Files
clean-architecture-backend-…/docs/redis/upgrade-guide.md
T
DongHyeonka 50cd7154d0 refactor(compose)!: compose 파일을 개발 환경 하나와 lane fixture로 나눈다
문제:
- 루트 compose 6개와 infra/ 아래 12개가 개발 환경, 테스트 lane fixture,
  보조 파일을 가리지 않고 섞여 있다. 이름 모양도 셋이다.
- README 빠른 시작 scripts/bootstrap-local.sh는 docker-compose.yml과
  .local.yml만 합친 뒤, 둘 어디에도 없는 db 서비스를 띄우려다 멈춘다.
- compose smoke all-adapters lane은 앱이 mem_limit 512m를 넘어 커널에
  죽고(OOM), restart: unless-stopped로 다시 떠서 죽는 때가 smoke 요청과
  겹칠 때만 실패한다. smoke는 재시작을 보지 않아 나머지 실행은 거짓 통과다.

고른 방법(소유자 결정 2026-11-07):
- 루트에 개발용 compose.yaml 하나. app(local 런타임)과 뒷단 서비스를
  capability profile(jpa, redis, kafka, mongo, objectstorage, auth, mail)로
  고른다. 마운트하는 보조 파일은 compose/(keycloak, minio init, Redis ACL,
  mailpit 인증서 폴더).
- lane fixture는 lane을 가진 leaf로: dev·prod-smoke·tls overlay, fileserver
  overlay, smoke·스키마 적용 서비스(lane-services.yaml)와 스크립트, lane
  계약 JSON → src/app-bootstrap/compose-lanes/. Redis topology →
  redis-sdk/compose/, NFS → fileserver/compose/nfs/. infra/는 없어진다.
- 계약의 composeProfile(문자열)을 composeProfiles(목록)로 바꾼다. lane이
  backing profile과 lane profile을 함께 켠다. 18 lane의 서비스 집합은
  전과 같다(verify-compose-profile-contracts.sh가 정확히 대조).
- smoke: lane이 끝나기 전에 그 project의 모든 컨테이너가 재시작 0,
  OOMKilled false인지 보고 아니면 실패한다. 앱의 memory.peak를 증거로
  남긴다. all-adapters만 appMemoryLimit 1g(계약 칸, smoke가 overlay에 씀).
- bootstrap-local.sh는 compose.yaml로 db와 app을 띄운다(인자 없음).
- 개발 compose와 lane 바탕이 같은 파일이라, app healthcheck는 모든
  런타임에서 답하는 관리 포트 readiness로 둔다.
버린 대안: lane을 Testcontainers로 옮기기(설계 다-2, 별도 판단), lane마다
기본 메모리를 올리기(배포 기본값 512m 계약을 흐림).

위험·한계:
- 개발 compose의 app이 src/.env를 읽으므로, 개발자 src/.env가 dev·prod
  lane에도 들어간다. lane 값은 생성 overlay의 environment가 이기고, CI에는
  src/.env가 없다.
- MinIO 이미지 기본값이 compose.yaml과 lane-services.yaml 두 곳에 있다
  (YAML anchor는 파일을 넘지 못한다). ComposeMinioImageMatchesEvidenceTest가
  둘 다 evidence digest와 대조한다.
- env 키 검사(verifyEnvKeys)는 compose.yaml과 lane overlay를 읽고
  lane-services.yaml(smoke client 변수)은 뺀다.

검증(desktop, 트리 6d1e9f9d → 주석·문서만 고친 d0703901, ~/t1-layout-logs/):
- verify-compose-profile-contracts.sh rc=0 (18 lane 정확히 일치)
- :app-bootstrap:check :app-bootstrap:documentContractTest
  :app-bootstrap:verifyEnvKeys와 objectstorage·cache-redis·redis-sdk·
  redis-runtime·fileserver·composition:fileserver-platform check rc=0
- :app-bootstrap:integrationTest(ComposeMergeCharacterizationTest 포함),
  :adapter:outbound:cache-redis:redisContainerContractTest rc=0
- -p build-logic check rc=0, :verifyStrictLaneStages --dry-run rc=0
- scripts/run-compose-runtime-smoke.sh --matrix(18 lane) rc=0, 794초.
  모든 lane 컨테이너 재시작 0·OOMKilled false. 앱 memory.peak 최고
  all-adapters 532MiB(512m 한도를 넘는 값이라 1g의 근거), 나머지 244~422MiB
- scripts/bootstrap-local.sh rc=0(db·app 기동, /api/healthcheck 200 UP).
  기준 ac03192d의 base+local 조합은 config --services가 app뿐이고
  up --wait db가 "no such service: db"로 rc=1
- 정리할 때 `docker compose down`은 profile 서비스(db)를 남겨
  README·compose.yaml 안내를 `--profile '*' down`으로 고치고 다시 확인
- 주석·문서 수정 뒤 :app-bootstrap:repositoryContractTest
  :app-bootstrap:documentContractTest, verify-compose rc=0

Refs: design-docs repo-structure/README.md 3.1, repo-structure/01-backend-layout.md 6.3
BREAKING CHANGE: 루트 docker-compose*.yml과 infra/가 없어지고 compose.yaml(profile)·compose/·leaf 안 lane fixture로 옮겨졌다. lane 계약은 src/app-bootstrap/compose-lanes/compose-profile-contracts.json, composeProfile → composeProfiles. docs/upgrade/2026-11-maturity-review.md B-51.
2026-11-07 12:52:36 +09:00

67 lines
3.2 KiB
Markdown

# Redis and client upgrade gate
Changing the Redis server version or the Lettuce version is not a dependency bump. Both change what
commands exist, what they reply, and what an ACL account is allowed to do — all three are things this
SDK encodes as fixed decisions. The checks below must pass before either version moves, and each one
exists because skipping it produces a specific failure that only shows up in production.
## 1. Command metadata diff
Compare `COMMAND DOCS` / `COMMAND INFO` from the new server with the policy by hand: every command
the server reports must be classified in
`src/adapter/outbound/redis-sdk/src/main/resources/redis-sdk/redis-command-policy.yml`, and every
key specification, ACL category and deprecation the policy relies on must still match. There is no
automated diff: `RedisCommandMetadataDiff` compared against server metadata nothing ever fetched,
and was removed (see `docs/capability-status.md`).
*Why:* an unclassified command is refused by `CommandPolicyGuard`, so a server that grew a command
does not create a hole — but a command whose **risk changed upstream** and is still classified R1
here does. The comparison is what surfaces that.
## 2. ACL regression
Re-run `ACL DRYRUN` for every account against every command the SDK can issue, using
`RedisAdminOperations.aclDryRun`.
*Why:* a permit never widens an ACL account, so the account is the last boundary. A new server
version that moved a command into a different ACL category silently turns a working call into a
runtime refusal on the first request that needs it.
## 3. Serializer golden bytes
Compare the encoded form of every registered codec against the stored golden bytes.
*Why:* a value written by the old version must still decode after the upgrade. A codec change that
looks harmless in a round-trip test is not harmless against data already in the instance.
## 4. Support matrix
Update `docs/redis/support-matrix.md`. `RedisSupportMatrixTest` fails when a module or capability is
missing, and the certified-version table must not claim a version until its topology lane has
actually run.
## 5. Topology suite
Run the standalone, Sentinel, and Cluster lanes declared in `src/adapter/outbound/redis-sdk/compose/`. A version is
certified by the lane passing, not by the version number being newer.
*Why:* failover certainty and cross-slot behaviour are the two things the in-memory fixture cannot
prove. `ExecutionCertainty` and `RedisSlotCalculator` are classification and arithmetic; whether the
driver actually behaves that way during a promotion or a resharding is only observable on a real
topology.
## 6. Rollback
Before the upgrade, record the previous server version, the previous Lettuce version, and the
`SCRIPT LOAD` digests of every registered script. A rollback is not complete until the digests
resolve again on the restored version.
*Why:* digests are cached per process and invalidated by `SCRIPT FLUSH` and by restarts. A rollback
that leaves a process holding digests the restored server does not know produces `NOSCRIPT` on
every scripted call until the cache is dropped.
## What this gate does not cover
Data migration. Nothing here moves or reshapes stored values; a change that alters what is stored,
rather than how it is addressed, needs its own plan.