chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 17 Packages Projects Releases Wiki Activity

Fix wait-for-stage jobs running when call-gate fails (#17443)

Browse Source
This commit is contained in:
Alison Shao
2026-01-20 17:40:05 -08:00
committed by GitHub
parent 1e309030e3
commit 648aab0ce3
1 changed files with 29 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/pr-test.yml
+29 -17
View File
@@ -188,13 +188,18 @@ jobs:
# For scheduled runs: wait jobs are skipped, enabling parallel execution for easier retry.
wait-for-stage-a:
needs: [check-changes]
needs: [check-changes, call-gate]
# Only run for PRs (not scheduled) and when not targeting a specific stage
# Skip if call-gate failed (stage-a jobs will be skipped, nothing to wait for)
# !cancelled() ensures this job respects workflow cancellation from concurrency group
if: |
always() &&
!cancelled() &&
github.event_name == 'pull_request' &&
!inputs.target_stage &&
inputs.test_parallel_dispatch != true &&
(needs.check-changes.outputs.main_package == 'true' || needs.check-changes.outputs.sgl_kernel == 'true')
(needs.check-changes.outputs.main_package == 'true' || needs.check-changes.outputs.sgl_kernel == 'true') &&
(needs.call-gate.result == 'success' || needs.call-gate.result == 'skipped')
runs-on: ubuntu-latest
outputs:
stage_a_result: ${{ steps.wait.outputs.result }}
@@ -205,7 +210,7 @@ jobs:
with:
script: |
const maxWaitMinutes = 60;
const pollIntervalSeconds = 30;
const pollIntervalSeconds = 120; // 2 minutes to reduce GH API calls
const maxAttempts = (maxWaitMinutes * 60) / pollIntervalSeconds;
for (let attempt = 0; attempt < maxAttempts; attempt++) {
@@ -222,7 +227,7 @@ jobs:
console.log(`stage-a-test-1 status: ${stageAJob.status}, conclusion: ${stageAJob.conclusion}`);
if (stageAJob.status === 'completed') {
if (stageAJob.conclusion === 'success' || stageAJob.conclusion === 'skipped' || stageAJob.conclusion === 'cancelled') {
if (stageAJob.conclusion === 'success' || stageAJob.conclusion === 'skipped') {
core.setOutput('result', stageAJob.conclusion === 'success' ? 'success' : 'skipped');
return;
} else {
@@ -243,15 +248,19 @@ jobs:
core.setOutput('result', 'timeout');
wait-for-stage-b:
needs: [check-changes, wait-for-stage-a]
needs: [check-changes, call-gate, wait-for-stage-a]
# Only run for PRs (not scheduled) and when not targeting a specific stage
# Skip if call-gate failed (stage-b jobs will be skipped, nothing to wait for)
# !cancelled() ensures this job respects workflow cancellation from concurrency group
if: |
always() &&
!cancelled() &&
github.event_name == 'pull_request' &&
!inputs.target_stage &&
inputs.test_parallel_dispatch != true &&
(needs.check-changes.outputs.main_package == 'true' || needs.check-changes.outputs.sgl_kernel == 'true') &&
(needs.wait-for-stage-a.result == 'success' || needs.wait-for-stage-a.result == 'skipped')
(needs.wait-for-stage-a.result == 'success' || needs.wait-for-stage-a.result == 'skipped') &&
(needs.call-gate.result == 'success' || needs.call-gate.result == 'skipped')
runs-on: ubuntu-latest
outputs:
stage_b_result: ${{ steps.wait.outputs.result }}
@@ -262,20 +271,24 @@ jobs:
with:
script: |
const maxWaitMinutes = 90;
const pollIntervalSeconds = 30;
const pollIntervalSeconds = 120; // 2 minutes to reduce GH API calls
const maxAttempts = (maxWaitMinutes * 60) / pollIntervalSeconds;
// Stage-b jobs to wait for
// Use exact prefix + ( or end-of-string to avoid matching -performance/-accuracy variants
// Stage-b jobs to wait for (all stage-b tests including performance and accuracy)
const stageBJobs = [
{ prefix: 'stage-b-test-small-1-gpu', expectedCount: 8 }, // partitions 0-7
{ prefix: 'stage-b-test-large-1-gpu', expectedCount: 12 }, // partitions 0-11
{ prefix: 'stage-b-test-large-2-gpu', expectedCount: 2 }, // partitions 0-1
{ prefix: 'stage-b-test-4-gpu-b200', expectedCount: 1 }
{ prefix: 'stage-b-test-small-1-gpu', expectedCount: 8 }, // partitions 0-7
{ prefix: 'stage-b-test-large-1-gpu', expectedCount: 12 }, // partitions 0-11
{ prefix: 'stage-b-test-large-2-gpu', expectedCount: 2 }, // partitions 0-1
{ prefix: 'stage-b-test-4-gpu-b200', expectedCount: 1 },
{ prefix: 'stage-b-test-small-1-gpu-performance', expectedCount: 1 },
{ prefix: 'stage-b-test-large-1-gpu-performance', expectedCount: 2 }, // partitions 0-1
{ prefix: 'stage-b-test-large-2-gpu-performance', expectedCount: 1 },
{ prefix: 'stage-b-test-small-1-gpu-accuracy', expectedCount: 1 },
{ prefix: 'stage-b-test-large-2-gpu-accuracy', expectedCount: 1 }
];
const totalExpectedJobs = stageBJobs.reduce((sum, j) => sum + j.expectedCount, 0);
const totalExpectedJobs = stageBJobs.reduce((sum, j) => sum + j.expectedCount, 0); // 29 total
// Helper to match job names exactly (prefix + optional " (N)" suffix)
// Helper to match job names exactly (prefix alone or prefix + " (N)" for matrix jobs)
const matchesPrefix = (jobName, prefix) => {
return jobName === prefix || jobName.startsWith(prefix + ' (');
};
@@ -306,8 +319,7 @@ jobs:
allCompleted = false;
} else {
completedCount++;
// cancelled is not a failure - it means the workflow was stopped (e.g., by concurrency)
if (job.conclusion !== 'success' && job.conclusion !== 'skipped' && job.conclusion !== 'cancelled') {
if (job.conclusion !== 'success' && job.conclusion !== 'skipped') {
anyFailed = true;
failedJobs.push(job.name);
}