Repository navigation
Update sglang-ci-monitor for current CI shape - #1
Merged
Merged
Conversation
Reflect upstream CI changes since the skill was first written:
- Stage rename: NVIDIA `stage-{a,b,c}-test-*` → `base-{a,b,c}-test-*`
(PR sgl-project/sglang#25420). AMD jobs still use the `stage-*` prefix.
- `check-stage-health` action → `check-pr-test-health`. Annotation
messages still use `Fast-fail: skipping — root cause job(s): <names>`
and `Fast-fail: lint check failed`, but they are now mutually
exclusive (lint is checked first and returns early).
- `/rerun-stage` is deprecated (PR #25322): handler replies `-1` plus a
deprecation comment, does nothing. Removed from decision tree and
command table; documented as a do-not-use.
- New `/rerun-group <group>` command (PR #24023): expands
`test/registered/<group>/test_*.py` into a `/rerun-test` batch.
- New `pr-test-extra.yml` workflow (label-gated extras, stages
`extra-a/b-test-*`) double-gated on `run-ci` + `run-ci-extra`.
- New labels: `run-ci-extra` opts in to extra CI; `bypass-fastfail`
skips the jobs-failed cascade (lint still cascades);
`bypass-maintenance` bypasses the rebase-required / maintenance-pause
gates inside `/rerun-test` and `check-maintenance`.
- New shortcut for grabbing `<RUN_ID>`: read the PR States block at the
bottom of every PR body (maintained by `pr-states.yml`), which links
the latest pr-test and pr-test-extra runs and is refreshed on
workflow_run events so slash-command reruns refresh it too.
- `base-c-test-8-gpu-h20` is excluded from the fast-fail root-cause set
(flaky h20 noise); document the implication.
- `/rerun-failed-ci` now iterates every workflow run on the head SHA
(pr-test, pr-test-extra, pr-test-sgl-kernel, jit-kernel,
multimodal-gen…), so sibling-workflow failures are picked up in one
command.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Refresh the
sglang-ci-monitorskill to match how sgl-project/sglang's CI actually works today. The skill had drifted: it still referenced the oldstage-{a,b,c}-test-*job names, leaned on/rerun-stageas a primary tool, and didn't know aboutpr-test-extra.yml,/rerun-group, thebypass-*labels, or the PR States block.What changed
base-{a,b,c}-test-*([CI] Rename basic CIstage-a/b/c->base-a/b/cfor symmetry with extra CI sgl-project/sglang#25420). AMD is unchanged — stillstage-*. Filter pattern is still*-amd*so behavior is unaffected, but the prose and examples were updated.check-stage-health→check-pr-test-health(composite action renamed). The two annotation shapes (Fast-fail: lint check failedvsFast-fail: skipping — root cause job(s): <names>) are now documented as mutually exclusive, since lint is checked first and returns early./rerun-stageis deprecated (Deprecate /rerun-stage; scrub CUDA target_stage infra sgl-project/sglang#25322) — handler replies with a-1reaction and a deprecation comment, then does nothing. Removed from the decision tree and command table; called out as a do-not-use in the Execution policy and What to not do sections./rerun-group <group>command (ci: add /rerun-group to rerun all registered tests in a group sgl-project/sglang#24023): expandstest/registered/<group>/test_*.pyinto a/rerun-testbatch. Added to the command table and pre-approved list.pr-test-extra.ymlis a distinct workflow with its own run id, gated onrun-ciANDrun-ci-extra, with stagesextra-{a,b}-test-*. Skill now distinguishes base CI from extra CI throughout.run-ci-extra,bypass-fastfail(skips jobs-failed cascade, lint still cascades),bypass-maintenance(bypasses/rerun-test's pre-dispatch maintenance gate).pr-states.yml) is now the recommended fastest way to grab<RUN_ID>. Refreshed onworkflow_runevents so slash-command reruns refresh it too.base-c-test-8-gpu-h20exclusion: flaky-h20 noise is excluded from the fast-fail root-cause set, so an h20-only FAILURE won't cascade — documented so it isn't misread as a non-failure./rerun-failed-cimulti-workflow sweep: clarified that it iterates every workflow run on the head SHA (pr-test, pr-test-extra, pr-test-sgl-kernel, jit-kernel, multimodal-gen…), not justpr-test.yml.Test plan
.github/workflows/{pr-gate,pr-test,pr-test-extra,pr-states,lint}.yml,.github/actions/check-pr-test-health/action.yml,scripts/ci/utils/slash_command_handler.py.>-block-scalar form as commit3e50268("Fix skill frontmatter YAML") to keep the description on one logical line without breaking the YAML parser./sglang-ci-monitoron a real PR and confirm the rollup, fast-fail, and rerun branches match the documented flow.