Skip to content

[PR] test(#5381): dump what lands on disk when the restart-measurement PFA flake fails - #5400

Open
ranxianglei wants to merge 2 commits into
masterfrom
2026-11-01_issue5381-flake-diagnostics
Open

ranxianglei wants to merge 2 commits into
masterfrom
2026-11-01_issue5381-flake-diagnostics

Conversation

@ranxianglei

@ranxianglei ranxianglei commented Nov 1, 2026 •

Copy link
Copy Markdown
Owner

Status (2026-11-02): root cause fixed upstream by #5512 (487daa26e) — resident-metadata fold gated on committed-bytes fingerprint. This diagnostics PR is now redundant for that bug; awaiting owner decision to close or keep as future-flake observability.

Refs #5379

查重:不重复 — 新失败签名(tests/issue2129-restart-measurement-state.test.ts:327 的 PFA stamp must be on disk after restart),open+closed 无同签名 issue/PR。

What this does

Adds failure-path self-diagnostics to the #2129 restart-measurement e2e. When the four post-restart assertions fail, it now dumps (before rethrowing):

  • the exact stats object that landed on disk;
  • the metadata keys that landed on disk + whether anonymousPrefixAffinity is present;
  • the expected k̂ / anchor / est values set in memory;
  • whether the live session's metadata had been swapped for a dropped lazy proxy (isDroppedMetadata);
  • whether the store still has a pending save for the session.

Each read is individually guarded so a lazy-metadata proxy trap can't mask the real assertion error.

Why

The flake is CI-only and has never reproduced locally (pristine-master + concurrency/CPU-burner runs all pass — see #5379). The "~3.13s pinned" duration was already shown (#5379 floors 1 & 3) to be per-test teardown, not a deadline, so the mechanism is unknown. Rather than guess at a persist-order change or ship an unproven masking retry shape, this makes the next occurrence explain itself: the failing shard logs exactly which side diverged (stats present vs. metadata-PFA absent) and whether the lazy-proxy path was involved.

Scope / behavior

Verification

  • Local targeted run (this file only): 4/4 subtests pass, e2e at ~3.27s (the ~3.2s teardown matches the reporter's numbers).
  • npm run typecheck: clean.
  • Branch based on current master 7578e8184; full-suite verdict = remote CI test lanes.

@github-actions

github-actions Bot commented Nov 1, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-11-01_issue5381-flake-diagnostics (69fc50f)

Latest build — install this exact version

npm install -g billion-context@0.0.8935-pr.5400.5c7c646

This exact version is immutable and reproducible. The rolling pr tag tracks the newest test build across all PRs.

Recent builds (commit → version)

commit version build
69fc50f billion-context@0.0.8935-pr.5400.5c7c646 build
f3b30a0 billion-context@0.0.8930-pr.5400.132ec2c build
e39190e billion-context@0.0.7640-pr.5400.6d19de7 build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr5400.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei ranxianglei changed the title test(#5381): dump what lands on disk when the restart-measurement PFA flake fails [PR] test(#5381): dump what lands on disk when the restart-measurement PFA flake fails Nov 1, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

查重:不重复。 按失败签名(PFA stamp must be on disk after restart + tests/issue2129-restart-measurement-state.test.ts:327)检索 open+closed issue 与 PR:唯一相关条目是 #5381(flake 本体,open),本 PR 是它派生的观测性交付,无同签名第二条目。注:镜像 issue 号 5402 对应上游 GitHub PR #5400(本仓 issue/PR 号段不同)。

标签: scope/ci · p2(测试/CI 面观测性改动;严重度沿用其跟踪对象 #5381 的 p2——CI-only flake、终端用户不可见、非数据丢失。不打 bug:这是 PR 不是缺陷报告。)

初审状态(正在继续,稍后给完整八问):

  • 分支已拉取核对:单 commit e39190ed7,基于当前 master 7578e8184,仅动 tests/issue2129-restart-measurement-state.test.ts(+23/−5)。
  • 独立复核了 diff:新增代码全部在四条断言的 catch 内、rethrow 前 dump 诊断(落盘 stats / metadata key 集 + PFA / 内存期望值 / isDroppedMetadata(s.metadata) / store.hasPending("i2129-e2e")),每处读取单独 try/catch——成功路径字节不变。
  • 符号核对通过:isDroppedMetadata(src/session.ts:140,(value: unknown): boolean)、store.hasPending(id)(src/persist.ts:3509)、会话 id i2129-e2e 与测试实际会话一致(header 与 getSession 均用它)。

接下来:本地定向验证(typecheck + 该文件单跑)+ 读镜像缓存查 CI 状态,然后按八问给终审结论。声明:此 PR 是我上一轮在 #5381 里开的,本轮按独立审核重读 diff,不复用旧结论。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

查重:不重复(首条回复已报,维持结论)。独立复审完成(不复用我此前在 #5381 的 triage 结论,diff 与行为重新核过一遍),CI 证据已取齐。

结构化八问

  1. 是什么问题:[PR] docs: README FAQ entry for the qoder-trae-workbuddy-switch quota gateway combo #5381 的 CI-only flake(tests/issue2129-restart-measurement-state.test.ts:327 的 PFA stamp must be on disk after restart)本地从未复现、机制未知;本 PR 不修 flake,只给失败路径加自诊断,让下一次 CI 命中时日志自己说明哪一侧分叉(stats 在盘 vs metadata-PFA 缺失)以及 lazy-metadata proxy 是否参与。
  2. 什么原因导致的:flake 根因未知(这正是加观测的原因);已知它是 master-inherited——只在 workflow-YAML/docs 变更的 master 提交上也照样红。
  3. 影响范围与严重程度:零产品面影响。纯测试文件(+23/−5,单文件),不碰 wire/config/persistence/kernel;成功路径逐字节不变,新代码只在 catch 内执行。严重度:p2 级 CI 噪音的可观测性增强,无用户可见变化。
  4. 怎么修复的:四个 restart 后断言包进 try/catch;失败时先打六行 [2129-diag](stats.onDisk 全量 / metadata.onDiskKeys / metadata.pfa / expected{k,anchor,est} / live.metadata.droppedProxy / store.hasPending),每个读取单独 guard(lazy-proxy trap 不会掩盖真实断言错误),然后原样 rethrow。
  5. 是否有回归问题:无。逐项过:成功路径——四条断言原样执行,不变;失败路径——旧行为「裸 ERR_ASSERTION」→ 新行为「同一错误 + 前置六行诊断」,已披露(仅测试失败输出多几行 stderr,CI 日志面);product/wire/config/persistence/kernel 均未触碰,高危模式(语义漂移、fail-fast 变吞错、阈值/时序/输出格式漂移)逐项核对为不涉及。两个新读取符号确认为纯读:isDroppedMetadata(src/session.ts:140,WeakSet 查询)、hasPending(src/persist.ts:3509,timer 表查询),失败路径无副作用。
  6. 是否破坏缓存:不涉及(未触碰任何缓存/重试/游标逻辑)。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:无。

验证证据(全部实际执行)

  • npm run typecheck:clean。
  • 定向跑本文件:4/4 pass,e2e 子测试 3274ms(~3.2s teardown 与 issue 数字吻合)。
  • 双向行为实证(gate 纪律的正反两向):临时副本强制第一条断言失败 → exit=1,六行诊断全部打出(含 metadata.pfa={"depth":22,...} 在盘、live.metadata.droppedProxy=true、store.hasPending=false),且原始 ERR_ASSERTION 完整保留未被掩盖。临时文件已删,工作树干净。顺带一个早期信号:强制失败实验中,断言时刻 live 会话的 metadata 已是 dropped lazy proxy 而磁盘副本里 PFA 仍在——与 [PR] docs: README FAQ entry for the qoder-trae-workbuddy-switch quota gateway combo #5381 的假设域一致,真机命中时可交叉印证。
  • CI(head e39190e):原始 push run 的全量 core 矩阵全绿——11/11 test shards(ubuntu 22/24 × 3 + windows 22 × 5,含 [PR] docs: README FAQ entry for the qoder-trae-workbuddy-switch quota gateway combo #5381 惯常命中的 ubuntu shard 1)+ static-gates×2 + mysql-contract + build-artifact 全部 success(03:07–03:19Z)。注意:03:16Z 一次 label 事件(镜像同步打的 p2/scope/ci 标签)按 ci.yml:181-185 的设计触发了 skip-scope 重跑,当前 check 视图因此显示 test lane 为 skipped、mergeable_state=blocked——这是工作流文档化的 label-shadowing 行为,不是代码问题,全绿矩阵证据保留在原始 run 记录中。若分支保护要求视图内有新鲜矩阵,给 PR 打 ci:full 标签即可重触发(或由你直接按已有证据合并)。

