结果:owner_crash_takeover 的 interrupted 轮询改为有界等待 + 显式失败(点名未观察到的 run 状态序列、轮询实例、实际耗时与预算、实例日志),未收口时不再落到后续 POST 的误导性 409;restart_invariants 在断言消息数之前新增「恢复视图」有界等待(存在 interrupted run 且无 pending/running/waiting_hitl),未恢复时给出点名判据的失败信息;其余 t.Fatal→t.Fatalf 仅补失败信息,断言条件未变。新增 runStatuses 与 e.instanceOf 两个小helper 供等待条件与失败信息复用。
验证:PASS——真实 PostgreSQL(127.0.0.1:54331,fsync=on)下空闲机器 15.1s 通过、96 忙循环压满 32 核 17.4s 通过(新增失败路径未误触发);gofmt/go vet 干净;独立 Review(pi-glm,PASS 无阻塞项)报告 .task/reviews/2026-10-21-mi-diagnostics-review.md,6 项非阻塞发现中第 1 项如实指出恢复视图判据的 interrupted 半边由历史 run 平凡满足(不检出 B 重启对账回归,但改动前同样不检出,覆盖无回退)。
影响与剩余项:不改生产代码、不弱化任何断言、未放大任何超时预算、包级 go test -timeout 未动。Review 其余非阻塞项(busy 白名单对新增非终态 status 不设防、轮询只附 logA、instanceOf 守卫不可达、顺序耦合、瞬态 5xx 即失败)均为既有共享弱点或死代码,不阻塞。
Why: establish the authorized task workspace and its initial baseline.
Evidence: preserved precheck receipt in .task/fix-multi-instance-test-failure-diagnostics.md; R0=5360884e9283d962caf945d944df05b5e5059177.
Impact: task metadata only; the feature workspace starts at 5360884e92.
Refs: .task/fix-multi-instance-test-failure-diagnostics.md
Why: the test-hygiene change needs its scope, the review verdict and the
real-PostgreSQL evidence recorded before delivery.
Evidence:
/tmp/mi-diag-impl-r1-20261021.json: cursor, success (single file)
/tmp/mi-diag-review-r1-20261021.json: pi-glm, success (PASS, 6 non-blocking)
Real-PG runs: idle 15.1s and 96-load 17.4s, both ok (fsync=on cluster
on 127.0.0.1:54331); the Backend fix is already in main via PR #237/#674.
Impact: card metadata only; no behaviour change.
Refs: PR #679; fix/multi-instance-test-failure-diagnostics
Why: close the task card inside its own PR so landing it needs no post-merge backfill or separate close PR.
Evidence: .task/fix-multi-instance-test-failure-diagnostics.md -> .task/archive/fix-multi-instance-test-failure-diagnostics.md; status 已完成(合入前闭合); PR #678https://git-home.u1680832.nyat.app:30319/weichen/Agent-System-Dev/pulls/678.
Impact: task metadata only; the card is archived within PR #678 and lands with its merge commit.
Refs: task.sh close --self
weichen
merged commit e634608d97 into main2026-10-21 12:39:17 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
fix/multi-instance-test-failure-diagnostics 任务卡
任务主记录:<PR 链接,交付前填写>;状态:进行中;owner:weichen
权威仓:ssh://git@192.168.3.2:2222/weichen/Agent-System-Dev.git;目标分支:main;R0(initial_base_sha)=
5360884e92工作区:/home/weichen/projects/Agent-System-Dev-wt-fix-multi-instance-test-failure-diagnostics(worktree);分支:fix/multi-instance-test-failure-diagnostics
分支类型:standalone;root_target:main;immediate_target:main;accepting_requirements:false
需求分流:verdict=NEW;selected_initiative=无;requirement_sha256=e2453ebdabc4cc9c61bb25a3535fbe7cf0d0fe687f2ea7bc565b799dcf0fcb6d;triage_receipt=/tmp/triage-mi-test-hygiene-20261021.json;receipt_sha256=4f0a610213a5f6363e303e7129cd09a630ae1a04acbaa5116635c804b2f7f56c
授权与交付终点:修复
tests/integration/multi-instance两处测试卫生缺陷(轮询静默超时、断言早于恢复视图),让失败直指真正原因。不得改 Backend 或任何生产代码(其根因已由 Backend PR #237 修复)、不得弱化断言、不得放大超时预算。交付模式:goal-first(目标、范围和交付终点一次确认;范围内连续推进)
实施执行:agent-per-unit(配置:AGENT-EXECUTION.md)
合入执行:agent(已授权交付范围内,Review 通过、冲突解决且当前 HEAD 的 CI 通过后合入)
流程依据:仓库根目录 policy/delivery.md(归档后仍按仓库根目录定位)
目标与非目标
<预期结果、受影响入口/模块/契约;明确不做什么。>
Agent 实施收据
路由与 fail-stop 规则见根目录 AGENT-EXECUTION.md 与 policy/delivery.md;每次尝试一行,原始日志留在源码外收据路径。
同步记录
5360884e925360884e92证据(Evidence)
结果与验证(完成后填写)
t.Fatal→t.Fatalf仅补失败信息,断言条件未变。新增runStatuses与e.instanceOf两个小helper 供等待条件与失败信息复用。gofmt/go vet干净;独立 Review(pi-glm,PASS 无阻塞项)报告.task/reviews/2026-10-21-mi-diagnostics-review.md,6 项非阻塞发现中第 1 项如实指出恢复视图判据的 interrupted 半边由历史 run 平凡满足(不检出 B 重启对账回归,但改动前同样不检出,覆盖无回退)。go test -timeout未动。Review 其余非阻塞项(busy 白名单对新增非终态 status 不设防、轮询只附 logA、instanceOf守卫不可达、顺序耦合、瞬态 5xx 即失败)均为既有共享弱点或死代码,不阻塞。合入与清理
.task/reviews/2026-10-21-mi-diagnostics-review.md;无阻塞项,6 项非阻塞发现;重放依据:host 用真实 PG 在空闲与 96 负载两种条件下各跑一次多实例验收,均通过)