Files
clean-architecture-backend-…/.github/workflows/integration-main.yml
T
DongHyeonka de070e16a7 ci: 다른 단계가 이미 돌리는 workflow job을 지운다
workflow 15개 가운데 모듈 모양 파일들이 PR·main·nightly 단계가 이미
계획하는 Gradle 태스크를 한 번 더 돌린다(설계 design-docs
repo-structure/03-ci.md 3.1절 「중복」).

- fileserver-pr.yml 다섯 job: 모두 leaf `test`나 그 부분집합이다.
  `ci`가 매 PR 돌린다.
- fileserver-nightly.yml의 --rerun-tasks job 셋: 같은 `test`의
  부분집합이다. NFS job만 남긴다.
- jpa-nightly.yml의 failure·query plan·security·pool job: 모두
  integration()·performance() lane이라 :integrationCheck·
  :performanceCheck가 돈다. PostgreSQL major 행렬만 남긴다.
- notification-platform.yml의 pr job: `ci`와 :integrationCheck의
  태스크다. 경로 필터가 docs/adr/로 옮긴 알림 ADR을 보지 못했다.
  chaos job만 남긴다.
- object-storage-qualification.yml의 MinIO job 둘: integration lane이다.
  보호된 AWS job만 남긴다.
- link-check.yml: 문서 검사라 main 단계로(소유자 결정 2026-09-16).
  integration-main.yml의 links job으로 옮긴다.

fileserver 지원 매트릭스는 지운 job id 대신 그 증거를 실제로 만드는
테스트 클래스·모듈 test·남은 job을 적고, FileserverDocumentation
CoverageTest가 행마다 그 이름이 있는지 본다. job이 사라진 시험 문장
(PoolLaneClaimTest의 nightly job 검사, lychee 경로 필터 검사)은
지우거나 새 위치로 맞춘다.

위험: PR에서 경로로 켜지던 컨테이너 lane(notification의 실 PostgreSQL
두 lane, MinIO)은 이제 main push와 nightly에서만 돈다. 어댑터 컨테이너
lane을 main 단계로 옮긴 소유자 결정(2026-11-07 ②)과 같은 쪽이다.

검증: 기준 2959eff0의 desktop worktree에서 workflow의 Gradle 명령 50개를
--dry-run해 태스크 집합을 만들었다. 지운·옮긴 job 27개의 태스크는 모두
:ci·:mainCheck·:integrationCheck·:performanceCheck·:releaseCheck 계획
안에 있다(단계 집계 밖 0). 로그 desktop ~/t2a-backend-ci-logs/wave1-base/.
python3 scripts/check-docs.py, check-work-docs.py 통과.

Refs: design-docs repo-structure/03-ci.md 7.3절 Wave 1
2026-11-08 17:00:49 +09:00

269 lines
14 KiB
YAML

name: integration-main
# Stage 2: is the merged state healthy.
#
# The question this stage answers is different from stage 1's. Stage 1 asks whether a diff is safe
# and blocks a merge; stage 2 asks whether main is healthy and does not — the merge has already
# happened. That difference is the point, and it is what lets a control exist without being an
# obstacle: a gate here still fails loudly, it just fails after the thing it is reporting on.
#
# Two kinds of work live here.
#
# 1. The lanes that need a machine that is not simultaneously compiling something else — load,
# abuse, graceful shutdown, TCP fault injection, resource bounds. They were web-nightly.yml and
# httpclient-nightly.yml, two module-shaped files whose only real difference was the cadence they
# shared. They now run on every push to main as well as nightly, which is strictly more often
# than before.
#
# 2. The container-backed lanes, run as the root category aggregates `:integrationCheck` and
# `:performanceCheck` rather than one workflow line per lane. A lane nobody runs is not coverage;
# it is a file that looks like coverage — and naming lanes here one by one is how fourteen of
# them came to run nowhere (review 2026-11-01 08-01). The pull-request gate's
# `verifyStrictLaneStages` now fails when a strict lane is in no stage a workflow executes. They
# are here rather than in stage 1 because every one of them starts containers or real servers,
# and the pull-request budget is minutes for the whole gate.
#
# What is deliberately NOT here: the web "Advanced capability" nightly lane that used to exist as
# web-advanced-nightly.yml (websocket-advanced-nightly.yml went with the websocket Advanced code,
# deleted 2026-09-25). The web leaf's build file says it outright — "They also run inside `test`,
# deliberately ... excluding them from the PR gate to make this lane look meaningful would mean the
# PR gate stopped covering a fifth of the leaf" — so `webAdvancedTest` selects tagged tests that
# `:<leaf>:test` already runs, and `:<leaf>:test` runs inside the root `check` on every pull request
# and every push to main. The strict lane itself survives in release.yml, where its
# fail-on-nothing-discovered guard is worth a job.
on:
push:
branches: ["main"]
schedule:
# 03:00 UTC. Late enough that the day's merges are in, early enough that a failure is triaged
# before the next working day starts.
- cron: '0 3 * * *'
workflow_dispatch:
permissions:
contents: read
concurrency:
group: ${{ github.workflow }}-${{ github.ref }}
cancel-in-progress: false
# The documentation-drift gates that used to run here are gone rather than demoted.
#
# They were four hand-written parsers: README shell blocks compared against the Gradle task graph,
# runbook identifiers compared against every declared Java type, a leaf count written in prose
# compared against the registry, and a Markdown table compared against the declared source sets.
# Each was a custom parser for a file format nobody controls, and each made a documentation edit a
# precondition for a build. A stale sentence is a defect, but it is not one a build can be failed
# for, and the `links` job below answers the one documentation question with a stable machine
# answer: does this link resolve.
jobs:
# The main-branch stage of the Gradle gates (`:mainCheck`): the environment contract, full-boot
# qualification, the public-path baseline, the conditional transport
# lanes, the messaging contract lanes and each leaf's document-contract lanes. API surfaces are
# in the pull-request `check` (owner decision 20). The rest left the pull-request gate (review
# 2026-09-24 BUILD, item 12); none of them is deleted.
main-check:
runs-on: ubuntu-latest
timeout-minutes: 60
steps:
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # actions/checkout@v4.2.2
- name: Require the committed public-path security baseline
run: |
set -euo pipefail
readonly snapshot='docs/security/public-paths-snapshot.txt'
if [[ ! -s "${snapshot}" ]] || ! git ls-files --error-unmatch "${snapshot}" >/dev/null 2>&1; then
echo "::error::${snapshot} is missing, empty or not committed; verifyPublicPathSnapshot must not create its own baseline."
exit 1
fi
- uses: ./.github/actions/setup-gradle-java
- name: Run the main-branch stage
working-directory: src
# `--continue`: one failed lane (an external image that cannot be pulled, say) must not stop
# the other main-stage gates from reporting (review 2026-11-01 08-04).
run: ./gradlew :mainCheck --continue --warning-mode=fail --stacktrace
# Storability only: the `--dry-run` below plans `integrationCheck` without running a lane. The
# lanes run in `integration-check`.
#
# Two questions. Can every task in the repository-wide gates be *stored* (a property of the task
# graph, so `--dry-run` answers it for all of them in seconds)? And is a stored entry *reused* while
# the build still executes its work (checked on the two core leaves)?
configuration-cache:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # actions/checkout@v4.2.2
- uses: ./.github/actions/setup-gradle-java
- name: Require every repository-wide gate to be storable
working-directory: src
shell: bash
run: |
set -euo pipefail
for gate in ci mainCheck integrationCheck releaseCheck; do
echo "::group::${gate}"
./gradlew "${gate}" --dry-run \
--configuration-cache --configuration-cache-problems=fail --stacktrace
output="$({ ./gradlew "${gate}" --dry-run \
--configuration-cache --configuration-cache-problems=fail --stacktrace; } 2>&1)"
printf '%s\n' "${output}"
echo '::endgroup::'
grep -Fq 'Reusing configuration cache.' <<<"${output}" || {
echo "::error::Gradle did not reuse the configuration cache for ${gate}."
exit 1
}
done
- name: Store configuration cache for the everyday core build
working-directory: src
run: >-
./gradlew
:domain-core:check
:application-core:check
--configuration-cache
--configuration-cache-problems=fail
--stacktrace
- name: Require configuration-cache reuse
working-directory: src
shell: bash
run: |
set -euo pipefail
output="$({ ./gradlew :domain-core:check :application-core:check \
--configuration-cache --configuration-cache-problems=fail --stacktrace; } 2>&1)"
printf '%s\n' "${output}"
grep -Fq 'Reusing configuration cache.' <<<"${output}" || {
echo '::error::Gradle did not reuse the configuration cache on the second identical build.'
exit 1
}
# The integration stage: every integration-category strict lane (`:integrationCheck`).
#
# This job replaced six jobs that named lanes one by one — the MongoDB lanes, app-bootstrap's
# Testcontainers lane, the moved messaging/object-storage/sample contracts, six of the
# persistence-jpa PostgreSQL contracts and the httpclient fault injection. Naming lanes was how
# eleven persistence-jpa readiness lanes ended up running nowhere: a lane added to Gradle without a
# workflow line was declared, failed closed, and executed on no machine (review 2026-11-01 08-01).
# The aggregate is the list now. A lane joins it by declaring `integration()`, and the root
# `verifyStrictLaneStages` (pull-request gate) fails when a strict lane is in no stage a workflow
# executes.
#
# `--continue` so that one red lane does not hide the others: every failed lane is named in the
# same run. Lanes that need an input this job does not have (the Redis topology coordinates, the
# protected AWS sandbox) are `explicitOnly`; the aggregate prints which ones it left out.
# The container lanes take the build-wide `exclusiveRealServerLane` slot one at a time, so the
# job's length is the sum of the lanes. Split the aggregate across jobs when it outgrows its
# timeout rather than going back to naming lanes.
#
# Measured length (one local run of `:integrationCheck`, 2026-11-01, 37 lanes, images already
# pulled, machine load 24): 794 s for the command, of which the lanes' test JVMs took about 624 s
# in sum — mongoSecurityIntegrationTest 89, jpaPlatformContractTest 104, messagingCertificationTest
# 70, mongoCompatibilityTest 69, jpaPlatformMigrationTest 66, mongoFailoverTest 59, the eleven JPA
# readiness lanes 57 together, httpClientFailureInjectionTest 27, sample integrationTest 23,
# messaging starter integrationTest 21, the rest under 10 each — and the remainder was compilation
# and container start-up between lanes. A hosted runner adds cold image pulls (about 5 GB
# uncompressed across the PostgreSQL, MongoDB, Kafka, nginx, httpbin and Toxiproxy images) and a
# cold Gradle cache; expect 25 to 40 minutes (an estimate, not a measured hosted run). The 240-minute timeout is a ceiling for a hung container, not a budget.
integration-check:
runs-on: ubuntu-latest
timeout-minutes: 240
env:
# Reuse would hand the failover lane a replica set another lane had already faulted.
TESTCONTAINERS_REUSE_ENABLE: "false"
steps:
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # actions/checkout@v4.2.2
- uses: ./.github/actions/setup-gradle-java
- name: Run every integration-category strict lane
working-directory: src
run: ./gradlew :integrationCheck --continue --stacktrace
- name: Publish the integration lane reports
if: always()
uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # actions/upload-artifact@v4.6.2
with:
name: integration-check-reports
path: src/**/build/reports/tests/
if-no-files-found: warn
# The performance stage: every performance-category strict lane (`:performanceCheck`) — Tomcat and
# Jetty load, slow-client and graceful shutdown, the Hikari pool contract, the httpclient pool,
# streaming, retry and rotation bounds, the MongoDB resource budget. They measure behaviour that
# degrades gradually rather than breaking outright, which is the kind of regression a per-PR gate
# never catches. This job replaced `web-load-abuse-and-shutdown`, which despite its name ran
# `:adapter:inbound:web:test` and `webJettyCompatTest` — two lanes the pull-request and main stages
# already run — and never the load lanes (review 2026-11-01 08-01), and `httpclient-performance`.
# No assertions switch: the strict-lane convention asserts every performance lane's bounds, so this
# job and a person on a workstation run the same command and get the same bounds.
performance-check:
runs-on: ubuntu-latest
timeout-minutes: 120
env:
TESTCONTAINERS_REUSE_ENABLE: "false"
steps:
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # actions/checkout@v4.2.2
- uses: ./.github/actions/setup-gradle-java
- name: Run every performance-category strict lane
working-directory: src
run: ./gradlew :performanceCheck --continue --stacktrace
- name: Publish the performance lane reports
if: always()
uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # actions/upload-artifact@v4.6.2
with:
name: performance-check-reports
path: src/**/build/reports/tests/
if-no-files-found: warn
httpclient-http3-experimental:
runs-on: ubuntu-latest
timeout-minutes: 30
# Experimental by design (D-08): the result is reported, never used to block a merge. Registered
# advisory so that "this job cannot fail the
# build" is written down rather than inferred from a field two hundred lines into a workflow.
continue-on-error: true
steps:
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # actions/checkout@v4.2.2
- uses: ./.github/actions/setup-gradle-java
- name: Exercise the experimental HTTP/3 opt-in
working-directory: src
run: ./gradlew :adapter:outbound:httpclient:test -Phttp3.tests.enabled=true --stacktrace
# Dynamic half of the compose-profile contract. Every blocking lane the matrix declares runs on
# main and nightly; a partial lane run is reproduction only and cannot satisfy this job. The count
# is not written here on purpose — a number in prose drifts the moment a lane is added, and the
# matrix file is the thing this job reads.
#
# The lanes boot the qualification image (src/Dockerfile COMPOSITION=qualification: every
# capability module present). The script selects it; release.yml builds the default image.
compose-runtime-smoke:
runs-on: ubuntu-latest
timeout-minutes: 180
steps:
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # actions/checkout@v4.2.2
- name: Run every blocking Compose runtime lane
run: bash scripts/run-compose-runtime-smoke.sh --matrix src/app-bootstrap/compose-lanes/compose-profile-contracts.json
- name: Publish sanitized runtime-smoke evidence
if: always()
uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # actions/upload-artifact@v4.6.2
with:
name: compose-runtime-smoke-evidence
path: src/app-bootstrap/build/evidence/runtime-smoke
if-no-files-found: warn
# Every committed Markdown link, external URLs included (lychee). A documentation check, so the main
# stage rather than the pull request (owner decision 2026-09-16); the relative links of every living
# document are also checked on each pull request by scripts/check-docs.py, which needs no network.
links:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # actions/checkout@v4.2.2
- name: Check committed documentation links
uses: lycheeverse/lychee-action@7cd0af4c74a61395d455af97419279d86aafaede # lycheeverse/lychee-action@v2.0.2
with:
args: >-
--no-progress
--root-dir .
README.md
src/README.md
'src/**/README.md'
'src/**/CLAUDE.md'
'docs/**/*.md'
'.github/**/*.md'
fail: true