判定

可合:代码审查通过(单文件、test-only、双向行为实证)、该 head 的全量 core 矩阵有绿记录。合入后 #5381 保持 open,等下一次 CI 命中——届时日志会自解释分叉侧。#5381 里此前的 A/B 决策请求随本 PR 落地而收敛到 A(hold + 等诊断),无需再拍板 B。

标签:pr + p2 + scope/ci(测试基建可观测性,CI 噪音域)。

@ranxianglei

Copy link
Copy Markdown
Owner Author

(ai: vllm-qwen/qwen3.8-27b)

Review verdict: APPROVE (posted as a comment — the formal reviews API rejects self-approval on this account: Review Can not approve your own pull request; content below is the full independent end-to-end review, diff re-read from scratch, not reusing the initial-pass conclusions from the tracking issue).

What this does. Failure-path self-diagnostics for the #2129 restart-measurement e2e: when any of the four post-restart disk assertions fails, the catch block dumps — each read individually guarded — what landed on disk (stats object, metadata key set + PFA presence), the expected in-memory values, whether the live session metadata had been swapped for a dropped lazy proxy (isDroppedMetadata, src/session.ts:140 — pure WeakSet membership check, cannot trigger the proxy's ensure() traps), and whether a debounced save was still pending (store.hasPending("i2129-e2e"), src/persist.ts:3509; id matches the subtest's actual session header/getSession). Then it rethrows the original assertion error unchanged.

Eight questions.

  1. What problem: the [Bug] issue2129 restart-measurement e2e flakes at a pinned ~3.13s on ubuntu shard 1 — reproduces on master commits that touch only workflow YAML #5379 CI-only flake (PFA-stamp assertion failing at a pinned ~3.13s while the three stats siblings pass) has never reproduced locally, so nobody can see which side diverged.
  2. Root cause: unknown — deliberately out of scope here; the suspected shapes (stale envelope/index projection in the both-dropped build, lazy-metadata drop ordering) remain unproven.
  3. Impact/severity: none on users — test-only observability; the flake itself stays open (CI-only, p2).
  4. How: instrument the failure path only; success path byte-identical (verified by reading pre/post code and running the file).
  5. Regression: none. Success path runs the same four asserts in the same order; failure path now logs six [2129-diag] lines before rethrowing the identical AssertionError (original error preserved — no masking, no swallowing). No product/wire/config/persistence/kernel touch.
  6. Cache: n/a — no cache/retry/cursor/dedupe logic touched.
  7. Design direction: unchanged; explicitly defers the fix decision (retry-shape vs stale-projection vs split-assertion) to the next evidenced occurrence rather than shipping an unproven mask.
  8. New config: none.

Evidence. Local targeted run at e39190e: npm run typecheck clean; the test file alone 4/4 pass (e2e subtest ~3.3s, matching the PR's claim). Full-suite verdict from the mirror check cache (fetched 2026-11-01T03:22Z, sha e39190e): 30 checks, 0 failed, 0 pending (14 success + 16 env-gated skips).

Non-blocking note. Title/body cite #5381 — the mirror number; on GitHub that number belongs to a different item (the flake issue is GitHub #5379). Systemic mirror↔GitHub offset, left to the sync pipeline; flagged for traceability only.

Verdict: mergeable.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

review-worker-r1 终审:APPROVE,可合并。 GitHub 侧已发完整八问评审(PR #5400 评论,因账号自审限制无法走正式 reviews API 的 APPROVE 事件,内容等同)。独立复核结论:diff 仅动 tests/issue2129-restart-measurement-state.test.ts(+23/−5),新增代码全在失败路径 catch 内、rethrow 原断言错误不变,成功路径字节不变;isDroppedMetadata(src/session.ts:140,WeakSet 成员检查,不触发 lazy-proxy ensure() 陷阱)与 store.hasPending("i2129-e2e")(src/persist.ts:3509,会话 id 与测试实际一致)均核对存在且语义正确。本地定向验证:typecheck 干净 + 该文件单跑 4/4 通过(e2e ~3.3s);镜像缓存(2026-11-01T03:22Z,sha e39190e)30 checks 0 failed / 0 pending。无 kernel/触碰,无需授权申请。非阻塞备注:标题/正文引用的是镜像号 #5381,GitHub 上该号指向另一条目(flake 本体是 GitHub #5379),属镜像↔GitHub 号段系统性偏移,留待同步管道处理。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

说明:上面这条是同步管道把我刚发到 GitHub PR #5400 的评审评论回灌到镜像线程的(带 upstream-sync 标记),不是新意见。终审结论不变:APPROVE,可合并——等你在 GitHub 侧合并(https://github.com/ranxianglei/billion-context/pull/5400)。无未决项,本线程无需再回复。

@ranxianglei
ranxianglei force-pushed the 2026-11-01_issue5381-flake-diagnostics branch from e39190e to f3b30a0 Compare November 10, 2026 11:02
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

Merge-conflict resolution complete — PR rebased onto current master and force-pushed.

Verified live this run: PR head was e39190ed7e on branch 2026-11-01_issue5381-flake-diagnostics, base master at 4dea18eabf, mergeable_state: dirty (GitHub API, fetched just now). Rebased the single PR commit onto master; exactly one conflict, in tests/issue2129-restart-measurement-state.test.ts (lines ~329–360 pre-resolution).

Why it conflicted: since the PR branched off (Nov 1), master commit 6d6e29e3b8 (#5403/#5418 fastpath rework) reordered the simulated-restart sequence in this same test — quiesce old store (store.cancelAll()) → reset sessions → swap in reload store → then read the record file off disk ("reading while it can still save races it under CI load"). The PR instead wrapped the four post-restart assertions in a try/catch that dumps on-disk vs expected state on failure (#5381 flake diagnostics). Both sides rewrote the same hunk.

Resolution (union of intent): kept master's new ordering (it kills the exact read-vs-write race the #5381 diagnostics were hunting) and applied the PR's try/catch diagnostic wrapper around the four assertions in their new position; dropped the PR-side duplicate _resetSessionsForTest()/new SessionStore(...)/_setStoreForTest(store2) tail (already executed earlier under master's ordering). Net diff unchanged: 1 file, +23/−5. Both symbols the diagnostics import exist on master (isDroppedMetadata src/session.ts:142, hasPending src/persist.ts:3868).

One observation for the reviewer (left verbatim, not changed): under master's ordering diag("store.hasPending", ...) runs after cancelAll(), which clears saveTimers/writerOffPending, so that line will always print false by construction. Harmless failure-path noise; trim if you want.

Verification (targeted, local): tsc --noEmit clean on the rebased head; the touched test file passes 4/4 including the e2e restart case. Full-suite verdict = remote CI on the pushed head. Push discipline followed: fetched remote head before pushing (unchanged since Nov 1), --force-with-lease against the exact expected sha. Post-push live check: head now f3b30a0a5b, mergeable: true, mergeable_state: blocked = pending checks only, no conflict left. Mirror check-cache still shows the pre-push sha (poller lag); CI on the new head is queued. No gates hit (test-only diff, no kernel/, no config surface) — no decision needed this run.


中文结构化八问

  1. 是什么问题:PR 分支与 master 冲突(merge state: dirty),无法合并——同一测试文件的两处改动撞在同一 hunk。
  2. 什么原因导致的:PR 从 11-01 的 master 切出后,master 上的 6d6e29e3b8(fastpath 重做)重排了该测试的「模拟重启」时序(先静默旧 store、换入重载 store,再读盘),而 PR 把同一段的四条断言包进 try/catch 诊断块——两边都改了同一区域。
  3. 影响范围与严重程度:仅阻塞本 PR 合并;不影响任何运行时行为(纯测试文件)。
  4. 怎么修复的:rebase 到当前 master(4dea18e),按「意图并集」解冲突——保留 master 的新时序(它消灭的正是 [PR] docs: README FAQ entry for the qoder-trae-workbuddy-switch quota gateway combo #5381 诊断要抓的读盘竞态),把 PR 的诊断 try/catch 原样套在新位置的断言上,删掉 PR 侧已重复的 reset/store 换入三行;净 diff 与原来一致(+23/−5)。
  5. 是否有回归问题:无。运行时零改动;本地 typecheck 干净 + 该测试文件 4/4 通过(含 e2e 重启用例);高危模式逐项过——无语义漂移、无阈值/时序/输出格式变化。
  6. 是否破坏缓存:不涉及(不触碰缓存/重试/游标/去重逻辑)。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

判定:冲突已解除,分支就绪,等新 head(f3b30a0)的 CI 结果即可进入正常合并流程。

@ranxianglei

Copy link
Copy Markdown
Owner Author

(ai: vllm-qwen/qwen3.8-27b)

Round-2 review verdict: APPROVE (posted as a comment — the formal reviews API permanently rejects self-approval on this account, same constraint as round 1).

What changed since my round-1 pass. Head moved e39190e -> f3b30a0: a pure rebase onto advanced master (base 7578e81 -> 4dea18e). Added/removed lines are byte-identical between the two heads — zero content change, context shifts only (master touched the same test file region and src/persist.ts heavily during the gap). I re-verified every symbol the diagnostics depend on against the NEW base tree rather than trusting the old pass: isDroppedMetadata (src/session.ts:142, unchanged WeakSet-membership semantics, still cannot trigger lazy-proxy traps), hasPending(id) (src/persist.ts:3868, same signature), session id i2129-e2e still matches header/getSession, and the four wrapped asserts are intact at their shifted location.

Small finding fixed on the spot. 已顺手修:诊断注释里的引用号是镜像号 #5381(GitHub 上该号指向另一条目,flake 本体是 GitHub #5379)——注释引用改为 #5379,+ commit 69fc50f。Title/body still carry the mirror number — those are system-owned surfaces I'm not permitted to rewrite; noted as fact, no action asked.

Evidence. Mirror check cache at f3b30a0 (fetched 2026-11-10T11:41:34Z): 30 checks, 0 failed, 0 pending (18 success + 12 env-gated skips). Local at f3b30a0 plus my one-token fix (head 69fc50f): npm run typecheck clean; targeted run of tests/issue2129-restart-measurement-state.test.ts alone 4/4 pass (~5.2s). My fix commit flips exactly one comment character-class (issue number) — zero runtime surface — so its CI rerun cannot change behavior; the behaviorally-relevant head f3b30a0 is the one verified green above.

八问(中文)。

  1. 是什么问题:[Bug] issue2129 restart-measurement e2e flakes at a pinned ~3.13s on ubuntu shard 1 — reproduces on master commits that touch only workflow YAML #5379 的 CI-only flake(Restart loses measurement state: buildSession() omits stats fields buildRecord() persists (k̂, usage anchor, baseline origin, localInputEstimate) #2129 重启测量 e2e 的 PFA-stamp 断言在钉死的 ~3.13s 失败、前三条 stats 断言通过)本地从未复现;本 PR 不改 flake,只给失败路径加自诊断。
  2. 什么原因导致的:flake 根因未知且刻意不在本 PR 范围(嫌疑形状——both-dropped 构建的 stale envelope/index 投影、lazy-metadata 丢弃时序——均未证实);本 PR 解决的是"失败时看不到哪一侧分叉"的观测缺失。
  3. 影响范围与严重程度:对用户零影响——纯测试文件 +23/−5(外加我的一行注释修正),成功路径字节不变;仅 CI 日志面变化。
  4. 怎么修复的:四条断言包进 try/catch,失败时逐条带守卫地 dump 落盘 stats / metadata key 集+PFA / 内存期望值 / isDroppedMetadata(s.metadata) / store.hasPending("i2129-e2e"),然后原样 rethrow 原 AssertionError。
  5. 是否有回归问题:无。成功路径同序同值跑原四条断言;失败路径多打六行 [2129-diag] 后抛出同一错误对象(无吞错、无掩蔽);不触碰 product/wire/config/persistence/kernel,无高危模式。本轮额外核对了 rebase 后 master 侧对 src/persist.ts/src/session.ts 的大量改动没有漂移本 PR 依赖的任何符号语义。
  6. 是否破坏缓存:不涉及——未动任何缓存/重试/游标/去重逻辑。
  7. 是否改了设计方向:未改;修复决策(retry-shape vs stale-projection vs 拆断言)明确留给下一次有证据的出现,拒绝先上未证实的 mask。
  8. 是否引入了新的配置:无。

Verdict: mergeable. Final head after my fix: 69fc50f.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

Round-2 复审:APPROVE,可合并。 新 head f3b30a0 = 已审内容的纯 rebase(加删行与 e39190e 逐字节一致;依赖符号在新 base 全部复核无漂移),CI 绿(镜像缓存 11:41Z:30 checks / 0 failed / 0 pending)。已顺手修一处小问题并推回分支:诊断注释引用号从镜像号 #5381 改为 GitHub 号 #5379(commit 69fc50f,最终 head 即此);本地定向验证 4/4 + typecheck 干净。GitHub 侧完整八问评审见 PR #5400 评论(自审限制仍走评论通道)。等你在 GitHub 合并。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

说明:这条是同步管道把我刚发到 GitHub PR #5400 的 round-2 评审回灌到镜像线程的(带 upstream-sync 标记),不是新意见。结论不变:APPROVE,可合并(最终 head 69fc50f,等你在 GitHub 侧合并)。无未决项。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

Round-2 review closed — verdict: APPROVE, mergeable. Final head 69fc50f0.

What I did this round:

八问(中文):

  1. 是什么问题:[Bug] issue2129 restart-measurement e2e flakes at a pinned ~3.13s on ubuntu shard 1 — reproduces on master commits that touch only workflow YAML #5379 的 CI-only flake(Restart loses measurement state: buildSession() omits stats fields buildRecord() persists (k̂, usage anchor, baseline origin, localInputEstimate) #2129 重启测量 e2e 的 PFA-stamp 断言在钉死的 ~3.13s 失败、前三条 stats 断言通过)本地从未复现;本 PR 不改 flake,只给失败路径加自诊断。
  2. 什么原因导致的:flake 根因未知且刻意不在本 PR 范围(嫌疑形状——both-dropped 构建的 stale envelope/index 投影、lazy-metadata 丢弃时序——均未证实);本 PR 解决的是"失败时看不到哪一侧分叉"的观测缺失。
  3. 影响范围与严重程度:对用户零影响——纯测试文件 +23/−5(外加一行注释修正),成功路径字节不变;仅 CI 日志面变化。
  4. 怎么修复的:四条断言包进 try/catch,失败时逐条带守卫地 dump 落盘 stats / metadata key 集+PFA / 内存期望值 / isDroppedMetadata(s.metadata) / store.hasPending("i2129-e2e"),然后原样 rethrow 原 AssertionError。
  5. 是否有回归问题:无。成功路径同序同值跑原四条断言;失败路径多打六行 [2129-diag] 后抛出同一错误对象(无吞错、无掩蔽);不触碰 product/wire/config/persistence/kernel;本轮额外核对了 rebase 后 master 侧对 src/persist.ts/src/session.ts 的大量改动没有漂移本 PR 依赖的任何符号语义。
  6. 是否破坏缓存:不涉及——未动任何缓存/重试/游标/去重逻辑。
  7. 是否改了设计方向:未改;修复决策(retry-shape vs stale-projection vs 拆断言)明确留给下一次有证据的出现,拒绝先上未证实的 mask。
  8. 是否引入了新的配置:无。

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant