diff --git a/docs/MDTODO/observability-trace-reliability.md b/docs/MDTODO/observability-trace-reliability.md index 09715af9..b1cb50be 100644 --- a/docs/MDTODO/observability-trace-reliability.md +++ b/docs/MDTODO/observability-trace-reliability.md @@ -42,3 +42,23 @@ ### R3.4 [completed] 独立溯源 Web Probe Sentinel CI 从历史可成功构建到连续 BuildKit readiness 失败的首个回归:对齐成功/失败 PipelineRun、镜像 digest、TaskSpec、节点环境与 Git 提交/PR/作者,区分环境启动变慢、官方 daemonless 默认重试窗口不足和 PR #1939 裸 buildkitd 新回归,输出可复核的时间线与责任变更,不修改运行面或人工触发 PipelineRun,完成任务后将详细报告写入[任务报告](./details/observability-trace-reliability/R3.4_Task_Report.md)。 + +## R4 [in_progress] + +解决 [AgentRun #352](https://github.com/pikasTech/agentrun/issues/352)、[HWLAB #2525](https://github.com/pikasTech/HWLAB/issues/2525) 与 [UniDesk #1949](https://github.com/pikasTech/unidesk/issues/1949):消除 trace trc_f389d0493c0c489e 的 Kafka outbox 队头阻塞与 17 分钟伪耗时,恢复 agentrun.event.v1→transactional projector→hwlab.event.v1→实时/回放 SSE 单一权威;禁止 HTTP 补链、页面轮询、读侧修复、进程内 finalizer、双 authority、额外锁/租约/状态库或降低 Kafka/SSE 能力,完成任务后将详细报告写入[任务报告](./details/observability-trace-reliability/R4_Task_Report.md)。 + +### R4.1 [in_progress] + +执行 [AgentRun #352](https://github.com/pikasTech/agentrun/issues/352):按 key 保序、跨 key 有界并行排空 transactional outbox,保留 durable at-least-once 并补 backlog/oldest/rate/typed error 可见性,以 NC01/nc01-v02 真实积压归零和新事件实时发布验收,完成任务后将详细报告写入[任务报告](./details/observability-trace-reliability/R4.1_Task_Report.md)。 + +### R4.2 [in_progress] + +执行 [HWLAB #2525](https://github.com/pikasTech/HWLAB/issues/2525):恢复固定 group transactional projector、原子 facts/checkpoint/browser outbox、hwlab.event.v1 relay 与同 cursor 实时/回放 SSE;耗时取 canonical 时间,禁止 direct/live 双写、HTTP 补链和页面轮询,完成任务后将详细报告写入[任务报告](./details/observability-trace-reliability/R4.2_Task_Report.md)。 + +### R4.3 [in_progress] + +执行 [UniDesk #1949](https://github.com/pikasTech/unidesk/issues/1949):由 config/hwlab-node-lanes.yaml 选择 transactionalProjector、projectionOutboxRelay、projectionRealtime 单一产品 authority,关闭 live-only direct authority,保持 renderer/status/preflight YAML-first,完成任务后将详细报告写入[任务报告](./details/observability-trace-reliability/R4.3_Task_Report.md)。 + +### R4.4 + +依赖 R4.1-R4.3,主代理完成 PR 架构审核、顺序合并、自动 CI/CD 和 NC01/v03 原 Workbench 新 hi turn 验收;核对 command/run 实际耗时、Kafka backlog/lag、运行中增量、terminal/final、刷新 replay、公开 result 与 OTel 一致后收口,完成任务后将详细报告写入[任务报告](./details/observability-trace-reliability/R4.4_Task_Report.md)。