(*wp).config).callback_data;
let api = callback_data as *mut Box<dyn ApiImpl>;
(*api).confirm_wal_streamed(&mut (*wp), lsn)
}
}
extern "C" fn log_internal(
wp: *mut WalProposer,
level: ::std::os::raw::c_int,
@@ -340,6 +332,14 @@ extern "C" fn log_internal(
}
}
extern "C" fn after_election(wp: *mut WalProposer) {
unsafe {
let callback_data = (*(*wp).config).callback_data;
let api = callback_data as *mut Box<dyn ApiImpl>;
(*api).after_election(&mut (*wp))
}
}
#[derive(Debug)]
pub enum Level {
Debug5,
@@ -398,20 +398,20 @@ pub(crate) fn create_api() -> walproposer_api {
conn_async_write: Some(conn_async_write),
conn_blocking_write: Some(conn_blocking_write),
recovery_download: Some(recovery_download),
wal_reader_allocate: Some(wal_reader_allocate),
wal_read: Some(wal_read),
wal_reader_events: Some(wal_reader_events),
wal_reader_allocate: Some(wal_reader_allocate),
free_event_set: Some(free_event_set),
init_event_set: Some(init_event_set),
update_event_set: Some(update_event_set),
active_state_update_event_set: Some(active_state_update_event_set),
add_safekeeper_event_set: Some(add_safekeeper_event_set),
rm_safekeeper_event_set: Some(rm_safekeeper_event_set),
wait_event_set: Some(wait_event_set),
strong_random: Some(strong_random),
get_redo_start_lsn: Some(get_redo_start_lsn),
finish_sync_safekeepers: Some(finish_sync_safekeepers),
process_safekeeper_feedback: Some(process_safekeeper_feedback),
confirm_wal_streamed: Some(confirm_wal_streamed),
log_internal: Some(log_internal),
after_election: Some(after_election),
}
}

View File

@@ -6,8 +6,8 @@ use utils::id::TenantTimelineId;
use crate::{
api_bindings::{create_api, take_vec_u8, Level},
bindings::{
NeonWALReadResult, Safekeeper, WalProposer, WalProposerConfig, WalProposerCreate,
WalProposerFree, WalProposerStart,
Safekeeper, WalProposer, WalProposerConfig, WalProposerCreate, WalProposerFree,
WalProposerStart,
},
};
@@ -86,19 +86,19 @@ pub trait ApiImpl {
todo!()
}
fn recovery_download(&self, _wp: &mut WalProposer, _sk: &mut Safekeeper) -> bool {
fn recovery_download(&self, _sk: &mut Safekeeper, _startpos: u64, _endpos: u64) -> bool {
todo!()
}
fn wal_reader_allocate(&self, _sk: &mut Safekeeper) -> NeonWALReadResult {
fn wal_read(&self, _sk: &mut Safekeeper, _buf: &mut [u8], _startpos: u64) {
todo!()
}
fn wal_read(&self, _sk: &mut Safekeeper, _buf: &mut [u8], _startpos: u64) -> NeonWALReadResult {
fn wal_reader_allocate(&self, _sk: &mut Safekeeper) {
todo!()
}
fn wal_reader_events(&self, _sk: &mut Safekeeper) -> u32 {
fn free_event_set(&self, _wp: &mut WalProposer) {
todo!()
}
@@ -110,18 +110,10 @@ pub trait ApiImpl {
todo!()
}
fn active_state_update_event_set(&self, _sk: &mut Safekeeper) {
todo!()
}
fn add_safekeeper_event_set(&self, _sk: &mut Safekeeper, _events_mask: u32) {
todo!()
}
fn rm_safekeeper_event_set(&self, _sk: &mut Safekeeper) {
todo!()
}
fn wait_event_set(&self, _wp: &mut WalProposer, _timeout_millis: i64) -> WaitResult {
todo!()
}
@@ -142,6 +134,10 @@ pub trait ApiImpl {
todo!()
}
fn confirm_wal_streamed(&self, _wp: &mut WalProposer, _lsn: u64) {
todo!()
}
fn log_internal(&self, _wp: &mut WalProposer, _level: Level, _msg: &str) {
todo!()
}
@@ -244,7 +240,6 @@ impl Drop for Wrapper {
#[cfg(test)]
mod tests {
use core::panic;
use std::{
cell::Cell,
sync::{atomic::AtomicUsize, mpsc::sync_channel},
@@ -252,7 +247,7 @@ mod tests {
use utils::id::TenantTimelineId;
use crate::{api_bindings::Level, bindings::NeonWALReadResult, walproposer::Wrapper};
use crate::{api_bindings::Level, walproposer::Wrapper};
use super::ApiImpl;
@@ -360,17 +355,12 @@ mod tests {
true
}
fn recovery_download(
&self,
_wp: &mut crate::bindings::WalProposer,
_sk: &mut crate::bindings::Safekeeper,
) -> bool {
true
fn wal_reader_allocate(&self, _: &mut crate::bindings::Safekeeper) {
println!("wal_reader_allocate")
}
fn wal_reader_allocate(&self, _: &mut crate::bindings::Safekeeper) -> NeonWALReadResult {
println!("wal_reader_allocate");
crate::bindings::NeonWALReadResult_NEON_WALREAD_SUCCESS
fn free_event_set(&self, _: &mut crate::bindings::WalProposer) {
println!("free_event_set")
}
fn init_event_set(&self, _: &mut crate::bindings::WalProposer) {
@@ -393,13 +383,6 @@ mod tests {
self.wait_events.set(WaitEventsData { sk, event_mask });
}
fn rm_safekeeper_event_set(&self, sk: &mut crate::bindings::Safekeeper) {
println!(
"rm_safekeeper_event_set, sk={:?}",
sk as *mut crate::bindings::Safekeeper
);
}
fn wait_event_set(
&self,
_: &mut crate::bindings::WalProposer,
@@ -425,7 +408,7 @@ mod tests {
}
fn log_internal(&self, _wp: &mut crate::bindings::WalProposer, level: Level, msg: &str) {
println!("wp_log[{}] {}", level, msg);
println!("walprop_log[{}] {}", level, msg);
}
fn after_election(&self, _wp: &mut crate::bindings::WalProposer) {

View File

@@ -63,7 +63,6 @@ thiserror.workspace = true
tokio = { workspace = true, features = ["process", "sync", "fs", "rt", "io-util", "time"] }
tokio-io-timeout.workspace = true
tokio-postgres.workspace = true
tokio-stream.workspace = true
tokio-util.workspace = true
toml_edit = { workspace = true, features = [ "serde" ] }
tracing.workspace = true

View File

@@ -13,7 +13,6 @@ use bytes::{Buf, Bytes};
use pageserver::{
config::PageServerConf, repository::Key, walrecord::NeonWalRecord, walredo::PostgresRedoManager,
};
use pageserver_api::shard::TenantShardId;
use utils::{id::TenantId, lsn::Lsn};
use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion};
@@ -27,9 +26,9 @@ fn redo_scenarios(c: &mut Criterion) {
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 tenant_id = TenantId::generate();
let manager = PostgresRedoManager::new(conf, tenant_shard_id);
let manager = PostgresRedoManager::new(conf, tenant_id);
let manager = Arc::new(manager);

View File

@@ -1,22 +0,0 @@
[package]
name = "pageserver_client"
version = "0.1.0"
edition.workspace = true
license.workspace = true
[dependencies]
pageserver_api.workspace = true
thiserror.workspace = true
async-trait.workspace = true
reqwest.workspace = true
utils.workspace = true
serde.workspace = true
workspace_hack = { version = "0.1", path = "../../workspace_hack" }
tokio-postgres.workspace = true
tokio-stream.workspace = true
tokio.workspace = true
futures.workspace = true
tokio-util.workspace = true
anyhow.workspace = true
postgres.workspace = true
bytes.workspace = true

View File

@@ -1,2 +0,0 @@
pub mod mgmt_api;
pub mod page_service;

View File

@@ -1,214 +0,0 @@
use pageserver_api::{models::*, shard::TenantShardId};
use reqwest::{IntoUrl, Method};
use utils::{
http::error::HttpErrorBody,
id::{TenantId, TimelineId},
};
pub mod util;
#[derive(Debug)]
pub struct Client {
mgmt_api_endpoint: String,
authorization_header: Option<String>,
client: reqwest::Client,
}
#[derive(thiserror::Error, Debug)]
pub enum Error {
#[error("receive body: {0}")]
ReceiveBody(reqwest::Error),
#[error("receive error body: {0}")]
ReceiveErrorBody(String),
#[error("pageserver API: {0}")]
ApiError(String),
}
pub type Result<T> = std::result::Result<T, Error>;
#[async_trait::async_trait]
pub trait ResponseErrorMessageExt: Sized {
async fn error_from_body(self) -> Result<Self>;
}
#[async_trait::async_trait]
impl ResponseErrorMessageExt for reqwest::Response {
async fn error_from_body(mut self) -> Result<Self> {
let status = self.status();
if !(status.is_client_error() || status.is_server_error()) {
return Ok(self);
}
let url = self.url().to_owned();
Err(match self.json::<HttpErrorBody>().await {
Ok(HttpErrorBody { msg }) => Error::ApiError(msg),
Err(_) => {
Error::ReceiveErrorBody(format!("Http error ({}) at {}.", status.as_u16(), url))
}
})
}
}
impl Client {
pub fn new(mgmt_api_endpoint: String, jwt: Option<&str>) -> Self {
Self {
mgmt_api_endpoint,
authorization_header: jwt.map(|jwt| format!("Bearer {jwt}")),
client: reqwest::Client::new(),
}
}
pub async fn list_tenants(&self) -> Result<Vec<pageserver_api::models::TenantInfo>> {
let uri = format!("{}/v1/tenant", self.mgmt_api_endpoint);
let resp = self.get(&uri).await?;
resp.json().await.map_err(Error::ReceiveBody)
}
pub async fn tenant_details(
&self,
tenant_id: TenantId,
) -> Result<pageserver_api::models::TenantDetails> {
let uri = format!("{}/v1/tenant/{tenant_id}", self.mgmt_api_endpoint);
self.get(uri)
.await?
.json()
.await
.map_err(Error::ReceiveBody)
}
pub async fn list_timelines(
&self,
tenant_id: TenantId,
) -> Result<Vec<pageserver_api::models::TimelineInfo>> {
let uri = format!("{}/v1/tenant/{tenant_id}/timeline", self.mgmt_api_endpoint);
self.get(&uri)
.await?
.json()
.await
.map_err(Error::ReceiveBody)
}
pub async fn timeline_info(
&self,
tenant_id: TenantId,
timeline_id: TimelineId,
) -> Result<pageserver_api::models::TimelineInfo> {
let uri = format!(
"{}/v1/tenant/{tenant_id}/timeline/{timeline_id}",
self.mgmt_api_endpoint
);
self.get(&uri)
.await?
.json()
.await
.map_err(Error::ReceiveBody)
}
pub async fn keyspace(
&self,
tenant_id: TenantId,
timeline_id: TimelineId,
) -> Result<pageserver_api::models::partitioning::Partitioning> {
let uri = format!(
"{}/v1/tenant/{tenant_id}/timeline/{timeline_id}/keyspace",
self.mgmt_api_endpoint
);
self.get(&uri)
.await?
.json()
.await
.map_err(Error::ReceiveBody)
}
async fn get<U: IntoUrl>(&self, uri: U) -> Result<reqwest::Response> {
self.request(Method::GET, uri, ()).await
}
async fn request<B: serde::Serialize, U: reqwest::IntoUrl>(
&self,
method: Method,
uri: U,
body: B,
) -> Result<reqwest::Response> {
let req = self.client.request(method, uri);
let req = if let Some(value) = &self.authorization_header {
req.header(reqwest::header::AUTHORIZATION, value)
} else {
req
};
let res = req.json(&body).send().await.map_err(Error::ReceiveBody)?;
let response = res.error_from_body().await?;
Ok(response)
}
pub async fn status(&self) -> Result<()> {
let uri = format!("{}/v1/status", self.mgmt_api_endpoint);
self.get(&uri).await?;
Ok(())
}
pub async fn tenant_create(&self, req: &TenantCreateRequest) -> Result<TenantId> {
let uri = format!("{}/v1/tenant", self.mgmt_api_endpoint);
self.request(Method::POST, &uri, req)
.await?
.json()
.await
.map_err(Error::ReceiveBody)
}
pub async fn tenant_config(&self, req: &TenantConfigRequest) -> Result<()> {
let uri = format!("{}/v1/tenant/config", self.mgmt_api_endpoint);
self.request(Method::PUT, &uri, req).await?;
Ok(())
}
pub async fn tenant_secondary_download(&self, tenant_id: TenantShardId) -> Result<()> {
let uri = format!(
"{}/v1/tenant/{}/secondary/download",
self.mgmt_api_endpoint, tenant_id
);
self.request(Method::POST, &uri, ())
.await?
.error_for_status()
.map(|_| ())
.map_err(|e| Error::ApiError(format!("{}", e)))
}
pub async fn location_config(
&self,
tenant_id: TenantId,
config: LocationConfig,
flush_ms: Option<std::time::Duration>,
) -> Result<()> {
let req_body = TenantLocationConfigRequest { tenant_id, config };
let path = format!(
"{}/v1/tenant/{}/location_config",
self.mgmt_api_endpoint, tenant_id
);
let path = if let Some(flush_ms) = flush_ms {
format!("{}?flush_ms={}", path, flush_ms.as_millis())
} else {
path
};
self.request(Method::PUT, &path, &req_body).await?;
Ok(())
}
pub async fn timeline_create(
&self,
tenant_id: TenantId,
req: &TimelineCreateRequest,
) -> Result<TimelineInfo> {
let uri = format!(
"{}/v1/tenant/{}/timeline",
self.mgmt_api_endpoint, tenant_id
);
self.request(Method::POST, &uri, req)
.await?
.json()
.await
.map_err(Error::ReceiveBody)
}
}

View File

@@ -1,49 +0,0 @@
//! Helpers to do common higher-level tasks with the [`Client`].
use std::sync::Arc;
use tokio::task::JoinSet;
use utils::id::{TenantId, TenantTimelineId};
use super::Client;
/// Retrieve a list of all of the pageserver's timelines.
///
/// Fails if there are sharded tenants present on the pageserver.
pub async fn get_pageserver_tenant_timelines_unsharded(
api_client: &Arc<Client>,
) -> anyhow::Result<Vec<TenantTimelineId>> {
let mut timelines: Vec<TenantTimelineId> = Vec::new();
let mut tenants: Vec<TenantId> = Vec::new();
for ti in api_client.list_tenants().await? {
if !ti.id.is_unsharded() {
anyhow::bail!(
"only unsharded tenants are supported at this time: {}",
ti.id
);
}
tenants.push(ti.id.tenant_id)
}
let mut js = JoinSet::new();
for tenant_id in tenants {
js.spawn({
let mgmt_api_client = Arc::clone(api_client);
async move {
(
tenant_id,
mgmt_api_client.tenant_details(tenant_id).await.unwrap(),
)
}
});
}
while let Some(res) = js.join_next().await {
let (tenant_id, details) = res.unwrap();
for timeline_id in details.timelines {
timelines.push(TenantTimelineId {
tenant_id,
timeline_id,
});
}
}
Ok(timelines)
}

View File

@@ -1,144 +0,0 @@
use std::pin::Pin;
use futures::SinkExt;
use pageserver_api::{
models::{
PagestreamBeMessage, PagestreamFeMessage, PagestreamGetPageRequest,
PagestreamGetPageResponse,
},
reltag::RelTag,
};
use tokio::task::JoinHandle;
use tokio_postgres::CopyOutStream;
use tokio_stream::StreamExt;
use tokio_util::sync::CancellationToken;
use utils::{
id::{TenantId, TimelineId},
lsn::Lsn,
};
pub struct Client {
client: tokio_postgres::Client,
cancel_on_client_drop: Option<tokio_util::sync::DropGuard>,
conn_task: JoinHandle<()>,
}
pub struct BasebackupRequest {
pub tenant_id: TenantId,
pub timeline_id: TimelineId,
pub lsn: Option<Lsn>,
pub gzip: bool,
}
impl Client {
pub async fn new(connstring: String) -> anyhow::Result<Self> {
let (client, connection) = tokio_postgres::connect(&connstring, postgres::NoTls).await?;
let conn_task_cancel = CancellationToken::new();
let conn_task = tokio::spawn({
let conn_task_cancel = conn_task_cancel.clone();
async move {
tokio::select! {
_ = conn_task_cancel.cancelled() => { }
res = connection => {
res.unwrap();
}
}
}
});
Ok(Self {
cancel_on_client_drop: Some(conn_task_cancel.drop_guard()),
conn_task,
client,
})
}
pub async fn pagestream(
self,
tenant_id: TenantId,
timeline_id: TimelineId,
) -> anyhow::Result<PagestreamClient> {
let copy_both: tokio_postgres::CopyBothDuplex<bytes::Bytes> = self
.client
.copy_both_simple(&format!("pagestream {tenant_id} {timeline_id}"))
.await?;
let Client {
cancel_on_client_drop,
conn_task,
client: _,
} = self;
Ok(PagestreamClient {
copy_both: Box::pin(copy_both),
conn_task,
cancel_on_client_drop,
})
}
pub async fn basebackup(&self, req: &BasebackupRequest) -> anyhow::Result<CopyOutStream> {
let BasebackupRequest {
tenant_id,
timeline_id,
lsn,
gzip,
} = req;
let mut args = Vec::with_capacity(5);
args.push("basebackup".to_string());
args.push(format!("{tenant_id}"));
args.push(format!("{timeline_id}"));
if let Some(lsn) = lsn {
args.push(format!("{lsn}"));
}
if *gzip {
args.push("--gzip".to_string())
}
Ok(self.client.copy_out(&args.join(" ")).await?)
}
}
/// Create using [`Client::pagestream`].
pub struct PagestreamClient {
copy_both: Pin<Box<tokio_postgres::CopyBothDuplex<bytes::Bytes>>>,
cancel_on_client_drop: Option<tokio_util::sync::DropGuard>,
conn_task: JoinHandle<()>,
}
pub struct RelTagBlockNo {
pub rel_tag: RelTag,
pub block_no: u32,
}
impl PagestreamClient {
pub async fn shutdown(mut self) {
let _ = self.cancel_on_client_drop.take();
self.conn_task.await.unwrap();
}
pub async fn getpage(
&mut self,
req: PagestreamGetPageRequest,
) -> anyhow::Result<PagestreamGetPageResponse> {
let req = PagestreamFeMessage::GetPage(req);
let req: bytes::Bytes = req.serialize();
// let mut req = tokio_util::io::ReaderStream::new(&req);
let mut req = tokio_stream::once(Ok(req));
self.copy_both.send_all(&mut req).await?;
let next: Option<Result<bytes::Bytes, _>> = self.copy_both.next().await;
let next: bytes::Bytes = next.unwrap()?;
let msg = PagestreamBeMessage::deserialize(next)?;
match msg {
PagestreamBeMessage::GetPage(p) => Ok(p),
PagestreamBeMessage::Error(e) => anyhow::bail!("Error: {:?}", e),
PagestreamBeMessage::Exists(_)
| PagestreamBeMessage::Nblocks(_)
| PagestreamBeMessage::DbSize(_) => {
anyhow::bail!(
"unexpected be message kind in response to getpage request: {}",
msg.kind()
)
}
}
}
}

View File

@@ -1,26 +0,0 @@
[package]
name = "pagebench"
version = "0.1.0"
edition.workspace = true
license.workspace = true
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
anyhow.workspace = true
clap.workspace = true
futures.workspace = true
hdrhistogram.workspace = true
humantime.workspace = true
humantime-serde.workspace = true
rand.workspace = true
serde.workspace = true
serde_json.workspace = true
tracing.workspace = true
tokio.workspace = true
pageserver = { path = ".." }
pageserver_client.workspace = true
pageserver_api.workspace = true
utils = { path = "../../libs/utils/" }
workspace_hack = { version = "0.1", path = "../../workspace_hack" }

View File

@@ -1,272 +0,0 @@
use anyhow::Context;
use pageserver_client::page_service::BasebackupRequest;
use utils::id::TenantTimelineId;
use utils::lsn::Lsn;
use rand::prelude::*;
use tokio::sync::Barrier;
use tokio::task::JoinSet;
use tracing::{debug, info, instrument};
use std::collections::HashMap;
use std::num::NonZeroUsize;
use std::ops::Range;
use std::sync::atomic::{AtomicU64, AtomicUsize, Ordering};
use std::sync::{Arc, Mutex};
use std::time::Instant;
use crate::util::tokio_thread_local_stats::AllThreadLocalStats;
use crate::util::{request_stats, tokio_thread_local_stats};
/// basebackup@LatestLSN
#[derive(clap::Parser)]
pub(crate) struct Args {
#[clap(long, default_value = "http://localhost:9898")]
mgmt_api_endpoint: String,
#[clap(long, default_value = "localhost:64000")]
page_service_host_port: String,
#[clap(long)]
pageserver_jwt: Option<String>,
#[clap(long, default_value = "1")]
num_clients: NonZeroUsize,
#[clap(long, default_value = "1.0")]
gzip_probability: f64,
#[clap(long)]
runtime: Option<humantime::Duration>,
#[clap(long)]
limit_to_first_n_targets: Option<usize>,
targets: Option<Vec<TenantTimelineId>>,
}
#[derive(Debug, Default)]
struct LiveStats {
completed_requests: AtomicU64,
}
impl LiveStats {
fn inc(&self) {
self.completed_requests.fetch_add(1, Ordering::Relaxed);
}
}
struct Target {
timeline: TenantTimelineId,
lsn_range: Option<Range<Lsn>>,
}
#[derive(serde::Serialize)]
struct Output {
total: request_stats::Output,
}
tokio_thread_local_stats::declare!(STATS: request_stats::Stats);
pub(crate) fn main(args: Args) -> anyhow::Result<()> {
tokio_thread_local_stats::main!(STATS, move |thread_local_stats| {
main_impl(args, thread_local_stats)
})
}
async fn main_impl(
args: Args,
all_thread_local_stats: AllThreadLocalStats<request_stats::Stats>,
) -> 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(),
));
// 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 js = JoinSet::new();
for timeline in &timelines {
js.spawn({
let timeline = *timeline;
// FIXME: this triggers initial logical size calculation
// https://github.com/neondatabase/neon/issues/6168
let info = mgmt_api_client
.timeline_info(timeline.tenant_id, timeline.timeline_id)
.await
.unwrap();
async move {
anyhow::Ok(Target {
timeline,
// TODO: support lsn_range != latest LSN
lsn_range: Some(info.last_record_lsn..(info.last_record_lsn + 1)),
})
}
});
}
let mut all_targets: Vec<Target> = Vec::new();
while let Some(res) = js.join_next().await {
all_targets.push(res.unwrap().unwrap());
}
let live_stats = Arc::new(LiveStats::default());
let num_client_tasks = timelines.len();
let num_live_stats_dump = 1;
let num_work_sender_tasks = 1;
let start_work_barrier = Arc::new(tokio::sync::Barrier::new(
num_client_tasks + num_live_stats_dump + num_work_sender_tasks,
));
let all_work_done_barrier = Arc::new(tokio::sync::Barrier::new(num_client_tasks));
tokio::spawn({
let stats = Arc::clone(&live_stats);
let start_work_barrier = Arc::clone(&start_work_barrier);
async move {
start_work_barrier.wait().await;
loop {
let start = std::time::Instant::now();
tokio::time::sleep(std::time::Duration::from_secs(1)).await;
let completed_requests = stats.completed_requests.swap(0, Ordering::Relaxed);
let elapsed = start.elapsed();
info!(
"RPS: {:.0}",
completed_requests as f64 / elapsed.as_secs_f64()
);
}
}
});
let mut work_senders = HashMap::new();
let mut tasks = Vec::new();
for tl in &timelines {
let (sender, receiver) = tokio::sync::mpsc::channel(1); // TODO: not sure what the implications of this are
work_senders.insert(tl, sender);
tasks.push(tokio::spawn(client(
args,
*tl,
Arc::clone(&start_work_barrier),
receiver,
Arc::clone(&all_work_done_barrier),
Arc::clone(&live_stats),
)));
}
let work_sender = async move {
start_work_barrier.wait().await;
loop {
let (timeline, work) = {
let mut rng = rand::thread_rng();
let target = all_targets.choose(&mut rng).unwrap();
let lsn = target.lsn_range.clone().map(|r| rng.gen_range(r));
(
target.timeline,
Work {
lsn,
gzip: rng.gen_bool(args.gzip_probability),
},
)
};
let sender = work_senders.get(&timeline).unwrap();
// TODO: what if this blocks?
sender.send(work).await.ok().unwrap();
}
};
if let Some(runtime) = args.runtime {
match tokio::time::timeout(runtime.into(), work_sender).await {
Ok(()) => unreachable!("work sender never terminates"),
Err(_timeout) => {
// this implicitly drops the work_senders, making all the clients exit
}
}
} else {
work_sender.await;
unreachable!("work sender never terminates");
}
for t in tasks {
t.await.unwrap();
}
let output = Output {
total: {
let mut agg_stats = request_stats::Stats::new();
for stats in all_thread_local_stats.lock().unwrap().iter() {
let stats = stats.lock().unwrap();
agg_stats.add(&stats);
}
agg_stats.output()
},
};
let output = serde_json::to_string_pretty(&output).unwrap();
println!("{output}");
anyhow::Ok(())
}
#[derive(Copy, Clone)]
struct Work {
lsn: Option<Lsn>,
gzip: bool,
}
#[instrument(skip_all)]
async fn client(
args: &'static Args,
timeline: TenantTimelineId,
start_work_barrier: Arc<Barrier>,
mut work: tokio::sync::mpsc::Receiver<Work>,
all_work_done_barrier: Arc<Barrier>,
live_stats: Arc<LiveStats>,
) {
start_work_barrier.wait().await;
let client = pageserver_client::page_service::Client::new(crate::util::connstring::connstring(
&args.page_service_host_port,
args.pageserver_jwt.as_deref(),
))
.await
.unwrap();
while let Some(Work { lsn, gzip }) = work.recv().await {
let start = Instant::now();
let copy_out_stream = client
.basebackup(&BasebackupRequest {
tenant_id: timeline.tenant_id,
timeline_id: timeline.timeline_id,
lsn,
gzip,
})
.await
.with_context(|| format!("start basebackup for {timeline}"))
.unwrap();
use futures::StreamExt;
let size = Arc::new(AtomicUsize::new(0));
copy_out_stream
.for_each({
|r| {
let size = Arc::clone(&size);
async move {
let size = Arc::clone(&size);
size.fetch_add(r.unwrap().len(), Ordering::Relaxed);
}
}
})
.await;
debug!("basebackup size is {} bytes", size.load(Ordering::Relaxed));
let elapsed = start.elapsed();
live_stats.inc();
STATS.with(|stats| {
stats.borrow().lock().unwrap().observe(elapsed).unwrap();
});
}
all_work_done_barrier.wait().await;
}

