Repository navigation
[PR] release v0.1.209 - #4982
Conversation
(ai: vllm-qwen/qwen3.8-27b) 查重: 本 PR 不是重复修复,但发现一个需要你知晓的情况 —— v0.1.209 有两个并行的 release PR:
已核实这两块内容 master 上都不存在(master 的 release-notes 仍停在 0.1.203 critical;该测试文件还没有 retrieve case)。只合 #4982 会丢掉它们 —— ≤0.1.204 用户升级后看不到「大存储被 0.1.207 弄坏」的 critical advisory。 建议(等你拍板):先把 #4838 那两个内容 commit port 进 #4982(或先落 master),再合 #4982,#4838 之后可关(我不能关 PR)。或者 rebase #4838 解冲突后合它——都行,推荐前者(冲突面小)。 标签: pr + scope/ci + p3(例行发版审核单,构建/发版域,非缺陷)。 审计已开始:窗口 = v0.1.206..本 PR head,477 commits / 466 files(v0.1.207/.208 是 lane/应急车道发布,其修复已分别落 master,本窗覆盖全部新增)。按政策分段执行:diff --stat 总览 → 高危域(自更新/dsh/lockfile/systemd/网络路径)逐行精读 → 其余文件按目录分批通读 → 逐项回归检查 → 行为变更清单。完成后按八问回复。 |
📦 Built Package ArtifactBranch: Latest build — install this exact versionnpm install -g billion-context@0.0.6658-pr.4982.e45b04dThis exact version is immutable and reproducible. The rolling Recent builds (commit → version)
Download artifact
tar xzf billion-context-pr4982.tgz
npm install -g packageInstall instructions are refreshed on each push. |
回归 Review 结论(v0.1.208 → v0.1.209,458 files / +51122 −5491)总体:1 个 MAJOR(fork clone 顺序,建议修复后再发版);其余无阻断性回归。 方法:5 路并行深审(kernel/压缩核心、dsh/客户端 lanes、存储、server/wire/缓存、web/CLI/MCP)覆盖全 diff(非只看 release 条目)+ bili-gate 真实模型回归 + 关键链路人工逐点复核。 🔴 MAJOR:fork clone 顺序使 #4316 PR3/3「number-preserving clone」在生产中完全失效链路已逐环验证(每行都在本分支确认过):
后果:
修复方向(供参考):把 clone 移到 publishForkSession 的 flush 之前(S12a 契约本来就是 clone-first;注意 ingest 与 clone 的交互 —— UNIQUE(session_id,msg_id,content_hash) + delta refusal),或把 guard 改成与「期望的 parent 行数」比较而非仅判存在。另注意:naive 改顺序会激活下面的 latent minor(lineage 无条件覆写)。 🟡 Minor(建议本 release 或紧随其后处理)
🟠 默认行为变化(均已文档化、可回退;请确认意图)
✅ 回归核查要点(未发现问题的部分)
🧪 bili-gate 真实模型回归结论:零产品回归证据。 所有红项均机械归因(上游饱和 ×2、窗口/模型标定 ×2),且 Lane A 的 9 项断言在多次运行中全部拿到过绿证据(明细见下)。
🔧 review 期间发现并已修复
本地测试证据(pr4982 分支,均绿)served-markers 11/11(+split-brain-surface 5/5)、pi-msgids 12/12、opencode-msgids 5/5、omp-msgids 6/6、msgid-lane-flow 6/6、dsh-msgids 6/6、native-intercept 66/66、issue4516 13/13、web-blogfeed / web-blog-route / web-i18n / web-blog-refresh / visibility-markers 39/39、issue4460、issue4666-migration-dedup、storage-state-fields parity、seq-ledger-pr3、ccr-restamp、seq-interval-* 等。 |
(ai: vllm-qwen/qwen3.8-27b) Merge guard sync done: merged origin/master ( |
Unblocks PR #5004's test (windows-latest, 22, 2) red on head da25207: three abort cells (proxy-openai-sse / proxy-responses-sse / proxy-json) asserted closed=0 after exhausting the flat 3s (30x100ms) poll on a loaded Windows runner -- the same signature already recorded under #4599 for #4982/#4774/#4683 (rerun green / in-branch budget bump). Sibling cell abort x proxy-anthropic-sse passed in 176ms on the same shard and every other windows/ubuntu shard ran this file green on the same head, so this is runner load, not a code regression; the PR's diff touches none of these paths. Labeled CI-unblock commit; duplicates #4683's pending identical change -- drop this commit when rebasing if #4683 lands first.
fix: draining proxy no longer claims its freed port — same-lane newcomer ladders instead of dying twin-collision (#4982)
|
review代码 |
…n misclassification no longer kills same-lane newcomers (ranxianglei#4982) A proxy entering graceful shutdown releases its listening socket immediately but kept its registry marker until the drain finished (seconds-to-minutes later). During that window the marker still claimed the port while a different instance could bind it; a same-lane newcomer hitting EADDRINUSE then misclassified the draining predecessor as its twin and died with the adopt signal (exit 3), stranding its launcher behind a cross-lane stranger it is forbidden to attach to (ranxianglei#1225). CI hit this in the ranxianglei#3203 concurrent-sessions e2e (release v0.1.209, run 40574417718). Two complementary fixes: - shutdown() stamps portReleasedAt on its own marker the moment the listener closes; classifySameLanePeer skips stamped entries, so a draining instance can never be mistaken for the port holder again (predecessor-wait consumers keep counting them, preserving ranxianglei#3044/ranxianglei#1723). - the late EADDRINUSE path now verifies the claimed twin's socket ownership via /__bili/health (pid + instance identity, the ranxianglei#4341 criterion) before declaring twin-collision; an unverified claim falls through to the plain foreign-holder policy and ladders. tests/issue4982-draining-twin-ladder.test.ts pins both halves end to end: a real draining predecessor + silent squatter (newcomer must ladder, not die) and an unstamped marker variant (verification demotes the claim, logged as 'no longer answers as itself').
…te in LIVE_FILES (#4982)
[bot] 进度一行(自动修复车道 attempt 2/2):ci-fixes(哨兵 LIVE_FILES 登记修复 02ef0d1)已 merge 进 release 分支并推送 sha= |
[bot] 进度一行:新 run( |
…anxianglei#4626) Resolution notes (3 conflicted files): * src/served-markers.ts — UNION of the two independent fixes for the same root cause (ranxianglei#4626 false split-brain on graceful handover): master's ranxianglei#4982 line stamps stoppedAt on dispose(), ranxianglei#4630 stamps disposedAt. dispose() now sets BOTH (stoppedAt=Date.now() then writeMarker(true)); reader-side contestedServedAt treats either stamp as known-dead, so markers written by old builds (stoppedAt only) and new builds (both) are mutually recognizable across versions. d-4630-only capability kept in full: peerStillContests fail-closed three-state check + runtime-status self-clearing of fired split-brain verdicts when the contesting peer provably stopped. * src/server.ts — import-line unions (peerStillContests, registerSplitBrainEvidenceCheck added on top of master's existing list); registration call already landed conflict-free at startServer. * tests/served-markers.test.ts — ALL FOUR graceful-stop tests kept (2x stoppedAt family asserting body.stoppedAt / alive-pid skip, 2x disposedAt family asserting tombstone shape + e2e rotation silence) plus display-fix tests in admin/web; each branch's assertions pass against the union implementation because both stamps are now always co-written.
…rsistence adopt-exit (option A) Ruling (owner-approved 2026-10-31): for an unverified unstamped same-lane claim the early pre-persistence gate (#4977) wins over #5016's late ladder. The second row of tests/issue4982-draining-twin-ladder.test.ts now asserts the code-3 twin-collision adopt-exit BEFORE index.db is opened instead of ladder-off; the draining (portReleasedAt-stamped) row keeps asserting the ladder. Fixture cleanup hygiene was already finally-wrapped upstream, nothing to add. Refs #5018 #5004 #4982 #5016
Auto-fallback from the one-click release workflow: direct push to master was blocked by branch protection. Merging this PR publishes via the standard release.yml flow.
Changes since v0.1.206
contentnext to top-level startId/endId — plus U+2212 dash silently folds the wrong range #4252 follow-up)autoFoldTargetTokenscauses a fold storm, andexternalSummary.*file edits do not take effect until aPUT /__bili/config#3444 benefit gate (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 sentinel)testshard is red —tests/issue3890-parent-gone-confirm-drain.test.tsspawns processes but is missing from LIVE_FILES #4604)unused-exportsis red —ClientBuildCheckis exported but only used inside its own file #4602)unused-exportsis red —ClientBuildCheckis exported but only used inside its own file #4602)unused-exportsis red —ClientBuildCheckis exported but only used inside its own file #4602/[PR] fix(ci): clear the two master-red gates in one branch — unused-exports (#4602) + the #3890 LIVE_FILES tripwire (#4604) #4606, same one-liner as [PR] fix(ci): clear the two master-red gates in one branch — unused-exports (#4602) + the #3890 LIVE_FILES tripwire (#4604) #4606 commit 1)unused-exportsis red —ClientBuildCheckis exported but only used inside its own file #4602, same one-liner as [PR] fix(ci): drop the unused export of ClientBuildCheck #4603/[PR] fix(pi): inline mode now intercepts codex Responses WebSocket traffic (#4526) #4559)tool_callsarray | MITM/流式代理: 上游 chunk 携带空 tool_calls 数组时第一轮文本被静默丢弃 #3926)autoFoldTargetTokenscauses a fold storm, andexternalSummary.*file edits do not take effect until aPUT /__bili/config#3444)providers.<url>.models."<model>".outputcan never be read for path-style lane requests; the output ceiling remains unk | providers.<url>.models."<model>".output 对路径式 lane 请求永远读不到,output ceiling 始终 unknown(#3438 警告无法消除) #4501)choicesframe with an empty delta #4488)#2383Structured digest produces "zero-conclusion" compression blocks under extreme overflow: one block = one messag | ##2383结构化 digest 在极端溢出下产出「零结论」压缩块:一块 = 一条消息一行、每请求只折一块,87 块常驻 22.9% 窗口(DSH · deepseek-flash · v0.1.201/0.1.202) #4224 structure line merges tool results with their calls under one verb#2383Structured digest produces "zero-conclusion" compression blocks under extreme overflow: one block = one messag | ##2383结构化 digest 在极端溢出下产出「零结论」压缩块:一块 = 一条消息一行、每请求只折一块,87 块常驻 22.9% 窗口(DSH · deepseek-flash · v0.1.201/0.1.202) #4224 emergency digest quality + convergence (multi-fold, dynamic trim, structural conclusion)stripAcpTagson the same text will additionally remove the dead refere | 标签清理不幂等:同一文本第二次 stripAcpTags 会额外删掉死引用闭合(m3117</cPa) #4106): pin the pass-cap residue corner; state the bounded fixed-point guarantee in the docstring (review finding)stripAcpTagson the same text will additionally remove the dead refere | 标签清理不幂等:同一文本第二次 stripAcpTags 会额外删掉死引用闭合(m3117</cPa) #4106): make tag cleanup idempotent — input-side gate attestation + fixed-point walk<transcript>appears in the history, compression for the entire session is | isAutoModeClassifier() 误判:历史中出现 <transcript> 后,整个会话的压缩被静默跳过 #3908)<transcript>appears in the history, compression for the entire session is | isAutoModeClassifier() 误判:历史中出现 <transcript> 后,整个会话的压缩被静默跳过 #3908)