fix(migration): 실패한 동시 인덱스 마이그레이션 복구와 검사기 설명을 실제 동작에 맞춘다

문제 (재리뷰 round7 영속성 RR7-03-01~05)
- 실패한 동시 인덱스 마이그레이션의 복구를 가이드·검사기·학습 문서가
  "DROP INDEX 하나"로 안내한다. Flyway는 트랜잭션 밖에서 실패한 버전의
  이력 행(success = false)을 남기고, 다음 배포의 validate가 그 행 때문에
  다시 멈춘다. runbook에는 이 경우가 없고 이력 수정은 금지다.
  docs/jpa/runbooks.md "An invalid index exists"도 같은 이유로 재적용에서
  멈춘다.
- 검사기 Javadoc·학습 문서가 "섞인 파일은 부분 적용되지 않는다"고 쓴다.
  가이드가 요구하는 executeInTransaction=false를 붙이면 Flyway가 문장마다
  커밋해 앞 문장이 남는다.
- 가이드·검사기 Javadoc·실패 메시지·학습 문서가 "Flyway는 REINDEX …
  CONCURRENTLY를 트랜잭션 안에서 돌린다"고 쓴다. REINDEX SCHEMA|DATABASE
  CONCURRENTLY는 Flyway가 스스로 트랜잭션 밖에서 돌린다.
- LockWaitMigrationInspector는 .sql.conf가 executeInTransaction=true면
  SET LOCAL 뒤의 VACUUM FULL·REINDEX SCHEMA를 통과시킨다. Javadoc은
  "늘 거부"라고 쓴다.
- ledger 잠금 대기 IT가 동시 인덱스 파일을 "이 테스트에 예외로 적으라"고
  안내하지만 예외를 적을 곳이 없다. 같은 원인으로 ledger의 다른 IT 두 개
  (GrpcOperationLedgerPostgresIT, JpaGrpcOperationLedgerContractIT)도
  모든 파일을 한 문자열로 실행해, 동시 인덱스 파일이 생기면 "cannot be
  executed within a pipeline"으로 실패한다.

고른 방법
- 복구(RR7-03-03): 소유자 결정(2026-11-06)대로 migration-failed.md에
  "Recover a failed concurrent index migration" 절을 둔다. 진행 중 빌드
  확인 → 무효 인덱스 DROP INDEX CONCURRENTLY → 원인 정리 → 그 버전의
  실패 행만 DELETE → 재배포. "Do not"에 이 한 가지 예외를 적고
  flyway repair는 계속 금지한다. SQL 세 블록에 runbook-sql 표시를 달고
  postgresqlConcurrentIndexMigrationIntegrationTest가 runbook에서 읽어
  실 PostgreSQL에서 돌린다(outbox runbook과 같은 방식). 가이드·검사기
  Javadoc·메시지·학습 문서·jpa/runbooks.md는 그 절을 가리킨다.
- 섞인 파일(RR7-03-04)·REINDEX(RR7-03-02): 문장을 고치고, 같은 IT와
  잠금 대기 IT에 사실을 고정한다(섞인 파일 두 경우, REINDEX 다섯 형태).
- 검사기(RR7-03-05): 트랜잭션 안에서 돌 수 없는 잠금 문장은 .sql.conf와
  상관없이 거부한다(리뷰 권장안). 버린 안: Javadoc만 좁히기 — 그 파일은
  PostgreSQL이 배포 때 25001로 거부하므로 검사기가 먼저 알려 주는 쪽이
  맞다. 잠그지 않는 문장(plain VACUUM 등)은 이 검사기의 범위가 아니라
  그대로 둔다.
- ledger(RR7-03-01): IT에 RUN_OUTSIDE_A_TRANSACTION(파일 → 이유) 목록을
  두고 잠금 대기 케이스에서 뺀다. 목록의 파일은 같은 IT가 동시 인덱스
  문장 하나·executeInTransaction=false인지 ConcurrentIndexMigration
  Inspector로 본다. optional 빌드는 testkit에 의존할 수 없어
  sharedMigrationFileSources가 그 파일과 MigrationResource를 복사한다
  (VersionedMigrationFiles와 같은 방식). ledger IT 셋은 마이그레이션을
  파일마다 따로 실행한다(Flyway처럼). 버린 안: 안내만 "다룰 수 없다"로
  바꾸기 — 다음 동시 인덱스 파일이 생길 때 같은 질문이 다시 남는다.

위험·한계
- 검사기 변경으로 서비스의 자기 스트림 검사가 새로 실패할 수 있다.
  걸리는 파일은 PostgreSQL이 배포 때 거부하는 파일뿐이라 배포된 적이
  없다(업그레이드 안내 4.1에 적음).
- runbook의 DELETE는 운영자가 version을 옮겨 적는다. NOT success라
  성공한 행은 지우지 못하지만, 다른 버전의 실패 행을 잘못 적으면 그 행이
  지워진다. 한 행만 지워야 하고 0행이면 다시 분류하라고 적었다.
- ledger 복사 파일 두 개도 JDK만 써야 한다(각 Javadoc에 적음).
- Spotless 검사는 같은 worktree에서 spotlessApply가 변경 없이 끝난 직후라
  UP-TO-DATE로 처리됐다.

검증 (desktop, 공유 잠금, 로그 ~/r8-f3-logs/)
RED — 기준 73d39212 + 이 커밋의 테스트(+ 기준 문서 주장을 단언하는 임시
메서드, + 임시 ledger V004):
- persistence-jpa test --tests LockWaitMigrationInspectorTest: 80건 중
  2건 실패, 검사기가 빈 목록(통과)을 돌려줌 (red-unit.log)
- postgresqlMigrationLockWaitIntegrationTest: 47건 중 4건 실패 — 옛 주장
  대로면 거부여야 할 REINDEX SCHEMA|DATABASE CONCURRENTLY가 applied,
  conf=true VACUUM FULL·REINDEX SCHEMA를 PostgreSQL은 25001로 거부하는데
  검사기는 통과 (red2-lockwait.log)
- postgresqlConcurrentIndexMigrationIntegrationTest: 9건 중 3건 실패 —
  옛 절차(DROP INDEX) 뒤 재배포가 "Detected failed migration to version
  2", conf를 붙인 섞인 파일이 행 3001·무효 인덱스·실패 이력을 남김,
  runbook 블록이 없어 복구 테스트 준비 실패(RED로 세지 않음)
  (red2-concurrent.log)
- ledger integrationTest(V004 동시 인덱스): 6건 중 4건 실패 — 잠금 대기
  IT "name it as an exception here", 다른 IT 둘 "cannot be executed within
  a pipeline" (red-ledger.log)
GREEN — 이 커밋의 트리 b49369e7:
- :adapter:outbound:persistence-jpa:check rc=0 (test 666건 실패 0),
  :testQualityCheck rc=0
- postgresqlMigrationLockWaitIntegrationTest rc=0 (42건 실패 0)
- postgresqlConcurrentIndexMigrationIntegrationTest rc=0 (7건 실패 0)
- -p optional-platforms :grpc:grpc-operation-ledger-jpa:check
  integrationTest testQualityCheck rc=0 (IT 3+2+42건 실패 0)
- :app-bootstrap:documentContractTest rc=0 (66건 실패 0)
- 두 빌드의 verifyStrictLaneStages --dry-run rc=0
- ledger 탐침(임시 V004): 목록에 없으면 잠금 대기 케이스만 실패하고
  메시지가 RUN_OUTSIDE_A_TRANSACTION을 가리킴(48건 중 1건), 목록에 넣으면
  rc=0, 목록에 넣고 SET LOCAL을 섞으면 목록 검사가 "mixes a concurrent
  index statement"로 거부
- python3 scripts/check-work-docs.py --links docs: 깨진 링크 0

Refs: docs/reviews/2026-11-03-rereview/round7/03-persistence.md
This commit is contained in:
DongHyeonka
2026-11-07 01:14:09 +09:00
parent 73d39212f5
commit 6ec0c4f7fa
18 changed files with 587 additions and 95 deletions
+35 -11
View File
@@ -47,6 +47,10 @@ applied one, and the interesting question is which change is missing from this d
makes that question unaskable. It exists only as an explicit admin operation with an operator, a
reason, and an approval (design §8.4).
The one change to the history an operator makes is narrower than repair: after a concurrent index
migration fails, the failed row of that one version is deleted so that the version can run again
("Concurrent index builds" below). Every applied row, and its checksum, stays as it was.
Only Flyway's structured error codes reach the exception. Its messages embed the script path and
part of the failing statement.
@@ -84,7 +88,9 @@ Flyway's list: Flyway runs it in the transaction, PostgreSQL refuses it there, a
`VACUUM FULL`, a non-concurrent `REINDEX SCHEMA|DATABASE|SYSTEM` — can never be bounded this way and
does not belong in a migration: run it by hand as an operations step, outside a deploy, in a psql
session that sets `SET lock_timeout = '5s';` first (the session ends with the step, so the plain
`SET` is the bound there).
`SET` is the bound there). A `.sql.conf` with `executeInTransaction=true` does not change that: it
keeps the statement and its `SET LOCAL` in the transaction only for PostgreSQL to refuse it at
deploy, and the inspector below refuses it as well.
Enforced, not remembered: `LockWaitMigrationInspector` (persistence-jpa testkit) reads a script —
and the `.sql.conf` beside it, as Flyway reads it (a properties file, `false` in any case), and
@@ -112,18 +118,25 @@ optional build, which cannot depend on the testkit: its `GrpcOperationLedgerMigr
runs every ledger migration after V001, read from the tree, against a held ledger instead, so its
next migration is checked on real PostgreSQL. It finds and orders that tree with the testkit's own
`VersionedMigrationFiles`, which the ledger's `build.gradle` copies into its integration-test sources
at build time (`sharedMigrationFileSources`), so both builds read one definition. The rule came from
V21 having the bound and V22, in the same commit, not having it (re-review 2026-11-03 round 2,
at build time (`sharedMigrationFileSources`), so both builds read one definition. A concurrent index
build in that tree runs outside a transaction and is no case for a held ledger: it goes in the
test's `RUN_OUTSIDE_A_TRANSACTION` with its reason, and the same test holds each file listed there to
`ConcurrentIndexMigrationInspector`, copied the same way ("Concurrent index builds"). Every ledger
lane applies the tree one file at a time, as Flyway does, so such a file is not refused inside the
implicit transaction of one string holding many. The rule came
from V21 having the bound and V22, in the same commit, not having it (re-review 2026-11-03 round 2,
RR-02-09; round 3, RR3-02-01, RR3-02-03, RR3-02-04; round 4, RR4-02-01; round 6, RR6-03-02,
RR6-03-03).
RR6-03-03; round 7, RR7-03-01).
## Concurrent index builds
`CREATE INDEX CONCURRENTLY`, `DROP INDEX CONCURRENTLY` and `REINDEX … CONCURRENTLY` cannot run
inside a transaction block. Flyway recognizes `CREATE`/`DROP INDEX CONCURRENTLY` and runs such a
script outside a transaction on its own (above), but not `REINDEX … CONCURRENTLY`, which it runs in
the transaction for PostgreSQL to refuse. Write the companion configuration for every concurrent
statement anyway, so the file says what it needs and no reader has to know Flyway's list:
inside a transaction block. Flyway runs a script outside a transaction on its own when it holds
`CREATE`/`DROP INDEX CONCURRENTLY` or `REINDEX SCHEMA|DATABASE`, with or without `CONCURRENTLY`
(above). It runs `REINDEX INDEX|TABLE … CONCURRENTLY` — `REINDEX (CONCURRENTLY) TABLE` included — in
the transaction, where PostgreSQL refuses it. `PostgreSqlMigrationLockWaitIntegrationTest` shows the
five forms on real PostgreSQL. Write the companion configuration for every concurrent statement
anyway, so the file says what it needs and no reader has to know Flyway's list:
```conf
# V42__order_index.sql.conf
@@ -144,9 +157,20 @@ open holds the build, and a file run outside a transaction cannot bound that wai
Look for old transactions (`pg_stat_activity.xact_start`) before deploying one.
The testkit's `ConcurrentIndexMigrationInspector` checks a script for this in a migration test
(nothing checks it at startup), and additionally requires the migration to contain nothing else. A failed concurrent build leaves an invalid index behind;
recovering is a single `DROP INDEX` when the migration did nothing else, and a manual reconstruction
of partial state when it did.
(nothing checks it at startup), and additionally requires the migration to contain nothing else.
Next to another statement, a concurrent one makes Flyway refuse the file before any of it runs when
nothing configures it; with `executeInTransaction=false` Flyway commits each statement on its own,
and a failed build leaves the statements before it applied — state to reconstruct by hand.
Alone, a failed build leaves an invalid index and a history row with `success = false` for its
version, and Flyway refuses every later deploy ("Detected failed migration to version …") until that
row is gone: dropping the index is not enough. The recovery is
[migration-failed.md](../runbooks/migration-failed.md) "Recover a failed concurrent index migration":
confirm no build is running, drop the invalid index, remove the cause, delete that version's failed
row and nothing else, deploy again. It is not `flyway repair`, which would also rewrite the
remaining rows' checksums. `PostgreSqlConcurrentIndexMigrationIntegrationTest`
(`postgresqlConcurrentIndexMigrationIntegrationTest`) runs the runbook's SQL, read from the runbook,
after a unique build that failed on a duplicate, and shows both mixed files above.
An invalid index is not merely useless — the planner ignores it while every write still maintains
it. [runbooks.md](runbooks.md) has the query that finds them. Nothing drops them automatically: an
+5 -1
View File
@@ -73,7 +73,11 @@ A concurrent build failed. The index is ignored by the planner and maintained by
1. Confirm no build is currently running. An in-progress build looks identical in the catalog.
2. Run `DROP INDEX CONCURRENTLY IF EXISTS <invalid_index>;` outside a migration.
3. Re-apply the index migration.
3. If a migration built it, Flyway recorded that version as failed (`success = false`) and refuses
every later deploy until that row is gone: dropping the index does not clear it. Follow
[migration-failed.md](../runbooks/migration-failed.md) "Recover a failed concurrent index
migration", which removes the cause and deletes that one row.
4. Re-apply the index migration.
The platform does not drop these automatically: on a rolling deploy every instance would race to
drop an index another instance was about to finish building.