View File

@@ -1,351 +0,0 @@
use anyhow::Context;
use futures::future::join_all;
use pageserver::pgdatadir_mapping::key_to_rel_block;
use pageserver::repository;
use pageserver_api::key::is_rel_block_key;
use pageserver_api::models::PagestreamGetPageRequest;
use utils::id::TenantTimelineId;
use utils::lsn::Lsn;
use rand::prelude::*;
use tokio::sync::Barrier;
use tokio::task::JoinSet;
use tracing::{info, instrument};
use std::collections::HashMap;
use std::future::Future;
use std::num::NonZeroUsize;
use std::pin::Pin;
use std::sync::atomic::{AtomicU64, Ordering};
use std::sync::{Arc, Mutex};
use std::time::{Duration, Instant};
use crate::util::tokio_thread_local_stats::AllThreadLocalStats;
use crate::util::{request_stats, tokio_thread_local_stats};
/// GetPage@LatestLSN, uniformly distributed across the compute-accessible keyspace.
#[derive(clap::Parser)]
pub(crate) struct Args {
#[clap(long, default_value = "http://localhost:9898")]
mgmt_api_endpoint: String,
#[clap(long, default_value = "postgres://postgres@localhost:64000")]
page_service_connstring: String,
#[clap(long)]
pageserver_jwt: Option<String>,
#[clap(long, default_value = "1")]
num_clients: NonZeroUsize,
#[clap(long)]
runtime: Option<humantime::Duration>,
#[clap(long)]
per_target_rate_limit: Option<usize>,
/// Probability for sending `latest=true` in the request (uniform distribution).
#[clap(long, default_value = "1")]
req_latest_probability: f64,
#[clap(long)]
limit_to_first_n_targets: Option<usize>,
targets: Option<Vec<TenantTimelineId>>,
}
#[derive(Debug, Default)]
struct LiveStats {
completed_requests: AtomicU64,
}
impl LiveStats {
fn inc(&self) {
self.completed_requests.fetch_add(1, Ordering::Relaxed);
}
}
#[derive(Clone)]
struct KeyRange {
timeline: TenantTimelineId,
timeline_lsn: Lsn,
start: i128,
end: i128,
}
impl KeyRange {
fn len(&self) -> i128 {
self.end - self.start
}
}
#[derive(serde::Serialize)]
struct Output {
total: request_stats::Output,
}
tokio_thread_local_stats::declare!(STATS: request_stats::Stats);
pub(crate) fn main(args: Args) -> anyhow::Result<()> {
tokio_thread_local_stats::main!(STATS, move |thread_local_stats| {
main_impl(args, thread_local_stats)
})
}
async fn main_impl(
args: Args,
all_thread_local_stats: AllThreadLocalStats<request_stats::Stats>,
) -> 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(),
));
// 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 js = JoinSet::new();
for timeline in &timelines {
js.spawn({
let mgmt_api_client = Arc::clone(&mgmt_api_client);
let timeline = *timeline;
async move {
let partitioning = mgmt_api_client
.keyspace(timeline.tenant_id, timeline.timeline_id)
.await?;
let lsn = partitioning.at_lsn;
let ranges = partitioning
.keys
.ranges
.iter()
.filter_map(|r| {
let start = r.start;
let end = r.end;
// filter out non-relblock keys
match (is_rel_block_key(&start), is_rel_block_key(&end)) {
(true, true) => Some(KeyRange {
timeline,
timeline_lsn: lsn,
start: start.to_i128(),
end: end.to_i128(),
}),
(true, false) | (false, true) => {
unimplemented!("split up range")
}
(false, false) => None,
}
})
.collect::<Vec<_>>();
anyhow::Ok(ranges)
}
});
}
let mut all_ranges: Vec<KeyRange> = Vec::new();
while let Some(res) = js.join_next().await {
all_ranges.extend(res.unwrap().unwrap());
}
let live_stats = Arc::new(LiveStats::default());
let num_client_tasks = timelines.len();
let num_live_stats_dump = 1;
let num_work_sender_tasks = 1;
let start_work_barrier = Arc::new(tokio::sync::Barrier::new(
num_client_tasks + num_live_stats_dump + num_work_sender_tasks,
));
let all_work_done_barrier = Arc::new(tokio::sync::Barrier::new(num_client_tasks));
tokio::spawn({
let stats = Arc::clone(&live_stats);
let start_work_barrier = Arc::clone(&start_work_barrier);
async move {
start_work_barrier.wait().await;
loop {
let start = std::time::Instant::now();
tokio::time::sleep(std::time::Duration::from_secs(1)).await;
let completed_requests = stats.completed_requests.swap(0, Ordering::Relaxed);
let elapsed = start.elapsed();
info!(
"RPS: {:.0}",
completed_requests as f64 / elapsed.as_secs_f64()
);
}
}
});
let mut work_senders = HashMap::new();
let mut tasks = Vec::new();
for tl in &timelines {
let (sender, receiver) = tokio::sync::mpsc::channel(10); // TODO: not sure what the implications of this are
work_senders.insert(tl, sender);
tasks.push(tokio::spawn(client(
args,
*tl,
Arc::clone(&start_work_barrier),
receiver,
Arc::clone(&all_work_done_barrier),
Arc::clone(&live_stats),
)));
}
let work_sender: Pin<Box<dyn Send + Future<Output = ()>>> = match args.per_target_rate_limit {
None => Box::pin(async move {
let weights = rand::distributions::weighted::WeightedIndex::new(
all_ranges.iter().map(|v| v.len()),
)
.unwrap();
start_work_barrier.wait().await;
loop {
let (timeline, req) = {
let mut rng = rand::thread_rng();
let r = &all_ranges[weights.sample(&mut rng)];
let key: i128 = rng.gen_range(r.start..r.end);
let key = repository::Key::from_i128(key);
let (rel_tag, block_no) =
key_to_rel_block(key).expect("we filter non-rel-block keys out above");
(
r.timeline,
PagestreamGetPageRequest {
latest: rng.gen_bool(args.req_latest_probability),
lsn: r.timeline_lsn,
rel: rel_tag,
blkno: block_no,
},
)
};
let sender = work_senders.get(&timeline).unwrap();
// TODO: what if this blocks?
sender.send(req).await.ok().unwrap();
}
}),
Some(rps_limit) => Box::pin(async move {
let period = Duration::from_secs_f64(1.0 / (rps_limit as f64));
let make_timeline_task: &dyn Fn(
TenantTimelineId,
)
-> Pin<Box<dyn Send + Future<Output = ()>>> = &|timeline| {
let sender = work_senders.get(&timeline).unwrap();
let ranges: Vec<KeyRange> = all_ranges
.iter()
.filter(|r| r.timeline == timeline)
.cloned()
.collect();
let weights = rand::distributions::weighted::WeightedIndex::new(
ranges.iter().map(|v| v.len()),
)
.unwrap();
Box::pin(async move {
let mut ticker = tokio::time::interval(period);
ticker.set_missed_tick_behavior(
/* TODO review this choice */
tokio::time::MissedTickBehavior::Burst,
);
loop {
ticker.tick().await;
let req = {
let mut rng = rand::thread_rng();
let r = &ranges[weights.sample(&mut rng)];
let key: i128 = rng.gen_range(r.start..r.end);
let key = repository::Key::from_i128(key);
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),
lsn: r.timeline_lsn,
rel: rel_tag,
blkno: block_no,
}
};
sender.send(req).await.ok().unwrap();
}
})
};
let tasks: Vec<_> = work_senders
.keys()
.map(|tl| make_timeline_task(**tl))
.collect();
start_work_barrier.wait().await;
join_all(tasks).await;
}),
};
if let Some(runtime) = args.runtime {
match tokio::time::timeout(runtime.into(), work_sender).await {
Ok(()) => unreachable!("work sender never terminates"),
Err(_timeout) => {
// this implicitly drops the work_senders, making all the clients exit
}
}
} else {
work_sender.await;
unreachable!("work sender never terminates");
}
for t in tasks {
t.await.unwrap();
}
let output = Output {
total: {
let mut agg_stats = request_stats::Stats::new();
for stats in all_thread_local_stats.lock().unwrap().iter() {
let stats = stats.lock().unwrap();
agg_stats.add(&stats);
}
agg_stats.output()
},
};
let output = serde_json::to_string_pretty(&output).unwrap();
println!("{output}");
anyhow::Ok(())
}
#[instrument(skip_all)]
async fn client(
args: &'static Args,
timeline: TenantTimelineId,
start_work_barrier: Arc<Barrier>,
mut work: tokio::sync::mpsc::Receiver<PagestreamGetPageRequest>,
all_work_done_barrier: Arc<Barrier>,
live_stats: Arc<LiveStats>,
) {
start_work_barrier.wait().await;
let client = pageserver_client::page_service::Client::new(args.page_service_connstring.clone())
.await
.unwrap();
let mut client = client
.pagestream(timeline.tenant_id, timeline.timeline_id)
.await
.unwrap();
while let Some(req) = work.recv().await {
let start = Instant::now();
client
.getpage(req)
.await
.with_context(|| format!("getpage for {timeline}"))
.unwrap();
let elapsed = start.elapsed();
live_stats.inc();
STATS.with(|stats| {
stats.borrow().lock().unwrap().observe(elapsed).unwrap();
});
}
all_work_done_barrier.wait().await;
}

View File

@@ -1,85 +0,0 @@
use std::sync::Arc;
use humantime::Duration;
use tokio::task::JoinSet;
use utils::id::TenantTimelineId;
#[derive(clap::Parser)]
pub(crate) struct Args {
#[clap(long, default_value = "http://localhost:9898")]
mgmt_api_endpoint: String,
#[clap(long, default_value = "localhost:64000")]
page_service_host_port: String,
#[clap(long)]
pageserver_jwt: Option<String>,
#[clap(
long,
help = "if specified, poll mgmt api to check whether init logical size calculation has completed"
)]
poll_for_completion: Option<Duration>,
#[clap(long)]
limit_to_first_n_targets: Option<usize>,
targets: Option<Vec<TenantTimelineId>>,
}
pub(crate) fn main(args: Args) -> anyhow::Result<()> {
let rt = tokio::runtime::Builder::new_multi_thread()
.enable_all()
.build()
.unwrap();
let main_task = rt.spawn(main_impl(args));
rt.block_on(main_task).unwrap()
}
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(),
));
// 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?;
// kick it off
let mut js = JoinSet::new();
for tl in timelines {
let mgmt_api_client = Arc::clone(&mgmt_api_client);
js.spawn(async move {
// TODO: API to explicitly trigger initial logical size computation.
// Should probably also avoid making it a side effect of timeline details to trigger initial logical size calculation.
// => https://github.com/neondatabase/neon/issues/6168
let info = mgmt_api_client
.timeline_info(tl.tenant_id, tl.timeline_id)
.await
.unwrap();
if let Some(period) = args.poll_for_completion {
let mut ticker = tokio::time::interval(period.into());
ticker.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
let mut info = info;
while !info.current_logical_size_is_accurate {
ticker.tick().await;
info = mgmt_api_client
.timeline_info(tl.tenant_id, tl.timeline_id)
.await
.unwrap();
}
}
});
}
while let Some(res) = js.join_next().await {
let _: () = res.unwrap();
}
Ok(())
}

View File

@@ -1,48 +0,0 @@
use clap::Parser;
use utils::logging;
/// Re-usable pieces of code that aren't CLI-specific.
mod util {
pub(crate) mod connstring;
pub(crate) mod request_stats;
#[macro_use]
pub(crate) mod tokio_thread_local_stats;
/// Re-usable pieces of CLI-specific code.
pub(crate) mod cli {
pub(crate) mod targets;
}
}
/// The pagebench CLI sub-commands, dispatched in [`main`] below.
mod cmd {
pub(super) mod basebackup;
pub(super) mod getpage_latest_lsn;
pub(super) mod trigger_initial_size_calculation;
}
/// Component-level performance test for pageserver.
#[derive(clap::Parser)]
enum Args {
Basebackup(cmd::basebackup::Args),
GetPageLatestLsn(cmd::getpage_latest_lsn::Args),
TriggerInitialSizeCalculation(cmd::trigger_initial_size_calculation::Args),
}
fn main() {
logging::init(
logging::LogFormat::Plain,
logging::TracingErrorLayerEnablement::Disabled,
logging::Output::Stderr,
)
.unwrap();
let args = Args::parse();
match args {
Args::Basebackup(args) => cmd::basebackup::main(args),
Args::GetPageLatestLsn(args) => cmd::getpage_latest_lsn::main(args),
Args::TriggerInitialSizeCalculation(args) => {
cmd::trigger_initial_size_calculation::main(args)
}
}
.unwrap()
}

View File

@@ -1,34 +0,0 @@
use std::sync::Arc;
use pageserver_client::mgmt_api;
use tracing::info;
use utils::id::TenantTimelineId;
pub(crate) struct Spec {
pub(crate) limit_to_first_n_targets: Option<usize>,
pub(crate) targets: Option<Vec<TenantTimelineId>>,
}
pub(crate) async fn discover(
api_client: &Arc<mgmt_api::Client>,
spec: Spec,
) -> anyhow::Result<Vec<TenantTimelineId>> {
let mut timelines = if let Some(targets) = spec.targets {
targets
} else {
mgmt_api::util::get_pageserver_tenant_timelines_unsharded(api_client).await?
};
if let Some(limit) = spec.limit_to_first_n_targets {
timelines.sort(); // for determinism
timelines.truncate(limit);
if timelines.len() < limit {
anyhow::bail!("pageserver has less than limit_to_first_n_targets={limit} tenants");
}
}
info!("timelines:\n{:?}", timelines);
info!("number of timelines:\n{:?}", timelines.len());
Ok(timelines)
}

View File

@@ -1,8 +0,0 @@
pub(crate) fn connstring(host_port: &str, jwt: Option<&str>) -> String {
let colon_and_jwt = if let Some(jwt) = jwt {
format!(":{jwt}") // TODO: urlescape
} else {
String::new()
};
format!("postgres://postgres{colon_and_jwt}@{host_port}")
}

View File

@@ -1,88 +0,0 @@
use std::time::Duration;
use anyhow::Context;
pub(crate) struct Stats {
latency_histo: hdrhistogram::Histogram<u64>,
}
impl Stats {
pub(crate) fn new() -> Self {
Self {
// Initialize with fixed bounds so that we panic at runtime instead of resizing the histogram,
// which would skew the benchmark results.
latency_histo: hdrhistogram::Histogram::new_with_bounds(1, 1_000_000_000, 3).unwrap(),
}
}
pub(crate) fn observe(&mut self, latency: Duration) -> anyhow::Result<()> {
let micros: u64 = latency
.as_micros()
.try_into()
.context("latency greater than u64")?;
self.latency_histo
.record(micros)
.context("add to histogram")?;
Ok(())
}
pub(crate) fn output(&self) -> Output {
let latency_percentiles = std::array::from_fn(|idx| {
let micros = self
.latency_histo
.value_at_percentile(LATENCY_PERCENTILES[idx]);
Duration::from_micros(micros)
});
Output {
request_count: self.latency_histo.len(),
latency_mean: Duration::from_micros(self.latency_histo.mean() as u64),
latency_percentiles: LatencyPercentiles {
latency_percentiles,
},
}
}
pub(crate) fn add(&mut self, other: &Self) {
let Self {
ref mut latency_histo,
} = self;
latency_histo.add(&other.latency_histo).unwrap();
}
}
impl Default for Stats {
fn default() -> Self {
Self::new()
}
}
const LATENCY_PERCENTILES: [f64; 4] = [95.0, 99.00, 99.90, 99.99];
struct LatencyPercentiles {
latency_percentiles: [Duration; 4],
}
impl serde::Serialize for LatencyPercentiles {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
use serde::ser::SerializeMap;
let mut ser = serializer.serialize_map(Some(LATENCY_PERCENTILES.len()))?;
for p in LATENCY_PERCENTILES {
ser.serialize_entry(
&format!("p{p}"),
&format!(
"{}",
&humantime::format_duration(self.latency_percentiles[0])
),
)?;
}
ser.end()
}
}
#[derive(serde::Serialize)]
pub(crate) struct Output {
request_count: u64,
#[serde(with = "humantime_serde")]
latency_mean: Duration,
latency_percentiles: LatencyPercentiles,
}

View File

@@ -1,45 +0,0 @@
pub(crate) type ThreadLocalStats<T> = Arc<Mutex<T>>;
pub(crate) type AllThreadLocalStats<T> = Arc<Mutex<Vec<ThreadLocalStats<T>>>>;
macro_rules! declare {
($THREAD_LOCAL_NAME:ident: $T:ty) => {
thread_local! {
pub static $THREAD_LOCAL_NAME: std::cell::RefCell<crate::util::tokio_thread_local_stats::ThreadLocalStats<$T>> = std::cell::RefCell::new(
std::sync::Arc::new(std::sync::Mutex::new(Default::default()))
);
}
};
}
use std::sync::{Arc, Mutex};
pub(crate) use declare;
macro_rules! main {
($THREAD_LOCAL_NAME:ident, $main_impl:expr) => {{
let main_impl = $main_impl;
let all = Arc::new(Mutex::new(Vec::new()));
let rt = tokio::runtime::Builder::new_multi_thread()
.on_thread_start({
let all = Arc::clone(&all);
move || {
// pre-initialize the thread local stats by accessesing them
// (some stats like requests_stats::Stats are quite costly to initialize,
// we don't want to pay that cost during the measurement period)
$THREAD_LOCAL_NAME.with(|stats| {
let stats: Arc<_> = Arc::clone(&*stats.borrow());
all.lock().unwrap().push(stats);
});
}
})
.enable_all()
.build()
.unwrap();
let main_task = rt.spawn(main_impl(all));
rt.block_on(main_task).unwrap()
}};
}
pub(crate) use main;

View File

@@ -23,7 +23,6 @@ use tracing::*;
use tokio_tar::{Builder, EntryType, Header};
use crate::context::RequestContext;
use crate::pgdatadir_mapping::Version;
use crate::tenant::Timeline;
use pageserver_api::reltag::{RelTag, SlruKind};
@@ -175,7 +174,7 @@ where
] {
for segno in self
.timeline
.list_slru_segments(kind, Version::Lsn(self.lsn), self.ctx)
.list_slru_segments(kind, self.lsn, self.ctx)
.await?
{
self.add_slru_segment(kind, segno).await?;
@@ -193,7 +192,7 @@ where
// Otherwise only include init forks of unlogged relations.
let rels = self
.timeline
.list_rels(spcnode, dbnode, Version::Lsn(self.lsn), self.ctx)
.list_rels(spcnode, dbnode, self.lsn, self.ctx)
.await?;
for &rel in rels.iter() {
// Send init fork as main fork to provide well formed empty
@@ -268,7 +267,7 @@ where
async fn add_rel(&mut self, src: RelTag, dst: RelTag) -> anyhow::Result<()> {
let nblocks = self
.timeline
.get_rel_size(src, Version::Lsn(self.lsn), false, self.ctx)
.get_rel_size(src, self.lsn, false, self.ctx)
.await?;
// If the relation is empty, create an empty file
@@ -289,7 +288,7 @@ 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, self.lsn, false, self.ctx)
.await?;
segment_data.extend_from_slice(&img[..]);
}
@@ -311,7 +310,7 @@ where
async fn add_slru_segment(&mut self, slru: SlruKind, segno: u32) -> anyhow::Result<()> {
let nblocks = self
.timeline
.get_slru_segment_size(slru, segno, Version::Lsn(self.lsn), self.ctx)
.get_slru_segment_size(slru, segno, self.lsn, self.ctx)
.await?;
let mut slru_buf: Vec<u8> = Vec::with_capacity(nblocks as usize * BLCKSZ as usize);
@@ -353,7 +352,7 @@ where
let relmap_img = if has_relmap_file {
let img = self
.timeline
.get_relmap_file(spcnode, dbnode, Version::Lsn(self.lsn), self.ctx)
.get_relmap_file(spcnode, dbnode, self.lsn, self.ctx)
.await?;
ensure!(
@@ -400,7 +399,7 @@ where
if !has_relmap_file
&& self
.timeline
.list_rels(spcnode, dbnode, Version::Lsn(self.lsn), self.ctx)
.list_rels(spcnode, dbnode, self.lsn, self.ctx)
.await?
.is_empty()
{

View File

@@ -31,7 +31,6 @@ use pageserver::{
virtual_file,
};
use postgres_backend::AuthType;
use utils::failpoint_support;
use utils::logging::TracingErrorLayerEnablement;
use utils::signals::ShutdownSignals;
use utils::{
@@ -127,7 +126,7 @@ fn main() -> anyhow::Result<()> {
}
// Initialize up failpoints support
let scenario = failpoint_support::init();
let scenario = pageserver::failpoint_support::init();
// Basic initialization of things that don't change after startup
virtual_file::init(conf.max_file_descriptors);

View File

@@ -37,12 +37,10 @@ use crate::tenant::{
TENANTS_SEGMENT_NAME, TENANT_DELETED_MARKER_FILE_NAME, TIMELINES_SEGMENT_NAME,
};
use crate::{
IGNORED_TENANT_FILE_NAME, METADATA_FILE_NAME, TENANT_CONFIG_NAME, TENANT_HEATMAP_BASENAME,
TENANT_LOCATION_CONFIG_NAME, TIMELINE_DELETE_MARK_SUFFIX, TIMELINE_UNINIT_MARK_SUFFIX,
IGNORED_TENANT_FILE_NAME, METADATA_FILE_NAME, TENANT_CONFIG_NAME, TENANT_LOCATION_CONFIG_NAME,
TIMELINE_DELETE_MARK_SUFFIX, TIMELINE_UNINIT_MARK_SUFFIX,
};
use self::defaults::DEFAULT_CONCURRENT_TENANT_WARMUP;
pub mod defaults {
use crate::tenant::config::defaults::*;
use const_format::formatcp;
@@ -63,8 +61,6 @@ pub mod defaults {
pub const DEFAULT_LOG_FORMAT: &str = "plain";
pub const DEFAULT_CONCURRENT_TENANT_WARMUP: usize = 8;
pub const DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES: usize =
super::ConfigurableSemaphore::DEFAULT_INITIAL.get();
@@ -75,9 +71,6 @@ pub mod defaults {
pub const DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY: &str = "10s";
pub const DEFAULT_HEATMAP_UPLOAD_CONCURRENCY: usize = 8;
pub const DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY: usize = 1;
pub const DEFAULT_INGEST_BATCH_SIZE: u64 = 100;
///
/// Default built-in configuration file.
@@ -91,7 +84,6 @@ pub mod defaults {
#wait_lsn_timeout = '{DEFAULT_WAIT_LSN_TIMEOUT}'
#wal_redo_timeout = '{DEFAULT_WAL_REDO_TIMEOUT}'
#page_cache_size = {DEFAULT_PAGE_CACHE_SIZE}
#max_file_descriptors = {DEFAULT_MAX_FILE_DESCRIPTORS}
# initial superuser role name to use when creating a new tenant
@@ -102,7 +94,6 @@ pub mod defaults {
#log_format = '{DEFAULT_LOG_FORMAT}'
#concurrent_tenant_size_logical_size_queries = '{DEFAULT_CONCURRENT_TENANT_SIZE_LOGICAL_SIZE_QUERIES}'
#concurrent_tenant_warmup = '{DEFAULT_CONCURRENT_TENANT_WARMUP}'
#metric_collection_interval = '{DEFAULT_METRIC_COLLECTION_INTERVAL}'
#cached_metric_collection_interval = '{DEFAULT_CACHED_METRIC_COLLECTION_INTERVAL}'
@@ -112,8 +103,6 @@ pub mod defaults {
#background_task_maximum_delay = '{DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY}'
#ingest_batch_size = {DEFAULT_INGEST_BATCH_SIZE}
[tenant_config]
#checkpoint_distance = {DEFAULT_CHECKPOINT_DISTANCE} # in bytes
#checkpoint_timeout = {DEFAULT_CHECKPOINT_TIMEOUT}
@@ -131,7 +120,6 @@ pub mod defaults {
#gc_feedback = false
#heatmap_upload_concurrency = {DEFAULT_HEATMAP_UPLOAD_CONCURRENCY}
#secondary_download_concurrency = {DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY}
[remote_storage]
@@ -192,11 +180,6 @@ pub struct PageServerConf {
pub log_format: LogFormat,
/// Number of tenants which will be concurrently loaded from remote storage proactively on startup,
/// does not limit tenants loaded in response to client I/O. A lower value implicitly deprioritizes
/// loading such tenants, vs. other work in the system.
pub concurrent_tenant_warmup: ConfigurableSemaphore,
/// Number of concurrent [`Tenant::gather_size_inputs`](crate::tenant::Tenant::gather_size_inputs) allowed.
pub concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore,
/// Limit of concurrent [`Tenant::gather_size_inputs`] issued by module `eviction_task`.
@@ -240,13 +223,6 @@ pub struct PageServerConf {
/// How many heatmap uploads may be done concurrency: lower values implicitly deprioritize
/// heatmap uploads vs. other remote storage operations.
pub heatmap_upload_concurrency: usize,
/// How many remote storage downloads may be done for secondary tenants concurrently. Implicitly
/// deprioritises secondary downloads vs. remote storage operations for attached tenants.
pub secondary_download_concurrency: usize,
/// Maximum number of WAL records to be ingested and committed at the same time
pub ingest_batch_size: u64,
}
/// We do not want to store this in a PageServerConf because the latter may be logged
@@ -307,7 +283,6 @@ struct PageServerConfigBuilder {
log_format: BuilderValue<LogFormat>,
concurrent_tenant_warmup: BuilderValue<NonZeroUsize>,
concurrent_tenant_size_logical_size_queries: BuilderValue<NonZeroUsize>,
metric_collection_interval: BuilderValue<Duration>,
@@ -328,9 +303,6 @@ struct PageServerConfigBuilder {
control_plane_emergency_mode: BuilderValue<bool>,
heatmap_upload_concurrency: BuilderValue<usize>,
secondary_download_concurrency: BuilderValue<usize>,
ingest_batch_size: BuilderValue<u64>,
}
impl Default for PageServerConfigBuilder {
@@ -368,8 +340,6 @@ impl Default for PageServerConfigBuilder {
.expect("cannot parse default keepalive interval")),
log_format: Set(LogFormat::from_str(DEFAULT_LOG_FORMAT).unwrap()),
concurrent_tenant_warmup: Set(NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP)
.expect("Invalid default constant")),
concurrent_tenant_size_logical_size_queries: Set(
ConfigurableSemaphore::DEFAULT_INITIAL,
),
@@ -403,9 +373,6 @@ impl Default for PageServerConfigBuilder {
control_plane_emergency_mode: Set(false),
heatmap_upload_concurrency: Set(DEFAULT_HEATMAP_UPLOAD_CONCURRENCY),
secondary_download_concurrency: Set(DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY),
ingest_batch_size: Set(DEFAULT_INGEST_BATCH_SIZE),
}
}
}
@@ -486,10 +453,6 @@ impl PageServerConfigBuilder {
self.log_format = BuilderValue::Set(log_format)
}
pub fn concurrent_tenant_warmup(&mut self, u: NonZeroUsize) {
self.concurrent_tenant_warmup = BuilderValue::Set(u);
}
pub fn concurrent_tenant_size_logical_size_queries(&mut self, u: NonZeroUsize) {
self.concurrent_tenant_size_logical_size_queries = BuilderValue::Set(u);
}
@@ -554,18 +517,7 @@ impl PageServerConfigBuilder {
self.heatmap_upload_concurrency = BuilderValue::Set(value)
}
pub fn secondary_download_concurrency(&mut self, value: usize) {
self.secondary_download_concurrency = BuilderValue::Set(value)
}
pub fn ingest_batch_size(&mut self, ingest_batch_size: u64) {
self.ingest_batch_size = BuilderValue::Set(ingest_batch_size)
}
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!(
@@ -618,7 +570,6 @@ impl PageServerConfigBuilder {
.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,
),
@@ -660,15 +611,10 @@ impl PageServerConfigBuilder {
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"))?,
})
}
}
@@ -726,11 +672,6 @@ impl PageServerConf {
.join(TENANT_LOCATION_CONFIG_NAME)
}
pub(crate) fn tenant_heatmap_path(&self, tenant_shard_id: &TenantShardId) -> Utf8PathBuf {
self.tenant_path(tenant_shard_id)
.join(TENANT_HEATMAP_BASENAME)
}
pub fn timelines_path(&self, tenant_shard_id: &TenantShardId) -> Utf8PathBuf {
self.tenant_path(tenant_shard_id)
.join(TIMELINES_SEGMENT_NAME)
@@ -866,11 +807,6 @@ impl PageServerConf {
"log_format" => builder.log_format(
LogFormat::from_config(&parse_toml_string(key, item)?)?
),
"concurrent_tenant_warmup" => builder.concurrent_tenant_warmup({
let input = parse_toml_string(key, item)?;
let permits = input.parse::<usize>().context("expected a number of initial permits, not {s:?}")?;
NonZeroUsize::new(permits).context("initial semaphore permits out of range: 0, use other configuration to disable a feature")?
}),
"concurrent_tenant_size_logical_size_queries" => builder.concurrent_tenant_size_logical_size_queries({
let input = parse_toml_string(key, item)?;
let permits = input.parse::<usize>().context("expected a number of initial permits, not {s:?}")?;
@@ -916,10 +852,6 @@ impl PageServerConf {
"heatmap_upload_concurrency" => {
builder.heatmap_upload_concurrency(parse_toml_u64(key, item)? as usize)
},
"secondary_download_concurrency" => {
builder.secondary_download_concurrency(parse_toml_u64(key, item)? as usize)
},
"ingest_batch_size" => builder.ingest_batch_size(parse_toml_u64(key, item)?),
_ => bail!("unrecognized pageserver option '{key}'"),
}
}
@@ -972,10 +904,6 @@ impl PageServerConf {
broker_endpoint: storage_broker::DEFAULT_ENDPOINT.parse().unwrap(),
broker_keepalive_interval: Duration::from_secs(5000),
log_format: LogFormat::from_str(defaults::DEFAULT_LOG_FORMAT).unwrap(),
concurrent_tenant_warmup: ConfigurableSemaphore::new(
NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP)
.expect("Invalid default constant"),
),
concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore::default(),
eviction_task_immitated_concurrent_logical_size_queries: ConfigurableSemaphore::default(
),
@@ -991,8 +919,6 @@ impl PageServerConf {
control_plane_api_token: None,
control_plane_emergency_mode: false,
heatmap_upload_concurrency: defaults::DEFAULT_HEATMAP_UPLOAD_CONCURRENCY,
secondary_download_concurrency: defaults::DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY,
ingest_batch_size: defaults::DEFAULT_INGEST_BATCH_SIZE,
}
}
}
@@ -1196,9 +1122,6 @@ background_task_maximum_delay = '334 s'
storage_broker::DEFAULT_KEEPALIVE_INTERVAL
)?,
log_format: LogFormat::from_str(defaults::DEFAULT_LOG_FORMAT).unwrap(),
concurrent_tenant_warmup: ConfigurableSemaphore::new(
NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP).unwrap()
),
concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore::default(),
eviction_task_immitated_concurrent_logical_size_queries:
ConfigurableSemaphore::default(),
@@ -1221,9 +1144,7 @@ background_task_maximum_delay = '334 s'
control_plane_api: None,
control_plane_api_token: None,
control_plane_emergency_mode: false,
heatmap_upload_concurrency: defaults::DEFAULT_HEATMAP_UPLOAD_CONCURRENCY,
secondary_download_concurrency: defaults::DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY,
ingest_batch_size: defaults::DEFAULT_INGEST_BATCH_SIZE,
heatmap_upload_concurrency: defaults::DEFAULT_HEATMAP_UPLOAD_CONCURRENCY
},
"Correct defaults should be used when no config values are provided"
);
@@ -1267,9 +1188,6 @@ background_task_maximum_delay = '334 s'
broker_endpoint: storage_broker::DEFAULT_ENDPOINT.parse().unwrap(),
broker_keepalive_interval: Duration::from_secs(5),
log_format: LogFormat::Json,
concurrent_tenant_warmup: ConfigurableSemaphore::new(
NonZeroUsize::new(DEFAULT_CONCURRENT_TENANT_WARMUP).unwrap()
),
concurrent_tenant_size_logical_size_queries: ConfigurableSemaphore::default(),
eviction_task_immitated_concurrent_logical_size_queries:
ConfigurableSemaphore::default(),
@@ -1284,9 +1202,7 @@ background_task_maximum_delay = '334 s'
control_plane_api: None,
control_plane_api_token: None,
control_plane_emergency_mode: false,
heatmap_upload_concurrency: defaults::DEFAULT_HEATMAP_UPLOAD_CONCURRENCY,
secondary_download_concurrency: defaults::DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY,
ingest_batch_size: 100,
heatmap_upload_concurrency: defaults::DEFAULT_HEATMAP_UPLOAD_CONCURRENCY
},
"Should be able to parse all basic config values correctly"
);
@@ -1516,7 +1432,6 @@ threshold = "20m"
period: Duration::from_secs(10),
#[cfg(feature = "testing")]
mock_statvfs: None,
eviction_order: crate::disk_usage_eviction_task::EvictionOrder::AbsoluteAccessed,
})
);
match &conf.default_tenant_conf.eviction_policy {

View File

@@ -74,45 +74,6 @@ pub struct DiskUsageEvictionTaskConfig {
pub period: Duration,
#[cfg(feature = "testing")]
pub mock_statvfs: Option<crate::statvfs::mock::Behavior>,
/// Select sorting for evicted layers
#[serde(default)]
pub eviction_order: EvictionOrder,
}
/// Selects the sort order for eviction candidates *after* per tenant `min_resident_size`
/// partitioning.
#[derive(Default, Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(tag = "type", content = "args")]
pub enum EvictionOrder {
/// Order the layers to be evicted by how recently they have been accessed in absolute
/// time.
///
/// This strategy is unfair when some tenants grow faster than others towards the slower
/// growing.
#[default]
AbsoluteAccessed,
/// Order the layers to be evicted by how recently they have been accessed relatively within
/// the set of resident layers of a tenant.
///
/// This strategy will evict layers more fairly but is untested.
RelativeAccessed {
#[serde(default)]
highest_layer_count_loses_first: bool,
},
}
impl EvictionOrder {
/// Return true, if with [`Self::RelativeAccessed`] order the tenants with the highest layer
/// counts should be the first ones to have their layers evicted.
fn highest_layer_count_loses_first(&self) -> bool {
match self {
EvictionOrder::AbsoluteAccessed => false,
EvictionOrder::RelativeAccessed {
highest_layer_count_loses_first,
} => *highest_layer_count_loses_first,
}
}
}
#[derive(Default)]
@@ -231,14 +192,7 @@ async fn disk_usage_eviction_task_iteration(
) -> anyhow::Result<()> {
let usage_pre = filesystem_level_usage::get(tenants_dir, task_config)
.context("get filesystem-level disk usage before evictions")?;
let res = disk_usage_eviction_task_iteration_impl(
state,
storage,
usage_pre,
task_config.eviction_order,
cancel,
)
.await;
let res = disk_usage_eviction_task_iteration_impl(state, storage, usage_pre, cancel).await;
match res {
Ok(outcome) => {
debug!(?outcome, "disk_usage_eviction_iteration finished");
@@ -324,7 +278,6 @@ pub(crate) async fn disk_usage_eviction_task_iteration_impl<U: Usage>(
state: &State,
_storage: &GenericRemoteStorage,
usage_pre: U,
eviction_order: EvictionOrder,
cancel: &CancellationToken,
) -> anyhow::Result<IterationOutcome<U>> {
// use tokio's mutex to get a Sync guard (instead of std::sync::Mutex)
@@ -344,7 +297,7 @@ pub(crate) async fn disk_usage_eviction_task_iteration_impl<U: Usage>(
"running disk usage based eviction due to pressure"
);
let candidates = match collect_eviction_candidates(eviction_order, cancel).await? {
let candidates = match collect_eviction_candidates(cancel).await? {
EvictionCandidates::Cancelled => {
return Ok(IterationOutcome::Cancelled);
}
@@ -354,16 +307,16 @@ pub(crate) async fn disk_usage_eviction_task_iteration_impl<U: Usage>(
// Debug-log the list of candidates
let now = SystemTime::now();
for (i, (partition, candidate)) in candidates.iter().enumerate() {
let nth = i + 1;
let desc = candidate.layer.layer_desc();
let total_candidates = candidates.len();
let size = desc.file_size;
let rel = candidate.relative_last_activity;
debug!(
"cand {nth}/{total_candidates}: size={size}, rel_last_activity={rel}, no_access_for={}us, partition={partition:?}, {}/{}/{}",
"cand {}/{}: size={}, no_access_for={}us, partition={:?}, {}/{}/{}",
i + 1,
candidates.len(),
desc.file_size,
now.duration_since(candidate.last_activity_ts)
.unwrap()
.as_micros(),
partition,
desc.tenant_shard_id,
desc.timeline_id,
candidate.layer,
@@ -506,7 +459,6 @@ struct EvictionCandidate {
timeline: Arc<Timeline>,
layer: Layer,
last_activity_ts: SystemTime,
relative_last_activity: finite_f32::FiniteF32,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
@@ -526,24 +478,24 @@ enum EvictionCandidates {
/// order. A caller that evicts in that order, until pressure is relieved, implements
/// the eviction policy outlined in the module comment.
///
/// # Example with EvictionOrder::AbsoluteAccessed
/// # Example
///
/// Imagine that there are two tenants, A and B, with five layers each, a-e.
/// Each layer has size 100, and both tenant's min_resident_size is 150.
/// The eviction order would be
///
/// ```text
/// partition last_activity_ts tenant/layer
/// Above 18:30 A/c
/// Above 19:00 A/b
/// Above 18:29 B/c
/// Above 19:05 B/b
/// Above 20:00 B/a
/// Above 20:03 A/a
/// Below 20:30 A/d
/// Below 20:40 B/d
/// Below 20:45 B/e
/// Below 20:58 A/e
/// partition last_activity_ts tenant/layer
/// Above 18:30 A/c
/// Above 19:00 A/b
/// Above 18:29 B/c
/// Above 19:05 B/b
/// Above 20:00 B/a
/// Above 20:03 A/a
/// Below 20:30 A/d
/// Below 20:40 B/d
/// Below 20:45 B/e
/// Below 20:58 A/e
/// ```
///
/// Now, if we need to evict 300 bytes to relieve pressure, we'd evict `A/c, A/b, B/c`.
@@ -553,77 +505,7 @@ enum EvictionCandidates {
/// `A/c, A/b, B/c, B/b, B/a, A/a, A/d, B/d, B/e`, reaching into the `Below` partition
/// after exhauting the `Above` partition.
/// So, we did not respect each tenant's min_resident_size.
///
/// # Example with EvictionOrder::RelativeAccessed
///
/// ```text
/// partition relative_age last_activity_ts tenant/layer
/// Above 0/4 18:30 A/c
/// Above 0/4 18:29 B/c
/// Above 1/4 19:00 A/b
/// Above 1/4 19:05 B/b
/// Above 2/4 20:00 B/a
/// Above 2/4 20:03 A/a
/// Below 3/4 20:30 A/d
/// Below 3/4 20:40 B/d
/// Below 4/4 20:45 B/e
/// Below 4/4 20:58 A/e
/// ```
///
/// With tenants having the same number of layers the picture does not change much. The same with
/// A having many more layers **resident** (not all of them listed):
///
/// ```text
/// Above 0/100 18:30 A/c
/// Above 0/4 18:29 B/c
/// Above 1/100 19:00 A/b
/// Above 2/100 20:03 A/a
/// Above 3/100 20:03 A/nth_3
/// Above 4/100 20:03 A/nth_4
/// ...
/// Above 1/4 19:05 B/b
/// Above 25/100 20:04 A/nth_25
/// ...
/// Above 2/4 20:00 B/a
/// Above 50/100 20:10 A/nth_50
/// ...
/// Below 3/4 20:40 B/d
/// Below 99/100 20:30 A/nth_99
/// Below 4/4 20:45 B/e
/// Below 100/100 20:58 A/nth_100
/// ```
///
/// Now it's easier to see that because A has grown fast it has more layers to get evicted. What is
/// difficult to see is what happens on the next round assuming the evicting 23 from the above list
/// relieves the pressure (22 A layers gone, 1 B layers gone) but a new fast growing tenant C has
/// appeared:
///
/// ```text
/// Above 0/87 20:04 A/nth_23
/// Above 0/3 19:05 B/b
/// Above 0/50 20:59 C/nth_0
/// Above 1/87 20:04 A/nth_24
/// Above 1/50 21:00 C/nth_1
/// Above 2/87 20:04 A/nth_25
/// ...
/// Above 16/50 21:02 C/nth_16
/// Above 1/3 20:00 B/a
/// Above 27/87 20:10 A/nth_50
/// ...
/// Below 2/3 20:40 B/d
/// Below 49/50 21:05 C/nth_49
/// Below 86/87 20:30 A/nth_99
/// Below 3/3 20:45 B/e
/// Below 50/50 21:05 C/nth_50
/// Below 87/87 20:58 A/nth_100
/// ```
///
/// Now relieving pressure with 23 layers would cost:
/// - tenant A 14 layers
/// - tenant B 1 layer
/// - tenant C 8 layers
async fn collect_eviction_candidates(
eviction_order: EvictionOrder,
cancel: &CancellationToken,
) -> anyhow::Result<EvictionCandidates> {
// get a snapshot of the list of tenants
@@ -709,63 +591,12 @@ async fn collect_eviction_candidates(
tenant_candidates
.sort_unstable_by_key(|(_, layer_info)| std::cmp::Reverse(layer_info.last_activity_ts));
let mut cumsum: i128 = 0;
// keeping the -1 or not decides if every tenant should lose their least recently accessed
// layer OR if this should happen in the order of having highest layer count:
let fudge = if eviction_order.highest_layer_count_loses_first() {
// relative_age vs. tenant layer count:
// - 0.1..=1.0 (10 layers)
// - 0.01..=1.0 (100 layers)
// - 0.001..=1.0 (1000 layers)
//
// leading to evicting less of the smallest tenants.
0
} else {
// use full 0.0..=1.0 range, which means even the smallest tenants could always lose a
// layer. the actual ordering is unspecified: for 10k tenants on a pageserver it could
// be that less than 10k layer evictions is enough, so we would not need to evict from
// all tenants.
//
// as the tenant ordering is now deterministic this could hit the same tenants
// disproportionetly on multiple invocations. alternative could be to remember how many
// layers did we evict last time from this tenant, and inject that as an additional
// fudge here.
1
};
let total = tenant_candidates
.len()
.checked_sub(fudge)
.filter(|&x| x > 0)
// support 0 or 1 resident layer tenants as well
.unwrap_or(1);
let divider = total as f32;
for (i, (timeline, layer_info)) in tenant_candidates.into_iter().enumerate() {
for (timeline, layer_info) in tenant_candidates.into_iter() {
let file_size = layer_info.file_size();
// as we iterate this reverse sorted list, the most recently accessed layer will always
// be 1.0; this is for us to evict it last.
let relative_last_activity = if matches!(
eviction_order,
EvictionOrder::RelativeAccessed { .. }
) {
// another possibility: use buckets, like (256.0 * relative_last_activity) as u8 or
// similarly for u16. unsure how it would help.
finite_f32::FiniteF32::try_from_normalized((total - i) as f32 / divider)
.unwrap_or_else(|val| {
tracing::warn!(%fudge, "calculated invalid relative_last_activity for i={i}, total={total}: {val}");
finite_f32::FiniteF32::ZERO
})
} else {
finite_f32::FiniteF32::ZERO
};
let candidate = EvictionCandidate {
timeline,
last_activity_ts: layer_info.last_activity_ts,
layer: layer_info.layer,
relative_last_activity,
};
let partition = if cumsum > min_resident_size as i128 {
MinResidentSizePartition::Above
@@ -779,19 +610,8 @@ async fn collect_eviction_candidates(
debug_assert!(MinResidentSizePartition::Above < MinResidentSizePartition::Below,
"as explained in the function's doc comment, layers that aren't in the tenant's min_resident_size are evicted first");
match eviction_order {
EvictionOrder::AbsoluteAccessed => {
candidates.sort_unstable_by_key(|(partition, candidate)| {
(*partition, candidate.last_activity_ts)
});
}
EvictionOrder::RelativeAccessed { .. } => {
candidates.sort_unstable_by_key(|(partition, candidate)| {
(*partition, candidate.relative_last_activity)
});
}
}
candidates
.sort_unstable_by_key(|(partition, candidate)| (*partition, candidate.last_activity_ts));
Ok(EvictionCandidates::Finished(candidates))
}
@@ -820,66 +640,6 @@ impl std::ops::Deref for TimelineKey {
}
}
/// A totally ordered f32 subset we can use with sorting functions.
mod finite_f32 {
/// A totally ordered f32 subset we can use with sorting functions.
#[derive(Clone, Copy, PartialEq)]
pub struct FiniteF32(f32);
impl std::fmt::Debug for FiniteF32 {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
std::fmt::Debug::fmt(&self.0, f)
}
}
impl std::fmt::Display for FiniteF32 {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
std::fmt::Display::fmt(&self.0, f)
}
}
impl std::cmp::Eq for FiniteF32 {}
impl std::cmp::PartialOrd for FiniteF32 {
fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
Some(self.cmp(other))
}
}
impl std::cmp::Ord for FiniteF32 {
fn cmp(&self, other: &Self) -> std::cmp::Ordering {
self.0.total_cmp(&other.0)
}
}
impl TryFrom<f32> for FiniteF32 {
type Error = f32;
fn try_from(value: f32) -> Result<Self, Self::Error> {
if value.is_finite() {
Ok(FiniteF32(value))
} else {
Err(value)
}
}
}
impl FiniteF32 {
pub const ZERO: FiniteF32 = FiniteF32(0.0);
pub fn try_from_normalized(value: f32) -> Result<Self, f32> {
if (0.0..=1.0).contains(&value) {
// -0.0 is within the range, make sure it is assumed 0.0..=1.0
let value = value.abs();
Ok(FiniteF32(value))
} else {
Err(value)
}
}
}
}
mod filesystem_level_usage {
use anyhow::Context;
use camino::Utf8Path;
@@ -961,7 +721,6 @@ mod filesystem_level_usage {
#[test]
fn max_usage_pct_pressure() {
use super::EvictionOrder;
use super::Usage as _;
use std::time::Duration;
use utils::serde_percent::Percent;
@@ -973,7 +732,6 @@ mod filesystem_level_usage {
period: Duration::MAX,
#[cfg(feature = "testing")]
mock_statvfs: None,
eviction_order: EvictionOrder::default(),
},
total_bytes: 100_000,
avail_bytes: 0,

View File

@@ -1,14 +1,3 @@
//! Failpoint support code shared between pageserver and safekeepers.
use crate::http::{
error::ApiError,
json::{json_request, json_response},
};
use hyper::{Body, Request, Response, StatusCode};
use serde::{Deserialize, Serialize};
use tokio_util::sync::CancellationToken;
use tracing::*;
/// use with fail::cfg("$name", "return(2000)")
///
/// The effect is similar to a "sleep(2000)" action, i.e. we sleep for the
@@ -36,7 +25,7 @@ pub use __failpoint_sleep_millis_async as sleep_millis_async;
// Helper function used by the macro. (A function has nicer scoping so we
// don't need to decorate everything with "::")
#[doc(hidden)]
pub async fn failpoint_sleep_helper(name: &'static str, duration_str: String) {
pub(crate) async fn failpoint_sleep_helper(name: &'static str, duration_str: String) {
let millis = duration_str.parse::<u64>().unwrap();
let d = std::time::Duration::from_millis(millis);
@@ -82,7 +71,7 @@ pub fn init() -> fail::FailScenario<'static> {
scenario
}
pub fn apply_failpoint(name: &str, actions: &str) -> Result<(), String> {
pub(crate) fn apply_failpoint(name: &str, actions: &str) -> Result<(), String> {
if actions == "exit" {
fail::cfg_callback(name, exit_failpoint)
} else {
@@ -95,45 +84,3 @@ fn exit_failpoint() {
tracing::info!("Exit requested by failpoint");
std::process::exit(1);
}
pub type ConfigureFailpointsRequest = Vec<FailpointConfig>;
/// Information for configuring a single fail point
#[derive(Debug, Serialize, Deserialize)]
pub struct FailpointConfig {
/// Name of the fail point
pub name: String,
/// List of actions to take, using the format described in `fail::cfg`
///
/// We also support `actions = "exit"` to cause the fail point to immediately exit.
pub actions: String,
}
/// Configure failpoints through http.
pub async fn failpoints_handler(
mut request: Request<Body>,
_cancel: CancellationToken,
) -> Result<Response<Body>, ApiError> {
if !fail::has_failpoints() {
return Err(ApiError::BadRequest(anyhow::anyhow!(
"Cannot manage failpoints because storage was compiled without failpoints support"
)));
}
let failpoints: ConfigureFailpointsRequest = json_request(&mut request).await?;
for fp in failpoints {
info!("cfg failpoint: {} {}", fp.name, fp.actions);
// We recognize one extra "action" that's not natively recognized
// by the failpoints crate: exit, to immediately kill the process
let cfg_result = apply_failpoint(&fp.name, &fp.actions);
if let Err(err_msg) = cfg_result {
return Err(ApiError::BadRequest(anyhow::anyhow!(
"Failed to configure failpoints: {err_msg}"
)));
}
}
json_response(StatusCode::OK, ())
}

View File

@@ -1,2 +1,4 @@
pub mod routes;
pub use routes::make_router;
pub use pageserver_api::models;

View File

@@ -159,12 +159,6 @@ paths:
application/json:
schema:
$ref: "#/components/schemas/ConflictError"
"412":
description: Deletion may not proceed, tenant is not in Active state
content:
application/json:
schema:
$ref: "#/components/schemas/PreconditionFailedError"
"500":
description: Generic operation error
content:

View File

@@ -14,7 +14,6 @@ use hyper::header;
use hyper::StatusCode;
use hyper::{Body, Request, Response, Uri};
use metrics::launch_timestamp::LaunchTimestamp;
use pageserver_api::models::TenantDetails;
use pageserver_api::models::{
DownloadRemoteLayersTaskSpawnRequest, LocationConfigMode, TenantAttachRequest,
TenantLoadRequest, TenantLocationConfigRequest,
@@ -25,18 +24,20 @@ use tenant_size_model::{SizeResult, StorageModel};
use tokio_util::sync::CancellationToken;
use tracing::*;
use utils::auth::JwtAuth;
use utils::failpoint_support::failpoints_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};
use super::models::{
StatusResponse, TenantConfigRequest, TenantCreateRequest, TenantCreateResponse, TenantInfo,
TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
};
use crate::context::{DownloadBehavior, RequestContext};
use crate::deletion_queue::DeletionQueueClient;
use crate::metrics::{StorageTimeOperation, STORAGE_TIME_GLOBAL};
use crate::pgdatadir_mapping::LsnForTimestamp;
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,
@@ -49,10 +50,6 @@ use crate::tenant::timeline::Timeline;
use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError, TenantSharedResources};
use crate::{config::PageServerConf, tenant::mgr};
use crate::{disk_usage_eviction_task, tenant};
use pageserver_api::models::{
StatusResponse, TenantConfigRequest, TenantCreateRequest, TenantCreateResponse, TenantInfo,
TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
};
use utils::{
auth::SwappableJwtAuth,
generation::Generation,
@@ -67,10 +64,8 @@ use utils::{
lsn::Lsn,
};
// For APIs that require an Active tenant, how long should we block waiting for that state?
// This is not functionally necessary (clients will retry), but avoids generating a lot of
// failed API calls while tenants are activating.
const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(5000);
// Imports only used for testing APIs
use super::models::ConfigureFailpointsRequest;
pub struct State {
conf: &'static PageServerConf,
@@ -152,7 +147,6 @@ impl From<PageReconstructError> for ApiError {
PageReconstructError::AncestorStopping(_) => {
ApiError::ResourceUnavailable(format!("{pre}").into())
}
PageReconstructError::AncestorLsnTimeout(e) => ApiError::Timeout(format!("{e}").into()),
PageReconstructError::WalRedo(pre) => ApiError::InternalServerError(pre),
}
}
@@ -239,19 +233,6 @@ impl From<GetTenantError> for ApiError {
}
}
impl From<GetActiveTenantError> for ApiError {
fn from(e: GetActiveTenantError) -> ApiError {
match e {
GetActiveTenantError::WillNotBecomeActive(_) => ApiError::Conflict(format!("{}", e)),
GetActiveTenantError::Cancelled => ApiError::ShuttingDown,
GetActiveTenantError::NotFound(gte) => gte.into(),
GetActiveTenantError::WaitForActiveTimeout { .. } => {
ApiError::ResourceUnavailable(format!("{}", e).into())
}
}
}
}
impl From<SetNewTenantConfigError> for ApiError {
fn from(e: SetNewTenantConfigError) -> ApiError {
match e {
@@ -307,7 +288,6 @@ impl From<crate::tenant::delete::DeleteTenantError> for ApiError {
SlotUpsertError(e) => e.into(),
Other(o) => ApiError::InternalServerError(o),
e @ InvalidState(_) => ApiError::PreconditionFailed(e.to_string().into_boxed_str()),
Cancelled => ApiError::ShuttingDown,
}
}
}
@@ -455,10 +435,7 @@ async fn timeline_create_handler(
let state = get_state(&request);
async {
let tenant = state.tenant_manager.get_attached_tenant_shard(tenant_shard_id, false)?;
tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
let tenant = state.tenant_manager.get_attached_tenant_shard(tenant_shard_id, true)?;
match tenant.create_timeline(
new_timeline_id,
request_data.ancestor_timeline_id.map(TimelineId::from),
@@ -593,6 +570,8 @@ async fn get_lsn_by_timestamp_handler(
)));
}
let version: Option<u8> = parse_query_param(&request, "version")?;
let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
let timestamp_raw = must_get_query_param(&request, "timestamp")?;
let timestamp = humantime::parse_rfc3339(&timestamp_raw)
@@ -605,18 +584,31 @@ async fn get_lsn_by_timestamp_handler(
let result = timeline
.find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
.await?;
#[derive(serde::Serialize)]
struct Result {
lsn: Lsn,
kind: &'static str,
if version.unwrap_or(0) > 1 {
#[derive(serde::Serialize)]
struct Result {
lsn: Lsn,
kind: &'static str,
}
let (lsn, kind) = match result {
LsnForTimestamp::Present(lsn) => (lsn, "present"),
LsnForTimestamp::Future(lsn) => (lsn, "future"),
LsnForTimestamp::Past(lsn) => (lsn, "past"),
LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
};
json_response(StatusCode::OK, Result { lsn, kind })
} else {
// FIXME: this is a temporary crutch not to break backwards compatibility
// See https://github.com/neondatabase/neon/pull/5608
let result = match result {
LsnForTimestamp::Present(lsn) => format!("{lsn}"),
LsnForTimestamp::Future(_lsn) => "future".into(),
LsnForTimestamp::Past(_lsn) => "past".into(),
LsnForTimestamp::NoData(_lsn) => "nodata".into(),
};
json_response(StatusCode::OK, result)
}
let (lsn, kind) = match result {
LsnForTimestamp::Present(lsn) => (lsn, "present"),
LsnForTimestamp::Future(lsn) => (lsn, "future"),
LsnForTimestamp::Past(lsn) => (lsn, "past"),
LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
};
json_response(StatusCode::OK, Result { lsn, kind })
}
async fn get_timestamp_of_lsn_handler(
@@ -702,23 +694,11 @@ async fn timeline_delete_handler(
let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
check_permission(&request, Some(tenant_shard_id.tenant_id))?;
let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
let state = get_state(&request);
let tenant = state
.tenant_manager
.get_attached_tenant_shard(tenant_shard_id, false)
.map_err(|e| {
match e {
// GetTenantError has a built-in conversion to ApiError, but in this context we don't
// want to treat missing tenants as 404, to avoid ambiguity with successful deletions.
GetTenantError::NotFound(_) => ApiError::PreconditionFailed(
"Requested tenant is missing".to_string().into_boxed_str(),
),
e => e.into(),
}
})?;
tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
tenant.delete_timeline(timeline_id).instrument(info_span!("timeline_delete", tenant_id=%tenant_shard_id.tenant_id, shard=%tenant_shard_id.shard_slug(), %timeline_id))
state.tenant_manager.delete_timeline(tenant_shard_id, timeline_id, &ctx)
.instrument(info_span!("timeline_delete", tenant_id=%tenant_shard_id.tenant_id, shard=%tenant_shard_id.shard_slug(), %timeline_id))
.await?;
json_response(StatusCode::ACCEPTED, ())
@@ -858,14 +838,11 @@ async fn tenant_status(
}
let state = tenant.current_state();
Result::<_, ApiError>::Ok(TenantDetails {
tenant_info: TenantInfo {
id: tenant_shard_id,
state: state.clone(),
current_physical_size: Some(current_physical_size),
attachment_status: state.attachment_status(),
},
timelines: tenant.list_timeline_ids(),
Result::<_, ApiError>::Ok(TenantInfo {
id: tenant_shard_id,
state: state.clone(),
current_physical_size: Some(current_physical_size),
attachment_status: state.attachment_status(),
})
}
.instrument(info_span!("tenant_status_handler",
@@ -886,9 +863,7 @@ async fn tenant_delete_handler(
let state = get_state(&request);
state
.tenant_manager
.delete_tenant(tenant_shard_id, ACTIVE_TENANT_TIMEOUT)
mgr::delete_tenant(state.conf, state.remote_storage.clone(), tenant_shard_id)
.instrument(info_span!("tenant_delete_handler",
tenant_id = %tenant_shard_id.tenant_id,
shard = %tenant_shard_id.shard_slug()
@@ -1161,10 +1136,7 @@ async fn tenant_create_handler(
// We created the tenant. Existing API semantics are that the tenant
// is Active when this function returns.
if let res @ Err(_) = new_tenant
.wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
.await
{
if let res @ Err(_) = new_tenant.wait_to_become_active().await {
// This shouldn't happen because we just created the tenant directory
// in tenant::mgr::create_tenant, and there aren't any remote timelines
// to load, so, nothing can really fail during load.
@@ -1274,23 +1246,6 @@ async fn put_tenant_location_config_handler(
// which is not a 400 but a 409.
.map_err(ApiError::BadRequest)?;
if let Some(_flush_ms) = flush {
match state
.secondary_controller
.upload_tenant(tenant_shard_id)
.await
{
Ok(()) => {
tracing::info!("Uploaded heatmap during flush");
}
Err(e) => {
tracing::warn!("Failed to flush heatmap: {e}");
}
}
} else {
tracing::info!("No flush requested when configuring");
}
json_response(StatusCode::OK, ())
}
@@ -1309,6 +1264,34 @@ async fn handle_tenant_break(
json_response(StatusCode::OK, ())
}
async fn failpoints_handler(
mut request: Request<Body>,
_cancel: CancellationToken,
) -> Result<Response<Body>, ApiError> {
if !fail::has_failpoints() {
return Err(ApiError::BadRequest(anyhow!(
"Cannot manage failpoints because pageserver was compiled without failpoints support"
)));
}
let failpoints: ConfigureFailpointsRequest = json_request(&mut request).await?;
for fp in failpoints {
info!("cfg failpoint: {} {}", fp.name, fp.actions);
// We recognize one extra "action" that's not natively recognized
// by the failpoints crate: exit, to immediately kill the process
let cfg_result = crate::failpoint_support::apply_failpoint(&fp.name, &fp.actions);
if let Err(err_msg) = cfg_result {
return Err(ApiError::BadRequest(anyhow!(
"Failed to configure failpoints: {err_msg}"
)));
}
}
json_response(StatusCode::OK, ())
}
// Run GC immediately on given timeline.
async fn timeline_gc_handler(
mut request: Request<Body>,
@@ -1504,6 +1487,69 @@ async fn timeline_collect_keyspace(
let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
check_permission(&request, Some(tenant_shard_id.tenant_id))?;
struct Partitioning {
keys: crate::keyspace::KeySpace,
at_lsn: Lsn,
}
impl serde::Serialize for Partitioning {
fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
use serde::ser::SerializeMap;
let mut map = serializer.serialize_map(Some(2))?;
map.serialize_key("keys")?;
map.serialize_value(&KeySpace(&self.keys))?;
map.serialize_key("at_lsn")?;
map.serialize_value(&WithDisplay(&self.at_lsn))?;
map.end()
}
}
struct WithDisplay<'a, T>(&'a T);
impl<'a, T: std::fmt::Display> serde::Serialize for WithDisplay<'a, T> {
fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
serializer.collect_str(&self.0)
}
}
struct KeySpace<'a>(&'a crate::keyspace::KeySpace);
impl<'a> serde::Serialize for KeySpace<'a> {
fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
use serde::ser::SerializeSeq;
let mut seq = serializer.serialize_seq(Some(self.0.ranges.len()))?;
for kr in &self.0.ranges {
seq.serialize_element(&KeyRange(kr))?;
}
seq.end()
}
}
struct KeyRange<'a>(&'a std::ops::Range<crate::repository::Key>);
impl<'a> serde::Serialize for KeyRange<'a> {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: serde::Serializer,
{
use serde::ser::SerializeTuple;
let mut t = serializer.serialize_tuple(2)?;
t.serialize_element(&WithDisplay(&self.0.start))?;
t.serialize_element(&WithDisplay(&self.0.end))?;
t.end()
}
}
let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
async {
@@ -1515,9 +1561,7 @@ async fn timeline_collect_keyspace(
.await
.map_err(|e| ApiError::InternalServerError(e.into()))?;
let res = pageserver_api::models::partitioning::Partitioning { keys, at_lsn };
json_response(StatusCode::OK, res)
json_response(StatusCode::OK, Partitioning { keys, at_lsn })
}
.instrument(info_span!("timeline_collect_keyspace", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
.await
@@ -1557,22 +1601,19 @@ async fn disk_usage_eviction_run(
struct Config {
/// How many bytes to evict before reporting that pressure is relieved.
evict_bytes: u64,
#[serde(default)]
eviction_order: crate::disk_usage_eviction_task::EvictionOrder,
}
#[derive(Debug, Clone, Copy, serde::Serialize)]
struct Usage {
// remains unchanged after instantiation of the struct
evict_bytes: u64,
config: Config,
// updated by `add_available_bytes`
freed_bytes: u64,
}
impl crate::disk_usage_eviction_task::Usage for Usage {
fn has_pressure(&self) -> bool {
self.evict_bytes > self.freed_bytes
self.config.evict_bytes > self.freed_bytes
}
fn add_available_bytes(&mut self, bytes: u64) {
@@ -1583,7 +1624,7 @@ async fn disk_usage_eviction_run(
let config = json_request::<Config>(&mut r).await?;
let usage = Usage {
evict_bytes: config.evict_bytes,
config,
freed_bytes: 0,
};
@@ -1598,11 +1639,7 @@ async fn disk_usage_eviction_run(
let state = state.disk_usage_eviction_state.clone();
let res = crate::disk_usage_eviction_task::disk_usage_eviction_task_iteration_impl(
&state,
storage,
usage,
config.eviction_order,
&cancel,
&state, storage, usage, &cancel,
)
.await;
@@ -1628,21 +1665,6 @@ async fn secondary_upload_handler(
json_response(StatusCode::OK, ())
}
async fn secondary_download_handler(
request: Request<Body>,
_cancel: CancellationToken,
) -> 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)?;
json_response(StatusCode::OK, ())
}
async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
json_response(
StatusCode::NOT_FOUND,
@@ -1911,9 +1933,6 @@ pub fn make_router(
.put("/v1/deletion_queue/flush", |r| {
api_handler(r, deletion_queue_flush)
})
.post("/v1/tenant/:tenant_shard_id/secondary/download", |r| {
api_handler(r, secondary_download_handler)
})
.put("/v1/tenant/:tenant_shard_id/break", |r| {
testing_api_handler("set tenant state to broken", r, handle_tenant_break)
})

View File

@@ -21,7 +21,6 @@ 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;
@@ -313,16 +312,13 @@ async fn import_wal(
waldecoder.feed_bytes(&buf);
let mut nrecords = 0;
let mut modification = tline.begin_modification(last_lsn);
let mut modification = tline.begin_modification(endpoint);
let mut decoded = DecodedWALRecord::default();
while last_lsn <= endpoint {
if let Some((lsn, recdata)) = waldecoder.poll_decode()? {
walingest
.ingest_record(recdata, lsn, &mut modification, &mut decoded, ctx)
.await?;
WAL_INGEST.records_committed.inc();
modification.commit(ctx).await?;
last_lsn = lsn;
nrecords += 1;
@@ -452,14 +448,13 @@ pub async fn import_wal_from_tar(
waldecoder.feed_bytes(&bytes[offset..]);
let mut modification = tline.begin_modification(last_lsn);
let mut modification = tline.begin_modification(end_lsn);
let mut decoded = DecodedWALRecord::default();
while last_lsn <= end_lsn {
if let Some((lsn, recdata)) = waldecoder.poll_decode()? {
walingest
.ingest_record(recdata, lsn, &mut modification, &mut decoded, ctx)
.await?;
modification.commit(ctx).await?;
last_lsn = lsn;
debug!("imported record at {} (end {})", lsn, end_lsn);

View File

@@ -1,12 +1,11 @@
use crate::repository::{key_range_size, singleton_range, Key};
use postgres_ffi::BLCKSZ;
use std::ops::Range;
use crate::key::Key;
///
/// Represents a set of Keys, in a compact form.
///
#[derive(Clone, Debug, Default, PartialEq, Eq)]
#[derive(Clone, Debug, Default)]
pub struct KeySpace {
/// Contiguous ranges of keys that belong to the key space. In key order,
/// and with no overlap.
@@ -124,9 +123,6 @@ impl KeySpaceAccum {
if range.start == accum.end {
accum.end = range.end;
} else {
// TODO: to efficiently support small sharding stripe sizes, we should avoid starting
// a new range here if the skipped region was all keys that don't belong on this shard.
// (https://github.com/neondatabase/neon/issues/6247)
assert!(range.start > accum.end);
self.ranges.push(accum.clone());
*accum = range;
@@ -190,33 +186,6 @@ impl KeySpaceRandomAccum {
}
}
pub fn key_range_size(key_range: &Range<Key>) -> u32 {
let start = key_range.start;
let end = key_range.end;
if end.field1 != start.field1
|| end.field2 != start.field2
|| end.field3 != start.field3
|| end.field4 != start.field4
{
return u32::MAX;
}
let start = (start.field5 as u64) << 32 | start.field6 as u64;
let end = (end.field5 as u64) << 32 | end.field6 as u64;
let diff = end - start;
if diff > u32::MAX as u64 {
u32::MAX
} else {
diff as u32
}
}
pub fn singleton_range(key: Key) -> Range<Key> {
key..key.next()
}
#[cfg(test)]
mod tests {
use super::*;

View File

@@ -10,7 +10,7 @@ pub mod deletion_queue;
pub mod disk_usage_eviction_task;
pub mod http;
pub mod import_datadir;
pub use pageserver_api::keyspace;
pub mod keyspace;
pub mod metrics;
pub mod page_cache;
pub mod page_service;
@@ -25,6 +25,8 @@ pub mod walingest;
pub mod walrecord;
pub mod walredo;
pub mod failpoint_support;
use crate::task_mgr::TaskKind;
use camino::Utf8Path;
use deletion_queue::DeletionQueue;
@@ -117,10 +119,6 @@ pub const TENANT_CONFIG_NAME: &str = "config";
/// Full path: `tenants/<tenant_id>/config`.
pub 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";
/// 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";

View File

@@ -29,7 +29,7 @@ const CRITICAL_OP_BUCKETS: &[f64] = &[
// Metrics collected on operations on the storage repository.
#[derive(Debug, EnumVariantNames, IntoStaticStr)]
#[strum(serialize_all = "kebab_case")]
pub(crate) enum StorageTimeOperation {
pub enum StorageTimeOperation {
#[strum(serialize = "layer flush")]
LayerFlush,
@@ -55,7 +55,7 @@ pub(crate) enum StorageTimeOperation {
CreateTenant,
}
pub(crate) static STORAGE_TIME_SUM_PER_TIMELINE: Lazy<CounterVec> = Lazy::new(|| {
pub static STORAGE_TIME_SUM_PER_TIMELINE: Lazy<CounterVec> = Lazy::new(|| {
register_counter_vec!(
"pageserver_storage_operations_seconds_sum",
"Total time spent on storage operations with operation, tenant and timeline dimensions",
@@ -64,7 +64,7 @@ pub(crate) static STORAGE_TIME_SUM_PER_TIMELINE: Lazy<CounterVec> = Lazy::new(||
.expect("failed to define a metric")
});
pub(crate) static STORAGE_TIME_COUNT_PER_TIMELINE: Lazy<IntCounterVec> = Lazy::new(|| {
pub static STORAGE_TIME_COUNT_PER_TIMELINE: Lazy<IntCounterVec> = Lazy::new(|| {
register_int_counter_vec!(
"pageserver_storage_operations_seconds_count",
"Count of storage operations with operation, tenant and timeline dimensions",
@@ -150,7 +150,7 @@ pub(crate) static MATERIALIZED_PAGE_CACHE_HIT: Lazy<IntCounter> = Lazy::new(|| {
.expect("failed to define a metric")
});
pub(crate) struct PageCacheMetricsForTaskKind {
pub struct PageCacheMetricsForTaskKind {
pub read_accesses_materialized_page: IntCounter,
pub read_accesses_immutable: IntCounter,
@@ -159,7 +159,7 @@ pub(crate) struct PageCacheMetricsForTaskKind {
pub read_hits_materialized_page_older_lsn: IntCounter,
}
pub(crate) struct PageCacheMetrics {
pub struct PageCacheMetrics {
map: EnumMap<TaskKind, EnumMap<PageContentKind, PageCacheMetricsForTaskKind>>,
}
@@ -181,7 +181,7 @@ static PAGE_CACHE_READ_ACCESSES: Lazy<IntCounterVec> = Lazy::new(|| {
.expect("failed to define a metric")
});
pub(crate) static PAGE_CACHE: Lazy<PageCacheMetrics> = Lazy::new(|| PageCacheMetrics {
pub static PAGE_CACHE: Lazy<PageCacheMetrics> = Lazy::new(|| PageCacheMetrics {
map: EnumMap::from_array(std::array::from_fn(|task_kind| {
let task_kind = <TaskKind as enum_map::Enum>::from_usize(task_kind);
let task_kind: &'static str = task_kind.into();
@@ -243,9 +243,10 @@ impl PageCacheMetrics {
}
}
pub(crate) struct PageCacheSizeMetrics {
pub struct PageCacheSizeMetrics {
pub max_bytes: UIntGauge,
pub current_bytes_ephemeral: UIntGauge,
pub current_bytes_immutable: UIntGauge,
pub current_bytes_materialized_page: UIntGauge,
}
@@ -259,26 +260,31 @@ static PAGE_CACHE_SIZE_CURRENT_BYTES: Lazy<UIntGaugeVec> = Lazy::new(|| {
.expect("failed to define a metric")
});
pub(crate) static PAGE_CACHE_SIZE: Lazy<PageCacheSizeMetrics> =
Lazy::new(|| PageCacheSizeMetrics {
max_bytes: {
register_uint_gauge!(
"pageserver_page_cache_size_max_bytes",
"Maximum size of the page cache in bytes"
)
.expect("failed to define a metric")
},
current_bytes_immutable: {
PAGE_CACHE_SIZE_CURRENT_BYTES
.get_metric_with_label_values(&["immutable"])
.unwrap()
},
current_bytes_materialized_page: {
PAGE_CACHE_SIZE_CURRENT_BYTES
.get_metric_with_label_values(&["materialized_page"])
.unwrap()
},
});
pub static PAGE_CACHE_SIZE: Lazy<PageCacheSizeMetrics> = Lazy::new(|| PageCacheSizeMetrics {
max_bytes: {
register_uint_gauge!(
"pageserver_page_cache_size_max_bytes",
"Maximum size of the page cache in bytes"
)
.expect("failed to define a metric")
},
current_bytes_ephemeral: {
PAGE_CACHE_SIZE_CURRENT_BYTES
.get_metric_with_label_values(&["ephemeral"])
.unwrap()
},
current_bytes_immutable: {
PAGE_CACHE_SIZE_CURRENT_BYTES
.get_metric_with_label_values(&["immutable"])
.unwrap()
},
current_bytes_materialized_page: {
PAGE_CACHE_SIZE_CURRENT_BYTES
.get_metric_with_label_values(&["materialized_page"])
.unwrap()
},
});
pub(crate) mod page_cache_eviction_metrics {
use std::num::NonZeroUsize;
@@ -516,18 +522,14 @@ pub(crate) mod initial_logical_size {
impl StartCalculation {
pub(crate) fn first(&self, circumstances: StartCircumstances) -> OngoingCalculationGuard {
let circumstances_label: &'static str = circumstances.into();
self.0
.with_label_values(&["first", circumstances_label])
.inc();
self.0.with_label_values(&["first", circumstances_label]);
OngoingCalculationGuard {
inc_drop_calculation: Some(DROP_CALCULATION.first.clone()),
}
}
pub(crate) fn retry(&self, circumstances: StartCircumstances) -> OngoingCalculationGuard {
let circumstances_label: &'static str = circumstances.into();
self.0
.with_label_values(&["retry", circumstances_label])
.inc();
self.0.with_label_values(&["retry", circumstances_label]);
OngoingCalculationGuard {
inc_drop_calculation: Some(DROP_CALCULATION.retry.clone()),
}
@@ -682,65 +684,25 @@ pub static STARTUP_IS_LOADING: Lazy<UIntGauge> = Lazy::new(|| {
.expect("Failed to register pageserver_startup_is_loading")
});
/// Metrics related to the lifecycle of a [`crate::tenant::Tenant`] object: things
/// like how long it took to load.
///
/// Note that these are process-global metrics, _not_ per-tenant metrics. Per-tenant
/// metrics are rather expensive, and usually fine grained stuff makes more sense
/// at a timeline level than tenant level.
pub(crate) struct TenantMetrics {
/// How long did tenants take to go from construction to active state?
pub(crate) activation: Histogram,
pub(crate) preload: Histogram,
pub(crate) attach: Histogram,
/// How many tenants are included in the initial startup of the pagesrever?
pub(crate) startup_scheduled: IntCounter,
pub(crate) startup_complete: IntCounter,
}
pub(crate) static TENANT: Lazy<TenantMetrics> = Lazy::new(|| {
TenantMetrics {
activation: register_histogram!(
/// How long did tenants take to go from construction to active state?
pub(crate) static TENANT_ACTIVATION: Lazy<Histogram> = Lazy::new(|| {
register_histogram!(
"pageserver_tenant_activation_seconds",
"Time taken by tenants to activate, in seconds",
CRITICAL_OP_BUCKETS.into()
)
.expect("Failed to register metric"),
preload: register_histogram!(
"pageserver_tenant_preload_seconds",
"Time taken by tenants to load remote metadata on startup/attach, in seconds",
CRITICAL_OP_BUCKETS.into()
)
.expect("Failed to register metric"),
attach: register_histogram!(
"pageserver_tenant_attach_seconds",
"Time taken by tenants to intialize, after remote metadata is already loaded",
CRITICAL_OP_BUCKETS.into()
)
.expect("Failed to register metric"),
startup_scheduled: register_int_counter!(
"pageserver_tenant_startup_scheduled",
"Number of tenants included in pageserver startup (doesn't count tenants attached later)"
).expect("Failed to register metric"),
startup_complete: register_int_counter!(
"pageserver_tenant_startup_complete",
"Number of tenants that have completed warm-up, or activated on-demand during initial startup: \
should eventually reach `pageserver_tenant_startup_scheduled_total`. Does not include broken \
tenants: such cases will lead to this metric never reaching the scheduled count."
).expect("Failed to register metric"),
}
.expect("Failed to register pageserver_tenant_activation_seconds metric")
});
/// Each `Timeline`'s [`EVICTIONS_WITH_LOW_RESIDENCE_DURATION`] metric.
#[derive(Debug)]
pub(crate) struct EvictionsWithLowResidenceDuration {
pub struct EvictionsWithLowResidenceDuration {
data_source: &'static str,
threshold: Duration,
counter: Option<IntCounter>,
}
pub(crate) struct EvictionsWithLowResidenceDurationBuilder {
pub struct EvictionsWithLowResidenceDurationBuilder {
data_source: &'static str,
threshold: Duration,
}
@@ -1003,7 +965,7 @@ pub enum SmgrQueryType {
}
#[derive(Debug)]
pub(crate) struct SmgrQueryTimePerTimeline {
pub struct SmgrQueryTimePerTimeline {
metrics: [GlobalAndPerTimelineHistogram; SmgrQueryType::COUNT],
}
@@ -1017,62 +979,12 @@ static SMGR_QUERY_TIME_PER_TENANT_TIMELINE: Lazy<HistogramVec> = Lazy::new(|| {
.expect("failed to define a metric")
});
static SMGR_QUERY_TIME_GLOBAL_BUCKETS: Lazy<Vec<f64>> = Lazy::new(|| {
[
1,
10,
20,
40,
60,
80,
100,
200,
300,
400,
500,
600,
700,
800,
900,
1_000, // 1ms
2_000,
4_000,
6_000,
8_000,
10_000, // 10ms
20_000,
40_000,
60_000,
80_000,
100_000,
200_000,
400_000,
600_000,
800_000,
1_000_000, // 1s
2_000_000,
4_000_000,
6_000_000,
8_000_000,
10_000_000, // 10s
20_000_000,
50_000_000,
100_000_000,
200_000_000,
1_000_000_000, // 1000s
]
.into_iter()
.map(Duration::from_micros)
.map(|d| d.as_secs_f64())
.collect()
});
static SMGR_QUERY_TIME_GLOBAL: Lazy<HistogramVec> = Lazy::new(|| {
register_histogram_vec!(
"pageserver_smgr_query_seconds_global",
"Time spent on smgr query handling, aggregated by query type.",
&["smgr_query_type"],
SMGR_QUERY_TIME_GLOBAL_BUCKETS.clone(),
CRITICAL_OP_BUCKETS.into(),
)
.expect("failed to define a metric")
});
@@ -1175,8 +1087,8 @@ static COMPUTE_STARTUP_BUCKETS: Lazy<[f64; 28]> = Lazy::new(|| {
.map(|ms| (ms as f64) / 1000.0)
});
pub(crate) struct BasebackupQueryTime(HistogramVec);
pub(crate) static BASEBACKUP_QUERY_TIME: Lazy<BasebackupQueryTime> = Lazy::new(|| {
pub struct BasebackupQueryTime(HistogramVec);
pub static BASEBACKUP_QUERY_TIME: Lazy<BasebackupQueryTime> = Lazy::new(|| {
BasebackupQueryTime({
register_histogram_vec!(
"pageserver_basebackup_query_seconds",
@@ -1196,7 +1108,7 @@ impl DurationResultObserver for BasebackupQueryTime {
}
}
pub(crate) static LIVE_CONNECTIONS_COUNT: Lazy<IntGaugeVec> = Lazy::new(|| {
pub static LIVE_CONNECTIONS_COUNT: Lazy<IntGaugeVec> = Lazy::new(|| {
register_int_gauge_vec!(
"pageserver_live_connections",
"Number of live network connections",
@@ -1363,8 +1275,6 @@ pub(crate) struct SecondaryModeMetrics {
pub(crate) upload_heatmap: IntCounter,
pub(crate) upload_heatmap_errors: IntCounter,
pub(crate) upload_heatmap_duration: Histogram,
pub(crate) download_heatmap: IntCounter,
pub(crate) download_layer: IntCounter,
}
pub(crate) static SECONDARY_MODE: Lazy<SecondaryModeMetrics> = Lazy::new(|| SecondaryModeMetrics {
upload_heatmap: register_int_counter!(
@@ -1382,16 +1292,6 @@ pub(crate) static SECONDARY_MODE: Lazy<SecondaryModeMetrics> = Lazy::new(|| Seco
"Time to build and upload a heatmap, including any waiting inside the S3 client"
)
.expect("failed to define a metric"),
download_heatmap: register_int_counter!(
"pageserver_secondary_download_heatmap",
"Number of downloads of heatmaps by secondary mode locations"
)
.expect("failed to define a metric"),
download_layer: register_int_counter!(
"pageserver_secondary_download_layer",
"Number of downloads of layers by secondary mode locations"
)
.expect("failed to define a metric"),
});
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
@@ -1661,7 +1561,7 @@ pub(crate) static WAL_REDO_PROCESS_COUNTERS: Lazy<WalRedoProcessCounters> =
Lazy::new(WalRedoProcessCounters::default);
/// Similar to `prometheus::HistogramTimer` but does not record on drop.
pub(crate) struct StorageTimeMetricsTimer {
pub struct StorageTimeMetricsTimer {
metrics: StorageTimeMetrics,
start: Instant,
}
@@ -1686,7 +1586,7 @@ impl StorageTimeMetricsTimer {
/// Timing facilities for an globally histogrammed metric, which is supported by per tenant and
/// timeline total sum and count.
#[derive(Clone, Debug)]
pub(crate) struct StorageTimeMetrics {
pub struct StorageTimeMetrics {
/// Sum of f64 seconds, per operation, tenant_id and timeline_id
timeline_sum: Counter,
/// Number of oeprations, per operation, tenant_id and timeline_id
@@ -1725,7 +1625,7 @@ impl StorageTimeMetrics {
}
#[derive(Debug)]
pub(crate) struct TimelineMetrics {
pub struct TimelineMetrics {
tenant_id: String,
shard_id: String,
timeline_id: String,
@@ -1933,7 +1833,7 @@ impl Drop for PerTimelineRemotePhysicalSizeGauge {
}
}
pub(crate) struct RemoteTimelineClientMetrics {
pub struct RemoteTimelineClientMetrics {
tenant_id: String,
timeline_id: String,
remote_physical_size_gauge: Mutex<Option<PerTimelineRemotePhysicalSizeGauge>>,
@@ -2231,7 +2131,7 @@ impl Drop for RemoteTimelineClientMetrics {
/// Wrapper future that measures the time spent by a remote storage operation,
/// and records the time and success/failure as a prometheus metric.
pub(crate) trait MeasureRemoteOp: Sized {
pub trait MeasureRemoteOp: Sized {
fn measure_remote_op(
self,
tenant_id: TenantId,
@@ -2256,7 +2156,7 @@ pub(crate) trait MeasureRemoteOp: Sized {
impl<T: Sized> MeasureRemoteOp for T {}
pin_project! {
pub(crate) struct MeasuredRemoteOp<F>
pub struct MeasuredRemoteOp<F>
{
#[pin]
inner: F,
@@ -2313,9 +2213,6 @@ pub fn preinitialize_metrics() {
// Deletion queue stats
Lazy::force(&DELETION_QUEUE);
// Tenant stats
Lazy::force(&TENANT);
// Tenant manager stats
Lazy::force(&TENANT_MANAGER);

View File

@@ -25,7 +25,6 @@ use postgres_backend::{self, is_expected_io_error, AuthType, PostgresBackend, Qu
use pq_proto::framed::ConnectionError;
use pq_proto::FeStartupPacket;
use pq_proto::{BeMessage, FeMessage, RowDescriptor};
use std::borrow::Cow;
use std::io;
use std::net::TcpListener;
use std::pin::pin;
@@ -54,7 +53,7 @@ use crate::context::{DownloadBehavior, RequestContext};
use crate::import_datadir::import_wal_from_tar;
use crate::metrics;
use crate::metrics::LIVE_CONNECTIONS_COUNT;
use crate::pgdatadir_mapping::{rel_block_to_key, Version};
use crate::pgdatadir_mapping::rel_block_to_key;
use crate::task_mgr;
use crate::task_mgr::TaskKind;
use crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id;
@@ -62,9 +61,6 @@ use crate::tenant::mgr;
use crate::tenant::mgr::get_active_tenant_with_timeout;
use crate::tenant::mgr::GetActiveTenantError;
use crate::tenant::mgr::ShardSelector;
use crate::tenant::timeline::WaitLsnError;
use crate::tenant::GetTimelineError;
use crate::tenant::PageReconstructError;
use crate::tenant::Timeline;
use crate::trace::Tracer;
@@ -287,64 +283,6 @@ struct PageServerHandler {
connection_ctx: RequestContext,
}
#[derive(thiserror::Error, Debug)]
enum PageStreamError {
/// We encountered an error that should prompt the client to reconnect:
/// in practice this means we drop the connection without sending a response.
#[error("Reconnect required: {0}")]
Reconnect(Cow<'static, str>),
/// We were instructed to shutdown while processing the query
#[error("Shutting down")]
Shutdown,
/// Something went wrong reading a page: this likely indicates a pageserver bug
#[error("Read error: {0}")]
Read(PageReconstructError),
/// Ran out of time waiting for an LSN
#[error("LSN timeout: {0}")]
LsnTimeout(WaitLsnError),
/// The entity required to serve the request (tenant or timeline) is not found,
/// or is not found in a suitable state to serve a request.
#[error("Not found: {0}")]
NotFound(std::borrow::Cow<'static, str>),
/// Request asked for something that doesn't make sense, like an invalid LSN
#[error("Bad request: {0}")]
BadRequest(std::borrow::Cow<'static, str>),
}
impl From<PageReconstructError> for PageStreamError {
fn from(value: PageReconstructError) -> Self {
match value {
PageReconstructError::Cancelled => Self::Shutdown,
e => Self::Read(e),
}
}
}
impl From<GetActiveTimelineError> for PageStreamError {
fn from(value: GetActiveTimelineError) -> Self {
match value {
GetActiveTimelineError::Tenant(GetActiveTenantError::Cancelled) => Self::Shutdown,
GetActiveTimelineError::Tenant(e) => Self::NotFound(format!("{e}").into()),
GetActiveTimelineError::Timeline(e) => Self::NotFound(format!("{e}").into()),
}
}
}
impl From<WaitLsnError> for PageStreamError {
fn from(value: WaitLsnError) -> Self {
match value {
e @ WaitLsnError::Timeout(_) => Self::LsnTimeout(e),
WaitLsnError::Shutdown => Self::Shutdown,
WaitLsnError::BadState => Self::Reconnect("Timeline is not active".into()),
}
}
}
impl PageServerHandler {
pub fn new(
conf: &'static PageServerConf,
@@ -490,7 +428,7 @@ impl PageServerHandler {
// Check that the timeline exists
let timeline = tenant
.get_timeline(timeline_id, true)
.map_err(|e| QueryError::NotFound(format!("{e}").into()))?;
.map_err(|e| anyhow::anyhow!(e))?;
// Avoid starting new requests if the timeline has already started shutting down,
// and block timeline shutdown until this request is complete, or drops out due
@@ -582,44 +520,32 @@ impl PageServerHandler {
}
};
match response {
Err(PageStreamError::Shutdown) => {
if let Err(e) = &response {
// Requests may fail as soon as we are Stopping, even if the Timeline's cancellation token wasn't fired yet,
// because wait_lsn etc will drop out
// is_stopping(): [`Timeline::flush_and_shutdown`] has entered
// is_canceled(): [`Timeline::shutdown`]` has entered
if timeline.cancel.is_cancelled() || timeline.is_stopping() {
// If we fail to fulfil a request during shutdown, which may be _because_ of
// shutdown, then do not send the error to the client. Instead just drop the
// connection.
span.in_scope(|| info!("dropping connection due to shutdown"));
span.in_scope(|| info!("dropped response during shutdown: {e:#}"));
return Err(QueryError::Shutdown);
}
Err(PageStreamError::Reconnect(reason)) => {
span.in_scope(|| info!("handler requested reconnect: {reason}"));
return Err(QueryError::Reconnect);
}
Err(e) if timeline.cancel.is_cancelled() || timeline.is_stopping() => {
// This branch accomodates code within request handlers that returns an anyhow::Error instead of a clean
// shutdown error, this may be buried inside a PageReconstructError::Other for example.
//
// Requests may fail as soon as we are Stopping, even if the Timeline's cancellation token wasn't fired yet,
// because wait_lsn etc will drop out
// is_stopping(): [`Timeline::flush_and_shutdown`] has entered
// is_canceled(): [`Timeline::shutdown`]` has entered
span.in_scope(|| info!("dropped error response during shutdown: {e:#}"));
return Err(QueryError::Shutdown);
}
r => {
let response_msg = r.unwrap_or_else(|e| {
// print the all details to the log with {:#}, but for the client the
// error message is enough. Do not log if shutting down, as the anyhow::Error
// here includes cancellation which is not an error.
span.in_scope(|| error!("error reading relation or page version: {:#}", e));
PagestreamBeMessage::Error(PagestreamErrorResponse {
message: e.to_string(),
})
});
pgb.write_message_noflush(&BeMessage::CopyData(&response_msg.serialize()))?;
self.flush_cancellable(pgb, &timeline.cancel).await?;
}
}
let response = response.unwrap_or_else(|e| {
// print the all details to the log with {:#}, but for the client the
// error message is enough. Do not log if shutting down, as the anyhow::Error
// here includes cancellation which is not an error.
span.in_scope(|| error!("error reading relation or page version: {:#}", e));
PagestreamBeMessage::Error(PagestreamErrorResponse {
message: e.to_string(),
})
});
pgb.write_message_noflush(&BeMessage::CopyData(&response.serialize()))?;
self.flush_cancellable(pgb, &timeline.cancel).await?;
}
Ok(())
}
@@ -766,7 +692,7 @@ impl PageServerHandler {
latest: bool,
latest_gc_cutoff_lsn: &RcuReadGuard<Lsn>,
ctx: &RequestContext,
) -> Result<Lsn, PageStreamError> {
) -> anyhow::Result<Lsn> {
if latest {
// Latest page version was requested. If LSN is given, it is a hint
// to the page server that there have been no modifications to the
@@ -797,19 +723,15 @@ impl PageServerHandler {
}
} else {
if lsn == Lsn(0) {
return Err(PageStreamError::BadRequest(
"invalid LSN(0) in request".into(),
));
anyhow::bail!("invalid LSN(0) in request");
}
timeline.wait_lsn(lsn, ctx).await?;
}
if lsn < **latest_gc_cutoff_lsn {
return Err(PageStreamError::BadRequest(format!(
"tried to request a page version that was garbage collected. requested at {} gc cutoff {}",
lsn, **latest_gc_cutoff_lsn
).into()));
}
anyhow::ensure!(
lsn >= **latest_gc_cutoff_lsn,
"tried to request a page version that was garbage collected. requested at {} gc cutoff {}",
lsn, **latest_gc_cutoff_lsn
);
Ok(lsn)
}
@@ -818,14 +740,14 @@ impl PageServerHandler {
timeline: &Timeline,
req: &PagestreamExistsRequest,
ctx: &RequestContext,
) -> Result<PagestreamBeMessage, PageStreamError> {
) -> anyhow::Result<PagestreamBeMessage> {
let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
let lsn =
Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
.await?;
let exists = timeline
.get_rel_exists(req.rel, Version::Lsn(lsn), req.latest, ctx)
.get_rel_exists(req.rel, lsn, req.latest, ctx)
.await?;
Ok(PagestreamBeMessage::Exists(PagestreamExistsResponse {
@@ -838,15 +760,13 @@ impl PageServerHandler {
timeline: &Timeline,
req: &PagestreamNblocksRequest,
ctx: &RequestContext,
) -> Result<PagestreamBeMessage, PageStreamError> {
) -> anyhow::Result<PagestreamBeMessage> {
let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
let lsn =
Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
.await?;
let n_blocks = timeline
.get_rel_size(req.rel, Version::Lsn(lsn), req.latest, ctx)
.await?;
let n_blocks = timeline.get_rel_size(req.rel, lsn, req.latest, ctx).await?;
Ok(PagestreamBeMessage::Nblocks(PagestreamNblocksResponse {
n_blocks,
@@ -858,20 +778,14 @@ impl PageServerHandler {
timeline: &Timeline,
req: &PagestreamDbSizeRequest,
ctx: &RequestContext,
) -> Result<PagestreamBeMessage, PageStreamError> {
) -> anyhow::Result<PagestreamBeMessage> {
let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
let lsn =
Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
.await?;
let total_blocks = timeline
.get_db_size(
DEFAULTTABLESPACE_OID,
req.dbnode,
Version::Lsn(lsn),
req.latest,
ctx,
)
.get_db_size(DEFAULTTABLESPACE_OID, req.dbnode, lsn, req.latest, ctx)
.await?;
let db_size = total_blocks as i64 * BLCKSZ as i64;
@@ -880,35 +794,30 @@ impl PageServerHandler {
}))
}
async fn do_handle_get_page_at_lsn_request(
&self,
timeline: &Timeline,
req: &PagestreamGetPageRequest,
ctx: &RequestContext,
) -> Result<PagestreamBeMessage, PageStreamError> {
let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
let lsn =
Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
.await?;
let page = timeline
.get_rel_page_at_lsn(req.rel, req.blkno, Version::Lsn(lsn), req.latest, ctx)
.await?;
Ok(PagestreamBeMessage::GetPage(PagestreamGetPageResponse {
page,
}))
}
async fn handle_get_page_at_lsn_request(
&self,
timeline: &Timeline,
req: &PagestreamGetPageRequest,
ctx: &RequestContext,
) -> Result<PagestreamBeMessage, PageStreamError> {
) -> anyhow::Result<PagestreamBeMessage> {
let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
let lsn =
Self::wait_or_get_last_lsn(timeline, req.lsn, req.latest, &latest_gc_cutoff_lsn, ctx)
.await?;
/*
// Add a 1s delay to some requests. The delay helps the requests to
// hit the race condition from github issue #1047 more easily.
use rand::Rng;
if rand::thread_rng().gen::<u8>() < 5 {
std::thread::sleep(std::time::Duration::from_millis(1000));
}
*/
let key = rel_block_to_key(req.rel, req.blkno);
if timeline.get_shard_identity().is_key_local(&key) {
self.do_handle_get_page_at_lsn_request(timeline, req, ctx)
.await
let page = if timeline.get_shard_identity().is_key_local(&key) {
timeline
.get_rel_page_at_lsn(req.rel, req.blkno, lsn, req.latest, ctx)
.await?
} else {
// The Tenant shard we looked up at connection start does not hold this particular
// key: look for other shards in this tenant. This scenario occurs if a pageserver
@@ -927,30 +836,30 @@ impl PageServerHandler {
Err(GetActiveTimelineError::Tenant(GetActiveTenantError::NotFound(_))) => {
// We already know this tenant exists in general, because we resolved it at
// start of connection. Getting a NotFound here indicates that the shard containing
// the requested page is not present on this node: the client's knowledge of shard->pageserver
// mapping is out of date.
tracing::info!("Page request routed to wrong shard: my identity {:?}, should go to shard {}, key {}",
timeline.get_shard_identity(), timeline.get_shard_identity().get_shard_number(&key).0, key);
// Closing the connection by returning ``::Reconnect` has the side effect of rate-limiting above message, via
// client's reconnect backoff, as well as hopefully prompting the client to load its updated configuration
// and talk to a different pageserver.
return Err(PageStreamError::Reconnect(
"getpage@lsn request routed to wrong shard".into(),
));
// the requested page is not present on this node.
// TODO: this should be some kind of structured error that the client will understand,
// so that it can block until its config is updated: this error is expected in the case
// that the Tenant's shards' placements are being updated and the client hasn't been
// informed yet.
//
// https://github.com/neondatabase/neon/issues/6038
return Err(anyhow::anyhow!("Request routed to wrong shard"));
}
Err(e) => return Err(e.into()),
};
// Take a GateGuard for the duration of this request. If we were using our main Timeline object,
// the GateGuard was already held over the whole connection.
let _timeline_guard = timeline
.gate
.enter()
.map_err(|_| PageStreamError::Shutdown)?;
let _timeline_guard = timeline.gate.enter().map_err(|_| QueryError::Shutdown)?;
timeline
.get_rel_page_at_lsn(req.rel, req.blkno, lsn, req.latest, ctx)
.await?
};
self.do_handle_get_page_at_lsn_request(&timeline, req, ctx)
.await
}
Ok(PagestreamBeMessage::GetPage(PagestreamGetPageResponse {
page,
}))
}
#[allow(clippy::too_many_arguments)]
@@ -1091,7 +1000,9 @@ impl PageServerHandler {
)
.await
.map_err(GetActiveTimelineError::Tenant)?;
let timeline = tenant.get_timeline(timeline_id, true)?;
let timeline = tenant
.get_timeline(timeline_id, true)
.map_err(|e| GetActiveTimelineError::Timeline(anyhow::anyhow!(e)))?;
Ok(timeline)
}
}
@@ -1513,15 +1424,14 @@ enum GetActiveTimelineError {
#[error(transparent)]
Tenant(GetActiveTenantError),
#[error(transparent)]
Timeline(#[from] GetTimelineError),
Timeline(anyhow::Error),
}
impl From<GetActiveTimelineError> for QueryError {
fn from(e: GetActiveTimelineError) -> Self {
match e {
GetActiveTimelineError::Tenant(GetActiveTenantError::Cancelled) => QueryError::Shutdown,
GetActiveTimelineError::Tenant(e) => e.into(),
GetActiveTimelineError::Timeline(e) => QueryError::NotFound(format!("{e}").into()),
GetActiveTimelineError::Timeline(e) => QueryError::Other(e),
}
}
}

View File

@@ -11,7 +11,7 @@ use crate::context::RequestContext;
use crate::keyspace::{KeySpace, KeySpaceAccum};
use crate::repository::*;
use crate::walrecord::NeonWalRecord;
use anyhow::{ensure, Context};
use anyhow::Context;
use bytes::{Buf, Bytes};
use pageserver_api::key::is_rel_block_key;
use pageserver_api::reltag::{RelTag, SlruKind};
@@ -147,7 +147,6 @@ impl Timeline {
{
DatadirModification {
tline: self,
pending_lsns: Vec::new(),
pending_updates: HashMap::new(),
pending_deletions: Vec::new(),
pending_nblocks: 0,
@@ -160,11 +159,11 @@ impl Timeline {
//------------------------------------------------------------------------------
/// Look up given page version.
pub(crate) async fn get_rel_page_at_lsn(
pub async fn get_rel_page_at_lsn(
&self,
tag: RelTag,
blknum: BlockNumber,
version: Version<'_>,
lsn: Lsn,
latest: bool,
ctx: &RequestContext,
) -> Result<Bytes, PageReconstructError> {
@@ -174,47 +173,44 @@ impl Timeline {
));
}
let nblocks = self.get_rel_size(tag, version, latest, ctx).await?;
let nblocks = self.get_rel_size(tag, lsn, latest, ctx).await?;
if blknum >= nblocks {
debug!(
"read beyond EOF at {} blk {} at {}, size is {}: returning all-zeros page",
tag,
blknum,
version.get_lsn(),
nblocks
tag, blknum, lsn, nblocks
);
return Ok(ZERO_PAGE.clone());
}
let key = rel_block_to_key(tag, blknum);
version.get(self, key, ctx).await
self.get(key, lsn, ctx).await
}
// Get size of a database in blocks
pub(crate) async fn get_db_size(
pub async fn get_db_size(
&self,
spcnode: Oid,
dbnode: Oid,
version: Version<'_>,
lsn: Lsn,
latest: bool,
ctx: &RequestContext,
) -> Result<usize, PageReconstructError> {
let mut total_blocks = 0;
let rels = self.list_rels(spcnode, dbnode, version, ctx).await?;
let rels = self.list_rels(spcnode, dbnode, lsn, ctx).await?;
for rel in rels {
let n_blocks = self.get_rel_size(rel, version, latest, ctx).await?;
let n_blocks = self.get_rel_size(rel, lsn, latest, ctx).await?;
total_blocks += n_blocks as usize;
}
Ok(total_blocks)
}
/// Get size of a relation file
pub(crate) async fn get_rel_size(
pub async fn get_rel_size(
&self,
tag: RelTag,
version: Version<'_>,
lsn: Lsn,
latest: bool,
ctx: &RequestContext,
) -> Result<BlockNumber, PageReconstructError> {
@@ -224,12 +220,12 @@ impl Timeline {
));
}
if let Some(nblocks) = self.get_cached_rel_size(&tag, version.get_lsn()) {
if let Some(nblocks) = self.get_cached_rel_size(&tag, lsn) {
return Ok(nblocks);
}
if (tag.forknum == FSM_FORKNUM || tag.forknum == VISIBILITYMAP_FORKNUM)
&& !self.get_rel_exists(tag, version, latest, ctx).await?
&& !self.get_rel_exists(tag, lsn, latest, ctx).await?
{
// FIXME: Postgres sometimes calls smgrcreate() to create
// FSM, and smgrnblocks() on it immediately afterwards,
@@ -239,7 +235,7 @@ impl Timeline {
}
let key = rel_size_to_key(tag);
let mut buf = version.get(self, key, ctx).await?;
let mut buf = self.get(key, lsn, ctx).await?;
let nblocks = buf.get_u32_le();
if latest {
@@ -250,16 +246,16 @@ impl Timeline {
// latest=true, then it can not cause cache corruption, because with latest=true
// pageserver choose max(request_lsn, last_written_lsn) and so cached value will be
// associated with most recent value of LSN.
self.update_cached_rel_size(tag, version.get_lsn(), nblocks);
self.update_cached_rel_size(tag, lsn, nblocks);
}
Ok(nblocks)
}
/// Does relation exist?
pub(crate) async fn get_rel_exists(
pub async fn get_rel_exists(
&self,
tag: RelTag,
version: Version<'_>,
lsn: Lsn,
_latest: bool,
ctx: &RequestContext,
) -> Result<bool, PageReconstructError> {
@@ -270,12 +266,12 @@ impl Timeline {
}
// first try to lookup relation in cache
if let Some(_nblocks) = self.get_cached_rel_size(&tag, version.get_lsn()) {
if let Some(_nblocks) = self.get_cached_rel_size(&tag, lsn) {
return Ok(true);
}
// fetch directory listing
let key = rel_dir_to_key(tag.spcnode, tag.dbnode);
let buf = version.get(self, key, ctx).await?;
let buf = self.get(key, lsn, ctx).await?;
match RelDirectory::des(&buf).context("deserialization failure") {
Ok(dir) => {
@@ -291,16 +287,16 @@ impl Timeline {
/// # Cancel-Safety
///
/// This method is cancellation-safe.
pub(crate) async fn list_rels(
pub async fn list_rels(
&self,
spcnode: Oid,
dbnode: Oid,
version: Version<'_>,
lsn: Lsn,
ctx: &RequestContext,
) -> Result<HashSet<RelTag>, PageReconstructError> {
// fetch directory listing
let key = rel_dir_to_key(spcnode, dbnode);
let buf = version.get(self, key, ctx).await?;
let buf = self.get(key, lsn, ctx).await?;
match RelDirectory::des(&buf).context("deserialization failure") {
Ok(dir) => {
@@ -319,7 +315,7 @@ impl Timeline {
}
/// Look up given SLRU page version.
pub(crate) async fn get_slru_page_at_lsn(
pub async fn get_slru_page_at_lsn(
&self,
kind: SlruKind,
segno: u32,
@@ -332,29 +328,29 @@ impl Timeline {
}
/// Get size of an SLRU segment
pub(crate) async fn get_slru_segment_size(
pub async fn get_slru_segment_size(
&self,
kind: SlruKind,
segno: u32,
version: Version<'_>,
lsn: Lsn,
ctx: &RequestContext,
) -> Result<BlockNumber, PageReconstructError> {
let key = slru_segment_size_to_key(kind, segno);
let mut buf = version.get(self, key, ctx).await?;
let mut buf = self.get(key, lsn, ctx).await?;
Ok(buf.get_u32_le())
}
/// Get size of an SLRU segment
pub(crate) async fn get_slru_segment_exists(
pub async fn get_slru_segment_exists(
&self,
kind: SlruKind,
segno: u32,
version: Version<'_>,
lsn: Lsn,
ctx: &RequestContext,
) -> Result<bool, PageReconstructError> {
// fetch directory listing
let key = slru_dir_to_key(kind);
let buf = version.get(self, key, ctx).await?;
let buf = self.get(key, lsn, ctx).await?;
match SlruSegmentDirectory::des(&buf).context("deserialization failure") {
Ok(dir) => {
@@ -372,7 +368,7 @@ impl Timeline {
/// so it's not well defined which LSN you get if there were multiple commits
/// "in flight" at that point in time.
///
pub(crate) async fn find_lsn_for_timestamp(
pub async fn find_lsn_for_timestamp(
&self,
search_timestamp: TimestampTz,
cancel: &CancellationToken,
@@ -452,7 +448,7 @@ impl Timeline {
/// Additionally, sets 'found_smaller'/'found_Larger, if encounters any commits
/// with a smaller/larger timestamp.
///
pub(crate) async fn is_latest_commit_timestamp_ge_than(
pub async fn is_latest_commit_timestamp_ge_than(
&self,
search_timestamp: TimestampTz,
probe_lsn: Lsn,
@@ -475,7 +471,7 @@ impl Timeline {
/// Obtain the possible timestamp range for the given lsn.
///
/// If the lsn has no timestamps, returns None. returns `(min, max, median)` if it has timestamps.
pub(crate) async fn get_timestamp_for_lsn(
pub async fn get_timestamp_for_lsn(
&self,
probe_lsn: Lsn,
ctx: &RequestContext,
@@ -505,11 +501,11 @@ impl Timeline {
mut f: impl FnMut(TimestampTz) -> ControlFlow<T>,
) -> Result<T, PageReconstructError> {
for segno in self
.list_slru_segments(SlruKind::Clog, Version::Lsn(probe_lsn), ctx)
.list_slru_segments(SlruKind::Clog, probe_lsn, ctx)
.await?
{
let nblocks = self
.get_slru_segment_size(SlruKind::Clog, segno, Version::Lsn(probe_lsn), ctx)
.get_slru_segment_size(SlruKind::Clog, segno, probe_lsn, ctx)
.await?;
for blknum in (0..nblocks).rev() {
let clog_page = self
@@ -532,36 +528,36 @@ impl Timeline {
}
/// Get a list of SLRU segments
pub(crate) async fn list_slru_segments(
pub async fn list_slru_segments(
&self,
kind: SlruKind,
version: Version<'_>,
lsn: Lsn,
ctx: &RequestContext,
) -> Result<HashSet<u32>, PageReconstructError> {
// fetch directory entry
let key = slru_dir_to_key(kind);
let buf = version.get(self, key, ctx).await?;
let buf = self.get(key, lsn, ctx).await?;
match SlruSegmentDirectory::des(&buf).context("deserialization failure") {
Ok(dir) => Ok(dir.segments),
Err(e) => Err(PageReconstructError::from(e)),
}
}
pub(crate) async fn get_relmap_file(
pub async fn get_relmap_file(
&self,
spcnode: Oid,
dbnode: Oid,
version: Version<'_>,
lsn: Lsn,
ctx: &RequestContext,
) -> Result<Bytes, PageReconstructError> {
let key = relmap_file_key(spcnode, dbnode);
let buf = version.get(self, key, ctx).await?;
let buf = self.get(key, lsn, ctx).await?;
Ok(buf)
}
pub(crate) async fn list_dbdirs(
pub async fn list_dbdirs(
&self,
lsn: Lsn,
ctx: &RequestContext,
@@ -575,7 +571,7 @@ impl Timeline {
}
}
pub(crate) async fn get_twophase_file(
pub async fn get_twophase_file(
&self,
xid: TransactionId,
lsn: Lsn,
@@ -586,7 +582,7 @@ impl Timeline {
Ok(buf)
}
pub(crate) async fn list_twophase_files(
pub async fn list_twophase_files(
&self,
lsn: Lsn,
ctx: &RequestContext,
@@ -600,7 +596,7 @@ impl Timeline {
}
}
pub(crate) async fn get_control_file(
pub async fn get_control_file(
&self,
lsn: Lsn,
ctx: &RequestContext,
@@ -608,7 +604,7 @@ impl Timeline {
self.get(CONTROLFILE_KEY, lsn, ctx).await
}
pub(crate) async fn get_checkpoint(
pub async fn get_checkpoint(
&self,
lsn: Lsn,
ctx: &RequestContext,
@@ -616,7 +612,7 @@ impl Timeline {
self.get(CHECKPOINT_KEY, lsn, ctx).await
}
pub(crate) async fn list_aux_files(
pub async fn list_aux_files(
&self,
lsn: Lsn,
ctx: &RequestContext,
@@ -656,10 +652,7 @@ impl Timeline {
let mut total_size: u64 = 0;
for (spcnode, dbnode) in dbdir.dbdirs.keys() {
for rel in self
.list_rels(*spcnode, *dbnode, Version::Lsn(lsn), ctx)
.await?
{
for rel in self.list_rels(*spcnode, *dbnode, lsn, ctx).await? {
if self.cancel.is_cancelled() {
return Err(CalculateLogicalSizeError::Cancelled);
}
@@ -699,7 +692,7 @@ impl Timeline {
result.add_key(rel_dir_to_key(spcnode, dbnode));
let mut rels: Vec<RelTag> = self
.list_rels(spcnode, dbnode, Version::Lsn(lsn), ctx)
.list_rels(spcnode, dbnode, lsn, ctx)
.await?
.into_iter()
.collect();
@@ -806,39 +799,18 @@ pub struct DatadirModification<'a> {
/// in the state in 'tline' yet.
pub tline: &'a Timeline,
/// Current LSN of the modification
lsn: Lsn,
/// Lsn assigned by begin_modification
pub lsn: Lsn,
// The modifications are not applied directly to the underlying key-value store.
// The put-functions add the modifications here, and they are flushed to the
// underlying key-value store by the 'finish' function.
pending_lsns: Vec<Lsn>,
pending_updates: HashMap<Key, Vec<(Lsn, Value)>>,
pending_deletions: Vec<(Range<Key>, Lsn)>,
pending_updates: HashMap<Key, Value>,
pending_deletions: Vec<Range<Key>>,
pending_nblocks: i64,
}
impl<'a> DatadirModification<'a> {
/// Get the current lsn
pub(crate) fn get_lsn(&self) -> Lsn {
self.lsn
}
/// Set the current lsn
pub(crate) fn set_lsn(&mut self, lsn: Lsn) -> anyhow::Result<()> {
ensure!(
lsn >= self.lsn,
"setting an older lsn {} than {} is not allowed",
lsn,
self.lsn
);
if lsn > self.lsn {
self.pending_lsns.push(self.lsn);
self.lsn = lsn;
}
Ok(())
}
/// Initialize a completely new repository.
///
/// This inserts the directory metadata entries that are assumed to
@@ -1012,9 +984,11 @@ impl<'a> DatadirModification<'a> {
dbnode: Oid,
ctx: &RequestContext,
) -> anyhow::Result<()> {
let req_lsn = self.tline.get_last_record_lsn();
let total_blocks = self
.tline
.get_db_size(spcnode, dbnode, Version::Modified(self), true, ctx)
.get_db_size(spcnode, dbnode, req_lsn, true, ctx)
.await?;
// Remove entry from dbdir
@@ -1103,11 +1077,8 @@ impl<'a> DatadirModification<'a> {
ctx: &RequestContext,
) -> anyhow::Result<()> {
anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
if self
.tline
.get_rel_exists(rel, Version::Modified(self), true, ctx)
.await?
{
let last_lsn = self.tline.get_last_record_lsn();
if self.tline.get_rel_exists(rel, last_lsn, true, ctx).await? {
let size_key = rel_size_to_key(rel);
// Fetch the old size first
let old_size = self.get(size_key, ctx).await?.get_u32_le();
@@ -1352,23 +1323,17 @@ impl<'a> DatadirModification<'a> {
let writer = self.tline.writer().await;
// Flush relation and SLRU data blocks, keep metadata.
let mut retained_pending_updates = HashMap::<_, Vec<_>>::new();
for (key, values) in self.pending_updates.drain() {
for (lsn, value) in values {
if is_rel_block_key(&key) || is_slru_block_key(key) {
// This bails out on first error without modifying pending_updates.
// That's Ok, cf this function's doc comment.
writer.put(key, lsn, &value, ctx).await?;
} else {
retained_pending_updates
.entry(key)
.or_default()
.push((lsn, value));
}
let mut retained_pending_updates = HashMap::new();
for (key, value) in self.pending_updates.drain() {
if is_rel_block_key(&key) || is_slru_block_key(key) {
// This bails out on first error without modifying pending_updates.
// That's Ok, cf this function's doc comment.
writer.put(key, self.lsn, &value, ctx).await?;
} else {
retained_pending_updates.insert(key, value);
}
}
self.pending_updates = retained_pending_updates;
self.pending_updates.extend(retained_pending_updates);
if pending_nblocks != 0 {
writer.update_current_logical_size(pending_nblocks * i64::from(BLCKSZ));
@@ -1385,28 +1350,18 @@ impl<'a> DatadirModification<'a> {
///
pub async fn commit(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
let writer = self.tline.writer().await;
let lsn = self.lsn;
let pending_nblocks = self.pending_nblocks;
self.pending_nblocks = 0;
if !self.pending_updates.is_empty() {
writer.put_batch(&self.pending_updates, ctx).await?;
self.pending_updates.clear();
for (key, value) in self.pending_updates.drain() {
writer.put(key, lsn, &value, ctx).await?;
}
for key_range in self.pending_deletions.drain(..) {
writer.delete(key_range, lsn).await?;
}
if !self.pending_deletions.is_empty() {
writer.delete_batch(&self.pending_deletions).await?;
self.pending_deletions.clear();
}
self.pending_lsns.push(self.lsn);
for pending_lsn in self.pending_lsns.drain(..) {
// Ideally, we should be able to call writer.finish_write() only once
// with the highest LSN. However, the last_record_lsn variable in the
// timeline keeps track of the latest LSN and the immediate previous LSN
// so we need to record every LSN to not leave a gap between them.
writer.finish_write(pending_lsn);
}
writer.finish_write(lsn);
if pending_nblocks != 0 {
writer.update_current_logical_size(pending_nblocks * i64::from(BLCKSZ));
@@ -1415,86 +1370,44 @@ impl<'a> DatadirModification<'a> {
Ok(())
}
pub(crate) fn len(&self) -> usize {
self.pending_updates.len() + self.pending_deletions.len()
pub(crate) fn is_empty(&self) -> bool {
self.pending_updates.is_empty() && self.pending_deletions.is_empty()
}
// Internal helper functions to batch the modifications
async fn get(&self, key: Key, ctx: &RequestContext) -> Result<Bytes, PageReconstructError> {
// Have we already updated the same key? Read the latest pending updated
// Have we already updated the same key? Read the pending updated
// version in that case.
//
// Note: we don't check pending_deletions. It is an error to request a
// value that has been removed, deletion only avoids leaking storage.
if let Some(values) = self.pending_updates.get(&key) {
if let Some((_, value)) = values.last() {
return if let Value::Image(img) = value {
Ok(img.clone())
} else {
// Currently, we never need to read back a WAL record that we
// inserted in the same "transaction". All the metadata updates
// work directly with Images, and we never need to read actual
// data pages. We could handle this if we had to, by calling
// the walredo manager, but let's keep it simple for now.
Err(PageReconstructError::from(anyhow::anyhow!(
"unexpected pending WAL record"
)))
};
if let Some(value) = self.pending_updates.get(&key) {
if let Value::Image(img) = value {
Ok(img.clone())
} else {
// Currently, we never need to read back a WAL record that we
// inserted in the same "transaction". All the metadata updates
// work directly with Images, and we never need to read actual
// data pages. We could handle this if we had to, by calling
// the walredo manager, but let's keep it simple for now.
Err(PageReconstructError::from(anyhow::anyhow!(
"unexpected pending WAL record"
)))
}
} else {
let lsn = Lsn::max(self.tline.get_last_record_lsn(), self.lsn);
self.tline.get(key, lsn, ctx).await
}
let lsn = Lsn::max(self.tline.get_last_record_lsn(), self.lsn);
self.tline.get(key, lsn, ctx).await
}
fn put(&mut self, key: Key, val: Value) {
let values = self.pending_updates.entry(key).or_default();
// Replace the previous value if it exists at the same lsn
if let Some((last_lsn, last_value)) = values.last_mut() {
if *last_lsn == self.lsn {
*last_value = val;
return;
}
}
values.push((self.lsn, val));
self.pending_updates.insert(key, val);
}
fn delete(&mut self, key_range: Range<Key>) {
trace!("DELETE {}-{}", key_range.start, key_range.end);
self.pending_deletions.push((key_range, self.lsn));
}
}
/// This struct facilitates accessing either a committed key from the timeline at a
/// specific LSN, or the latest uncommitted key from a pending modification.
/// During WAL ingestion, the records from multiple LSNs may be batched in the same
/// modification before being flushed to the timeline. Hence, the routines in WalIngest
/// need to look up the keys in the modification first before looking them up in the
/// timeline to not miss the latest updates.
#[derive(Clone, Copy)]
pub enum Version<'a> {
Lsn(Lsn),
Modified(&'a DatadirModification<'a>),
}
impl<'a> Version<'a> {
async fn get(
&self,
timeline: &Timeline,
key: Key,
ctx: &RequestContext,
) -> Result<Bytes, PageReconstructError> {
match self {
Version::Lsn(lsn) => timeline.get(key, *lsn, ctx).await,
Version::Modified(modification) => modification.get(key, ctx).await,
}
}
fn get_lsn(&self) -> Lsn {
match self {
Version::Lsn(lsn) => *lsn,
Version::Modified(modification) => modification.lsn,
}
self.pending_deletions.push(key_range);
}
}
@@ -1863,7 +1776,6 @@ pub fn is_inherited_key(key: Key) -> bool {
key != AUX_FILES_KEY
}
/// Guaranteed to return `Ok()` if [[is_rel_block_key]] returns `true` for `key`.
pub fn key_to_rel_block(key: Key) -> anyhow::Result<(RelTag, BlockNumber)> {
Ok(match key.field1 {
0x00 => (
@@ -1878,6 +1790,7 @@ pub fn key_to_rel_block(key: Key) -> anyhow::Result<(RelTag, BlockNumber)> {
_ => anyhow::bail!("unexpected value kind 0x{:02x}", key.field1),
})
}
pub fn is_rel_fsm_block_key(key: Key) -> bool {
key.field1 == 0x00 && key.field4 != 0 && key.field5 == FSM_FORKNUM && key.field6 != 0xffffffff
}

View File

@@ -2,11 +2,38 @@ use crate::walrecord::NeonWalRecord;
use anyhow::Result;
use bytes::Bytes;
use serde::{Deserialize, Serialize};
use std::ops::AddAssign;
use std::ops::{AddAssign, Range};
use std::time::Duration;
pub use pageserver_api::key::{Key, KEY_SIZE};
pub fn key_range_size(key_range: &Range<Key>) -> u32 {
let start = key_range.start;
let end = key_range.end;
if end.field1 != start.field1
|| end.field2 != start.field2
|| end.field3 != start.field3
|| end.field4 != start.field4
{
return u32::MAX;
}
let start = (start.field5 as u64) << 32 | start.field6 as u64;
let end = (end.field5 as u64) << 32 | end.field6 as u64;
let diff = end - start;
if diff > u32::MAX as u64 {
u32::MAX
} else {
diff as u32
}
}
pub fn singleton_range(key: Key) -> Range<Key> {
key..key.next()
}
/// A 'value' stored for a one Key.
#[derive(Debug, Clone, Serialize, Deserialize)]
#[cfg_attr(test, derive(PartialEq))]

View File

@@ -147,7 +147,7 @@ pub(crate) static BACKGROUND_RUNTIME_WORKER_THREADS: Lazy<usize> = Lazy::new(||
// else, but that has not been needed in a long time.
std::env::var("TOKIO_WORKER_THREADS")
.map(|s| s.parse::<usize>().unwrap())
.unwrap_or_else(|_e| usize::max(2, num_cpus::get()))
.unwrap_or_else(|_e| usize::max(1, num_cpus::get()))
});
#[derive(Debug, Clone, Copy)]
@@ -258,9 +258,6 @@ pub enum TaskKind {
/// See [`crate::disk_usage_eviction_task`].
DiskUsageEviction,
/// See [`crate::tenant::secondary`].
SecondaryDownloads,
/// See [`crate::tenant::secondary`].
SecondaryUploads,

View File

@@ -33,12 +33,9 @@ use tracing::*;
use utils::backoff;
use utils::completion;
use utils::crashsafe::path_with_suffix_extension;
use utils::failpoint_support;
use utils::fs_ext;
use utils::sync::gate::Gate;
use utils::sync::gate::GateGuard;
use utils::timeout::timeout_cancellable;
use utils::timeout::TimeoutCancellableError;
use self::config::AttachedLocationConfig;
use self::config::AttachmentMode;
@@ -56,14 +53,13 @@ use self::timeline::uninit::TimelineUninitMark;
use self::timeline::uninit::UninitializedTimeline;
use self::timeline::EvictionTaskTenantState;
use self::timeline::TimelineResources;
use self::timeline::WaitLsnError;
use crate::config::PageServerConf;
use crate::context::{DownloadBehavior, RequestContext};
use crate::deletion_queue::DeletionQueueClient;
use crate::deletion_queue::DeletionQueueError;
use crate::import_datadir;
use crate::is_uninit_mark;
use crate::metrics::TENANT;
use crate::metrics::TENANT_ACTIVATION;
use crate::metrics::{remove_tenant_metrics, TENANT_STATE_METRIC, TENANT_SYNTHETIC_SIZE_METRIC};
use crate::repository::GcResult;
use crate::task_mgr;
@@ -230,7 +226,7 @@ pub struct Tenant {
/// The value creation timestamp, used to measure activation delay, see:
/// <https://github.com/neondatabase/neon/issues/4025>
constructed_at: Instant,
loading_started_at: Instant,
state: watch::Sender<TenantState>,
@@ -280,11 +276,6 @@ pub struct Tenant {
eviction_task_tenant_state: tokio::sync::Mutex<EvictionTaskTenantState>,
/// If the tenant is in Activating state, notify this to encourage it
/// to proceed to Active as soon as possible, rather than waiting for lazy
/// background warmup.
pub(crate) activate_now_sem: tokio::sync::Semaphore,
pub(crate) delete_progress: Arc<tokio::sync::Mutex<DeleteTenantFlow>>,
// Cancellation token fires when we have entered shutdown(). This is a parent of
@@ -596,9 +587,10 @@ impl Tenant {
mode: SpawnMode,
ctx: &RequestContext,
) -> anyhow::Result<Arc<Tenant>> {
// TODO(sharding): make WalRedoManager shard-aware
let wal_redo_manager = Arc::new(WalRedoManager::from(PostgresRedoManager::new(
conf,
tenant_shard_id,
tenant_shard_id.tenant_id,
)));
let TenantSharedResources {
@@ -630,14 +622,6 @@ impl Tenant {
"attach tenant",
false,
async move {
// Is this tenant being spawned as part of process startup?
let starting_up = init_order.is_some();
scopeguard::defer! {
if starting_up {
TENANT.startup_complete.inc();
}
}
// Ideally we should use Tenant::set_broken_no_wait, but it is not supposed to be used when tenant is in loading state.
let make_broken =
|t: &Tenant, err: anyhow::Error| {
@@ -664,62 +648,8 @@ impl Tenant {
.as_mut()
.and_then(|x| x.initial_tenant_load_remote.take());
enum AttachType<'a> {
// During pageserver startup, we are attaching this tenant lazily in the background
Warmup(tokio::sync::SemaphorePermit<'a>),
// During pageserver startup, we are attaching this tenant as soon as we can,
// because a client tried to access it.
OnDemand,
// During normal operations after startup, we are attaching a tenant.
Normal,
}
// Before doing any I/O, wait for either or:
// - A client to attempt to access to this tenant (on-demand loading)
// - A permit to become available in the warmup semaphore (background warmup)
//
// Some-ness of init_order is how we know if we're attaching during startup or later
// in process lifetime.
let attach_type = if init_order.is_some() {
tokio::select!(
_ = tenant_clone.activate_now_sem.acquire() => {
tracing::info!("Activating tenant (on-demand)");
AttachType::OnDemand
},
permit_result = conf.concurrent_tenant_warmup.inner().acquire() => {
match permit_result {
Ok(p) => {
tracing::info!("Activating tenant (warmup)");
AttachType::Warmup(p)
}
Err(_) => {
// This is unexpected: the warmup semaphore should stay alive
// for the lifetime of init_order. Log a warning and proceed.
tracing::warn!("warmup_limit semaphore unexpectedly closed");
AttachType::Normal
}
}
}
_ = tenant_clone.cancel.cancelled() => {
// This is safe, but should be pretty rare: it is interesting if a tenant
// stayed in Activating for such a long time that shutdown found it in
// that state.
tracing::info!(state=%tenant_clone.current_state(), "Tenant shut down before activation");
return Ok(());
},
)
} else {
AttachType::Normal
};
let preload_timer = TENANT.preload.start_timer();
let preload = match mode {
SpawnMode::Create => {
// Don't count the skipped preload into the histogram of preload durations
preload_timer.stop_and_discard();
None
},
SpawnMode::Create => {None},
SpawnMode::Normal => {
match &remote_storage {
Some(remote_storage) => Some(
@@ -729,11 +659,7 @@ impl Tenant {
tracing::info_span!(parent: None, "attach_preload", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug()),
)
.await {
Ok(p) => {
preload_timer.observe_duration();
p
}
,
Ok(p) => p,
Err(e) => {
make_broken(&tenant_clone, anyhow::anyhow!(e));
return Ok(());
@@ -795,43 +721,15 @@ impl Tenant {
}
}
// We will time the duration of the attach phase unless this is a creation (attach will do no work)
let attach_timer = match mode {
SpawnMode::Create => None,
SpawnMode::Normal => {Some(TENANT.attach.start_timer())}
};
match tenant_clone.attach(preload, &ctx).await {
Ok(()) => {
info!("attach finished, activating");
if let Some(t)= attach_timer {t.observe_duration();}
tenant_clone.activate(broker_client, None, &ctx);
}
Err(e) => {
if let Some(t)= attach_timer {t.observe_duration();}
make_broken(&tenant_clone, anyhow::anyhow!(e));
}
}
// If we are doing an opportunistic warmup attachment at startup, initialize
// logical size at the same time. This is better than starting a bunch of idle tenants
// with cold caches and then coming back later to initialize their logical sizes.
//
// It also prevents the warmup proccess competing with the concurrency limit on
// logical size calculations: if logical size calculation semaphore is saturated,
// then warmup will wait for that before proceeding to the next tenant.
if let AttachType::Warmup(_permit) = attach_type {
let mut futs = FuturesUnordered::new();
let timelines: Vec<_> = tenant_clone.timelines.lock().unwrap().values().cloned().collect();
for t in timelines {
futs.push(t.await_initial_logical_size())
}
tracing::info!("Waiting for initial logical sizes while warming up...");
while futs.next().await.is_some() {
}
tracing::info!("Warm-up complete");
}
Ok(())
}
.instrument({
@@ -891,7 +789,7 @@ impl Tenant {
) -> anyhow::Result<()> {
span::debug_assert_current_span_has_tenant_id();
failpoint_support::sleep_millis_async!("before-attaching-tenant");
crate::failpoint_support::sleep_millis_async!("before-attaching-tenant");
let preload = match preload {
Some(p) => p,
@@ -1003,7 +901,7 @@ impl Tenant {
// IndexPart is the source of truth.
self.clean_up_timelines(&existent_timelines)?;
failpoint_support::sleep_millis_async!("attach-before-activate");
crate::failpoint_support::sleep_millis_async!("attach-before-activate");
info!("Done");
@@ -1145,9 +1043,10 @@ impl Tenant {
tenant_shard_id: TenantShardId,
reason: String,
) -> Arc<Tenant> {
// TODO(sharding): make WalRedoManager shard-aware
let wal_redo_manager = Arc::new(WalRedoManager::from(PostgresRedoManager::new(
conf,
tenant_shard_id,
tenant_shard_id.tenant_id,
)));
Arc::new(Tenant::new(
TenantState::Broken {
@@ -1552,10 +1451,6 @@ impl Tenant {
.collect()
}
pub fn list_timeline_ids(&self) -> Vec<TimelineId> {
self.timelines.lock().unwrap().keys().cloned().collect()
}
/// This is used to create the initial 'main' timeline during bootstrapping,
/// or when importing a new base backup. The caller is expected to load an
/// initial image of the datadir to the new timeline after this.
@@ -1759,15 +1654,7 @@ impl Tenant {
// decoding the new WAL might need to look up previous pages, relation
// sizes etc. and that would get confused if the previous page versions
// are not in the repository yet.
ancestor_timeline
.wait_lsn(*lsn, ctx)
.await
.map_err(|e| match e {
e @ (WaitLsnError::Timeout(_) | WaitLsnError::BadState) => {
CreateTimelineError::AncestorLsn(anyhow::anyhow!(e))
}
WaitLsnError::Shutdown => CreateTimelineError::ShuttingDown,
})?;
ancestor_timeline.wait_lsn(*lsn, ctx).await?;
}
self.branch_timeline(
@@ -1809,15 +1696,6 @@ impl Tenant {
Ok(loaded_timeline)
}
pub(crate) async fn delete_timeline(
self: Arc<Self>,
timeline_id: TimelineId,
) -> Result<(), DeleteTimelineError> {
DeleteTimelineFlow::run(&self, timeline_id, false).await?;
Ok(())
}
/// perform one garbage collection iteration, removing old data files from disk.
/// this function is periodically called by gc task.
/// also it can be explicitly requested through page server api 'do_gc' command.
@@ -1979,7 +1857,7 @@ impl Tenant {
);
*current_state = TenantState::Active;
let elapsed = self.constructed_at.elapsed();
let elapsed = self.loading_started_at.elapsed();
let total_timelines = timelines_accessor.len();
// log a lot of stuff, because some tenants sometimes suffer from user-visible
@@ -1994,7 +1872,7 @@ impl Tenant {
"activation attempt finished"
);
TENANT.activation.observe(elapsed.as_secs_f64());
TENANT_ACTIVATION.observe(elapsed.as_secs_f64());
});
}
}
@@ -2249,41 +2127,18 @@ impl Tenant {
self.state.subscribe()
}
/// The activate_now semaphore is initialized with zero units. As soon as
/// we add a unit, waiters will be able to acquire a unit and proceed.
pub(crate) fn activate_now(&self) {
self.activate_now_sem.add_permits(1);
}
pub(crate) async fn wait_to_become_active(
&self,
timeout: Duration,
) -> Result<(), GetActiveTenantError> {
pub(crate) async fn wait_to_become_active(&self) -> Result<(), GetActiveTenantError> {
let mut receiver = self.state.subscribe();
loop {
let current_state = receiver.borrow_and_update().clone();
match current_state {
TenantState::Loading | TenantState::Attaching | TenantState::Activating(_) => {
// in these states, there's a chance that we can reach ::Active
self.activate_now();
match timeout_cancellable(timeout, &self.cancel, receiver.changed()).await {
Ok(r) => {
r.map_err(
|_e: tokio::sync::watch::error::RecvError|
// Tenant existed but was dropped: report it as non-existent
GetActiveTenantError::NotFound(GetTenantError::NotFound(self.tenant_shard_id.tenant_id))
)?
}
Err(TimeoutCancellableError::Cancelled) => {
return Err(GetActiveTenantError::Cancelled);
}
Err(TimeoutCancellableError::Timeout) => {
return Err(GetActiveTenantError::WaitForActiveTimeout {
latest_state: Some(self.current_state()),
wait_time: timeout,
});
}
}
receiver.changed().await.map_err(
|_e: tokio::sync::watch::error::RecvError|
// Tenant existed but was dropped: report it as non-existent
GetActiveTenantError::NotFound(GetTenantError::NotFound(self.tenant_shard_id.tenant_id))
)?;
}
TenantState::Active { .. } => {
return Ok(());
@@ -2608,7 +2463,7 @@ impl Tenant {
conf,
// using now here is good enough approximation to catch tenants with really long
// activation times.
constructed_at: Instant::now(),
loading_started_at: Instant::now(),
tenant_conf: Arc::new(RwLock::new(attached_conf)),
timelines: Mutex::new(HashMap::new()),
timelines_creating: Mutex::new(HashSet::new()),
@@ -2620,7 +2475,6 @@ impl Tenant {
cached_logical_sizes: tokio::sync::Mutex::new(HashMap::new()),
cached_synthetic_tenant_size: Arc::new(AtomicU64::new(0)),
eviction_task_tenant_state: tokio::sync::Mutex::new(EvictionTaskTenantState::default()),
activate_now_sem: tokio::sync::Semaphore::new(0),
delete_progress: Arc::new(tokio::sync::Mutex::new(DeleteTenantFlow::default())),
cancel: CancellationToken::default(),
gate: Gate::new(format!("Tenant<{tenant_shard_id}>")),
@@ -2847,7 +2701,9 @@ impl Tenant {
}
};
failpoint_support::sleep_millis_async!("gc_iteration_internal_after_getting_gc_timelines");
crate::failpoint_support::sleep_millis_async!(
"gc_iteration_internal_after_getting_gc_timelines"
);
// If there is nothing to GC, we don't want any messages in the INFO log.
if !gc_timelines.is_empty() {
@@ -3140,7 +2996,6 @@ impl Tenant {
/// For unit tests, make this visible so that other modules can directly create timelines
#[cfg(test)]
#[tracing::instrument(fields(tenant_id=%self.tenant_shard_id.tenant_id, shard_id=%self.tenant_shard_id.shard_slug(), %timeline_id))]
pub(crate) async fn bootstrap_timeline_test(
&self,
timeline_id: TimelineId,
@@ -3204,7 +3059,6 @@ impl Tenant {
storage,
&self.tenant_shard_id,
&existing_initdb_timeline_id,
&self.cancel,
)
.await
.context("download initdb tar")?;
@@ -3245,7 +3099,6 @@ impl Tenant {
&timeline_id,
pgdata_zstd.try_clone().await?,
tar_zst_size,
&self.cancel,
)
.await
},
@@ -3253,7 +3106,9 @@ impl Tenant {
3,
u32::MAX,
"persist_initdb_tar_zst",
backoff::Cancel::new(self.cancel.clone(), || anyhow::anyhow!("Cancelled")),
backoff::Cancel::new(self.cancel.clone(), || {
anyhow::anyhow!("initdb upload cancelled")
}),
)
.await?;

View File

@@ -46,8 +46,6 @@ pub mod defaults {
pub const DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT: &str = "10 seconds";
pub const DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG: u64 = 10 * 1024 * 1024;
pub const DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD: &str = "24 hour";
pub const DEFAULT_INGEST_BATCH_SIZE: u64 = 100;
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]

View File

@@ -48,9 +48,6 @@ pub(crate) enum DeleteTenantError {
#[error("Timeline {0}")]
Timeline(#[from] DeleteTimelineError),
#[error("Cancelled")]
Cancelled,
#[error(transparent)]
Other(#[from] anyhow::Error),
}
@@ -74,7 +71,6 @@ async fn create_remote_delete_mark(
conf: &PageServerConf,
remote_storage: &GenericRemoteStorage,
tenant_shard_id: &TenantShardId,
cancel: &CancellationToken,
) -> Result<(), DeleteTenantError> {
let remote_mark_path = remote_tenant_delete_mark_path(conf, tenant_shard_id)?;
@@ -91,7 +87,8 @@ async fn create_remote_delete_mark(
FAILED_UPLOAD_WARN_THRESHOLD,
FAILED_REMOTE_OP_RETRIES,
"mark_upload",
backoff::Cancel::new(cancel.clone(), || anyhow::anyhow!("Cancelled")),
// TODO: use a cancellation token (https://github.com/neondatabase/neon/issues/5066)
backoff::Cancel::new(CancellationToken::new(), || unreachable!()),
)
.await
.context("mark_upload")?;
@@ -173,7 +170,6 @@ async fn remove_tenant_remote_delete_mark(
conf: &PageServerConf,
remote_storage: Option<&GenericRemoteStorage>,
tenant_shard_id: &TenantShardId,
cancel: &CancellationToken,
) -> Result<(), DeleteTenantError> {
if let Some(remote_storage) = remote_storage {
let path = remote_tenant_delete_mark_path(conf, tenant_shard_id)?;
@@ -183,7 +179,8 @@ async fn remove_tenant_remote_delete_mark(
FAILED_UPLOAD_WARN_THRESHOLD,
FAILED_REMOTE_OP_RETRIES,
"remove_tenant_remote_delete_mark",
backoff::Cancel::new(cancel.clone(), || anyhow::anyhow!("Cancelled")),
// TODO: use a cancellation token (https://github.com/neondatabase/neon/issues/5066)
backoff::Cancel::new(CancellationToken::new(), || unreachable!()),
)
.await
.context("remove_tenant_remote_delete_mark")?;
@@ -325,15 +322,9 @@ impl DeleteTenantFlow {
// Though sounds scary, different mark name?
// Detach currently uses remove_dir_all so in case of a crash we can end up in a weird state.
if let Some(remote_storage) = &remote_storage {
create_remote_delete_mark(
conf,
remote_storage,
&tenant.tenant_shard_id,
// Can't use tenant.cancel, it's already shut down. TODO: wire in an appropriate token
&CancellationToken::new(),
)
.await
.context("remote_mark")?
create_remote_delete_mark(conf, remote_storage, &tenant.tenant_shard_id)
.await
.context("remote_mark")?
}
fail::fail_point!("tenant-delete-before-create-local-mark", |_| {
@@ -533,14 +524,8 @@ impl DeleteTenantFlow {
.context("timelines dir not empty")?;
}
remove_tenant_remote_delete_mark(
conf,
remote_storage.as_ref(),
&tenant.tenant_shard_id,
// Can't use tenant.cancel, it's already shut down. TODO: wire in an appropriate token
&CancellationToken::new(),
)
.await?;
remove_tenant_remote_delete_mark(conf, remote_storage.as_ref(), &tenant.tenant_shard_id)
.await?;
fail::fail_point!("tenant-delete-before-cleanup-remaining-fs-traces", |_| {
Err(anyhow::anyhow!(
@@ -588,7 +573,7 @@ impl DeleteTenantFlow {
}
break;
}
TenantsMapRemoveResult::Occupied(TenantSlot::Secondary(_)) => {
TenantsMapRemoveResult::Occupied(TenantSlot::Secondary) => {
// This is unexpected: this secondary tenants should not have been created, and we
// are not in a position to shut it down from here.
tracing::warn!("Tenant transitioned to secondary mode while deleting!");

View File

@@ -28,7 +28,7 @@ use crate::control_plane_client::{
ControlPlaneClient, ControlPlaneGenerationsApi, RetryForeverError,
};
use crate::deletion_queue::DeletionQueueClient;
use crate::metrics::{TENANT, TENANT_MANAGER as METRICS};
use crate::metrics::TENANT_MANAGER as METRICS;
use crate::task_mgr::{self, TaskKind};
use crate::tenant::config::{
AttachedLocationConfig, AttachmentMode, LocationConf, LocationMode, TenantConfOpt,
@@ -44,7 +44,7 @@ use utils::generation::Generation;
use utils::id::{TenantId, TimelineId};
use super::delete::DeleteTenantError;
use super::secondary::SecondaryTenant;
use super::timeline::delete::DeleteTimelineFlow;
use super::TenantSharedResources;
/// For a tenant that appears in TenantsMap, it may either be
@@ -58,7 +58,7 @@ use super::TenantSharedResources;
/// having a properly acquired generation (Secondary doesn't need a generation)
pub(crate) enum TenantSlot {
Attached(Arc<Tenant>),
Secondary(Arc<SecondaryTenant>),
Secondary,
/// In this state, other administrative operations acting on the TenantId should
/// block, or return a retry indicator equivalent to HTTP 503.
InProgress(utils::completion::Barrier),
@@ -68,7 +68,7 @@ impl std::fmt::Debug for TenantSlot {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
match self {
Self::Attached(tenant) => write!(f, "Attached({})", tenant.current_state()),
Self::Secondary(_) => write!(f, "Secondary"),
Self::Secondary => write!(f, "Secondary"),
Self::InProgress(_) => write!(f, "InProgress"),
}
}
@@ -79,7 +79,7 @@ impl TenantSlot {
fn get_attached(&self) -> Option<&Arc<Tenant>> {
match self {
Self::Attached(t) => Some(t),
Self::Secondary(_) => None,
Self::Secondary => None,
Self::InProgress(_) => None,
}
}
@@ -131,7 +131,7 @@ impl TenantsMap {
/// A page service client sends a TenantId, and to look up the correct Tenant we must
/// resolve this to a fully qualified TenantShardId.
fn resolve_attached_shard(
fn resolve_shard(
&self,
tenant_id: &TenantId,
selector: ShardSelector,
@@ -141,27 +141,25 @@ impl TenantsMap {
TenantsMap::Initializing => None,
TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => {
for slot in m.range(TenantShardId::tenant_range(*tenant_id)) {
// Ignore all slots that don't contain an attached tenant
let tenant = match &slot.1 {
TenantSlot::Attached(t) => t,
_ => continue,
};
match selector {
ShardSelector::First => return Some(*slot.0),
ShardSelector::Zero if slot.0.shard_number == ShardNumber(0) => {
return Some(*slot.0)
}
ShardSelector::Page(key) => {
// First slot we see for this tenant, calculate the expected shard number
// for the key: we will use this for checking if this and subsequent
// slots contain the key, rather than recalculating the hash each time.
if want_shard.is_none() {
want_shard = Some(tenant.shard_identity.get_shard_number(&key));
}
if let Some(tenant) = slot.1.get_attached() {
// First slot we see for this tenant, calculate the expected shard number
// for the key: we will use this for checking if this and subsequent
// slots contain the key, rather than recalculating the hash each time.
if want_shard.is_none() {
want_shard = Some(tenant.shard_identity.get_shard_number(&key));
}
if Some(tenant.shard_identity.number) == want_shard {
return Some(*slot.0);
if Some(tenant.shard_identity.number) == want_shard {
return Some(*slot.0);
}
} else {
continue;
}
}
_ => continue,
@@ -432,13 +430,6 @@ pub async fn init_tenant_mgr(
let tenant_generations =
init_load_generations(conf, &tenant_configs, &resources, &cancel).await?;
tracing::info!(
"Attaching {} tenants at startup, warming up {} at a time",
tenant_configs.len(),
conf.concurrent_tenant_warmup.initial_permits()
);
TENANT.startup_scheduled.inc_by(tenant_configs.len() as u64);
// Construct `Tenant` objects and start them running
for (tenant_shard_id, location_conf) in tenant_configs {
let tenant_dir_path = conf.tenant_path(&tenant_shard_id);
@@ -467,18 +458,12 @@ pub async fn init_tenant_mgr(
*gen
} else {
match &location_conf.mode {
LocationMode::Secondary(secondary_config) => {
LocationMode::Secondary(_) => {
// We do not require the control plane's permission for secondary mode
// tenants, because they do no remote writes and hence require no
// generation number
info!(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), "Loaded tenant in secondary mode");
tenants.insert(
tenant_shard_id,
TenantSlot::Secondary(SecondaryTenant::new(
tenant_shard_id,
secondary_config,
)),
);
tenants.insert(tenant_shard_id, TenantSlot::Secondary);
}
LocationMode::Attached(_) => {
// TODO: augment re-attach API to enable the control plane to
@@ -523,7 +508,10 @@ pub async fn init_tenant_mgr(
&ctx,
) {
Ok(tenant) => {
tenants.insert(tenant_shard_id, TenantSlot::Attached(tenant));
tenants.insert(
TenantShardId::unsharded(tenant.tenant_id()),
TenantSlot::Attached(tenant),
);
}
Err(e) => {
error!(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), "Failed to start tenant: {e:#}");
@@ -670,14 +658,8 @@ async fn shutdown_all_tenants0(tenants: &std::sync::RwLock<TenantsMap>) {
total_attached += 1;
}
TenantSlot::Secondary(state) => {
// We don't need to wait for this individually per-tenant: the
// downloader task will be waited on eventually, this cancel
// is just to encourage it to drop out if it is doing work
// for this tenant right now.
state.cancel.cancel();
shutdown_state.insert(tenant_shard_id, TenantSlot::Secondary(state));
TenantSlot::Secondary => {
shutdown_state.insert(tenant_shard_id, TenantSlot::Secondary);
}
TenantSlot::InProgress(notify) => {
// InProgress tenants are not visible in TenantsMap::ShuttingDown: we will
@@ -860,26 +842,21 @@ impl TenantManager {
Some(TenantSlot::InProgress(_)) => {
Err(GetTenantError::NotActive(tenant_shard_id.tenant_id))
}
None | Some(TenantSlot::Secondary(_)) => {
None | Some(TenantSlot::Secondary) => {
Err(GetTenantError::NotFound(tenant_shard_id.tenant_id))
}
}
}
pub(crate) fn get_secondary_tenant_shard(
pub(crate) async fn delete_timeline(
&self,
tenant_shard_id: TenantShardId,
) -> Option<Arc<SecondaryTenant>> {
let locked = self.tenants.read().unwrap();
let peek_slot = tenant_map_peek_slot(&locked, &tenant_shard_id, TenantSlotPeekMode::Read)
.ok()
.flatten();
match peek_slot {
Some(TenantSlot::Secondary(s)) => Some(s.clone()),
_ => None,
}
timeline_id: TimelineId,
_ctx: &RequestContext,
) -> Result<(), DeleteTimelineError> {
let tenant = self.get_attached_tenant_shard(tenant_shard_id, true)?;
DeleteTimelineFlow::run(&tenant, timeline_id, false).await?;
Ok(())
}
#[instrument(skip_all, fields(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug()))]
@@ -893,15 +870,10 @@ impl TenantManager {
debug_assert_current_span_has_tenant_id();
info!("configuring tenant location to state {new_location_config:?}");
enum FastPathModified {
Attached(Arc<Tenant>),
Secondary(Arc<SecondaryTenant>),
}
// Special case fast-path for updates to existing slots: if our upsert is only updating configuration,
// Special case fast-path for updates to Tenant: if our upsert is only updating configuration,
// then we do not need to set the slot to InProgress, we can just call into the
// existng tenant.
let fast_path_taken = {
let modify_tenant = {
let locked = self.tenants.read().unwrap();
let peek_slot =
tenant_map_peek_slot(&locked, &tenant_shard_id, TenantSlotPeekMode::Write)?;
@@ -915,19 +887,12 @@ impl TenantManager {
new_location_config.clone(),
)?);
Some(FastPathModified::Attached(tenant.clone()))
Some(tenant.clone())
} else {
// Different generations, fall through to general case
None
}
}
(
LocationMode::Secondary(secondary_conf),
Some(TenantSlot::Secondary(secondary_tenant)),
) => {
secondary_tenant.set_config(secondary_conf);
Some(FastPathModified::Secondary(secondary_tenant.clone()))
}
_ => {
// Not an Attached->Attached transition, fall through to general case
None
@@ -936,51 +901,34 @@ impl TenantManager {
};
// Fast-path continued: having dropped out of the self.tenants lock, do the async
// phase of writing config and/or waiting for flush, before returning.
match fast_path_taken {
Some(FastPathModified::Attached(tenant)) => {
Tenant::persist_tenant_config(self.conf, &tenant_shard_id, &new_location_config)
.await
.map_err(SetNewTenantConfigError::Persist)?;
// Transition to AttachedStale means we may well hold a valid generation
// still, and have been requested to go stale as part of a migration. If
// the caller set `flush`, then flush to remote storage.
if let LocationMode::Attached(AttachedLocationConfig {
generation: _,
attach_mode: AttachmentMode::Stale,
}) = &new_location_config.mode
{
if let Some(flush_timeout) = flush {
match tokio::time::timeout(flush_timeout, tenant.flush_remote()).await {
Ok(Err(e)) => {
return Err(e);
}
Ok(Ok(_)) => return Ok(()),
Err(_) => {
tracing::warn!(
// phase of waiting for flush, before returning.
if let Some(tenant) = modify_tenant {
// Transition to AttachedStale means we may well hold a valid generation
// still, and have been requested to go stale as part of a migration. If
// the caller set `flush`, then flush to remote storage.
if let LocationMode::Attached(AttachedLocationConfig {
generation: _,
attach_mode: AttachmentMode::Stale,
}) = &new_location_config.mode
{
if let Some(flush_timeout) = flush {
match tokio::time::timeout(flush_timeout, tenant.flush_remote()).await {
Ok(Err(e)) => {
return Err(e);
}
Ok(Ok(_)) => return Ok(()),
Err(_) => {
tracing::warn!(
timeout_ms = flush_timeout.as_millis(),
"Timed out waiting for flush to remote storage, proceeding anyway."
)
}
}
}
}
}
return Ok(());
}
Some(FastPathModified::Secondary(_secondary_tenant)) => {
Tenant::persist_tenant_config(self.conf, &tenant_shard_id, &new_location_config)
.await
.map_err(SetNewTenantConfigError::Persist)?;
return Ok(());
}
None => {
// Proceed with the general case procedure, where we will shutdown & remove any existing
// slot contents and replace with a fresh one
}
};
return Ok(());
}
// General case for upserts to TenantsMap, excluding the case above: we will substitute an
// InProgress value to the slot while we make whatever changes are required. The state for
@@ -989,73 +937,65 @@ impl TenantManager {
// not do significant I/O, and shutdowns should be prompt via cancellation tokens.
let mut slot_guard = tenant_map_acquire_slot(&tenant_shard_id, TenantSlotAcquireMode::Any)?;
match slot_guard.get_old_value() {
Some(TenantSlot::Attached(tenant)) => {
// The case where we keep a Tenant alive was covered above in the special case
// for Attached->Attached transitions in the same generation. By this point,
// if we see an attached tenant we know it will be discarded and should be
// shut down.
let (_guard, progress) = utils::completion::channel();
if let Some(TenantSlot::Attached(tenant)) = slot_guard.get_old_value() {
// The case where we keep a Tenant alive was covered above in the special case
// for Attached->Attached transitions in the same generation. By this point,
// if we see an attached tenant we know it will be discarded and should be
// shut down.
let (_guard, progress) = utils::completion::channel();
match tenant.get_attach_mode() {
AttachmentMode::Single | AttachmentMode::Multi => {
// Before we leave our state as the presumed holder of the latest generation,
// flush any outstanding deletions to reduce the risk of leaking objects.
self.resources.deletion_queue_client.flush_advisory()
}
AttachmentMode::Stale => {
// If we're stale there's not point trying to flush deletions
}
};
info!("Shutting down attached tenant");
match tenant.shutdown(progress, false).await {
Ok(()) => {}
Err(barrier) => {
info!("Shutdown already in progress, waiting for it to complete");
barrier.wait().await;
}
match tenant.get_attach_mode() {
AttachmentMode::Single | AttachmentMode::Multi => {
// Before we leave our state as the presumed holder of the latest generation,
// flush any outstanding deletions to reduce the risk of leaking objects.
self.resources.deletion_queue_client.flush_advisory()
}
AttachmentMode::Stale => {
// If we're stale there's not point trying to flush deletions
}
};
info!("Shutting down attached tenant");
match tenant.shutdown(progress, false).await {
Ok(()) => {}
Err(barrier) => {
info!("Shutdown already in progress, waiting for it to complete");
barrier.wait().await;
}
slot_guard.drop_old_value().expect("We just shut it down");
}
Some(TenantSlot::Secondary(state)) => {
info!("Shutting down secondary tenant");
state.shutdown().await;
}
Some(TenantSlot::InProgress(_)) => {
// This should never happen: acquire_slot should error out
// if the contents of a slot were InProgress.
anyhow::bail!("Acquired an InProgress slot, this is a bug.")
}
None => {
// Slot was vacant, nothing needs shutting down.
}
slot_guard.drop_old_value().expect("We just shut it down");
}
let tenant_path = self.conf.tenant_path(&tenant_shard_id);
let timelines_path = self.conf.timelines_path(&tenant_shard_id);
// Directory structure is the same for attached and secondary modes:
// create it if it doesn't exist. Timeline load/creation expects the
// timelines/ subdir to already exist.
//
// Does not need to be fsync'd because local storage is just a cache.
tokio::fs::create_dir_all(&timelines_path)
.await
.with_context(|| format!("Creating {timelines_path}"))?;
// Before activating either secondary or attached mode, persist the
// configuration, so that on restart we will re-attach (or re-start
// secondary) on the tenant.
Tenant::persist_tenant_config(self.conf, &tenant_shard_id, &new_location_config)
.await
.map_err(SetNewTenantConfigError::Persist)?;
let new_slot = match &new_location_config.mode {
LocationMode::Secondary(secondary_config) => {
TenantSlot::Secondary(SecondaryTenant::new(tenant_shard_id, secondary_config))
LocationMode::Secondary(_) => {
// Directory doesn't need to be fsync'd because if we crash it can
// safely be recreated next time this tenant location is configured.
tokio::fs::create_dir_all(&tenant_path)
.await
.with_context(|| format!("Creating {tenant_path}"))?;
Tenant::persist_tenant_config(self.conf, &tenant_shard_id, &new_location_config)
.await
.map_err(SetNewTenantConfigError::Persist)?;
TenantSlot::Secondary
}
LocationMode::Attached(_attach_config) => {
let timelines_path = self.conf.timelines_path(&tenant_shard_id);
// Directory doesn't need to be fsync'd because we do not depend on
// it to exist after crashes: it may be recreated when tenant is
// re-attached, see https://github.com/neondatabase/neon/issues/5550
tokio::fs::create_dir_all(&tenant_path)
.await
.with_context(|| format!("Creating {timelines_path}"))?;
Tenant::persist_tenant_config(self.conf, &tenant_shard_id, &new_location_config)
.await
.map_err(SetNewTenantConfigError::Persist)?;
let shard_identity = new_location_config.shard;
let tenant = tenant_spawn(
self.conf,
@@ -1167,95 +1107,6 @@ impl TenantManager {
.collect(),
}
}
// Do some synchronous work for all tenant slots in Secondary state. The provided
// callback should be small and fast, as it will be called inside the global
// TenantsMap lock.
pub(crate) fn foreach_secondary_tenants<F>(&self, mut func: F)
where
// TODO: let the callback return a hint to drop out of the loop early
F: FnMut(&TenantShardId, &Arc<SecondaryTenant>),
{
let locked = self.tenants.read().unwrap();
let map = match &*locked {
TenantsMap::Initializing | TenantsMap::ShuttingDown(_) => return,
TenantsMap::Open(m) => m,
};
for (tenant_id, slot) in map {
if let TenantSlot::Secondary(state) = slot {
// Only expose secondary tenants that are not currently shutting down
if !state.cancel.is_cancelled() {
func(tenant_id, state)
}
}
}
}
pub(crate) async fn delete_tenant(
&self,
tenant_shard_id: TenantShardId,
activation_timeout: Duration,
) -> Result<(), DeleteTenantError> {
// We acquire a SlotGuard during this function to protect against concurrent
// changes while the ::prepare phase of DeleteTenantFlow executes, but then
// have to return the Tenant to the map while the background deletion runs.
//
// TODO: refactor deletion to happen outside the lifetime of a Tenant.
// Currently, deletion requires a reference to the tenants map in order to
// keep the Tenant in the map until deletion is complete, and then remove
// it at the end.
//
// See https://github.com/neondatabase/neon/issues/5080
let slot_guard =
tenant_map_acquire_slot(&tenant_shard_id, TenantSlotAcquireMode::MustExist)?;
// unwrap is safe because we used MustExist mode when acquiring
let tenant = match slot_guard.get_old_value().as_ref().unwrap() {
TenantSlot::Attached(tenant) => tenant.clone(),
_ => {
// Express "not attached" as equivalent to "not found"
return Err(DeleteTenantError::NotAttached);
}
};
match tenant.current_state() {
TenantState::Broken { .. } | TenantState::Stopping { .. } => {
// If a tenant is broken or stopping, DeleteTenantFlow can
// handle it: broken tenants proceed to delete, stopping tenants
// are checked for deletion already in progress.
}
_ => {
tenant
.wait_to_become_active(activation_timeout)
.await
.map_err(|e| match e {
GetActiveTenantError::WillNotBecomeActive(_) => {
DeleteTenantError::InvalidState(tenant.current_state())
}
GetActiveTenantError::Cancelled => DeleteTenantError::Cancelled,
GetActiveTenantError::NotFound(_) => DeleteTenantError::NotAttached,
GetActiveTenantError::WaitForActiveTimeout {
latest_state: _latest_state,
wait_time: _wait_time,
} => DeleteTenantError::InvalidState(tenant.current_state()),
})?;
}
}
let result = DeleteTenantFlow::run(
self.conf,
self.resources.remote_storage.clone(),
&TENANTS,
tenant,
)
.await;
// The Tenant goes back into the map in Stopping state, it will eventually be removed by DeleteTenantFLow
slot_guard.revert();
result
}
}
#[derive(Debug, thiserror::Error)]
@@ -1305,7 +1156,7 @@ pub(crate) fn get_tenant(
Some(TenantSlot::InProgress(_)) => {
Err(GetTenantError::NotActive(tenant_shard_id.tenant_id))
}
None | Some(TenantSlot::Secondary(_)) => {
None | Some(TenantSlot::Secondary) => {
Err(GetTenantError::NotFound(tenant_shard_id.tenant_id))
}
}
@@ -1357,11 +1208,9 @@ pub(crate) async fn get_active_tenant_with_timeout(
let locked = TENANTS.read().unwrap();
// Resolve TenantId to TenantShardId
let tenant_shard_id = locked
.resolve_attached_shard(&tenant_id, shard_selector)
.ok_or(GetActiveTenantError::NotFound(GetTenantError::NotFound(
tenant_id,
)))?;
let tenant_shard_id = locked.resolve_shard(&tenant_id, shard_selector).ok_or(
GetActiveTenantError::NotFound(GetTenantError::NotFound(tenant_id)),
)?;
let peek_slot = tenant_map_peek_slot(&locked, &tenant_shard_id, TenantSlotPeekMode::Read)
.map_err(GetTenantError::MapState)?;
@@ -1372,13 +1221,10 @@ pub(crate) async fn get_active_tenant_with_timeout(
// Fast path: we don't need to do any async waiting.
return Ok(tenant.clone());
}
_ => {
tenant.activate_now();
(WaitFor::Tenant(tenant.clone()), tenant_shard_id)
}
_ => (WaitFor::Tenant(tenant.clone()), tenant_shard_id),
}
}
Some(TenantSlot::Secondary(_)) => {
Some(TenantSlot::Secondary) => {
return Err(GetActiveTenantError::NotFound(GetTenantError::NotActive(
tenant_id,
)))
@@ -1429,10 +1275,63 @@ pub(crate) async fn get_active_tenant_with_timeout(
};
tracing::debug!("Waiting for tenant to enter active state...");
tenant
.wait_to_become_active(deadline.duration_since(Instant::now()))
.await?;
Ok(tenant)
match timeout_cancellable(
deadline.duration_since(Instant::now()),
cancel,
tenant.wait_to_become_active(),
)
.await
{
Ok(Ok(())) => Ok(tenant),
Ok(Err(e)) => Err(e),
Err(TimeoutCancellableError::Timeout) => {
let latest_state = tenant.current_state();
if latest_state == TenantState::Active {
Ok(tenant)
} else {
Err(GetActiveTenantError::WaitForActiveTimeout {
latest_state: Some(latest_state),
wait_time: timeout,
})
}
}
Err(TimeoutCancellableError::Cancelled) => Err(GetActiveTenantError::Cancelled),
}
}
pub(crate) async fn delete_tenant(
conf: &'static PageServerConf,
remote_storage: Option<GenericRemoteStorage>,
tenant_shard_id: TenantShardId,
) -> Result<(), DeleteTenantError> {
// We acquire a SlotGuard during this function to protect against concurrent
// changes while the ::prepare phase of DeleteTenantFlow executes, but then
// have to return the Tenant to the map while the background deletion runs.
//
// TODO: refactor deletion to happen outside the lifetime of a Tenant.
// Currently, deletion requires a reference to the tenants map in order to
// keep the Tenant in the map until deletion is complete, and then remove
// it at the end.
//
// See https://github.com/neondatabase/neon/issues/5080
// TODO(sharding): make delete API sharding-aware
let slot_guard = tenant_map_acquire_slot(&tenant_shard_id, TenantSlotAcquireMode::MustExist)?;
// unwrap is safe because we used MustExist mode when acquiring
let tenant = match slot_guard.get_old_value().as_ref().unwrap() {
TenantSlot::Attached(tenant) => tenant.clone(),
_ => {
// Express "not attached" as equivalent to "not found"
return Err(DeleteTenantError::NotAttached);
}
};
let result = DeleteTenantFlow::run(conf, remote_storage, &TENANTS, tenant).await;
// The Tenant goes back into the map in Stopping state, it will eventually be removed by DeleteTenantFLow
slot_guard.revert();
result
}
#[derive(Debug, thiserror::Error)]
@@ -1642,7 +1541,7 @@ pub(crate) async fn list_tenants() -> Result<Vec<(TenantShardId, TenantState)>,
Ok(m.iter()
.filter_map(|(id, tenant)| match tenant {
TenantSlot::Attached(tenant) => Some((*id, tenant.current_state())),
TenantSlot::Secondary(_) => None,
TenantSlot::Secondary => None,
TenantSlot::InProgress(_) => None,
})
.collect())
@@ -1899,7 +1798,11 @@ impl SlotGuard {
fn old_value_is_shutdown(&self) -> bool {
match self.old_value.as_ref() {
Some(TenantSlot::Attached(tenant)) => tenant.gate.close_complete(),
Some(TenantSlot::Secondary(secondary_tenant)) => secondary_tenant.gate.close_complete(),
Some(TenantSlot::Secondary) => {
// TODO: when adding secondary mode tenants, this will check for shutdown
// in the same way that we do for `Tenant` above
true
}
Some(TenantSlot::InProgress(_)) => {
// A SlotGuard cannot be constructed for a slot that was already InProgress
unreachable!()
@@ -2109,19 +2012,26 @@ where
let mut slot_guard =
tenant_map_acquire_slot_impl(&tenant_shard_id, tenants, TenantSlotAcquireMode::MustExist)?;
// allow pageserver shutdown to await for our completion
let (_guard, progress) = completion::channel();
// The SlotGuard allows us to manipulate the Tenant object without fear of some
// concurrent API request doing something else for the same tenant ID.
let attached_tenant = match slot_guard.get_old_value() {
Some(TenantSlot::Attached(tenant)) => {
Some(TenantSlot::Attached(t)) => Some(t),
_ => None,
};
// allow pageserver shutdown to await for our completion
let (_guard, progress) = completion::channel();
// If the tenant was attached, shut it down gracefully. For secondary
// locations this part is not necessary
match &attached_tenant {
Some(attached_tenant) => {
// whenever we remove a tenant from memory, we don't want to flush and wait for upload
let freeze_and_flush = false;
// shutdown is sure to transition tenant to stopping, and wait for all tasks to complete, so
// that we can continue safely to cleanup.
match tenant.shutdown(progress, freeze_and_flush).await {
match attached_tenant.shutdown(progress, freeze_and_flush).await {
Ok(()) => {}
Err(_other) => {
// if pageserver shutdown or other detach/ignore is already ongoing, we don't want to
@@ -2130,19 +2040,11 @@ where
return Err(TenantStateError::IsStopping(tenant_shard_id.tenant_id));
}
}
Some(tenant)
}
Some(TenantSlot::Secondary(secondary_state)) => {
tracing::info!("Shutting down in secondary mode");
secondary_state.shutdown().await;
None
None => {
// Nothing to wait on when not attached, proceed.
}
Some(TenantSlot::InProgress(_)) => {
// Acquiring a slot guarantees its old value was not InProgress
unreachable!();
}
None => None,
};
}
match tenant_cleanup
.await

Some files were not shown because too many files have changed in this diff Show More