Skip to content

[PR] fix(lanes): demote the wire clients' auxiliary calls to the side lane (discussions/6884) - #7083

Merged
ranxianglei merged 2 commits into
masterfrom
design/6884-wire-client-sidecars
Nov 18, 2026
Merged

ranxianglei merged 2 commits into
masterfrom
design/6884-wire-client-sidecars

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

What was broken

Our session-keying audit (discussions/6884, five client codebases cloned and read end to end) confirmed the good news first: every client on the verbatim session-id roster mints one id per logical conversation, and the id lifecycles are healthy.

The bad news: each of them also fires auxiliary LLM calls under the SAME id — title generation, compaction summaries, prompt suggestions, memory, conversation naming. Those requests have "one-shot" body shapes that today's side-request heuristics misclassify as ordinary conversation turns, so they enter the main session's kernel pipeline: junk turns in the snapshot, usage-baseline pollution, stats inflation.

Client Affected auxiliary paths (all under the main session id)
grok-shell /btw, compaction, title refresh, recap, memory, laziness — 7+ paths
mcode prompt suggestion (default on, every accepted answer), Goal evaluator, title-gen
goose naming, compaction, tool-pair summaries, labels, adversarial review, judge, apps, summarize
CodeBuddy compaction summary request
opencode title-gen in pure-proxy mode (no bili plugin)

The fix

One clause per client, each gated on the client's own wire signal (its session header / its compaction header / its constant User-Agent) — the same trust tier as the session-id roster itself. No clause identifies a client from body shape alone, so a non-client can never be mislabeled.

  • grok-shell: main turns always stamp x-grok-turn-idx (turn.rs:3002-3003); every auxiliary path stamps x-grok-session-id without it — grok's own test-support contract (README:32). Header present + turn index absent → side lane. Covers every present and future auxiliary path, since the contract is the discriminator, not a per-path list.
  • mcode: title-gen's entire tool surface is the single submit_session_title tool; the prompt suggestion and the Goal evaluator are tool-less single-user one-shots reusing the main turn's headers (prompt-suggestion-service.ts:655-692). Lane gate x-mavis-session-id + shape. Real mcode main turns always carry the full tool surface, so they can never match.
  • goose: every auxiliary has exactly the dsh single-shot shape (no tools + one user + system carrier holding the context), so the existing isDshSingleShotSidecar covers it as-is; the lane evidence is the roster header agent-session-id.
  • CodeBuddy: the compaction request itself carries x-compaction on the wire — now consumed. Deliberately not x-context-compacted, which also appears on ordinary post-compaction main turns and would have demoted the real mainline.
  • opencode: the built-in title agent fires concurrently with the first main turn of every root session, under the root's own id, no tools, no budget, single user framed by a fixed template (session/title.ts). The constant User-Agent is the lane evidence and the template text the shape — this closes the pure-proxy hole described in the opencode v2 title-gen requests get injected with compression philosophy (misclassified by isTitleGen) #1699 note in side-request.ts (that defense required the plugin's header).

Untouched: the isSideRequest signature (all truth-table pins stay green), the dsh purpose contract (#6880 stays byte-for-byte), the demote gate, and every non-roster client path.

Tests

  • tests/issue6884-wire-client-sidecars.test.ts (new): pins each clause's shape on the lane signal (goes side: zero kernel requests, forwarded verbatim) and off it (control requests keep the full pipeline), plus two pure-predicate pins.
  • tests/e2e-session-identity.test.ts: the mavis fixture gains the tool surface a real mcode main turn carries (identity coverage unchanged).
  • tests/side-lane-invariants.test.ts: the boundary allowlist admits the three new gate symbols.
  • Full adjacent regression set (24 lane/identity files) run with a master baseline: failure sets identical to pristine master — the only failures are pre-existing local-environment ones (opencode host-marker test cluster). Typecheck green.

@github-actions

github-actions Bot commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: design/6884-wire-client-sidecars (3186e54)

Latest build — install this exact version

npm install -g billion-context@0.0.10743-pr.7083.b4b4ceb

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
3186e54 billion-context@0.0.10743-pr.7083.b4b4ceb build
687941b billion-context@0.0.10696-pr.7083.2a3d38e build
0c90fcb billion-context@0.0.10688-pr.7083.06cdd53 build
c6455c7 billion-context@0.0.10665-pr.7083.7e316d5 build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr7083.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei ranxianglei changed the title fix(lanes): demote the wire clients' auxiliary calls to the side lane (discussions/6884) [PR] fix(lanes): demote the wire clients' auxiliary calls to the side lane (discussions/6884) Nov 17, 2026
@ranxianglei ranxianglei added bug Something isn't working codebuddy goose grok-shell grok-shell (xAI CLI) host lane mcode opencode p2 scope/wiring labels Nov 17, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

查重: 不重复。 Open PR/issue sweep for the wire-client sidecar topic: #7083 is the only PR carrying these clauses. #7077 (pi purpose declaration) is the sibling lane this PR explicitly follows on — different mechanism (pi dispatch purpose vs. wire-client roster headers), not a duplicate. #5590 (sidecar stub ref-holding lane), #6475 (degenerate-turn gate), #3746 (threads-lane reforge) are adjacent side-lane work on different failure domains.

标签 (applied on pickup): bug p2 scope/wiring + host lanes codebuddy goose mcode opencode + grok-shell (label did not exist, created). 理由: request-pipeline lane classification for five roster clients' auxiliary calls — functional defect with a workaround (sessions still work, compression state is polluted), not an availability or cache-leak defect, so p2 not p1.

Status (verified live): PR open, mergeable: true, mergeable_state: behind (master advanced by #6996's merge — a rebase is needed before merge, not a conflict). CI on head c6455c7: static-gates green (ubuntu 24 + windows 22), test shards in progress (2 already green). Full review follows.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Review complete — verdict: mergeable. CI on head c6455c7 is fully green (36 success + 24 skipped, 0 failures, verified live). Master's advance since the PR base (#6996, src/agent/pi.ts only) does not overlap any PR file, so mergeable_state: behind is not a blocker under the merge-commit flow.

What I verified independently (not from the PR body):

  • Read the full diff (6 files, +496/−7) against the PR head; traced every new clause through both decision points — sideRequestLike (raw-key/anchor discipline, src/server/handle.ts:1315-1356) and sideIntent → resolveSideLane (src/server/handle.ts:2745-2770) — and confirmed the dsh 回归防护:泳道判定真值表 + 侧请求不变量 + dsh e2e lane + 运行时金丝雀 + lane-guard(5 项) #2170 side-effect canary (handle.ts:2848) covers the new clauses via sideIntent.
  • Confirmed no clause identifies a client from body shape alone: each shape predicate is pure (src/server/side-request.ts:513-620) and gated on the client's own wire signal (roster headers / x-compaction / constant UA) — the same trust tier as the roster in src/session-id.ts:96.
  • Confirmed x-compaction has no other consumer in src/ (no interaction), x-grok-turn-idx is consumed only by the new clause, and the roster already lists every lane-evidence header (x-grok-session-id, x-mavis-session-id, agent-session-id, x-conversation-id).
  • Ran the targeted set locally on the PR head: tests/issue6884-wire-client-sidecars.test.ts + side-lane-truth-table + side-lane-invariants + lane-decisions-matrix = 154/154 pass; issue5135-claude-title-sidecar + issue5656-single-shot-sidecar + e2e-session-identity + issue6880-purpose-declaration = 40/40 pass; npm run typecheck green.
  • Test validity: every clause is pinned in BOTH directions (gate on → zero kernel requests + verbatim-forward deepEqual; gate off → full pipeline), plus three pure-predicate pin sets; the mavis e2e fixture correctly gains the tool surface a real mcode main turn carries (a tools-less mcode body is now exactly the demoted shape). Deterministic (port 0, hermetic tempdirs).
  • Diff cleanliness: every line on-topic, no lockfile/version churn, no whitespace noise; code style matches the file's per-clause self-contained predicate convention (the isMcodeOneShotAuxRequest/isDshSingleShotSidecar loop overlap is the documented intentional difference — no system-carrier requirement).

结构化八问:

  1. 是什么问题:roster 上五个客户端(grok-shell/mcode/goose/CodeBuddy/opencode)都在主会话同一个 id 下发辅助 LLM 调用(标题、压缩摘要、提示建议、记忆、命名等),这些 one-shot 请求的 body 形状骗过现有 isSideRequest 启发式(无 tools + 预算缺失/过小/过大),进入主会话 kernel 管道当普通对话轮——快照 junk turn、usage 基线污染、统计膨胀。
  2. 什么原因导致的:side-request 分类只有 body 形状启发式这一个维度,这些辅助调用对启发式"不可见";根因是分类缺少"客户端自身 wire 信号"这个 lane 证据维度——roster 已经信任客户端盖戳的 session 头(同一信任级),但 side-lane 决策没有用它。
  3. 影响范围与严重程度:仅五个 roster 客户端的会话;修复前其辅助调用污染主会话压缩状态,主对话本身可用(功能性缺陷、有绕行)→ p2,非可用性/缓存泄露类。非 roster 客户端零影响(所有子句都 gate 在客户端自己的头/UA 上)。
  4. 怎么修复的:每个客户端一条子句,各自 gate 在该客户端自己的 wire 信号上(与 roster 同信任级):grok = x-grok-session-id 有 + x-grok-turn-idx 无(grok 自身契约:主轮恒盖 turn index);mcode = x-mavis-session-id + (单一 submit_session_title 工具 | 无工具无预算单 user one-shot);goose = agent-session-id + 既有 isDshSingleShotSidecar 形状;CodeBuddy = x-compaction(刻意不消费 x-context-compacted——它出现在普通压缩后主轮上,消费了会把真主线降级);opencode = UA opencode/ + 钉死的标题模板形状。形状谓词纯函数化放 side-request.ts,handle.ts 里 gate 后 OR 进 sideRequestLike 与 sideIntent 两个决策点。
  5. 是否有回归问题:逐项审过既有行为——isSideRequest 签名与 truth-table 钉死:未动,154/154 绿;dsh purpose 契约([Design][dsh] Stop guessing dsh's auxiliary requests — the in-process plugin can declare each request's purpose explicitly #6880):未动,测试绿;demote gate:未动;dsh title/review/single-shot、ccTitle、server-tool/structured-output/audio-output 各子句:全部保留;非 roster 客户端:六个新布尔全 false,行为字节级不变;session-id roster:仅注释变更(补充 x-grok-conv-id 在 Responses 辅助路径上带 TAGGED 值的事实);dsh 回归防护:泳道判定真值表 + 侧请求不变量 + dsh e2e lane + 运行时金丝雀 + lane-guard(5 项) #2170 canary 经 sideIntent 自动覆盖新子句。已披露残留(代码注释内):mcode/goose 零工具配置下第一个无工具主轮该轮走 side lane(第 2 轮起重放历史即回归全管道,[PR] fix: unknown dsh one-shot sidecars steal the raw-key anchor and kill compression (context-explosion root cause) #5687 同类);goose permission judge 带一个工具不被降级。最高风险子句是 grok(纯头、无形状校验):若某主轮缺 x-grok-turn-idx,该轮 verbatim 直通(轮次本身仍可用,只是该轮不进 kernel 跟踪);契约声明(turn.rs:3002-3003 恒盖戳)已在 discussions/6884 对照 grok-shell 1.0.34 源码核实,信任级与 roster 本身一致。
  6. 是否破坏缓存:否。side passthrough 不碰 kernel 状态;主会话 prefix cache 反而更稳(辅助 junk 不再进会话);被降级请求 verbatim 转发(测试 deepEqual 钉死),其自身 one-shot 缓存不受影响。
  7. 是否改了设计方向:否。按 [Bug] 0.1.209: Claude Code CLI main conversation forks onto <id>|sub:<fp>; the session-title call claims the raw key (regression from #4774) #5135(ccTitleSidecar)/dsh 0.2.1 Alpha.2更新了检查适配情况 #2503/[Bug][dsh] 0.1.204: auto-review side request collides with the main session → false native-compaction rebase → aux usage skips #453 clamp → permanent upstream 400 #4438/DSH: If the first request of a session originates from a bypass call, compression for the entire session is permanently | DSH: If the first request of a session originates from a bypass call, compress for the entire session is permanently inv | DSH:会话首个… #5656 的既有模式给 side-lane 机制加证据门控子句,无架构/数据流转向。
  8. 是否引入了新的配置:否。无新 config 字段/env/CLI flag;子句在客户端自身 wire 信号出现时无条件生效(与 roster 设计一致——roster 本身也没有 per-client 开关)。

收工判定:可合。 CI 全绿、定向回归全过、无回归发现、diff 干净;behind 状态不构成阻塞(master 前进只碰 src/agent/pi.ts)。合并权在你。

@ranxianglei
ranxianglei force-pushed the design/6884-wire-client-sidecars branch from c6455c7 to 0c90fcb Compare November 18, 2026 00:59
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Conflict resolved — verified independently, no push needed this run.

What I did this run:

  1. Live state check. Fetched master and the PR branch. The dirty conflict was in src/server/handle.ts: [PR] fix(pi): declare every pi dispatch as the session's main conversation turn #7077 (pi purpose declaration, now on master) and this PR ([PR] fix(lanes): demote the wire clients' auxiliary calls to the side lane (discussions/6884) #7083) both modify the same sideRequestLike / sideIntent decision points.
  2. Independent rebase + resolution. I rebased the PR onto master and resolved the two conflict hunks as a union of intent — kept master's declaredPiMainTurn const and its isSideRequest(parsed, declaredPiMainTurn ? "main" : …) form, kept the PR's six wire-client sidecar consts and disjuncts. Verified the result: typecheck green; 160/160 targeted lane/identity tests (issue6884-wire-client-sidecars + issue6884-pi-purpose-declaration + side-lane-truth-table + side-lane-invariants + lane-decisions-matrix) plus 20/20 e2e-session-identity.
  3. Found the remote head was already rebased. On the pre-push fetch the branch head had moved to 0c90fcb252 (force-pushed by a prior session before 01:03 UTC). git diff between my independently computed tree and 0c90fcb252 is empty — byte-identical resolution. So I did not push: the remote already holds the exact final state, and pushing would only churn the SHA and re-trigger a full CI pass for a zero-content change (one-full-CI-pass-per-branch discipline).
  4. Current live state (verified this run). PR open, mergeable: true, mergeable_state: behind — the dirty conflict is gone. Master advanced again to 1a8be5db3b ([PR] ⚠️ MODEL-FACING TEXT CHANGE: The ref snapshot line in the compression failure receipt is changed to a point-in-time anno | ⚠️ MODEL-FACING TEXT CHANGE: 压缩失败回执的 ref 快照行改为时点标注(弱模型把陈旧副本误读成数据损坏 → 7 连发幻影终点) #7078: CHANGELOG.md + src/stream.ts + one test file) — zero overlap with this PR's 6 files, so behind is not a merge blocker under the merge-commit flow (same precedent as the original review, floor 4). CI on 0c90fcb252 is in progress (mirror cache: 12 in progress / 5 success / 13 skipped); the live check-runs lookup currently 422s on that SHA (commit-index quirk right after a force-push), so the final CI verdict is pending — if any lane goes red, the next auto-fix wake handles it.

结构化八问:

  1. 是什么问题:PR 分支与 master 在 src/server/handle.ts 上有文本冲突([PR] fix(pi): declare every pi dispatch as the session's main conversation turn #7077 与 [PR] fix(lanes): demote the wire clients' auxiliary calls to the side lane (discussions/6884) #7083 改同一组 sideRequestLike/sideIntent 决策点),PR 处于 dirty 不可合并状态。
  2. 什么原因导致的:两个 跨 agent 核查:「session id = 唯一会话身份」这条定律,谁满足谁违反(#6880 背景) #6884 车道姊妹 PR([PR] fix(pi): declare every pi dispatch as the session's main conversation turn #7077 pi purpose 声明、[PR] fix(lanes): demote the wire clients' auxiliary calls to the side lane (discussions/6884) #7083 wire-client sidecar)并行开发在同一决策点行上,[PR] fix(pi): declare every pi dispatch as the session's main conversation turn #7077 先合入 master 后 [PR] fix(lanes): demote the wire clients' auxiliary calls to the side lane (discussions/6884) #7083 的 base 过期。
  3. 影响范围与严重程度:仅影响本 PR 的可合并性,无用户可见行为影响;冲突是两处决策点行及相邻 const 块的纯文本重叠。
  4. 怎么修复的:把 [PR] fix(lanes): demote the wire clients' auxiliary calls to the side lane (discussions/6884) #7083 rebase 到 master,按「意图并集」解冲突——同时保留 declaredPiMainTurn(pi 主轮声明,[PR] fix(pi): declare every pi dispatch as the session's main conversation turn #7077)与六条 wire-client sidecar 子句([PR] fix(lanes): demote the wire clients' auxiliary calls to the side lane (discussions/6884) #7083),两个决策点同时应用 pi "main" 档与新 sidecar 析取项。我的独立解与远端已有的 0c90fcb252 逐字节一致,故未再 push。
  5. 是否有回归问题:无。合并是两个独立评审、独立测试变更的纯并集,无语义交互(pi 子句 gate 在 x-bili-plugin: pi 头上,wire-client 子句各自 gate 在 roster 头/UA 上,同一请求不可能同时命中两类);typecheck + 180/180 定向车道/身份/e2e 测试全绿。
  6. 是否破坏缓存:不涉及——未触碰任何缓存/重试/去重逻辑。
  7. 是否改了设计方向:不涉及——rebase 完整保留两个 PR 各自评审通过的设计行为。
  8. 是否引入了新的配置:不涉及。

收工判定:冲突已解,PR 现为 mergeable(behind,非 dirty),等 0c90fcb252 的 CI 跑完即具备合并条件;本轮未 push(远端已在验证过的终态)。

… (discussions/6884)

Source-review audit (discussions/6884, five client codebases cloned and
read end to end) found that the clients on the verbatim session-id roster
each mint ONE id per logical conversation and the lifecycles are healthy,
but every one of them fires auxiliary LLM calls under the SAME id --
titles, compaction summaries, prompt suggestions, memory, naming -- and
those shapes defeat isSideRequest, so today they enter the main
session's kernel pipeline as ordinary conversation turns.

Each new clause is gated on the client's OWN wire signal, the same trust
tier as the roster itself; no clause identifies a client from body shape
alone:

- grok-shell: main turns ALWAYS stamp x-grok-turn-idx (turn.rs); every
  auxiliary path stamps x-grok-session-id WITHOUT it (grok's own
  test-support contract). Session header present + turn index absent
  -> side. Covers every present and future auxiliary path.
- mcode: title-gen's whole tool surface is the single
  submit_session_title tool; the prompt suggestion (default on) and the
  Goal evaluator are tool-less single-user one-shots. Lane gate
  x-mavis-session-id + shape (no tools + exactly one user). Real mcode
  main turns always carry the full tool surface, so they never match.
- goose: every auxiliary is the dsh single-shot shape, so
  isDshSingleShotSidecar covers it as-is; the lane evidence is the
  roster header agent-session-id.
- CodeBuddy: the compaction request itself carries x-compaction on the
  wire -- consume it. Deliberately NOT x-context-compacted, which also
  appears on ordinary post-compaction main turns.
- opencode: the built-in title agent fires concurrently with the first
  main turn of every root session, under the root's own id, no tools,
  no budget, single user framed by a fixed template. The constant
  User-Agent is the lane evidence and the template text the shape
  (closes the pure-proxy hole the #1699 note in side-request.ts
  describes; that defense required the plugin's header).

Untouched: the isSideRequest signature (truth-table pins stay green),
the dsh purpose contract (#6880 stays byte-for-byte), the demote gate,
and every non-roster client path.

Tests: tests/issue6884-wire-client-sidecars.test.ts pins each clause's
shape on and off the lane signal plus two pure-predicate pins; the
e2e mavis fixture gains the tool surface a real mcode main turn
carries; the side-lane boundary allowlist admits the three new gate
symbols.
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Red CI resolved — both failures were runner-load flakes, rerun is green, no code change needed.

What I did this run:

  1. Re-verified live state first. Head still 0c90fcb252 (no [ework:ci-held] marker, no new commits). Mirror cache confirmed exactly the two red checks: test (windows-latest, 22, 4) and test (windows-latest, 22, 3) on run 44212326647.
  2. Fetched both job logs and identified the exact failing tests:
    • Shard 3: #5955 processBirthStamp: own pid resolves to a stable non-empty stamp (tests/issue5955-wedged-same-lane-takeover.test.ts:327) — assertion "birth stamp available on this platform", test duration 20288 ms = exactly 2× the 10 s powershell budget (src/launcher.ts:5148): both Get-Process cold starts timed out and the best-effort stamp degraded to undefined. The same run shows the runner under heavy stress: ENOSPC: no space left on device on dump writes, EPERM AV/Defender write refusals, and the runner's own flake-classifier had already tagged the phase-1 failure of shard 4 as the 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 teardown-lock family.
    • Shard 4: a held tail in the just-sealed generation survives the deferred reclamation (tests/log-retention-followup.test.ts:183) — "one still-pending drain must keep the retention gate closed". The test's premise ("only the callbacks open the pendingRetentionDrains gate") predates 18af0c6142 (settle the retention drain gate on stream close, not only end()): the gate now ALSO settles on the stream close event, which the test does not hold. Under load, one stream's close fires during the test's await and the counter reaches zero on the first callback release → the retention pass runs "early" from the test's point of view. A pre-existing timing race in the test, load-sensitive.
  3. Proved neither is a PR regression. The PR diff touches only src/server/handle.ts, src/server/side-request.ts, src/session-id.ts + 3 lane test files — zero overlap with src/logger.ts, src/launcher.ts, or either failing test file. Both failing test files pass locally on this PR head: log-retention-followup 4/4, issue5955-wedged-same-lane-takeover 18/18.
  4. Rerun outcome. A rerun of the two failed jobs was already in progress when I checked (started 01:39–01:40Z, before my own rerun attempt at 01:48Z, which the API rejected with "This workflow is already running" — so it was triggered externally, not by this session). It completed green: run 44212326647, attempt 2, conclusion success — both shards passed on the identical head with zero code changes, confirming flake.
  5. No push performed. Nothing to fix in code; pushing would only burn this branch's one full CI pass.

结构化八问:

  1. 是什么问题:PR 的两个 windows-22 测试分片(3、4)红了,阻塞合并。
  2. 什么原因导致的:不是代码问题,是 Windows runner 高负载下的测试 flake——①分片 3:processBirthStamp 的 powershell 冷启动超过 10s 预算(耗时恰好 2×10s,两次都超时,机制按设计降级为 undefined,测试却硬断言可用性);同一 run 里 runner 磁盘 ENOSPC、AV 写拒绝,负载证据充分。②分片 4:log-retention 测试的中间断言与 18af0c6142 新增的「close 事件也结算 drain 门」路径存在时序竞争——测试注释「只有回调开闸」已过时,负载下 close 事件提前触发即现形。
  3. 影响范围与严重程度:仅本 PR 的 CI 信号,无用户可见行为影响;两个测试在健康 runner 上均通过,属间歇性(负载敏感)而非确定性失败。
  4. 怎么修复的:无需改代码——核实两红均为 flake(零 diff 重叠 + 本地 22/22 定向通过 + 重跑同一 head 全绿)后,由已在进行中的 failed-jobs 重跑收敛;未 push 任何 commit。
  5. 是否有回归问题:无——未改任何代码;两红与 PR 的 6 个文件零重叠,且重跑(同 head、零改动)转绿,反证非回归。
  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。

遗留记录(已同步 #4599 总控):两个测试的负载敏感性签名已留档——processBirthStamp 的 10s powershell 预算在高负载 runner 上偏紧(测试对 best-effort 机制硬断言可用性);log-retention held-tail 测试需适配 close-settle 路径(测试设计修正,非本 PR 范围)。两者若再红,先查 runner 负载再怀疑代码。

收工判定:CI 全绿(run 44212326647 attempt 2 success),PR mergeable(behind,非阻塞),具备合并条件。

@ranxianglei
ranxianglei force-pushed the design/6884-wire-client-sidecars branch from 0c90fcb to 687941b Compare November 18, 2026 01:56
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] auto-fix stalled: test (windows-latest, 22, 4), test (windows-latest, 22, 3) — stopping for human review

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Status confirmation — work is done and fully verified (full 八问 analysis is in my previous comment above).

Verdict: ready to merge.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

CI triage on new head 687941b83d — the one red shard is an INHERITED MASTER RED, not this PR's fault. No code change, no push (policy: master-level fixes converge to ci-fixes via the sentinel, never duplicated on PR branches).

What I verified this run (all live):

  • Run 44219582379 (head 687941b83d, the rebase onto new master 0e475ceb4f8): 26/27 jobs green, the single red = test (windows-latest, 22, 5), job 134553269196.
  • Fetched the job log: the failure is tests/server-process-listeners.test.ts, and the harness's own escalation says it is deterministic: phase 1 (concurrency 4) → isolated serial re-run → "REPRODUCED in isolation (deterministic break)" → full retry → "genuine failure, not a flake".
  • Root cause: the test's 4th case (tests/server-process-listeners.test.ts:127) simulates Ctrl+C with process.kill(process.pid, "SIGINT"). On Windows, Node implements process.kill via TerminateProcess — the signal parameter is ignored and the process is killed outright (exit=1, no stderr, buffered TAP output lost). The test is written for POSIX signal-delivery semantics; every ubuntu shard is green. The test and the code under test came from master — [PR] batch-2: 4 held Afloat16 fixes (2 as-is, 2 with review-patch commits) #7101, commit 894835ac94 — not from this PR.
  • Inherited, proven: master's own CI run 44222455549 on 0e475ceb4f8 (this PR's parent) fails on the same test, same signature, same shard on both windows-22 and windows-24 (jobs 134527565567 / 134527565542).
  • PR diff (6 files: src/server/handle.ts, src/server/side-request.ts, src/session-id.ts + 3 test files) has zero overlap with the failing test or the process-listener code.
  • ci-fixes (head 7247410976) is currently an ancestor of master — it does not yet contain a fix for this, so merging it is a no-op; there is nothing to push.
  • Reported the full signature + suggested fix (platform-gate the test case on win32; a child-process variant is not viable on Windows either) to 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 to pick up.

State: the PR itself is clean — 26/27 jobs green, the one red is inherited from master and will stay red for every PR rebased onto 0e475ceb4f8 until the fix lands on ci-fixes/master. Once it does, this branch just needs a rebase + rerun. This lane will not touch the branch in the meantime.


八问(中文,供人工 reviewer):

  1. 是什么问题:新 head 上 windows-22 shard 5 的测试车道红,挡住合并。
  2. 什么原因导致的:不是本 PR 的问题——master 自身红。[PR] batch-2: 4 held Afloat16 fixes (2 as-is, 2 with review-patch commits) #7101 合并(894835a)引入的测试 tests/server-process-listeners.test.ts 在 Windows 上确定性挂:第 4 用例用 process.kill(process.pid, "SIGINT") 模拟 Ctrl+C,而 Windows 上 Node 的 process.kill 走 TerminateProcess(signal 参数被忽略),测试进程当场被杀(exit=1、无 stderr);测试按 POSIX 信号语义编写,ubuntu 全绿。PR diff(6 文件)与该测试及其被测代码零重叠。
  3. 影响范围与严重程度:所有 rebase 到新 master(0e475ce)的 PR 都继承此红(master 自身 CI 在 windows-22/24 shard 5 双红);仅 CI 信号,无用户可见影响。
  4. 怎么修复的:PR 分支未做修复——按 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 CI 协调政策,master 级红必须收敛到 ci-fixes(哨兵统一维护),禁止分支重复修;ci-fixes(7247410)目前是 master 祖先、尚不含此修复(merge = no-op,无可推);完整失败签名 + 根因 + 建议修法(win32 平台门控该用例)已报 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 待哨兵接手。
  5. 是否有回归问题:无——未改任何代码;PR run 其余 26 个 job 全绿。
  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:红为继承性 master 红(同测试、同签名),PR 本身干净;等哨兵把修复收敛到 ci-fixes/master 后 rebase + 重跑即绿,在此之前本车道不碰 PR 分支。

…led its own child on win32)

test 4 (added by 894835a, #7026 review) assumed POSIX signal
semantics: process.kill(self, SIGINT) was expected to dispatch to the
installed handler and drive the graceful-shutdown chain. On WINDOWS the
same call terminates the process (TerminateProcess) WITHOUT running the
handlers - Node docs: 'On Windows, the signal being used to terminate a
process cannot be changed. SIGTERM is sent.' Verified on a win10 box:
SIGINT and SIGTERM self-kills die the process with the handler never
invoked. The windows-latest CI child therefore died at the first kill
line: file-level failure, zero TAP output, deterministic.

Invoke the installed handler directly on win32 instead (the identical
shutdown() path a console Ctrl+C would deliver). Also: bounded wait for
the stubbed exit instead of a fixed 500ms sleep (each flush stage is
deadline-capped at 5s, so a fixed sleep races slow volumes on every
platform), and keep the exit stub in finally while the chain is still
pending (last test in the file; a late REAL exit(0) would kill the
runner before the TAP summary).

Verified: 4/4 pass on a real win10 box (node v22.23.2), 4/4 on Linux.
@ranxianglei
ranxianglei merged commit 548e396 into master Nov 18, 2026
52 checks passed
@github-actions

Copy link
Copy Markdown
Contributor

📦 Merged Build Published

This merge to master was published as 0.0.1231-master.548e396 (rolling master tag).

npm install -g billion-context@0.0.1231-master.548e396  # this exact build
npm install -g billion-context@master  # always the newest merged build

npm view billion-context@master prints the current version at any time.

ranxianglei pushed a commit that referenced this pull request Nov 18, 2026
ranxianglei pushed a commit that referenced this pull request Nov 18, 2026
…lict at 0.0.200

Master advanced past the PR base again (batch11 wave11 #6827, lean-compress
receipt #7105, wire-client-sidecars #7083, p1 batch b2/b3, #7087, #7101 ->
kernel 0.0.199, storage-render goldens regenerated for #6775 persisted
fields, 400 files). Content hunks auto-merged (kernel/src/compress.ts,
kernel/src/types.ts, all four storage-render goldens); the only conflicts
were the three kernel version lines, resolved at 0.0.200 = master's 0.0.199
+ this PR's kernel/src change, per the CI-enforced kernel version discipline.

Targeted local verification: tests/golden-storage-render.test.ts 5/5,
tests/storage-state-fields.test.ts 2/2, kernel issue6116/issue3769/
hide-consumed suites 27/27.
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