fix(library): 写集合台账与跨范围冲突预检
build / build (push) Successful in 1m34s

失败恢复会在写完之后按"目标租户里没人引用"清理资源行,于是历史孤儿、同租户其它
知识库暂未绑定的资源、被共享的资源都可能被删掉;而预备份里没有这些行,删了不可
恢复,随后整实例指纹才发现不一致,只能停在 rollback_failed。恢复校验依赖的整实例
证据又只存在内存里,worker 重启后只按快照 id 恢复,等于静默跳过比对。

改法:写第一笔之前把"本次明确写集合 + 写前存在状态 + 预备份时刻整实例内容证据"
算出来并按内容寻址落盘(文件名与治理库里的引用都是 sha256),摘要登记在预备份
版本证据里;恢复只删"写前不存在 ∧ 由本次写入 ∧ 不在恢复内容里"的精确 ID,并对
资源额外要求仍无绑定/正文引用;缺记录、记录被改、绑定对不上都直接失败、维护保持。
resources 的 upsert 补租户限定,跨范围同 ID 在写入之前就被预检拒绝。
整实例指纹从"身份集合"改成"逐行内容摘要",别的知识库同 ID 行被覆盖写坏也会暴露。

本地验证(已跑):
- make verify 15 项全 success(identity 因本地档案 skipped)
- go test ./internal/releaseadapter/v071/ -run Isolated(真实 PG + 真实 ES +
  真实 MinIO 对象存储,写失败按定位符注入):目标非空/同租户其它知识库/共享资源/
  历史孤儿/跨租户数据在失败恢复后逐行字节不变、本次残留 0;缺写集合拒绝恢复;
  跨范围 ID 冲突在写入前拒绝;wiki 阶段半写回滚干净
- SIGKILL 子进程后按盘上写集合接管恢复,目标原有数据逐行不变
- go test ./internal/libraryrelease/:写集合先于第一笔、规划失败/登记失败都不写目标、
  接管不重新规划
This commit is contained in:
姚章浩
2026-10-29 23:55:13 +08:00
parent 034c0b3a1c
commit e0768f2caa
16 changed files with 2103 additions and 99 deletions
+35
View File
@@ -187,6 +187,41 @@
"package": "knowledge-governance/internal/qawemail",
"test": "TestScheduledSendRequiresConfirmationAndSendsExactlyOnce",
"reason": "需要一个一次性 PostgreSQL 与测试主密钥(KG_WEEKLY_MAIL_TEST_DSN/KG_WEEKLY_MAIL_TEST_KEY),CI 不提供;本批已在隔离库真实跑通并通过"
},
{
"package": "knowledge-governance/internal/releaseadapter/v071",
"test": "TestLibraryRollbackKeepsPreExistingTargetDataIsolated",
"reason": "需要一次性隔离库 kg_library_test_*、随机 ES 索引与真实对象存储(KG_LIBRARY_TEST_S3_*);验证目标非空/同租户其它知识库/共享资源/历史孤儿/跨租户数据在失败恢复后逐行字节不变,默认 CI 无这些服务,本机已实跑通过"
},
{
"package": "knowledge-governance/internal/releaseadapter/v071",
"test": "TestLibraryRestoreRefusesWithoutWriteSetRecordIsolated",
"reason": "同上环境;验证缺少持久化写集合时恢复必须失败且不动目标,默认 CI 跳过不算验证"
},
{
"package": "knowledge-governance/internal/releaseadapter/v071",
"test": "TestLibraryPreflightRejectsCrossScopeIDConflictIsolated",
"reason": "同上环境;验证跨范围同 ID 资源冲突在任何写入之前被拒绝,默认 CI 跳过不算验证"
},
{
"package": "knowledge-governance/internal/releaseadapter/v071",
"test": "TestLibraryWikiStageHalfWriteRollbackIsolated",
"reason": "同上环境;验证 wiki 阶段半写(database/resources 已提交)后回滚不留残渣,默认 CI 跳过不算验证"
},
{
"package": "knowledge-governance/internal/releaseadapter/v071",
"test": "TestLibraryWriteSetRecordTamperRefusesRestoreIsolated",
"reason": "同上环境;验证写集合记录被篡改/删除后取回与恢复都失败,默认 CI 跳过不算验证"
},
{
"package": "knowledge-governance/internal/releaseadapter/v071",
"test": "TestLibraryKillWorkerThenResumeRecoversWriteSetIsolated",
"reason": "同上环境并用 SIGKILL 杀掉子进程模拟 worker 被杀,验证重启后只按持久写集合恢复且目标原有数据不变;需要本地真实 PG/ES/MinIO,默认 CI 跳过不算验证"
},
{
"package": "knowledge-governance/internal/releaseadapter/v071",
"test": "TestWriteSetKillHelper",
"reason": "上面 SIGKILL 用例的子进程 fixture,单独运行没有意义;默认 CI 跳过"
}
]
}