Repository navigation
service-messaging: a late HTTP ack from a REAPED claim overwrites the live re-claim on sys_http_delivery — IHttpOutbox.ack(id, …) carries no claim credential (the notification outbox fixed this shape in #11859) #17634
Description
现象
sys_http_delivery 上，一个已被回收（reap）的认领的迟到 ack，会覆盖另一个节点正在进行的重新认领。
来源：#17623 开发席位在 PR #17632 期间的一次性探针（针对构建后 dist 的 MemoryHttpOutbox）：
- 节点 A 领取一行，开始发送；A 的认领超过
claimTtlMs，被 reap 回pending - 节点 B 重新领取这一行：
in_flight，claimedBy = node-B，正在发送 - A 的发送结束，迟到的
ack落下 ⇒ 该行变成dead、claimedBy = null、attempts = 1——而 B 还在发送
原因（读码，origin/main）
packages/services/service-messaging/src/sql-http-outbox.ts：
async ack(id: string, result: HttpAckResult): Promise<void> {
… where: { id } … // 只按 id 写，不校验认领者IHttpOutbox.ack(id, result)的签名不携带认领凭据SqlHttpOutbox.ack的写条件只有{ id }（经dispatcherAckOptions）MemoryHttpOutbox.ack无条件写
通知侧早已修过同一形状：sql-outbox.ts 的 ack(claimed: ClaimedDeliveryRecord, result) 以 (claimedBy, claimedAt) 作为认领凭据做 compare-and-set，读写两半都校验归属（#11859），并对未认领行的 ack 做前置拒绝（#11453）。HTTP outbox 从未获得这层保护。
后果
- 一次正在进行、可能成功的投递被标成
dead；或者两次投递的结果互相覆盖，最终状态与实际发送结果不一致 - webhook 可能重复发送，或该送达的被错误记为失败
- 触发条件是「一次发送耗时超过
claimTtlMs，同时另一个调度器实例在轮询同一张表」。多节点部署天然满足后半条；单容器内若同一环境的旧 kernel（stale/draining）与新 kernel 同时运行各自的调度器，推测也可能满足——这一点未实测
既有缺陷：PR #17632 既未扩大也未修复它。
验收
IHttpOutbox.ack携带认领凭据，形状与通知侧ClaimedDeliveryRecord对齐（新增为兼容形式，不破坏既有调用方——参照check-adr-0087-registration对契约成员的要求）SqlHttpOutbox.ack：写条件含id+claimed_by+claimed_at+status = 'in_flight'；不匹配时不写，并以warn记录「认领已不再持有」（与通知侧ack refused, claim no longer held同口径）MemoryHttpOutbox同样校验- 测试复现上面的三步序列：迟到 ack 不改变 B 的认领；B 的 ack 正常生效
Activity
hotlong commentedon Sep 11, 2026
Claim: maintainer direct dispatch (cloud ops session, incident follow-up)
Session:
c5c0ce54-bb9c-478c-9e5b-cf44b80d4569Time: 2026-09-11T06:40Z
Branch:
claude/issue-17634-http-ack-claim-credentialWorktree:
~/Documents/GitHub/objectstack-issue-17634Domain:
domain:servicesFile surface:
packages/services/service-messaging/src/http-outbox.ts(ack contract),sql-http-outbox.ts,memory-http-outbox.ts,http-dispatcher.ts(ack call site) + testsSerial constraint cleared: PR #17632 (same files) merged at
a9096af48.Clause-②: no — HTTP outbox ack ownership check inside
service-messaging; nopackages/spec/src/**surface expected (path limb re-checked on the actual diff at review)hotlong commentedon Sep 11, 2026
os-dev-report
{
"issue": 17634,
"status": "done",
"branch": "claude/issue-17634-http-ack-claim-credential",
"pr": "#17641",
"premise_still_valid": true,
"summary": "Premise reproduced on origin/main 8f751cd (contains a9096af) for both stores before any fix. IHttpOutbox.ack gains an optional third parameter, claimed?: HttpClaimCredential ({claimedBy: string, claimedAt: number}, the pair ClaimedDeliveryRecord guarantees). Handed it, SqlHttpOutbox.ack and MemoryHttpOutbox.ack refuse unless the row is still in_flight under that pair: read-half refusal, then the same test re-stated in the write via dispatcherAckCasOptions (where id + status in_flight + claimed_by + claimed_at, updateMany), then a (status, attempts) read-back; a refusal writes nothing and throws the new HttpAckError (existing code DELIVERY_NOT_ELIGIBLE). HttpDispatcher.ackAttempt always passes the claimed row, absorbs only that code and warns 'http-dispatcher: ack refused, claim no longer held'. Compatible form chosen over a new optional ackClaimed() method (a subclass overriding a built-in store's ack() would silently stop seeing dispatcher acks; TickRecordingOutbox in this repo is one) and over a record-typed first parameter (not assignable from existing (id: string) implementations, or an object handed to them at run time); claim() keeps declaring HttpDelivery[] because narrowing it breaks subclass overrides (TS2416). Changeset minor, not declared breaking. No packages/spec/src/** in the diff and no spec accept/reject change (Clause-2 stays no). Tenant-audit census regenerated 222 to 223 plus 8 hand-written prose figures; webhooks.mdx section 12 gains a failure-mode row. Deviation noted: the PR body's first line is 'Closes #17634' as this dispatch specified, where os-dev.md prescribes 'Fixes #17634' (same closing effect). Assignee untouched; PR labels were auto-applied by the labeler (documentation, size/l, tests, tooling), none added by this seat.",
"tests": "BASE RED: repro test committed alone as f945fa8 (non-test diff vs 8f751cd empty); vitest run src/http-outbox-ack-claim-ownership.integration.test.ts: 2 failed | 2 passed (4), 'AssertionError: expected dead:-:-:1:410 to be in_flight:node-b:1060001:0:-' on MemoryHttpOutbox and on SqlHttpOutbox (real ObjectQL + SqlDriver better-sqlite3), negative controls green. FIX (final head b373361, origin/main merged at 2407046): new file 16/16 (dispatcher legs: card replay, no-reap control, two-parameter legacy outbox; store legs: card replay, control, same-node re-claim, reaped-not-reclaimed retry ack, incomplete credential; refusals assert name HttpAckError + code DELIVERY_NOT_ELIGIBLE); delivery-update-tenant-audit pins the HTTP ack on updateMany with where.id scalar + status in_flight + claimed_by n1, bypassTenantAudit true, zero by-id updates, updateMany positive control. pnpm --filter @objectstack/service-messaging build exit 0 (check-dts-emitted 2/2), typecheck exit 0 (tsc --listFiles includes the 3 touched test files), test 40 files / 423 passed. ABLATION A (on committed fix 06215b0, trap-restored): dispatcher stops passing the credential (removed text 1->0, injected 0->1) -> 2 failed | 14 passed (16), both card-replay dispatcher legs red with the base assertion; restore proven by HEAD blob f8e5db21f == hash-object after, git diff HEAD on the path empty. GATES on b373361: dispatch-gates --commands 94, all run with exit codes; --ran: '94 derived famil(ies) accounted for — 93 run, 1 NOT-MEASURED (1 DERIVED from a recorded exit 3)'. 92 exit 0 incl. check-tenant-audit-census OK 223 sites / 23 prose figures, check-adr-0087-registration 'this PR adds no declared-breaking changeset', check:doc-authoring (first sweep caught tracker ids in two HttpAckError strings; fixed in b373361), check:dispatcher-error-vocabulary, check:engine-double-contract, check:nul-bytes, check:cross-package-test-inputs, check:published-files, spec check:docs, check:skill-examples and check:i18n (green after the named 58-task prerequisite closure), check:type-check-debt ('re-measure: OK … none above its recorded number'). NOT MEASURED: check:dual-build-cjs-loads, exit 3 PREREQUISITE NOT MET (needs dist of every package, a workspace-wide build this seat does not run). RED NOT FROM THIS DIFF: check:merge-driver exit 1 (self-test case 'a gate whose RUNNER is not installed refuses the same way' + diagnosis leg); control leg in a detached origin/main 76c9fab worktree on this host exits 1 with the identical cases; worktree removed. os-verify-lock ran in declared UNLOCKED mode (no flock on host). CI on the PR not waited on.",
"mcp_calls": "0 — GitHub reads and writes went through the gh CLI",
"open_questions": [],
"out_of_scope_findings": [
"reported for PM to file (not filed by this seat): pnpm check:merge-driver self-test exits 1 on this macOS host at origin/main 76c9fab itself — failing case 'a gate whose RUNNER is not installed refuses the same way' and its 'naming the command the shell could not find' leg (node scripts/git-env.mjs --self-test && node scripts/git-merge-regen.mjs --self-test && node scripts/check-regen-pending.mjs --self-test); repro: that command in a detached origin/main worktree",
"noted, not filed: packages/spec/src/api/error-code-ledger.zod.ts prose under DELIVERY_NOT_ELIGIBLE lists its throw surfaces (IHttpOutbox.redeliver, INotificationOutbox.ack) and does not yet name IHttpOutbox.ack with a credential (HttpAckError); left unedited to keep spec out of the diff, no gate reads it; picked up by: none scheduled",
"noted, not filed: the two-argument IHttpOutbox.ack arity stays unchecked by design for compatibility; retiring it and making claimed required is next-major work; picked up by: none scheduled",
"noted, not filed: the (status, attempts) read-back that reports a missed conditional write is blind to one interleaving (reaped, re-claimed and acked by another node to the same status and attempt count between read and write) — identical to SqlNotificationOutbox.ack; not reproduced; picked up by: none",
"noted, not filed: HttpDispatcher still calls onAttempt after a refused ack (unchanged; no production subscriber in this repo)",
"noted, not filed: delivery-update-tenant-audit.integration.test.ts header still describes the redeliver assertions as a spy on SqlDriver.update although they moved to updateMany in #11009 (pre-existing prose drift)"
]
}
Generated by Claude Code
hotlong commentedon Sep 11, 2026
PM 收尾（Session c5c0ce54）：已落地
4be4e049b。HttpDispatcher确认投递时总会带上领取凭据（claimedBy+claimedAt）。SqlHttpOutbox和MemoryHttpOutbox只在这条投递仍由同一次领取持有时才写入结果；被回收后迟到的 ack 会被拒绝，并记一条 WARN。ack(id, result)行为不变，已标记为 deprecated。范围外：
check:merge-driver在 macOS 上自测失败，已由 #16717 跟踪。