ci: update ECS runner fleet labels - #10629
Conversation
Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com>
|
Thanks for the PR! Template looks good ✓ Problem: an observed operational issue, not a hypothesis — the Singapore and 64-core runner fleets are retired, so their labels in the update matrix leave Direction: routine CI fleet maintenance, well within scope. No product-surface or security implications — the workflow's permissions, triggers, and job scripts are untouched. Size: not applicable — no core package paths touched; 6 changed lines across two Approach: exactly the minimal change — swap the two retired labels for the two new HK labels in the update matrix, and keep the actionlint known-label list in sync so lint accepts hk-4/hk-5. A repo-wide search confirms these are the only two files referencing Risk: no high-risk paths matched; no elevated risk signals. Re-run note: the follow-up push is a merge of Moving on to code review. 🔍 中文说明感谢贡献! 模板完整 ✓ 问题:已观测到的运维问题,不是假设——新加坡和 64 核 Runner 集群已退役,更新矩阵里保留这些标签会让 方向:日常 CI 集群维护,完全在范围内。不涉及产品面或安全面——工作流的权限、触发条件、任务脚本都未改动。 规模:不适用——未触及核心包路径;两个 方案:正是最小改动——把更新矩阵里两个退役标签换成两个新的香港标签,并同步 actionlint 的已知标签列表使 lint 接受 hk-4/hk-5。全仓库搜索确认只有这两个文件引用 风险:未命中高风险路径;无升级风险信号。 重跑说明:后续推送只是把 进入代码审查 🔍 — Qwen Code · qwen3.8-max Reviewed at |
Code reviewMy independent take before reading the diff: update the No findings. The two files stay consistent (hk-1…hk-5 in both, ordering preserved), and a repo-wide search confirms they are the only references to the Test evidenceEvidence carried: the PR's own CI check results on the reviewed commit, read via the API (unattended run — no PR code executed). tmux N/A: this is CI config with no user-visible product behavior.
On the first head ( What CI cannot show here is the scheduling change itself — labels only resolve to live runners at workflow dispatch. The PR's own test plan covers that: trigger the workflow manually after merge and confirm five hk jobs queue and none target sg/64c. 中文说明代码审查看 diff 前我的独立思路:更新 无问题。两个文件保持一致(均为 hk-1…hk-5,顺序不变),全仓库搜索确认只有这两处引用 测试证据本节证据:审查提交上 PR 自身 CI 的检查结果,通过 API 读取(无人值守运行——未执行 PR 代码)。tmux 不适用:本 PR 为 CI 配置,无用户可见的产品行为。 在第一个 head( CI 无法体现的是调度变化本身——标签只有在派发工作流时才会解析到在线 Runner。PR 自带的测试计划已覆盖:合并后手动触发工作流,确认五个 hk 任务入队且没有 sg/64c 任务。 — Qwen Code · qwen3.8-max Reviewed at |
|
Confidence: 5/5 — clean across every stage: a minimal diff that matches the stated fleet change exactly, both label references updated consistently, zero security-surface changes, and the author owns the infrastructure being changed. This is fleet bookkeeping done right — two retired labels out, two active HK hosts in, actionlint kept in sync, nothing else touched. The red checks on the first head were the pre-existing Verdict: approve. The unit suite and the no-AK integration job were still running on the merged head at review time, so approval is deferred until CI lands green on 中文说明置信度:5/5——每个阶段都干净:最小 diff 与所述的集群变更完全一致,两处标签引用同步更新,安全面零改动,作者正是这套基础设施的负责人。 这是一次规范的集群维护——移除两个退役标签、纳入两台在用的香港宿主机、actionlint 保持同步、其他一概不动。第一个 head 上的红色检查是 #10623 跟踪的 结论:通过。审查时合并后 head 上的单元测试与 no-AK 集成任务仍在运行,批准将延迟到 — Qwen Code · qwen3.8-max Reviewed at |
qwen-code-dev-bot
left a comment
There was a problem hiding this comment.
Gate APPROVE at head c1c7f032, decided on the review conclusion per maintainer instruction. The change is a two-file consistent label swap: the retired ecs-update-sg/ecs-update-64c entries are removed from both the workflow matrix and the actionlint whitelist, hk-4/hk-5 are added to both, and a full-tree grep at this head shows no remaining reference to the retired labels anywhere else. Risk (each HK host must keep exactly one online runner with its update label; a missing runner just queues that matrix leg) is documented in the PR. 0 unresolved threads, chiga0 approved this exact head; Test (ubuntu) was still running at submission, non-red.
中文:在 head c1c7f032 按 Review 结论通过(CI 不作判据,按维护者指示)。矩阵与 actionlint 白名单两处同步替换、全树 grep 确认退役标签零残留引用,风险(hk-4/5 需各有一个带标在线 runner,否则该矩阵腿排队等待)已在 PR 中声明。线程零未解决,chiga0 同 head 已 approve;提交时 Test 在跑、非红。
Code Coverage Summary
CLI Package - Full Text ReportCore Package - Full Text ReportFor detailed HTML reports, please see the 'coverage-reports-22.x-ubuntu-latest' artifact from the main CI run. |
|
Released in v0.23.0. |
What this PR does
Routes the ECS Qwen updater to the five active Hong Kong runner hosts and stops targeting the retired Singapore and 64-core runner labels.
Why it's needed
The old Singapore and 64-core runner fleets are offline and ready for removal. Keeping their update labels in the matrix leaves updater jobs waiting for runners that will not return, while two active Hong Kong hosts are currently excluded from fleet updates.
Reviewer Test Plan
How to verify
Trigger the workflow manually with an explicit published version or leave the version blank. Confirm that one update job is scheduled for each label from
ecs-update-hk-1throughecs-update-hk-5, and that no job targetsecs-update-sgorecs-update-64c.Evidence (Before & After)
N/A
Tested on
actionlintEnvironment (optional)
actionlint .github/workflows/update-ecs-runner-qwen.ymlRisk & Scope
Linked Issues
N/A
中文说明
本 PR 的改动
将 ECS Qwen 更新任务切换到五台仍在使用的香港 Runner 宿主机,并停止调度已经退役的新加坡和 64 核 Runner 标签。
修改原因
旧的新加坡和 64 核 Runner 集群已经离线,可以进入清理阶段。如果矩阵继续保留这些更新标签,更新任务会一直等待不会恢复的 Runner;与此同时,两台正在使用的香港宿主机不会收到版本更新任务。
Reviewer 测试计划
验证方式
手动触发工作流,可以指定一个已发布版本,也可以留空使用 npm latest。确认
ecs-update-hk-1到ecs-update-hk-5每个标签都调度一个更新任务,并且不再向ecs-update-sg或ecs-update-64c调度任务。修改前后证据
不适用。
已测试平台
actionlint测试环境(可选)
actionlint .github/workflows/update-ecs-runner-qwen.yml风险与范围
关联 Issue
无。