Skip to content

[PR] fix: wave-12 batch — 19 community fixes (throttle isolation, Google thinking budgets, Responses warmup SSE, dsh carrier shapes, image byte trigger, keepalive aborts, …) - #7199

Open
ranxianglei wants to merge 32 commits into
masterfrom
batch12-wave12
Open

ranxianglei wants to merge 32 commits into
masterfrom
batch12-wave12

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

What

Wave-12 batch: 19 small community fixes (12× Afloat16, 6× resec, 1× myfear) + 1 carryover commit from #6766's post-merge author push, merged as-is with author commits preserved (no rebase, no squash). Each entry is a single-purpose regression fix with its own test suite; all were reviewed line-by-line before merging.

PR Author Fix Closes
#7187 Afloat16 429 throttle holds for query-key-authenticated Google accounts no longer share one anonymous hold (per-account isolation on native RPC paths; header credentials and conflicts unchanged) #7186
#7185 Afloat16 Google functionResponse.response.error maps to the structural isError flag → output steering keeps the requested thinking budget after failed tool calls instead of dropping to 128 #7183
#7182 Afloat16 Responses warmup SSE fallback reuses the shared SSE helpers (CRLF/lone-CR/data: framing, multiline data, UTF-8 flush) → warmup checkpoints commit again #7180
#7177 resec Pinned-attach target proven dead → fast-path origin released (model requests degrade to direct instead of every request hitting the dead origin); + hermetic test + LIVE_FILES registration #7176
#7174 Afloat16 Canonical affinity builders use null-prototype accumulators → own __proto__ JSON keys contribute to fingerprints (no cross-session exact-prefix reuse) #7172
#7171 Afloat16 Rejected (non-429) continuation response bodies are cancelled → no orphaned transport connections after a 503-style refusal #7170
#7166 Afloat16 64 KiB CONNECT response-header limit now enforced after completion too (was only checked while the delimiter was still missing) #7165
#7162 Afloat16 planTokenize normalizes each source once instead of per token match (+ reproducible bench script) #7160
#7161 resec bili dsh launcher overlay decision uses the profile being booted, not the machine-wide install probe (booted profile without the bundle gets the overlay + a helpful stderr line) #7155
#7159 resec Host-plane /compact stood down on bili-owned lanes in both patch carriers (the #1729 wire guard refuses its envelope unconditionally; dsh misreported that as a model failure) #7154
#7158 resec Durable-carrier classifiers match the emitters' real shape (system-reminder frame / job-id lead-in / settlement ending), not bare markers → a human paste quoting a marker no longer steals the carrier identity or loses the Rule-3 anchor (343-session check: human false positives 1 → 0) #7153
#7147 Afloat16 Byte-size trigger restored for ordinary within-limit images (pre-#3701 gate): a 262 KB same-dimension PNG re-encodes to 757 B; forced path and exemptions untouched #7146
#7143 Afloat16 Superseded/disarmed cache keepalive snapshots abort their in-flight ping (per-ping AbortController, deadline relayed without AbortSignal.any for Node 20) → slow upstreams stop occupying connections #7142
#7141 Afloat16 Protected read_original_image originals count toward the upstream image cap while remaining un-archivable (rescue no longer leaves the full request over the limit) #7140
#7138 Afloat16 ETIMEDOUT with explicit connect-phase evidence (direct syscall:"connect" or an all-connect aggregate) classifies as retryable connect-timeout instead of non-replayed upstream-timeout refs #7120
#7136 myfear OpenAI adapter keeps provider usage fields (credits/cost, cache details) in the synthesized completion and forwards the trailing usage-only frame when stream_options.include_usage is set; per-round reset prevents cost inheritance —
#7133 Afloat16 patchPnpmLock propagates structural failures from every map (validate-before-write contract; a malformed later section no longer reports success with a partial repair) #7132
#7129 resec Patch-layer disabled: read by value (YAML 1.1 spellings, quoted forms, trailing comments; !!js expressions conservatively count as disabled) → cost advisory and host-plugin notice stop ignoring non-literal disables #7128
#7127 resec A host with no config file is the default state: ENOENT stays silent; a present-but-unusable config gets one warning naming the reason #7126
(carryover) ci-check #6766 author commit 1fcc31d11 (pushed after the wave-11 snapshot): compat.dropFields log notes when it drops the output-budget field #6752 (already closed)

Review notes

Closes #7186
Closes #7183
Closes #7180
Closes #7176
Closes #7172
Closes #7170
Closes #7165
Closes #7160
Closes #7155
Closes #7154
Closes #7153
Closes #7146
Closes #7142
Closes #7140
Closes #7132
Closes #7128
Closes #7126

resec and others added 30 commits November 18, 2026 14:56
loadDeclaredModelEndpoints folded three different situations into one console.error: no config file at all (the documented default for a host that never declared custom endpoints), a file that exists but cannot be read, and a file that exists but is not a usable declaration. The first one fires on every fresh install's very first boot and reads like a failure, so a new user's first impression is a broken setup.

ENOENT is now silent; a present-but-unusable config keeps exactly one line, now naming the reason (read error code, or the JSON/parse complaint) so it points at the file. The never-throws and one-shot contracts are unchanged.

Tests split accordingly: absent file asserts zero stderr; malformed JSON and structurally invalid declarations assert one line each, including the file name and the offending index.
…e` (#7128)

parseDshPatchEntries is the single judgement for 'is this plugin turned off in cordis.patch.yml', and both consumers act on it: the cost advisory treats !disabled as 'enabled, warn about spend', and the host-plugin notice treats disabled as 'the user handled it, stay quiet'. The old scan matched only the literal `true`.

dsh's patch layer is not limited to that spelling. In one stock web-profile dump, 52 `disabled:` lines carry only 29 literal trues; the other 23 are `!!js` expressions, plus YAML 1.1 forms and quoted booleans. Every one of those read as ENABLED, so a user who disabled a billed plugin kept seeing its cost hint, and a user who disabled a host plugin could not dismiss the notice.

Read the value instead of the literal: YAML booleans (true/True/yes/on), quoted forms, and a trailing comment. A `!!js` expression cannot be evaluated from text, so it counts as 'turned off' -- the safe direction for both consumers (under-reporting a hint beats nagging someone who already acted). Block attribution and the truex-style non-boolean rejection are unchanged.

Measured on the stock dump: 29/52 -> 52/52 recognised. Negative control: restoring the literal-true regex turns the new case red; the two consumer suites (2462 advisory, 2861 notice) stay green.
Propagate structural failures from every dependency section before committing
ledger changes. Keep sections without the target dependency skippable.
Preserve the last upstream usage object when emitting the synthesized completion chunk. OpenAI clients that price a turn from provider fields (for example cost/credits) otherwise lose them.

Forward the trailing usage-only frame to clients that set stream_options.include_usage, including text-only rounds, and reset the remembered usage per stream round so a later round cannot inherit a previous cost.

Adds five regression tests in tests/openai-usage-frame-forward.test.ts.
Let each snapshot own cancellation through response-body consumption and release its upstream connection when replaced or disarmed. Preserve the existing timeout and failure policy, including early Node 20 support.

Fixes #7142
…7153)

A human message that merely quoted a durable marker was classified as a host
injection. Two consumers made that a correctness bug: the kernel's Rule-3
last-user anchor refuses a virtual-path message (so the live request could be
folded away silently), and durable/* goes into protectedLatestTools (so a
quote took the identity over from the real carrier and was pinned forever).

Shape gates on the four classifiers whose bare marker collides with human or
relay text: the durable pair must open the system-reminder frame both
producers emit, and the two one-shot notices must carry their emitter's real
shape (job id + detail/EOL, a settlement ending). Every emitter shape keeps
matching; a future emitter change degrades to a miss, never a false positive.

Existing fixtures that encoded the bare-marker shape were updated to the
producers' real output; assertions untouched.

Real-session check (343 dsh journals, 3009 user-role messages): 1061 -> 1060
classifications, human false positives 1 -> 0, and the misclassified relay
report now resolves as oneshot/agent-relay instead of stealing
durable/skill-catalog.
…lane (#7154)

/compact drives the host's own compaction call — the same envelope bili's wire
guard refuses unconditionally (#1729) — so on a bili-owned lane the command has
no outcome except failure. dsh reports that failure as "Compaction could not
produce a useful summary", blaming the model for a policy refusal, and every
retry fails identically because the refusal is terminal.

Disable the command entry in both patch carriers (persistent bundle + launcher
overlay) with the same static semantics as the adjacent compaction-basic
auto:false row. Escape hatch: a profile-level `disabled: false` for the same
id brings it back; invocations are still refused at the wire.
…7155)

dshNativeInstalled() answered for the whole machine ('any profile has the
bundle'), and the launcher used that answer for THIS launch. On a machine where
a sibling profile has the bundle but the booted one does not — a scratch
profile created after 'bili plugin install dsh', a profile whose dep was
removed by hand — the overlay was skipped and the session ran with zero bili
entries: no ACP tools, no takeover, native compaction armed (which the wire
guard then refuses turn after turn), while the launch note described the other
profile.

Take an optional profile argument (omitting it keeps the machine-wide answer
for status/tests), parse the booted profile from the dsh argv the way dsh does
(bare first argument abbreviates --profile), and say plainly when the booted
profile lacks the bundle.
…itter shape (#7153)

The #7153 carrier gates require the complete <system-reminder> frame around the
workspace-instructions label, but two runtime fixtures still carried the bare
label, so their guard message stopped classifying as durable/agents-md and was
no longer pinned: the wire handed out a real ref where null was asserted.

- tests/plugin-fork.test.ts (#2676): the durable-guard history row now carries
  the frame dsh-agent-instructions emits.
- tests/session-continuity-proof.test.ts (#629/#2620/#2625/#2676): GUARD_TEXT
  likewise; the scenario-B fork-copy assertions are unchanged.

CI was red on the one assertion in each file (ubuntu 22/3, ubuntu 24/3,
windows 22/1, windows 22/4).
… proven dead (#7176)

apply() primes the interceptor's origin plane with the launcher-provided attach
origin before any health verdict exists — deliberate, so a normal boot does not
pay a probe. When the verdict then lands and the target is dead,
verifyAttachAndRecover's pinned branch cleared the tool base but left that
plane bound. readyOrigin() short-circuits on it, so every later model request
was rewritten onto the dead origin and failed; the documented direct
(uncompressed) degradation was unreachable for the rest of the process, and the
user saw only a bare transport/503 per turn.

The sibling spawnFallback() tail already drops the same field (together with the
preset unfreeze it needs because it then spawns). This branch must NOT spawn —
per #1365 a second instance would serve tools while the model channel stayed
pinned, and every bili tool call would 404 — so only the plane is cleared: the
preset stays, maybeRetry keeps re-probing, and traffic degrades to direct until
the target answers again. One log line records the transition.

Measured with a local Messages mock on the dsh lane: with the origin dead, three
consecutive turns failed TRANSPORT and the upstream saw zero requests before
this change; after it, requests that follow the verdict take the no-origin
direct lane. The first request of a process that booted against an already-dead
target still races the verdict (fails once) — proving the origin before priming
the plane, or giving this lane the resolver lanes' per-request liveness check,
are the remaining options and are not part of this change.
…rigin hermetic

run-ci-tests.mjs refuses any .test.ts matching SPAWN_RE that is not in
scripts/live-files.mjs, and this suite's own identifier (_setSpawnForTest) plus
the header comment tripped it — every test shard failed before a single test
ran.

Registered at the end of the table (slot = index drives each file's zone port
band, so entries are append-only) with the concurrency note the table asks for:
no shared port, no child process, both arms reset the module-level origin plane
in finally.

While here, the dead arm no longer hardcodes a port. It binds :0, reads the
port, closes it and reuses the number, so the 'nothing listens here' premise is
guaranteed at runtime instead of assuming a fixed number stays free on a shard
runner, and no zone band is involved.
)

When compat.dropFields removes a budget field that restoreOutputBudget
just restored, the drop log now says so explicitly so the log is honest
(the restore fired, but the field is absent from the forwarded body).

One-line note appended to both drop-log sites (graph and string paths).

(cherry picked from commit 1fcc31d)
@ranxianglei ranxianglei changed the title fix: wave-12 batch — 19 community fixes (throttle isolation, Google thinking budgets, Responses warmup SSE, dsh carrier shapes, image byte trigger, keepalive aborts, …) [PR] fix: wave-12 batch — 19 community fixes (throttle isolation, Google thinking budgets, Responses warmup SSE, dsh carrier shapes, image byte trigger, keepalive aborts, …) Nov 18, 2026
@github-actions

github-actions Bot commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: batch12-wave12 (b329f58)

Latest build — install this exact version

npm install -g billion-context@0.0.11111-pr.7199.fda4ab9

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
b329f58 billion-context@0.0.11111-pr.7199.fda4ab9 build
1c8cac2 billion-context@0.0.10850-pr.7199.faf3249 build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr7199.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei ranxianglei added bug Something isn't working p1 labels Nov 18, 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 — #7199 是 wave-12 的合并载体:分支 batch12-wave12 = master + 19 个 merge commit,每个对应一个社区 PR(#7127/#7129/#7133/#7136/#7138/#7141/#7143/#7147/#7158/#7159/#7161/#7162/#7166/#7171/#7174/#7177/#7182/#7185/#7187),这些 PR 仍各自 open。兄弟批次 #7196(batch-4,13 个修复)是不相交的集合(#7115/#7116/#7117/#7066/#7094/…),无重叠。没有第二个并行载体装这 19 个修复。

标签: bug + p1(已加到 PR,保留既有 root-cause)。判 p1 因为批次内含 p1 级可用性修复:#7176(dead attach origin → 会话每轮失败直到 proxy 回来)与 #7182(warmup checkpoint 丢失 → Responses WS 车道会话连续性断裂)。

范围:53 文件,+2811/−167(24 src/scripts + 29 test)。无 kernel/、无 pi-subagents/、无 version 改动、无 config-schema 改动。19 个修复 + 1 个 carryover 全部对照 PR 分支代码逐处核验;29 个测试文件全部通读。

逐修复判定(全部 OK)

# 修复 判定
1 #7154 dsh /compact 在 bili-owned 车道停用(两层 patch 同步) OK,行为变更已披露,逃生门 = profile 自己 disabled: false
2 #7162 plan 文本 toLowerCase 从逐次(O(n²))移到 tag 剥离后一次 OK,正则状态机 lastIndex 循环 → 匹配序列逐字相同,更快
3 #7176 proven-dead pinned 分支释放 model-channel origin(降级直连,不 spawn) OK,state/persistClientEvent/pinned 均在作用域,置于 waitForProxyVersion(pinned)===undefined 之后,守 #1365 不 spawn
4 #7143 keepalive 被取代的 ping 用 AbortController 中止(非 AbortSignal.any,Node 20 兼容) OK,所有 registry.get(sessionId)!==entry 身份检查保留;try/finally 清 listener + pingInFlight + controller
5 #7171 被拒非 429 continuation 取消 response body OK,最小改动,无孤儿连接
6 #7166 CONNECT 64KiB 头限制在分隔符找到后也强制 OK,headerBytes = end<0 ? buffer.length : end+4
7 #7174 affinity 累加器 {}→Object.create(null),自有 __proto__ JSON key 进入指纹 OK,下游仅 JSON.stringify→hash,无 proto 方法调用,null-proto 安全
8 #7141 受保护原件计入图片 cap(计入上限但永不归档) OK,remaining=occurrences.length 含 exempt;归档遍历 if(o.exempt) continue;字节 shed worklist 不变
9 #7147 恢复字节触发的图片重编码(非 forced 超限图现在重编码) OK,行为变更已披露;forced 路径(:199)不变
10 #7158 carrier 分类器收紧到 emitter 形状(system-reminder 帧 / 锚定正则) OK,343 会话检查:human FP 1→0;emitter 形状仍匹配(测试钉死)
11 #7129 patchPnpmLock v6-v8 任一 section 结构失败即返回 fail(先验证后写) OK,调用方 :527 在任何磁盘写之前 if(patch.kind==="fail") return fail(...) → 零写契约成立
12 #7128 isDisabledValue(true/yes/on、!!js、去引号)替代字面 true OK,旧代码漏的带引号值现在能读;quoted 含 # 的假阴性旧代码也漏,无回归
13 #7155 dshProfileFromArgs + dshNativeInstalled(profile) 按启动 profile 判 overlay OK,undefined/""→machine-wide 向后兼容;stderr 提示用既有 console.error 模式
14 #7136 OpenAI 合成 completion 保留 provider usage 字段;include_usage 尾帧转发 OK,per-round reset 防成本继承;provider 字段(credits/cost)存活
15 #7138 连接阶段 ETIMEDOUT → connect-timeout(可重试),其余保守 OK,未知 syscall/空/混合 aggregate 仍 upstream-timeout 不可重试
16 #7187 Google 原生 RPC 无头凭据时按 query-key 隔离 throttle hold OK,头身份不变;冲突/空值 → 匿名 hold
17 #7185 Google functionResponse 失败判定用 Object.hasOwn(response,"error") OK,失败工具调用保留请求的 thinking budget 而非跌到 128
18 #7182 Responses warmup SSE 完整事件解析,EOF 残量也处理 OK,sseEventData 语义核验;if(done) break 移到循环尾 → 终帧不再丢
19 #7127 model-endpoints ENOENT 静默,其他读错 warnOnce OK,行为变更已披露(新装不再告警)
carry relay.ts compat.dropFields 日志补 output-budget override 注记 OK,纯日志,零行为变更

回归审查(既有行为清点)

逐路径审了 diff 触碰的既有行为:无意外回归。仅有的有意行为变更全部已在 PR 正文披露:#7154 停用 /compact、#7147 恢复字节触发重编码、#7127 ENOENT 不告警、#7162 plan 文本单次小写化。高危模式逐项过:

  • 语义漂移:无 — 每个修复只改它声明的那一处判定,相邻路径保持。
  • fail-fast 变吞错:无 — [PR] fix(dsh): read the patch layer's disabled value, not the literal true (#7128) #7129 反而把"部分写"从可能变不可能(先验证后写)。
  • 阈值/时序/输出格式漂移:仅披露的 4 处;RESCALE_TRIGGER_BYTES=100*1024、PING_TIMEOUT_MS=60_000、64*1024 头限等常量值未动。
  • 缓存/重试策略静默失效:无 — keepalive 是取消被取代 ping,不动缓存语义;throttle 隔离只细化 hold key。

测试有效性

29 个测试文件全部通读,全部真测试、双向断言(该触发时触发 + 不该触发时不触发):

  • 新回归套件 8 个(issue7120/7140/7142/7146/7153/7154/7155/7176)+ openai-usage-frame-forward + output-steering-google-errors + proxy-connect-header-limit,每个都钉住原 bug 的精确形状(边界字节、split-delimiter、5 种 SSE 编码、7 种 fixture 形状、4 失效触发×2 阶段)。
  • 关键 e2e:issue7140/7146 走真实 HTTP plugin+proxy 双模;output-steering 走真实 HTTP clean→128 / failed→8000。
  • 保守面钉死:issue7120 的 7 个负例(unknown/read/write/empty/sparse/mixed aggregate/non-error child)全部保持 upstream-timeout 不可重试;issue7142 删掉 AbortSignal.any 仍工作(早 Node 20)。
  • 既有套件同步更新:plugin-fork / session-continuity-proof 的 fixture 包进 system-reminder 帧(与 [PR] fix(dsh): match carriers on their emitted shape, not a bare marker (#7153) #7158 一致);upstream-timeout-diagnostics 反映新分类。

diff 整洁度 + 代码整洁

53 文件全部与 19 个修复 + 测试相关,无跑题文件、无生成物/锁文件连带、无纯空/重排。命名达意、无死代码、无重复、错误处理带上下文、常量具名、注释讲为什么、风格与仓库一致、无新依赖。scripts/live-files.mjs 注册新测试文件带 SPAWN_RE tripwire 注释(listen(0) 端口,无固定端口)。

CI

head 1c8cac2c30 共 100 个 check run:65 success、35 skipped、0 failed、0 pending。绿。


结构化八问

  1. 是什么问题:这是一个把 19 个独立社区 bug 修复(节流隔离、Google thinking budget、Responses warmup SSE、dsh carrier 形状、图片字节触发、keepalive 中止等)合并成单一载体的批次 PR,不是新功能、也不是单一根因修复。
  2. 什么原因导致的:每个修复各自针对一个独立根因(19 个不同机制);批次本身不引入新根因,只是把已逐个审过的修复合并。
  3. 影响范围与严重程度:跨多车道(dsh / 核心 proxy / openai / google / responses / egress / image / keepalive / throttle)。各修复 p1–p3 不等;最重的是 [Bug][dsh] a dead attach origin stays bound to the interceptor fast path — every turn fails instead of degrading to a direct send #7176(dead attach origin → 会话每轮失败)与 [PR] fix: parse complete SSE events during Responses warmup #7182(warmup checkpoint 丢失 → 会话连续性断裂),属 p1。
  4. 怎么修复的:逐项修复(策略/机制层面),每项配双向回归测试;批次只做合并,不附加语义改动。
  5. 是否有回归问题:审了 diff 触碰路径的既有行为——无意外回归。仅 4 处有意行为变更,全部已在 PR 正文披露([Bug][dsh] non-web profiles still mount /compact, but every attempt is refused by the #1729 guard — and the error blames the model #7154/[PR] fix: restore byte-triggered image re-encoding #7147/[PR] fix(dsh): a never-configured host is not a config error (#7126) #7127/[PR] fix: normalize plan text once per source #7162)。
  6. 是否破坏缓存:否。keepalive 修复([PR] fix: abort superseded cache keepalive requests #7143)是取消被取代的 ping,不改缓存语义;节流隔离([PR] fix: isolate Google query-key throttle holds #7187)只细化 hold key,不动缓存。
  7. 是否改了设计方向:否。无架构/边界/数据流转向,全部是 scoped bugfix。
  8. 是否引入了新的配置:否。无新 config key / env / 默认值。[Bug][dsh] non-web profiles still mount /compact, but every attempt is refused by the #1729 guard — and the error blames the model #7154 的 dsh.bundle.patch.yml 改动是 dsh patch 层,不是 bili config 面。

判定:可合(mergeable) — 19 个修复逐一核验通过、双向测试齐全、CI 绿、diff 干净、无未披露行为变更。批次载体与 19 个独立 PR 的合并关系清晰,与兄弟批次 #7196 无重叠。

….mjs zone-list conflict (union: keep #7156 + #7176 entries)
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Conflict resolved — PR head is now b329f580abb, merge state cleared (was dirty, now mergeable: true, blocked = waiting on CI + branch protection only).

What I did this run:

  1. Merged origin/master (b10c1a0d4ee) into the PR head (1c8cac2c303, branch batch12-wave12) — master had moved 72 commits past the PR's base.
  2. One conflict, resolved by union — scripts/live-files.mjs: both sides had appended a different entry to the CI live-files zone list at the end of the array. Kept BOTH: tests/storage-integrity-cache.test.ts ([Bug][storage] Takeover spawns before reclaim: old and new sidecars simultaneously hold index.db, causing corruption; em | [Bug][storage] 接管时先 spawn 后回收:新旧侧车同时持有 index.db 引发损坏;空 payload_json 又让历史会话静默变 0 #7156, from master) and tests/issue7176-dead-attach-origin-release.test.ts ([Bug][dsh] a dead attach origin stays bound to the interceptor fast path — every turn fails instead of degrading to a direct send #7176, from this PR). Both test files exist in the merged tree; the script parses and exports 72 entries.
  3. Verified locally before the final push: npm ci (root + kernel + pi-subagents), npm run typecheck clean, npm run build OK (build-info commit = the merge commit), and the two conflict-involved test files run green (15/15). Also checked the merge tree by set-algebra against both parents: every file changed on exactly one side took that side's content, the only file changed differently on both sides was the one conflict, and ~20 wave-12 files appear on "both sides" only because identical fixes already landed on master via other batch PRs — no silent loss from either side.
  4. Pushed the merge commit as the final head (no [ework:ci-held] marker → this is the branch's full CI pass).

Incident disclosure (honesty, per push discipline): my pre-push git pull --rebase origin/batch12-wave12 — which the remote had not moved, so it was meant to be a no-op — flattened the merge commit and re-applied master's 72 commits individually, stopping at the same known live-files.mjs conflict. A pipeline exit-code trap (| tail masking the failure) let the && chain push that mid-rebase head (14488b0aad6, 18 of 39 commits applied) to the branch. I aborted the rebase, confirmed by fresh fetch that the remote contained only my own artifact (the branch was still at the pre-task head before my push — no other agent's work involved), and force-pushed the verified merge commit b329f580abb back (the task-sanctioned conflict-rebase force-push). Final remote head verified = b329f580abb.

CI will run on the new head; the mirror check cache will reflect it within a minute or two of the run starting.


结构化八问(中文结论)

  1. 是什么问题:PR [PR] fix: wave-12 batch — 19 community fixes (throttle isolation, Google thinking budgets, Responses warmup SSE, dsh carrier shapes, image byte trigger, keepalive aborts, …) #7199 与 master 产生合并冲突(merge state: dirty),无法合并。
  2. 什么原因导致的:PR 分支基于 wave-11 合并点,master 其后前进 72 个提交;唯一真实冲突在 scripts/live-files.mjs —— PR 侧和 master 侧各自在数组末尾追加了不同的 CI live-files 条目([Bug][dsh] a dead attach origin stays bound to the interceptor fast path — every turn fails instead of degrading to a direct send #7176 vs [Bug][storage] Takeover spawns before reclaim: old and new sidecars simultaneously hold index.db, causing corruption; em | [Bug][storage] 接管时先 spawn 后回收:新旧侧车同时持有 index.db 引发损坏;空 payload_json 又让历史会话静默变 0 #7156),属「双方同位置追加」型冲突。
  3. 影响范围与严重程度:仅阻塞本 PR 合并;不影响任何用户面行为。冲突文件是 CI 测试端口分区登记表,漏掉任一条目会让对应测试套件失去端口隔离登记。
  4. 怎么修复的:merge master 进 PR 分支,冲突处取并集(两条目都保留),本地验证后以 merge commit b329f580abb 作为最终 head 推送(不带 ci-held 标记,触发完整 CI)。
  5. 是否有回归问题:无。逐文件核对合并树与双亲的差集:单侧修改的文件全部原样取该侧内容,双侧内容相同的文件无差异,唯一双侧不同内容文件即该冲突且已按并集解决;typecheck + build + 两个涉事测试文件(15/15)本地验证通过。
  6. 是否破坏缓存:不涉及(无任何缓存/重试/游标逻辑触碰)。
  7. 是否改了设计方向:不涉及,纯冲突消解,无行为变更。
  8. 是否引入了新的配置:不涉及。

收工判定:冲突已解决并推送,PR 可合并状态已恢复,等待 CI 在 b329f580abb 上跑完即可进入正常 review/merge 流程。

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