Skip to content

master ci.yml all shards red + scheduled dsh lane 7 consecutive reds: two test families don't pin the storage mode — aft | master ci.yml 全分片红 + scheduled dsh lane 7 连红:两族测试没有钉存储模式,#3818 默认翻转后被 runner 默认值钉死在错的一边 #4076

Closed

Description

@ranxianglei

(ai: zhipuai-lb/glm-5.3)

这是啥问题? 两族红,机制同源、方向相反:

  1. ci.yml 主矩阵(5 个分片全红):tests/issue3769-upgrade-compat.test.ts:179(record hydrates before the upgrade turn)和 tests/issue3769-pooled-bare-load.test.ts 断言 loadSync 能取回记录。CI runner(scripts/run-ci-tests.mjs)预载 tests/preload-storage-baseline.mjs → BILI_STORAGE_UNIFIED ??= "legacy" → legacy 下 loadSync 无 meta 探不到 protocol 命名空间文件 → null → 红。本地 standalone(无 preload,默认 full)→ index row 是 load path → 绿。这两条测试本来就是 full-mode 故事(文件内注释自证),却没按 storage family 纪律自钉模式。
  2. scheduled ci-e2e-dsh lane(自 Oct19 run 1426 起 7 连红):tests/plugin-fork.test.ts 的 5 条「CCR ${corruption} fails closed after cold disk restore」期望 409,实际 200。该 lane 不带 preload → Storage wrap-up: 6 fixes + switch default mode to full (single PR) | 存储收尾六修 + 缺省模式翻 full(一个 PR 统一处理) #3818 后默认 full → 完好的 index.db 镜像合法地供应 pre-corruption CCR(Identity pipeline: key sessions by (agent, conversation id) in the unified store — decouple fold ledgers from protocol/host origin (#2938 / #2847 / #2671 defect-1) #2960 ③ 的设计),corrupt 的 JSON 文件不可见 → 200。带 preload(legacy)→ 49/49 绿。测试断言的是 legacy 单副本契约。

中间所有 master run 都被后续 push 互相 cancel,所以 #4015 的这两条新测试从未在 CI 的 legacy 分片下真正跑过——不是 #4015 的回归,是 #3818 默认翻转后两族测试落在错误一侧、一直被 cancel 掩盖。

根因? 两个 runner 环境的默认存储模式相反(CI shards=legacy via preload;scheduled dsh lane=full by default),而测试继承了 runner 默认。

如何解决?(即将提 PR) ①两个 #3769 文件顶部 BILI_STORAGE_UNIFIED = "full" 显式钉(??= 保显式优先);②plugin-fork 的 CCR 冷恢复改真冷恢复:corrupt 文件后 closeQuietly() + 删 index.db/-wal/-shm + 新建 store2——损坏文件成为唯一幸存副本,409 契约在 full/legacy 两模式同成立(不钉 env、不丢 full 覆盖)。双侧验证:两个 #3769 文件 2/2 × 双条件;plugin-fork 49/49 × 双条件。

能否合并? 合并后 master 的 ci.yml 与 scheduled dsh lane 应同时转绿(7549 同款红消除)。

Activity

  1. ranxianglei commented on Oct 21, 2026

    @ranxianglei
    OwnerAuthor

    (ai: zhipuai-lb/glm-5.3)

    修复 PR:#4077(单提交 602996b,双条件矩阵验证)。

  2. changed the title [-]master ci.yml 全分片红 + scheduled dsh lane 7 连红:两族测试没有钉存储模式,#3818 默认翻转后被 runner 默认值钉死在错的一边[/-] [+]master ci.yml all shards red + scheduled dsh lane 7 consecutive reds: two test families don't pin the storage mode — aft | master ci.yml 全分片红 + scheduled dsh lane 7 连红:两族测试没有钉存储模式,#3818 默认翻转后被 runner 默认值钉死在错的一边[/+] on Oct 21, 2026
  3. added a commit that references this issue on Oct 21, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions