Skip to content

[PR] fix(overflow): bili's own oversize-gate refusal now triggers self-heal instead of a permanent 400 loop (#6122) - #6546

Open
ranxianglei wants to merge 19 commits into
masterfrom
2026-11-13_6122-gate-overflow-recognition
Open

ranxianglei wants to merge 19 commits into
masterfrom
2026-11-13_6122-gate-overflow-recognition

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

What was broken

In the #6122 field case (DSH fork child inheriting a 1.13M-token parent history), one of the rejection bodies flying through the relay was bili's own oversize-gate refusal — code: bili_gate_oversize_refusal_unattributed from src/agent/native-intercept.ts. The relay treated it as an opaque gateway rewrite: no parseable overflow marker, the #4452 CANNOT engage warning printed on bili's own body, and the overflow self-heal (same-request refold / emergency shrink) never armed. Every retry re-sent the identical oversized history → 400 forever.

This is not fork-specific: any oversized request that gets refused by the gate and surfaces back at the relay (fork children, oversized tool outputs, giant pastes) hits the same dead end.

Changes

  1. src/util.ts — CONTEXT_OVERFLOW_PATTERNS gains /bili_gate_oversize_refusal/i, covering all five oversize variants (bare / _fallback / _no_live_state / _unattributed / _unattributed_bytes). The compaction refusal (bili_gate_dsh_compaction_refused) is a different class — no proxy available, body fits — and is deliberately not matched. No window number is learnable from the refusal prose ("1,000,000 tokens" is a pricing floor, not the model's window), so the no-window arm engages: min(judging window, payload estimate), then preflight folds under it and re-sends (使用pi 压缩后就提示prompt is too long #1195 same-request refold where available).

  2. src/fork-adoption.ts — the [resume-inheritance] … inherited … line now names the straddling active blocks when non-zero (, N active block(s) straddled the fork point (adopted once fully present, #6122)). The field triage had to rebuild the ledger by hand to see that 4 of 5 active blocks were inherited only implicitly; the lineage-reconcile line already reported this (fix(sessions): wire-neutral affinity hashing for cross-protocol continuity (#2791) #2807). Byte-identical when nothing straddles.

Scope note

This is the F4 + F5 slice of #6122. F2/F3 (straddled-block coverage gate, parent-missing prefix fallback) are intentionally deferred: the DSH fork path (server side) is still in development, so those design decisions should be made against its final shape, not the current one.

Pre-flight

…s a parseable overflow marker (#6122)

The relay saw bili's OWN oversize refusal (bili_gate_oversize_refusal_*)
come back through the client as a 400 and logged the #4452
CANNOT-engage warning on its own body — self-heal (same-request refold /
emergency shrink) never armed and the oversized history was re-sent
every turn (400 forever). The gate code is bili's definitive
over-window proof: it now reads as an overflow with no learnable
window (the 1,000,000 in the prose is a pricing floor), so the
no-window arm (min of judging window and payload estimate) engages.
The compaction refusal is a different class and stays dark.
…ine (#6122)

A fork that inherited 1 of 5 active blocks logged only the adopted
count — the 4 straddling blocks were invisible, forcing manual ledger
archaeology in field triage. The inherited line now appends the
straddled count when non-zero (the lineage-reconcile line already
does, #2807); byte-identical when nothing straddles.
@ranxianglei ranxianglei changed the title fix(overflow): bili's own oversize-gate refusal now triggers self-heal instead of a permanent 400 loop (#6122) [PR] fix(overflow): bili's own oversize-gate refusal now triggers self-heal instead of a permanent 400 loop (#6122) Nov 13, 2026
@github-actions

github-actions Bot commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-11-13_6122-gate-overflow-recognition (6b03788)

Latest build — install this exact version

npm install -g billion-context@0.0.9947-pr.6546.f628cc9

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
6b03788 billion-context@0.0.9947-pr.6546.f628cc9 build
a570049 billion-context@0.0.9852-pr.6546.86fe7fc build
0a565fa billion-context@0.0.9687-pr.6546.decc26b build
cfa8ee4 billion-context@0.0.9644-pr.6546.ceeef94 build
e072f19 billion-context@0.0.9610-pr.6546.448a12e build
518b175 billion-context@0.0.9584-pr.6546.7382d45 build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr6546.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei ranxianglei added bug Something isn't working p1 scope/wiring labels Nov 13, 2026
ranxianglei and others added 3 commits November 14, 2026 02:08
…e default (#6480)

The suite-wide baseline (tests/preload-storage-baseline.mjs) pins
BILI_STORAGE_UNIFIED ??= "legacy" for the legacy-written corpus, but this
suite asserts the fresh-install default posture — the exact #6480 pure-sqlite
flip the storage-sqlite-default notice advertises. Under the CI preload chain
the pin hid the notice and two subtests failed deterministically (red in
every shard leg containing the file). Drop the inherited pin so the
production default applies, per the preload's own pin-yourself contract.
The suite baseline preload (preload-storage-baseline.mjs) pins
BILI_STORAGE_UNIFIED="legacy" for the broad corpus. After #6480 flipped the
storage default to pure-sqlite, the notice-registry suite's visibleNotices
assertions expect the pure-sqlite default notice, which is gated on
unifiedSelectionIsSqlite (unified "sqlite" or absent) — the preload's
"legacy" made the notice silent, so the suite was born-red on master. Pin
the selection explicitly (#4011 lesson: a test that depends on the production
default passes locally but fails in CI).
…-fixes form

The branch's own duplicate fix (cfa8ee4: delete BILI_STORAGE_UNIFIED so
the production default applies) and the sentinel's ci-fixes fix (c6369fb:
pin BILI_STORAGE_UNIFIED="sqlite") both resolve the #6480 visibleNotices
born-red and both satisfy unifiedSelectionIsSqlite. Per the CI coordination
policy (2026-10-24) master-level fixes converge on ci-fixes, so keep the
ci-fixes form and drop the branch-side duplicate — this also removes the
future merge conflict with the ci-fixes->master PR.

Verified: tests/notice-registry.test.ts 5/5 with the CI preload chain and
5/5 in a clean env.
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

CI fix run (attempt 1/3) — both reds diagnosed and resolved; branch pushed to 0a565facbd, new CI run queued.

Live re-verification first. The mirror cache was stale, so I did a single check-runs fetch for the old head cfa8ee4e0f: exactly 2 failures — test (windows-latest, 22, 1) and test (windows-latest, 22, 4); all other shards and every static gate green (ubuntu all green, win22-2/3/5 green).

Diagnosis (logs fetched for both jobs): every failure is EBUSY: resource busy or locked, unlink '…\bili-*\index.db' in the teardown path (tests/tmp-rm.ts:30 defaultRm → rmrf) — the test bodies themselves pass. win22-4: 3× tests/persist-metadata-fold-gate.test.ts (#5506 subtests); win22-1: 10× tests/persist-zstd.test.ts. The runner's own triage classifies this as the #2799/#2908/#4347 family — transient OS lock (AV/indexer) in teardown, "never an assertion regression". Zero intersection with this PR's diff (6 files: src/util.ts, src/fork-adoption.ts, 3 test files — none touch storage teardown). It is inherited: master's own run 43469650189 shows the identical signature (persist-metadata-fold-gate #5506 ×3 + persist-zstd ×10), and the #4599 sentinel (floor 1324) already ruled it: flake, rerun goes green, no ci-fixes change needed.

What I did:

  1. Merged origin/ci-fixes into the PR branch (CI coordination policy 2026-10-24: inherited red → merge the sentinel branch, don't re-diagnose).
  2. One conflict in tests/notice-registry.test.ts: the branch's own duplicate fix (cfa8ee4e0f: delete process.env.BILI_STORAGE_UNIFIED) vs the sentinel's ci-fixes fix (c6369fb6a1: process.env.BILI_STORAGE_UNIFIED = "sqlite"). Both satisfy unifiedSelectionIsSqlite (src/knobs.ts:1038 — the gate is the selection, not the resolved mode), so the swap is behavior-neutral. Resolved by keeping the ci-fixes form and dropping the branch-side duplicate — master-level fixes converge on ci-fixes, and this removes the future merge conflict with the ci-fixes→master PR.
  3. Targeted verification: tests/notice-registry.test.ts 5/5 with the CI preload chain (the regime that was born-red) and 5/5 clean env.
  4. Pushed cfa8ee4e0f..0a565facbd (fast-forward; remote head verified unchanged before push). Net diff old→new head = the one hunk above (5 ins/7 del).

Status now: new CI run 43508987326 on head 0a565facbdc3 is queued, waiting for shared runner slots (verified via single API check; mirror cache intermittently stale). The flaky shards will draw fresh runners. If the EBUSY flake recurs there, the remedy per #4599 sentinel is rerun-failed-jobs (single write) — no code change.


八问(中文结构化结论):

  1. 是什么问题:PR 分支 CI 两个 Windows 测试分片(22-1、22-4)红——不是断言失败,是测试收尾阶段删临时 SQLite 文件时被操作系统锁住(EBUSY unlink index.db),测试本体全部通过。
  2. 什么原因导致的:Windows runner 上的瞬时 OS 锁(AV/索引器占用临时文件),属 Windows CI: storage-engine-proofs 套件清理 temp 目录遇 EBUSY/EPERM,windows test leg 随机红 #2799/test(windows): driveWire golden-harness 清理期 index.db EBUSY 复发(#2799 族,#2808 后残留) #2908/windows-22-1 lane randomly red: during test cleanup, unlink index.db hit an EBUSY external lock, couldn't survive tmp-rm | windows-22-1 车道随机红:测试清理期 unlink index.db 遇 EBUSY 外部锁,熬不过 tmp-rm 的 9.6 秒重试梯子(#2799/#2908 族复发) #4347 已登记的 flake 家族;master 自身 run 43469650189 出现完全相同签名,是继承性失败,与本 PR 的 6 个改动文件零交集。
  3. 影响范围与严重程度:仅 CI 红,不阻塞代码正确性;PR 代码本身在全部其他分片(ubuntu 全绿、win22-2/3/5 绿)与全部静态门禁上通过。
  4. 怎么修复的:按 CI 协调政策(2026-10-24)merge 哨兵分支 ci-fixes 进 PR 分支;唯一冲突(notice-registry 存储 pin,分支与 ci-fixes 各有一份等价修复)收敛到 ci-fixes 形式,行为中性(两者都满足 unifiedSelectionIsSqlite 门);本地定向验证 5/5(CI preload 链 + 干净环境双通过);推送 0a565fa。
  5. 是否有回归问题:无。净 diff 仅 tests/notice-registry.test.ts 一个 hunk(5 增 7 删),把分支自带的重复修复换成 ci-fixes 的等价形式,不触碰任何 src/ 行为;既有行为(通知可见性判定)两种形式下判决一致,已双环境实测钉住。
  6. 是否破坏缓存:不涉及(纯测试环境 pin,与运行时缓存/重试/游标无关)。
  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:两个红均为继承性 Windows teardown flake(非本 PR 引入),已按政策 merge ci-fixes 并推送;新 run 排队中,若 flake 在新 runner 上复现,重跑失败 job 即绿(哨兵 #4599 floor 1324 既定处置)。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

CI fix run (2nd wake) — same 2 reds re-verified, confirmed as the known master-level Windows EBUSY family; rerun triggered, then the run was externally cancelled; stopping here per fleet coordination.

Live re-verification. Mirror cache was stale (frozen at 22:22Z), so a single check-runs fetch for head 0a565facbd confirmed exactly the 2 reported failures — test (windows-latest, 22, 1) and test (windows-latest, 22, 4) — with 35 checks green and 13 skipped.

Log analysis (both jobs fetched): identical signature to the previous run — every failure is EBUSY: resource busy or locked, unlink '…\bili-*\index.db' in the teardown path (tests/tmp-rm.ts:30 defaultRm → rmrf). win22-1: 10× tests/persist-zstd.test.ts (subtests 1457–1466); win22-4: 3× tests/persist-metadata-fold-gate.test.ts #5506 (subtests 1437–1439). Test bodies pass (job1: 2228 pass / 10 fail across 245 files). The in-run full retry on the same runner reproduced 10/10, and the runner's own verdict: "either a persistent external lock (AV/indexer) on this runner or a DETERMINISTIC handle leak in the code; rerunning the job draws a fresh runner, and if the lock reproduces there too the leaked-handle hypothesis is primary (#4347)". Zero intersection with this PR's diff.

Fleet context (#4599, floors 1325–1344): this is no longer a one-PR flake — 4+ PRs (#6546, #4883, #6301, #6551, #5855, #4629) now share the identical signature, and a new mechanism claim is under the sentinel's verification (floor 1327: deterministic on Windows since 5a80e2b8ea's pure-SQLite default flip — the test opens index.db eagerly and never closes the store, so the process's own SQLite handle locks the file and the ~9.4s retry ladder in tests/tmp-rm.ts cannot win). Sentinel floor 1343: the Windows EBUSY face "remains a pending ci-fixes convergence item… I'll confirm flake-vs-real before adding any ci-fixes entry".

What I did this run:

  1. Triggered rerun-failed-jobs on run 43508987326 (HTTP 201) — the established first step for this family; it draws fresh runners, which is exactly the flake-vs-leak discriminator the runner verdict calls for.
  2. The rerun started (win22-4 ran ~11.5 min, win22-1 15 s), but the entire run was cancelled at 00:07:58Z by an external actor — actor not identifiable from my vantage: no new push on the branch (head still 0a565facbd), no movement on ci-fixes (c6369fb6a1) or master (e1f31ad9d5), no owner comment on this PR.
  3. Did NOT re-trigger: an external cancellation during a fleet-wide red storm (multiple concurrent reruns in flight per CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 floors 1341/1342, shared runners strained) is a signal to stop, not to fight for the branch/run; and the durable fix is the ci-fixes convergence the sentinel is tracking, not more reruns. Logged the cancellation in CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 for the sentinel's flake-vs-real evidence.

State of the PR: all 35 non-Windows-EBUSY checks green (all ubuntu shards, win22-2/3/5, every static gate, all e2e lanes, build artifact). The only remaining red is the 2 inherited Windows EBUSY shards — a master-level issue owned by the #4599 sentinel. Once the ci-fixes convergence lands, merging it into this branch (or a re-trigger) will close the PR. No code on this branch needs any further change from my side.


八问(中文结构化结论):

  1. 是什么问题:PR 分支 CI 两个 Windows 测试分片(22-1、22-4)再次红——与上一轮完全相同:测试收尾删临时 SQLite 文件被 OS 锁住(EBUSY unlink index.db),测试本体全部通过,非断言失败。
  2. 什么原因导致的:master 级继承问题(windows-22-1 lane randomly red: during test cleanup, unlink index.db hit an EBUSY external lock, couldn't survive tmp-rm | windows-22-1 车道随机红:测试清理期 unlink index.db 遇 EBUSY 外部锁,熬不过 tmp-rm 的 9.6 秒重试梯子(#2799/#2908 族复发) #4347 家族)。CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 已有 4+ 个 PR 同签名;新机制假说(5a80e2b 纯 SQLite 默认翻转后,测试急开 index.db 且从不在 finally 关 store,进程自身的 SQLite 句柄在 Windows 上锁死文件,~9.4s 重试阶梯必输)正由哨兵做 flake-vs-real 判定,尚未落 ci-fixes。
  3. 影响范围与严重程度:仅 CI 红,波及当前全部在途 PR 的 Windows 分片(红风暴),不阻塞本 PR 代码正确性——其余 35 项检查全绿。
  4. 怎么修复的:本轮无 PR 分支代码可修(修复属 master 级,政策规定收敛到 ci-fixes 由哨兵统一维护)。已触发 rerun-failed-jobs 抽新 runner 做 flake-vs-leak 判别;rerun 启动后整个 run 被外部取消(取消方未知:分支无新 push、ci-fixes/master 未动、PR 无 owner 指示),未再重触发,已在 CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 留档。
  5. 是否有回归问题:无。本轮零代码改动;上一轮的唯一改动(notice-registry 存储 pin 收敛)已双环境 5/5 钉住,且全部相关检查绿。
  6. 是否破坏缓存:不涉及。
  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:本 PR 侧工作已完成(代码 + ci-fixes 收敛 + 验证全绿);剩余 2 红为 #4599 哨兵在管的 master 级 Windows EBUSY 问题,等其 ci-fixes 收敛落地后随重触发即绿。若 owner 刚才的取消是有意为之(例如准备直接落 ci-fixes 修复),本分支无需任何进一步动作。

…test

The dsh plugin lane (ci-e2e-dsh.yml) runs node --import tsx --test
WITHOUT the preload-storage-baseline.mjs preload, so BILI_STORAGE_UNIFIED
is unset and the storage mode defaults to 'full'. Under the pure-sqlite
default flip (5a80e2b, #6480), 'full' mode turns the JSON envelope
writer OFF — so the .content-store.json sidecar is never written, and
the test's readdirSync(namespace) hits ENOENT.

Pin BILI_STORAGE_JSON_WRITER=on in the test so the content-store
sidecar is written regardless of the storage mode. This keeps the
corrupted-store detection path exercised in both full and legacy modes.

Closes: dsh plugin lane red on master (run 43532980164, job 132255459726)
@ranxianglei

Copy link
Copy Markdown
Owner Author

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

APPROVE — F4+F5 slice of #6122 is correct, well-tested, and green.

What I verified

Correctness (F5 — the main fix). src/util.ts:414 adds /bili_gate_oversize_refusal/i to CONTEXT_OVERFLOW_PATTERNS. I traced the full path:

Correctness (F4 — the log line). src/fork-adoption.ts:536 adds a straddled-block count to the [resume-inheritance] line, gated on plan.straddled > 0. Additive; the pre-#6122 wording is preserved when zero (the test's negative case pins this).

Tests. All three suites are well-written and would catch regressions:

Regressions. None. The overflow-pattern addition is purely additive (one new entry at the end of the array). The fork-adoption log line is additive (new clause only when straddled > 0). The compaction refusal stays dark. No existing behavior is broken.

CI. Green on the pushed head (0a565fa): 25 success, 3 cancelled (normal meta-cancels), 2 skipped (gated e2e lanes — expected). No failures.

House conventions. No kernel/ changes (the kernel HARD RULE does not apply). No config-surface changes. No new env vars. Code quality is clean; the new comment in src/util.ts:401-413 is thorough and explains the rationale (why the floor is not a window, why the compaction refusal stays dark).

One minor scope note (non-blocking)

The notice-registry test-isolation pin (tests/notice-registry.test.ts:24, commits cfa8ee4 + c6369fb, converged to the ci-fixes form in 0a565fa) is a drive-by fix for #6480 (pure-sqlite default), not #6122. It's a legitimate test-isolation fix (the comment at lines 20-23 explains why the suite baseline preload's "legacy" pin breaks the pure-sqlite notice assertion), it's in separate commits, and it's clearly labeled. Not a blocker — just flagging that it's a different issue's scope riding along. If the owner wants strict one-issue-one-scope, it could be split, but it's not worth blocking this PR on.


8问审核清单

  1. 是什么问题:bili 自己的 oversize-gate 拒绝(bili_gate_oversize_refusal*)被 relay 当成不透明的网关改写——没有可解析的 overflow 标记,Upstream 400 bodies rewritten to a generic error by an intermediate gateway starve overflow calibration — repeated 400s persist on those routes (v0.1.203+) #4452 CANNOT-engage 警告打在 bili 自己的 body 上,overflow self-heal(同请求 refold / emergency shrink)从未武装。每次重试都重发同样的超大历史 → 永久 400 循环。
  2. 什么原因导致的:CONTEXT_OVERFLOW_PATTERNS 里没有 bili 自己 gate 拒绝码的模式。gate 拒绝体(422 生成,经客户端回流为 400)回到 relay 时,inspectContextOverflow 判 isOverflow=false,isMarkerlessOverflowReject 判 true(400 且无标记)→ 走 Upstream 400 bodies rewritten to a generic error by an intermediate gateway starve overflow calibration — repeated 400s persist on those routes (v0.1.203+) #4452 警告分支,self-heal 不武装。根因是 bili 自己的错误码没被自己的 overflow 识别器认出来。
  3. 影响范围与严重程度:任何被 gate 拒绝的超大请求(不限于 DSH fork 子进程)在回流到 relay 时都命中同一个死路。严重程度:完全无法使用(特定条件下——请求超大且被 gate 拒绝后回流)——会话永久 400,用户必须手动干预。
  4. 怎么修复的:在 CONTEXT_OVERFLOW_PATTERNS 末尾加一条 /bili_gate_oversize_refusal/i,覆盖全部 5 个 oversize 变体;compaction refusal(不同类)刻意不匹配。无窗口号可从 prose 学习("1,000,000 tokens" 是定价下限不是模型窗口),所以 no-window arm 武装(judging window 与 payload 估计的 min),然后 preflight 在其下 fold 并重发(使用pi 压缩后就提示prompt is too long #1195 同请求 refold)。F4 顺带在 resume-inheritance 日志行里命名 straddled 的活跃块(plan.straddled > 0 时)。
  5. 是否有回归问题:无。overflow 模式是纯增量(数组末尾加一条);fork-adoption 日志行是增量(straddled > 0 时才加新子句,零时保持 pre-[Bug][dsh] DSH fork() omits parentSession, so the child silently inherits 1 of 33 compression blocks and becomes permanently unusable (400) | DSH「开启分支对话」漏写 parentSession,子会话静默只继承 1/33 压缩块后永久 400 #6122 措辞——有负例测试钉住);compaction refusal 保持暗(有负例测试钉住)。高危模式逐项过:语义漂移(无——只加新模式不改旧模式)、fail-fast 变吞错(无——self-heal 从「不武装」变「武装」,是修复不是吞错)、阈值/时序/输出格式变化(无——日志行只在 straddled > 0 时多一个子句)。
  6. 是否破坏缓存:不涉及。overflow 识别与缓存无关;self-heal 的 refold 走的是既有的 使用pi 压缩后就提示prompt is too long #1195 路径,不改变缓存/重试/增量游标/去重键策略。
  7. 是否改了设计方向:不涉及。这是 F4+F5 切片([Bug][dsh] DSH fork() omits parentSession, so the child silently inherits 1 of 33 compression blocks and becomes permanently unusable (400) | DSH「开启分支对话」漏写 parentSession,子会话静默只继承 1/33 压缩块后永久 400 #6122 的 5 个切片中的 2 个);F2/F3(straddled-block coverage gate、parent-missing prefix fallback)刻意推迟(DSH fork 路径仍在开发中)。架构/边界/数据流没有转向。
  8. 是否引入了新的配置:不涉及。没有新 env/config/默认值。

判定:可合。 修复正确、测试充分、CI 绿、无回归、无设计方向变更、无新配置。notice-registry 的 drive-by 测试隔离修复是唯一的范围瑕疵,非阻塞。

ework-agent and others added 5 commits November 14, 2026 11:21
…er-r1

# Conflicts:
#	src/util.ts
#	tests/issue4452-markerless-reject.test.ts
The pure-sqlite default flip (5a80e2b, #6480) turns the JSON envelope
writer OFF, so the .json session file this lane asserts on is never
written and 'proxy must persist the session' / 'no persisted session
carried compressed blocks' fail after the 20s wait. Pin the writer ON in
the spawned client's env (the bootstrapped proxy inherits it) so the
envelope this test was written for is still produced.

Closes: native opencode lane red on master (run 43547593803, job 132302438610)
Same pure-sqlite default flip (5a80e2b, #6480) as the opencode lane:
the JSON envelope writer is OFF, so 'proxy must persist the session
within 20000ms' fails — the on-disk .json session file is never written.
Pin the writer ON in the spawned dsh client's env (the bootstrapped
proxy inherits it).

Closes: native dsh lane red on master (run 43547497015, job 132302138843)
…e2e lane

Same pure-sqlite default flip (5a80e2b, #6480) as the opencode/dsh
lanes: the JSON envelope writer is OFF, so 'session persisted' fails —
the on-disk .json session file is never written. Pin the writer ON in
the spawned pi client's env (the bootstrapped proxy inherits it).

Closes: pi codex-responses WS lane red on master (run 43547626291, job 132302540960)
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Merge conflict resolved — pushed a570049c7b (fast-forward, no force-push).

Master moved e1f31ad9d5 → e5e1647ba0 while this PR was in flight. I merged upstream/master into the PR branch and resolved every conflict:

  1. src/util.ts — both sides appended a new pattern to the end of CONTEXT_OVERFLOW_PATTERNS: master's DashScope range of input length should be [ ([BUG] DashScope compatible-mode 400 "Range of input length should be [1, N]" is not recognized as an overflow → the #2313 probe path livelocks #4262) and this PR's bili_gate_oversize_refusal ([Bug][dsh] DSH fork() omits parentSession, so the child silently inherits 1 of 33 compression blocks and becomes permanently unusable (400) | DSH「开启分支对话」漏写 parentSession,子会话静默只继承 1/33 压缩块后永久 400 #6122). Resolved as a union — both entries kept (master's first, this PR's after). No semantic interaction: the two patterns match disjoint bodies.
  2. tests/issue4452-markerless-reject.test.ts — master's 518f0cecd7 (Upstream 400 bodies rewritten to a generic error by an intermediate gateway starve overflow calibration — repeated 400s persist on those routes (v0.1.203+) #4452 follow-up) rewrote the suite to 413-only markerless semantics, colliding with this PR's added GATE_REJECTION_BODY + [Bug][dsh] DSH fork() omits parentSession, so the child silently inherits 1 of 33 compression blocks and becomes permanently unusable (400) | DSH「开启分支对话」漏写 parentSession,子会话静默只继承 1/33 压缩块后永久 400 #6122 test. Resolved by keeping master's rewrite and adding this PR's constant and test on top.
  3. tests/notice-registry.test.ts — auto-merge silently stacked BOTH pin blocks (this branch's BILI_STORAGE_UNIFIED = "sqlite" from the earlier ci-fixes convergence, and master's own 369be2dab7 delete form of the same [PR] storage: pure-SQLite default — JSON envelopes stop being written unless dual-write is requested (#6479) #6480 fix). Resolved to master's form only; the file is now byte-identical to master.

Semantic compatibility check (the one that mattered): master's 518f0cecd7 changed isMarkerlessOverflowReject from (400 || 413) to 413-only. This PR's fix is unaffected by that change — it works through the overflow path, not the markerless path: inspectContextOverflow(400, gateBody) still returns isOverflow: true (400 is in the plausible set, the pattern matches), so self-heal arms exactly as approved. The markerless predicate is only consulted when isOverflow is false. All of this PR's test assertions pass under the new semantics (verified, not assumed).

Local verification on the merged tree: targeted suites 58/58 green under the exact CI preload chain (notice-registry, context-overflow, issue4452-markerless-reject, issue6122-fork-straddled-log) + notice-registry 5/5 in clean env; npm run typecheck clean; kernel 0.0.197 + pi-subagents 0.2.4 rebuilt.

Live state after push: PR mergeable: true, mergeable_state: blocked (conflict cleared — now just waiting on checks + approval). New CI run 43554861708 queued on a570049c7b. Note: your APPROVE was recorded on the old head 0a565facbd; depending on branch protection it may need a re-approval on the new merge commit.

Two environment notes (not PR issues): the local Gitea mirror's git remote started redirecting to login this run, so fetch/push went directly to GitHub; the mirror's checks cache is also stale (fetched 02:47Z) — CI state above was confirmed via single direct API fetches.


八问

  1. 是什么问题:PR 分支与 master 冲突(dirty)——master 前进了约 20 个提交(含 [PR] fix: watermark fastpath (plugin full-resend watermark fastpath) silently disabled by input release mutation | fix: 水印快路径(plugin full-resend watermark fastpath)被输入释放改动静默禁用 #6640/[PR] release gate: approval-pending PR runs no longer kill the release CI gate 26 seconds in #6568/[BUG] DashScope compatible-mode 400 "Range of input length should be [1, N]" is not recognized as an overflow → the #2313 probe path livelocks #4262/Upstream 400 bodies rewritten to a generic error by an intermediate gateway starve overflow calibration — repeated 400s persist on those routes (v0.1.203+) #4452 follow-up 等),与 PR 触碰的 6 个文件中有 3 个发生文本冲突。
  2. 什么原因导致的:PR 在途期间 master 合入了多批修复;[BUG] DashScope compatible-mode 400 "Range of input length should be [1, N]" is not recognized as an overflow → the #2313 probe path livelocks #4262(DashScope 溢出模式)与本 PR([Bug][dsh] DSH fork() omits parentSession, so the child silently inherits 1 of 33 compression blocks and becomes permanently unusable (400) | DSH「开启分支对话」漏写 parentSession,子会话静默只继承 1/33 压缩块后永久 400 #6122 gate 模式)在 CONTEXT_OVERFLOW_PATTERNS 数组同一位置各加一条,Upstream 400 bodies rewritten to a generic error by an intermediate gateway starve overflow calibration — repeated 400s persist on those routes (v0.1.203+) #4452 follow-up 重写了 4452 测试套件,与 PR 新增的测试同区冲突。
  3. 影响范围与严重程度:仅 PR 分支与基线的可合并性;冲突未解决前 PR 无法合并。无用户可见影响,无行为变更。
  4. 怎么修复的:merge master 进 PR 分支(非 rebase,无需 force-push);三处冲突按「并集/取 master 权威形态」逐一解决——模式数组保留双方两条;4452 测试保留 master 的 413-only 重写 + PR 的 [Bug][dsh] DSH fork() omits parentSession, so the child silently inherits 1 of 33 compression blocks and becomes permanently unusable (400) | DSH「开启分支对话」漏写 parentSession,子会话静默只继承 1/33 压缩块后永久 400 #6122 测试;notice-registry 收敛为 master 的 delete 形态(文件与 master 逐字节一致)。
  5. 是否有回归问题:无。关键语义核查:master 把 isMarkerlessOverflowReject 从 (400||413) 收紧为仅 413,本 PR 的修复走的是 overflow 路径(inspectContextOverflow(400, gateBody) → isOverflow: true → 自愈武装),不经过 markerless 谓词,行为不变;PR 全部测试断言在新语义下逐一验证通过。定向套件 58/58 + 干净环境 5/5 + typecheck 全绿。
  6. 是否破坏缓存:不涉及(纯测试/模式表合并,无缓存、游标、去重键改动)。
  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:冲突已全部解决并推送(0a565facbd..a570049c7b),PR 恢复可合并状态;等待新 CI 运行(43554861708)转绿及(如需)owner 对新 head 的重新批准。

ework-agent added 5 commits November 14, 2026 13:18
…lane

4th lane of the pure-sqlite default flip (5a80e2b, #6480) family: the
JSON envelope writer is OFF, so 'proxy must persist the session' and
'at least one persisted session must carry compressed blocks' fail — the
on-disk .json session file is never written. Pin the writer ON in the
spawned pi client's env (the bootstrapped proxy inherits it).

Closes: pi package-native lane red on ci-fixes (run 43558179054, job 132336647955)
Same pure-sqlite default flip (5a80e2b, #6480) family: the JSON writer
is OFF, so the per-session CCR content-store sidecar (.content-store.json)
that 'child persists its own companion and survives a proxy restart'
asserts on is never written (0 !== 1). The earlier b1bc17b pin only
covered tests/plugin-fork.test.ts, not this file. Pin the writer ON at
module load (before any SessionStore is constructed).

Closes: dsh plugin lane red on ci-fixes (run 43558179102, job 132336519539)
…lane

Pure-sqlite default flip (5a80e2b, #6480) family: the JSON envelope
writer is OFF, so 'proxy must persist the session' fails — the on-disk
.json session file is never written. Pin the writer ON in the spawned
opencode client's env (the bootstrapped proxy inherits it).
Pure-sqlite default flip (5a80e2b, #6480) family: the JSON envelope
writer is OFF, so the on-disk .json session envelope this lane asserts on
(exactly ONE persisted session under the raw key) is never written. Pin
the writer ON in the spawned dsh client's env.
Pure-sqlite default flip (5a80e2b, #6480) family: the JSON envelope
writer is OFF, so the on-disk .json session file this lane asserts on
(state survival across the OLD→NEW update) is never written. Pin the
writer ON in the spawned proxy's env (both update phases share envBase).
@2270094969

Copy link
Copy Markdown

Field report: this PR covers the symptom for unattributed traffic, but the lane still can never be learned

I hit exactly the same 422 from a different entry point and spent a day on it, so here's my field data. This comment is not asking for anything in #6546 to change — the fix there is correct and I want it. I'm adding the part it doesn't reach, because after #6546 lands the request will stop 400-ing, but it will still never be compressed.

The case I hit

dsh-connect-workbuddy (a provider plugin, PiAiAdapter + loopback shim) + bili. Long session on a 1M-window model:

2026-11-14T15:56:22  sent DIRECT (uncompressed) — no active initiator attribution … this body 6,954,570 B
2026-11-14T15:57:46  sent DIRECT (uncompressed) — no active initiator attribution … this body 7,936,254 B
2026-11-14T15:58:17  unattributed oversized model request refused locally with a 422
                     (8,401,132 B > 8,388,608 B); would be sent uncompressed (#5405/#5408)

Overshoot by 12,524 bytes. The two earlier requests show the real cost: ~7 MB re-sent uncompressed every turn, growing.

Why attribution is structurally none here

DSH agent
  → PiAiAdapter (pi-ai, openai-completions)
  → fetch(http://127.0.0.1:<random port>/v2/chat/completions)   ← bili sees this one, ALS store OK
  → [shim HTTP request handler]                                ← new async resource, ALS store NOT inherited
  → fetch(https://copilot.tencent.com/v2/chat/completions)     ← this is the request that needs compressing

The shim is in-process (no child process), so the hop that loses the store is purely AsyncLocalStorage not surviving the socket callback boundary. attributionOf(ctx) returns { state: "none" } on the second hop.

On the plugin side options.sessionId is in hand at the adapter layer — the plugin knows exactly which session this is. It just has no sanctioned way to say so.

Why #6546 will not fix compression for this lane

I read the PR description carefully and I think this is out of its stated scope, so just confirming the consequence:

learnConversation(laneKey, sid, boundaryFp) is only called from the attr.state === "ok" branch of maybeRecallBeforeSend. With state === "none" on every request, the lane table for this host stays empty forever. Then maybeNegotiatedClaim short-circuits on its first line:

const laneKey = recallLaneKey(url);
if (!recallLaneHasEntries(laneKey)) return void 0;

So the negotiated-claim path never even runs. It is a bootstrap deadlock, independent of whether the refusal body is recognised as an overflow marker.

After #6546: DSH will see the refusal, refold, re-send. And the re-send still arrives as state: "none" with body possibly still over 8 MiB if the fold wasn't aggressive enough (or with the whole history unfolded, if the fold target is computed against a 1M window and the history is only ~150k tokens — see below). Recognising the refusal converts a permanent 400 into a retry loop; it does not restore compression to this lane.

A second, independent trap worth knowing about

Even the "refold" arm can no-op here. The natural target is window * TARGET_FRACTION, and the declared window for these models is 1,000,000, but the 8 MiB byte ceiling gets hit at roughly 150k tokens of mixed prose+images. So foldUntilFits measures the session, finds totalTokens ≈ 150k < 600k, and decides nothing needs folding — zero iterations, re-send identical body.

(That's #6122's "no window number is learnable from the refusal prose" note showing up from a different angle: the refusing thing is a byte gate, and the fold loop is driven by a token target. When threshold unit ≠ fold target unit, the loop can be vacuously satisfied.)

If a byte-gate refusal is going to drive a refold, the fold target probably has to be derived from the payload itself rather than from the model window — e.g. fold until the measured payload drops below a fraction of the gate, not until tokens drop below a fraction of the window.

What I tried that didn't work (so nobody repeats it)

Cross-lane recall by message fingerprint: in recallConversation, when the lane table is empty, scan all lanes and match on historyFps derived from message bodies, refusing on ambiguity (hits.length > 1). Patch did arm and did log:

cross-lane recall matched an unattributed provider lane back onto a known session
by message history — compression re-armed

But it only works if the session was previously attributed under some other provider. The affected sessions have blocks: 0 in /__bili/stats across every lane — bili has never successfully owned them — so there is nothing to match against. Rolled back.

What I think the sustainable fix is

Make attribution something a plugin can assert, instead of something bili has to infer:

// plugin start-up: traffic to this URL prefix belongs to the current session
bili.registerSessionAttribution({
  urlMatch: /copilot\.tencent\.com\/v2\/chat\/completions/,
  resolve: (ctx) => ctx.agents.currentInitiator()?.session?.id,
});

or per-call:

bili.withSessionAttribution(sessionId, () => fetch(url, init));

The state2.headersFor(url) gate is the choke point:

state2.headersFor = (_url) => {
  maybeRetry(ctx);
  if (!register.toolsReady) return void 0;
  const attr = attributionOf(ctx);
  if (attr.state !== "ok") return void 0;      // ← third-party adapters die here
  ...
  return { "x-bili-plugin": "dsh", "x-bili-plugin-conversation": sid, ... };
};

A registered resolver would let attributionOf stay the last word for first-party traffic while giving adapters a scoped override.

Environment

OS Windows 11, MINGW64_NT-10.0-22631
DSH 0.2.1-rc.3
billion-context 0.1.213, commit 133f3525c
provider plugin dsh-connect-workbuddy 5.21.3 (upstream copilot.tencent.com)
admin port 127.0.0.1:18787
/__bili/stats blindTunnels.total: 0, unrecognizedPaths.total: 0 — bili sees the plaintext fine, this is purely the attribution gate

Happy to produce a PR for the registration API if the shape works for you — I've already got the local reproduction and a small harness for it. 🙏

ework-agent and others added 2 commits November 14, 2026 16:52
# Conflicts:
#	tests/e2e/e2e-native-dsh-persona.test.ts
#	tests/e2e/e2e-native-dsh.test.ts
#	tests/e2e/e2e-native-opencode.test.ts
#	tests/e2e/e2e-native-pi.test.ts
#	tests/e2e/e2e-pi-codex-ws.test.ts
#	tests/e2e/e2e-subagent-sessions.test.ts
#	tests/notice-registry.test.ts
#	tests/plugin-fork.test.ts
…atest master

The sentinel confirmed the deterministic handle-leak hypothesis and landed
the fix in ci-fixes: persist-zstd + persist-metadata-fold-gate now close
their SQLite sidecars (closeQuietly) before the tmp-dir teardown, so
Windows can unlink index.db. This is the exact family red on win22-5
(EBUSY unlink .../index.db, #5506 x3 + persist-zstd x10).

Also brings master e5e1647 -> 9495c76 (release v0.1.214 + fixes) and
the sentinel's storage-open-busy-lock tripwire (acc7944).

Conflict in tests/notice-registry.test.ts resolved to ci-fixes's form
(HEAD was byte-identical to the merge base; ci-fixes adds the
BILI_STORAGE_JSON_WRITER unpin required by the dual-write baseline pins).
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

CI fix (win22-5) — resolved, pushed 6b037887d3.

Live re-verification first. The mirror cache was stale (frozen at 04:17Z and didn't even list the test shards), so I fetched the failing job directly: test (windows-latest, 22, 5), conclusion failure, head a570049c7b (the current PR head). Every failure is EBUSY: resource busy or locked, unlink '…\index.db' in the teardown path (tests/tmp-rm.ts:30 defaultRm → rmrf → rimrafSync) — the test bodies pass, cleanup fails. #5506 (persist-metadata-fold-gate, 3 subtests) + persist-zstd (10 subtests), both full passes. Zero intersection with this PR's diff (6 files: src/util.ts, src/fork-adoption.ts, 3 test files — none touch storage teardown).

Diagnosis + the sentinel's fix. This is the #4347 Windows-EBUSY family from the prior reds (floors 7/9). The #4599 sentinel has since confirmed the deterministic handle-leak hypothesis and landed the fix in ci-fixes: 7c152dc08a — persist-zstd + persist-metadata-fold-gate now close their SQLite sidecars (closeQuietly) before the tmp-dir teardown, so Windows can unlink index.db. ci-fixes (7b673c7) also carries the latest master (9495c76, release v0.1.214) and the storage-open-busy-lock tripwire (acc7944).

What I did:

  1. Merged origin/ci-fixes (7b673c7) into the PR branch.
  2. One conflict in tests/notice-registry.test.ts: HEAD was byte-identical to the merge base; ci-fixes adds the BILI_STORAGE_JSON_WRITER unpin (required by the dual-write baseline pins). Resolved to ci-fixes's form wholesale (file now matches ci-fixes exactly).
  3. Targeted verification: persist-zstd + persist-metadata-fold-gate + notice-registry → 25/25 under the exact CI preload chain; npm run typecheck clean.
  4. Pushed a570049c7b..6b037887d3 (fast-forward; remote head verified unchanged before push).

Status now: new CI run on head 6b037887d3 — all 24 checks pending (verified via the mirror cache, refreshed 09:43Z). The EBUSY shards will draw the fix; if the flake recurs on a fresh runner, the remedy per #4599 is rerun-failed-jobs (single write), no code change.

八问(中文结构化结论):

  1. 是什么问题:PR 分支 CI 的 Windows 测试分片(22-5)红——不是断言失败,是测试收尾删临时 SQLite 文件(index.db)时被操作系统锁住(EBUSY unlink),测试本体全部通过。
  2. 什么原因导致的:master 级继承问题(windows-22-1 lane randomly red: during test cleanup, unlink index.db hit an EBUSY external lock, couldn't survive tmp-rm | windows-22-1 车道随机红:测试清理期 unlink index.db 遇 EBUSY 外部锁,熬不过 tmp-rm 的 9.6 秒重试梯子(#2799/#2908 族复发) #4347 家族)。哨兵已确认这是确定性句柄泄漏:纯 SQLite 默认翻转后,persist-zstd / persist-metadata-fold-gate 急开 index.db 且从不在 finally 关 store,进程自身的 SQLite 句柄在 Windows 上锁死文件,~9.4s 重试阶梯必输。与本 PR 的 6 个改动文件零交集。
  3. 影响范围与严重程度:仅 CI 红,波及当前全部在途 PR 的 Windows 分片(红风暴),不阻塞本 PR 代码正确性——其余检查全绿。
  4. 怎么修复的:按 CI 协调政策(2026-10-24)merge 哨兵分支 ci-fixes(7b673c7)进 PR 分支,带入哨兵的 EBUSY 修复(7c152dc:teardown 前 closeQuietly 关 store)+ 最新 master(9495c76,含 release v0.1.214)+ storage-open-busy-lock 触发器。唯一冲突(notice-registry 存储 pin)收敛到 ci-fixes 权威形态(HEAD 与 merge base 逐字节一致,ci-fixes 新增 BILI_STORAGE_JSON_WRITER 解 pin),行为中性。本地定向验证 25/25(CI preload 链)+ typecheck 全绿;推送 6b03788。
  5. 是否有回归问题:无。净 diff 是 ci-fixes 的合并(哨兵已审的 master 级修复 + 测试隔离 pin);本 PR 自身的 6 个文件未被改动;notice-registry 收敛为 ci-fixes 形态(两种形态都满足存储选择门,已实测)。
  6. 是否破坏缓存:不涉及(纯测试 teardown 修复 + 测试环境 pin,与运行时缓存/重试/游标无关)。
  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(无新 env/config/默认值;BILI_STORAGE_JSON_WRITER 解 pin 是测试隔离,不是新配置面)。

判定:win22-5 红为继承性 Windows EBUSY teardown 问题(非本 PR 引入),已按政策 merge 哨兵 ci-fixes(含其确定性句柄泄漏修复)并推送;新 run 在 head 6b03788 上排队,EBUSY 分片将随修复转绿。本分支侧工作完成,无进一步代码改动。


On the field report (attribution bootstrap deadlock)

Thanks — this is a well-researched report and the key claims check out against the code:

  • learnConversation is only called from the attr.state === "ok" branch (src/agent/dsh-native.ts:1871-1878);
  • maybeNegotiatedClaim short-circuits on its first line when the lane table is empty (src/agent/dsh-native.ts:1832: if (!recallLaneHasEntries(laneKey)) return undefined;).

So the bootstrap deadlock you describe is real: with state === "none" on every request (the ALS store not surviving the shim's socket-callback boundary), the lane table never fills and the negotiated-claim path never runs. #6546 converts the permanent 400 into a retry loop but does not restore compression to this lane — that part is correctly out of #6546's scope (as you noted), and I agree the fix there is right.

Two notes on routing:

  1. This is a distinct failure domain (attribution bootstrap deadlock for third-party adapters that lose the ALS store), separate from [PR] fix(overflow): bili's own oversize-gate refusal now triggers self-heal instead of a permanent 400 loop (#6122) #6546 (recognizing bili's own gate refusal). It does not block [PR] fix(overflow): bili's own oversize-gate refusal now triggers self-heal instead of a permanent 400 loop (#6122) #6546.
  2. The proposed registration API is a design decision — a new public API surface plus an attribution trust-boundary change (letting a plugin assert session attribution instead of bili inferring it). That's owner territory, not something I'll unilaterally build as the CI engineer. The shape you sketched (registerSessionAttribution({ urlMatch, resolve }) / withSessionAttribution(sid, fn)) is reasonable and correctly scoped to the headersFor choke point you identified.

Suggested next step: file an issue carrying your repro + harness (you offered the local reproduction), state the proposed shape, and let the owner decide on the API surface. Your second trap (byte-gate refusal driving a token-target refold that no-ops when threshold unit ≠ fold target unit) is a good independent sub-point to include there. I won't file it or build it myself — it's your proposal and it needs the owner's sign-off on the shape; happy to help route it if the owner wants.

@ranxianglei

Copy link
Copy Markdown
Owner Author

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

Re-review on new head 6b03788 — APPROVE still holds.

The head moved from 0a565fa → 6b03788 via a ci-fixes merge (latest master + Windows EBUSY teardown fix). I verified the PR's own F4+F5 changes are byte-identical in the new head:

  • src/util.ts:423 — /bili_gate_oversize_refusal/i pattern (shifted from line 414 by the merge; content unchanged)
  • src/fork-adoption.ts:536 — straddled-block log clause (unchanged)
  • All three test files present and unchanged

The 132-file / +5974 diff between the two heads is entirely from the ci-fixes merge (test-isolation pins, storage-writer pins, Windows teardown fix, release v0.1.214, etc.) — none of it touches the PR's F4+F5 logic.

CI note: the ci-fixes merge re-triggered the full suite. As of this review, 14 checks are still in progress, 0 failures, 3 success, 13 skipped (gated e2e). No failures so far. The previous head (0a565fa) was fully green (25 success / 3 cancelled / 2 skipped). The PR's own changes are unchanged, so the green verdict carries over — but the new head's CI is still settling.

Verdict: APPROVE (unchanged from the first review). The F4+F5 slice is correct, well-tested, and regression-free. The ci-fixes rebase introduces no changes to the PR's logic.


8问审核清单(增量 — 新 head 6b03788)

  1. 是什么问题:同前——head 从 0a565fa 移到 6b03788(ci-fixes merge),需确认 PR 自身 F4+F5 变更未被 merge 破坏。
  2. 什么原因导致的:ci-fixes 分支合入了最新 master(含 Windows EBUSY teardown fix、release v0.1.214、多个 test-isolation pin),触发了全量 CI 重跑。
  3. 影响范围与严重程度:PR 自身逻辑无影响(F4+F5 变更逐字节验证不变);CI 重跑中(14 项 in progress,0 failure),不影响代码正确性。
  4. 怎么修复的:无需修复——ci-fixes merge 是标准的分支收敛操作,PR 自身变更未被触碰。
  5. 是否有回归问题:无。PR 自身变更逐字节验证不变(pattern 在 line 423、log line 在 line 536、三个测试文件均在)。ci-fixes merge 引入的变更(test pins、storage-writer pins、Windows teardown)均与 F4+F5 逻辑无关。
  6. 是否破坏缓存:不涉及。
  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(ci-fixes merge 未引入新配置;PR 自身也未引入)。

判定:可合(APPROVE 维持)。 PR 自身 F4+F5 变更在新 head 上逐字节验证不变,CI 重跑中无 failure。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 Re-review on new head 6b03788 — APPROVE still holds.

Head moved from 0a565fa → 6b03788 via a ci-fixes merge (latest master + Windows EBUSY teardown fix). Verified the PR's own F4+F5 changes are byte-identical in the new head (pattern at src/util.ts:423, log line at src/fork-adoption.ts:536, all three test files unchanged). The 132-file / +5974 diff between heads is entirely from the ci-fixes merge — none of it touches the F4+F5 logic.

CI: the ci-fixes merge re-triggered the full suite. As of this review, 14 checks in progress, 0 failures. The previous head was fully green. The PR's own changes are unchanged, so the green verdict carries over.

Full re-review with 8问: #6546 (comment)

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

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants