ci: one shared path classification decides which suites run (#14150)

This commit is contained in:
Hengfei Yang
2026-09-03 09:34:31 +00:00
committed by GitHub
parent 5624e04a26
commit 2ddf142f6e
7 changed files with 137 additions and 47 deletions
@@ -0,0 +1,29 @@
name: Classify changed paths
description: One shared answer to "which suites does this change need", replacing per-workflow path regexes
inputs:
base:
description: Ref to diff against; the caller must have fetched it (fetch-depth 0)
default: origin/${{ github.base_ref || 'main' }}
head:
description: Commit to diff to
default: ${{ github.sha }}
outputs:
run_unit_backend: { value: "${{ steps.c.outputs.run_unit_backend }}" }
run_unit_ui: { value: "${{ steps.c.outputs.run_unit_ui }}" }
run_api: { value: "${{ steps.c.outputs.run_api }}" }
run_db: { value: "${{ steps.c.outputs.run_db }}" }
run_playwright: { value: "${{ steps.c.outputs.run_playwright }}" }
run_enterprise: { value: "${{ steps.c.outputs.run_enterprise }}" }
changed_workflows: { value: "${{ steps.c.outputs.changed_workflows }}" }
unknown: { value: "${{ steps.c.outputs.unknown }}" }
runs:
using: composite
steps:
- id: c
shell: bash
env:
BASE: ${{ inputs.base }}
HEAD: ${{ inputs.head }}
run: |
CHANGED=$(git -c core.quotepath=false diff --name-only "$BASE" "$HEAD" 2>/dev/null || true)
printf '%s\n' "$CHANGED" | "$GITHUB_ACTION_PATH/classify.sh" | tee -a "$GITHUB_OUTPUT"
+45
View File
@@ -0,0 +1,45 @@
#!/usr/bin/env bash
# Classify a list of changed paths (one per line on stdin) into the suites that must run.
# Prints key=value lines. Unknown paths run everything: the merge queue uses the same rules,
# so an allow-list miss would mean a path that is never tested.
set -euo pipefail
docs=false ci=false build_support=false ops=false backend=false frontend=false
api_tests=false ui_tests=false db_tests=false test_data=false mobile=false unknown=false
workflows=""
n=0
while IFS= read -r f; do
[ -z "$f" ] && continue
n=$((n+1))
case "$f" in
*.md|docs/*|screenshots/*|qa-reports/*|*.png|*.jpg|*.jpeg|*.gif|*.svg|*.webp|LICENSE|.github/ISSUE_TEMPLATE/*) docs=true ;;
.github/workflows/*) ci=true; workflows="$workflows ${f#.github/workflows/}" ;;
.github/actions/*|.github/scripts/*|.github/protoc.sh|.github/dependabot.yml|.github/pr-title-checker-config.json) build_support=true ;;
scripts/*|deploy/*|cross/*|config/*|download.sh|downloadO2.sh|Cross.*.toml|deny.toml|.devcontainer/*|.agents/*|.claude/*|.opencode/*|opencode.jsonc|.claudeignore|.cursorignore|.typos.toml|.gitattributes|.gitignore|.env.example|openobserve.cdx.xml|package-lock.json) ops=true ;;
src/*|proto/*|Cargo.toml|Cargo.lock|build.rs|rust-toolchain.toml|.cargo/*|tests/integration_test.rs|coverage.sh|clippy.toml|rustfmt.toml) backend=true ;;
web/*) frontend=true ;;
tests/api-testing/*) api_tests=true ;;
tests/ui-testing/*) ui_tests=true ;;
tests/db-testing/*) db_tests=true ;;
tests/mobile-testing/*) mobile=true ;;
tests/test-data/*|tests/*.json|tests/sourcemaps.zip|tests/.gitignore) test_data=true ;;
*) unknown=true ;;
esac
done
# No diff at all (no base to compare against) is treated as unknown: run everything.
[ "$n" -eq 0 ] && unknown=true
or() { for v in "$@"; do [ "$v" = true ] && { echo true; return; }; done; echo false; }
run_unit_backend=$(or "$backend" "$build_support" "$unknown")
run_unit_ui=$(or "$frontend" "$unknown")
run_api=$(or "$backend" "$api_tests" "$test_data" "$unknown")
run_db=$(or "$backend" "$db_tests" "$unknown")
run_playwright=$(or "$backend" "$frontend" "$ui_tests" "$test_data" "$unknown")
run_enterprise=$(or "$backend" "$frontend" "$unknown")
for k in docs ci build_support ops backend frontend api_tests ui_tests db_tests test_data mobile unknown \
run_unit_backend run_unit_ui run_api run_db run_playwright run_enterprise; do
echo "$k=${!k}"
done
echo "changed_workflows=${workflows# }"
+10 -17
View File
@@ -49,27 +49,20 @@ jobs:
with:
fetch-depth: 0
# One shared classification for every suite (see .github/actions/classify-changes); a change to
# this workflow file re-tests itself on top of it.
- uses: ./.github/actions/classify-changes
id: classify
- id: check
env:
HAS_CHANGES: ${{ steps.classify.outputs.run_api }}
CHANGED_WORKFLOWS: ${{ steps.classify.outputs.changed_workflows }}
run: |
# Skip when a PR touches only docs, images, frontend/UI-test files, or unrelated
# workflow files (e.g. a dependabot bump to another workflow) — none affect the API.
# Source and this workflow (api-testing.yml) still run the suite.
IGNORE_RE='(\.md$|^docs/|^web/|^tests/ui-testing/|^tests/playwright-tests/|\.(png|jpe?g|gif|svg|webp)$|^\.github/workflows/)'
CHANGED=$(git diff --name-only origin/${{ github.base_ref || 'main' }} ${{ github.sha }})
# Files not matching IGNORE_RE. Captured (not `grep -qv`) so the check doesn't depend
# on grep -qv exit semantics, which differ across grep builds; the `|| true` keeps
# `set -e` from aborting when grep matches nothing (exit 1). The `grep -qE` below is a
# loop-safe `if`-condition term (`set -e` doesn't fire there), so it needs no `|| true`.
RELEVANT=$(printf '%s' "$CHANGED" | grep -vE "$IGNORE_RE" || true)
if [ -z "$CHANGED" ] || [ -n "$RELEVANT" ] || printf '%s' "$CHANGED" | grep -qE '^\.github/workflows/api-testing\.yml$'; then
echo "has_changes=true" >> $GITHUB_OUTPUT
if [ "$HAS_CHANGES" = "true" ] || echo " $CHANGED_WORKFLOWS " | grep -q ' api-testing.yml '; then
echo "has_changes=true" >> "$GITHUB_OUTPUT"
else
echo "has_changes=false" >> $GITHUB_OUTPUT
echo "has_changes=false" >> "$GITHUB_OUTPUT"
fi
# Build the exact PR/merge-group binary once, then fan it out to every API
# test shard. The artifact is scoped to this workflow run and SHA; Rust cache
# remains an input to this single build, not a substitute for the binary.
build_api_binary:
name: build_api_binary
needs: [check_changes]
+10 -7
View File
@@ -44,16 +44,19 @@ jobs:
with:
fetch-depth: 0
# One shared classification for every suite (see .github/actions/classify-changes); a change to
# this workflow file re-tests itself on top of it.
- uses: ./.github/actions/classify-changes
id: classify
- id: check
env:
HAS_CHANGES: ${{ steps.classify.outputs.run_db }}
CHANGED_WORKFLOWS: ${{ steps.classify.outputs.changed_workflows }}
run: |
CHANGED=$(git diff --name-only origin/${{ github.base_ref || 'main' }} ${{ github.sha }})
# Same contract as playwright/api: other workflows' files are irrelevant, this file re-tests itself.
if [ -z "$CHANGED" ] || echo "$CHANGED" | grep -qE '^\.github/workflows/db-testing\.yml$'; then
echo "has_changes=true" >> $GITHUB_OUTPUT
elif echo "$CHANGED" | grep -qvE '(\.md$|^docs/|^web/|^tests/ui-testing/|^tests/playwright-tests/|\.(png|jpe?g|gif|svg|webp)$|^\.github/workflows/)'; then
echo "has_changes=true" >> $GITHUB_OUTPUT
if [ "$HAS_CHANGES" = "true" ] || echo " $CHANGED_WORKFLOWS " | grep -q ' db-testing.yml '; then
echo "has_changes=true" >> "$GITHUB_OUTPUT"
else
echo "has_changes=false" >> $GITHUB_OUTPUT
echo "has_changes=false" >> "$GITHUB_OUTPUT"
fi
db_validation_tests:
@@ -21,11 +21,28 @@ jobs:
runs-on: ubuntu-latest
permissions:
pull-requests: write
contents: read
statuses: write
steps:
# Debounce: a push inside this window cancels the run (concurrency) before anything paid starts.
- if: github.event_name == 'pull_request'
run: sleep 60
- uses: actions/checkout@v5
with:
fetch-depth: 0
- uses: ./.github/actions/classify-changes
id: classify
# ent-build-status-check is required; when no backend or frontend code changed there is nothing to build,
# so report it as passed here instead of dispatching.
- name: Report not applicable
if: ${{ steps.classify.outputs.run_enterprise != 'true' }}
env:
GH_TOKEN: ${{ github.token }}
run: |
gh api -X POST "repos/${{ github.repository }}/statuses/${{ github.event.pull_request.head.sha }}" \
-f state=success -f context=ent-build-status-check -f description="no backend or frontend change; enterprise build not needed"
- name: Generate token
if: ${{ steps.classify.outputs.run_enterprise == 'true' }}
id: generate_token
uses: actions/create-github-app-token@v3.2.0
with:
@@ -34,6 +51,7 @@ jobs:
# Set the owner, so the token can be used in all repositories
owner: ${{ github.repository_owner }}
- name: Set build short sha
if: ${{ steps.classify.outputs.run_enterprise == 'true' }}
shell: bash
run: |
unset SHORT_COMMIT_ID
@@ -41,9 +59,11 @@ jobs:
echo "COMMIT_ID=${GITHUB_HEAD_SHA}" >> $GITHUB_ENV
- name: Confirm git commit SHA output
if: ${{ steps.classify.outputs.run_enterprise == 'true' }}
run: echo ${{ env.SHORT_COMMIT_ID }}
- name: Repository Dispatch
if: ${{ steps.classify.outputs.run_enterprise == 'true' }}
uses: peter-evans/repository-dispatch@v4
env:
TOKEN: ${{ steps.generate_token.outputs.token }}
+9 -16
View File
@@ -87,26 +87,19 @@ jobs:
with:
fetch-depth: 0
# One shared classification for every suite (see .github/actions/classify-changes); a change to
# this workflow file re-tests itself on top of it.
- uses: ./.github/actions/classify-changes
id: classify
- id: check
env:
BASE_REF: ${{ github.base_ref }}
HAS_CHANGES: ${{ steps.classify.outputs.run_playwright }}
CHANGED_WORKFLOWS: ${{ steps.classify.outputs.changed_workflows }}
run: |
# Skip when a PR touches only docs, images, or unrelated workflow files (e.g. a
# dependabot bump to another workflow) — none can affect the app. Source, web/, and
# this workflow (playwright.yml) still run the suite.
# Two-dot on purpose: base-branch drift over-reports has_changes, which keeps the
# merge_group full run firing whenever the merged tree differs from what PRs tested.
IGNORE_RE='(\.md$|^docs/|\.(png|jpe?g|gif|svg|webp)$|^\.github/workflows/)'
CHANGED=$(git diff --name-only "origin/${BASE_REF:-main}" ${{ github.sha }})
# Files not matching IGNORE_RE. Captured (not `grep -qv`) so the check doesn't depend
# on grep -qv exit semantics, which differ across grep builds; the `|| true` keeps
# `set -e` from aborting when grep matches nothing (exit 1). The `grep -qE` below is a
# loop-safe `if`-condition term (`set -e` doesn't fire there), so it needs no `|| true`.
RELEVANT=$(printf '%s' "$CHANGED" | grep -vE "$IGNORE_RE" || true)
if [ -z "$CHANGED" ] || [ -n "$RELEVANT" ] || printf '%s' "$CHANGED" | grep -qE '^\.github/workflows/playwright\.yml$'; then
echo "has_changes=true" >> $GITHUB_OUTPUT
if [ "$HAS_CHANGES" = "true" ] || echo " $CHANGED_WORKFLOWS " | grep -q ' playwright.yml '; then
echo "has_changes=true" >> "$GITHUB_OUTPUT"
else
echo "has_changes=false" >> $GITHUB_OUTPUT
echo "has_changes=false" >> "$GITHUB_OUTPUT"
fi
build_binary:
+14 -7
View File
@@ -41,18 +41,25 @@ jobs:
with:
fetch-depth: 0
# One shared classification for every suite (see .github/actions/classify-changes); a change to
# this workflow file re-tests itself on top of it.
- uses: ./.github/actions/classify-changes
id: classify
- id: check
env:
HAS_BACKEND_CHANGES: ${{ steps.classify.outputs.run_unit_backend }}
HAS_UI_CHANGES: ${{ steps.classify.outputs.run_unit_ui }}
CHANGED_WORKFLOWS: ${{ steps.classify.outputs.changed_workflows }}
run: |
CHANGED=$(git diff --name-only origin/${{ github.base_ref || 'main' }} ${{ github.sha }})
if [ -z "$CHANGED" ] || echo "$CHANGED" | grep -qE '^src/|^proto/|^\.cargo/|^Cargo\.(toml|lock)$|^build\.rs$|^rust-toolchain\.toml$|^coverage\.sh$|^\.github/workflows/unit-tests\.yml$'; then
echo "has_backend_changes=true" >> $GITHUB_OUTPUT
if [ "$HAS_BACKEND_CHANGES" = "true" ] || echo " $CHANGED_WORKFLOWS " | grep -q ' unit-tests.yml '; then
echo "has_backend_changes=true" >> "$GITHUB_OUTPUT"
else
echo "has_backend_changes=false" >> $GITHUB_OUTPUT
echo "has_backend_changes=false" >> "$GITHUB_OUTPUT"
fi
if [ -z "$CHANGED" ] || echo "$CHANGED" | grep -q '^web/'; then
echo "has_ui_changes=true" >> $GITHUB_OUTPUT
if [ "$HAS_UI_CHANGES" = "true" ] || echo " $CHANGED_WORKFLOWS " | grep -q ' unit-tests.yml '; then
echo "has_ui_changes=true" >> "$GITHUB_OUTPUT"
else
echo "has_ui_changes=false" >> $GITHUB_OUTPUT
echo "has_ui_changes=false" >> "$GITHUB_OUTPUT"
fi
backend_static_checks: