|
| 1 | +name: Retry Transient Workflow Failures |
| 2 | + |
| 3 | +on: |
| 4 | + workflow_run: |
| 5 | + workflows: |
| 6 | + - Project Board Automation |
| 7 | + - Changelog Automation |
| 8 | + - Pull Request Label Sync |
| 9 | + - Generate Reports and Deploy to GitHub Pages |
| 10 | + - Rigorous Pull Request Review |
| 11 | + - Run PHPUnit Tests |
| 12 | + - Maintain Wiki |
| 13 | + - Maintain Wiki Publication |
| 14 | + - Update Wiki Preview |
| 15 | + - Update Wiki |
| 16 | + types: |
| 17 | + - completed |
| 18 | + |
| 19 | +permissions: |
| 20 | + actions: write |
| 21 | + contents: read |
| 22 | + |
| 23 | +concurrency: |
| 24 | + group: retry-transient-run-${{ github.event.workflow_run.id }} |
| 25 | + cancel-in-progress: false |
| 26 | + |
| 27 | +jobs: |
| 28 | + retry: |
| 29 | + if: ${{ github.event.workflow_run.conclusion == 'failure' }} |
| 30 | + name: Retry Failed Jobs When GitHub Infrastructure Looks Transient |
| 31 | + runs-on: ubuntu-latest |
| 32 | + |
| 33 | + steps: |
| 34 | + - id: retry |
| 35 | + uses: actions/github-script@v8 |
| 36 | + with: |
| 37 | + github-token: ${{ secrets.GITHUB_TOKEN }} |
| 38 | + script: | |
| 39 | + const transientPatterns = [ |
| 40 | + /RPC failed; HTTP 5\d\d/i, |
| 41 | + /expected flush after ref listing/i, |
| 42 | + /expected 'packfile'/i, |
| 43 | + /remote:\s+Internal Server Error/i, |
| 44 | + /requested URL returned error:\s*5\d\d/i, |
| 45 | + /fatal:\s+unable to access 'https:\/\/github\.com\/.*': The requested URL returned error:\s*5\d\d/i, |
| 46 | + ]; |
| 47 | +
|
| 48 | + const [owner, repo] = process.env.GITHUB_REPOSITORY.split('/'); |
| 49 | + const runId = Number.parseInt(`${{ github.event.workflow_run.id }}`, 10); |
| 50 | + const runAttempt = Number.parseInt(`${{ github.event.workflow_run.run_attempt }}`, 10); |
| 51 | + const workflowName = `${{ github.event.workflow_run.name }}`; |
| 52 | + const maxRunAttempts = 2; |
| 53 | +
|
| 54 | + const buildSummary = ({ status, failedJobs = [], matchedJobs = [] }) => { |
| 55 | + const lines = [ |
| 56 | + '## Transient Failure Retry Summary', |
| 57 | + '', |
| 58 | + `- Workflow: \`${workflowName}\``, |
| 59 | + `- Run ID: \`${runId}\``, |
| 60 | + `- Run attempt: \`${runAttempt}\``, |
| 61 | + `- Retry status: \`${status}\``, |
| 62 | + ]; |
| 63 | +
|
| 64 | + if (failedJobs.length > 0) { |
| 65 | + lines.push(`- Failed jobs inspected: ${failedJobs.map((job) => `\`${job}\``).join(', ')}`); |
| 66 | + } |
| 67 | +
|
| 68 | + if (matchedJobs.length > 0) { |
| 69 | + lines.push(`- Jobs with transient GitHub failure signatures: ${matchedJobs.map((job) => `\`${job}\``).join(', ')}`); |
| 70 | + } |
| 71 | +
|
| 72 | + if (status === 'rerun-requested') { |
| 73 | + lines.push('- Action: Requested a rerun of failed jobs because every failed job matched transient GitHub-side error signatures.'); |
| 74 | + } |
| 75 | +
|
| 76 | + if (status === 'skipped-run-attempt-limit') { |
| 77 | + lines.push('- Action: Skipped rerun because the run already reached the configured retry limit.'); |
| 78 | + } |
| 79 | +
|
| 80 | + if (status === 'skipped-no-failed-jobs') { |
| 81 | + lines.push('- Action: Skipped rerun because the workflow reported failure without failed jobs to inspect.'); |
| 82 | + } |
| 83 | +
|
| 84 | + if (status === 'skipped-no-transient-match') { |
| 85 | + lines.push('- Action: Skipped rerun because at least one failed job did not match the transient GitHub-side signatures.'); |
| 86 | + } |
| 87 | +
|
| 88 | + return lines.join('\n'); |
| 89 | + }; |
| 90 | +
|
| 91 | + if (runAttempt >= maxRunAttempts) { |
| 92 | + const summary = buildSummary({ status: 'skipped-run-attempt-limit' }); |
| 93 | + core.setOutput('status', 'skipped-run-attempt-limit'); |
| 94 | + core.setOutput('summary', summary); |
| 95 | +
|
| 96 | + return; |
| 97 | + } |
| 98 | +
|
| 99 | + const jobsResponse = await github.rest.actions.listJobsForWorkflowRun({ |
| 100 | + owner, |
| 101 | + repo, |
| 102 | + run_id: runId, |
| 103 | + per_page: 100, |
| 104 | + }); |
| 105 | +
|
| 106 | + const failedJobs = jobsResponse.data.jobs.filter((job) => job.conclusion === 'failure'); |
| 107 | +
|
| 108 | + if (failedJobs.length === 0) { |
| 109 | + const summary = buildSummary({ status: 'skipped-no-failed-jobs' }); |
| 110 | + core.setOutput('status', 'skipped-no-failed-jobs'); |
| 111 | + core.setOutput('summary', summary); |
| 112 | +
|
| 113 | + return; |
| 114 | + } |
| 115 | +
|
| 116 | + const matchedJobs = []; |
| 117 | +
|
| 118 | + for (const job of failedJobs) { |
| 119 | + const logsResponse = await fetch(`https://api.github.com/repos/${owner}/${repo}/actions/jobs/${job.id}/logs`, { |
| 120 | + headers: { |
| 121 | + Accept: 'application/vnd.github+json', |
| 122 | + Authorization: `Bearer ${process.env.GITHUB_TOKEN}`, |
| 123 | + 'X-GitHub-Api-Version': '2022-11-28', |
| 124 | + }, |
| 125 | + redirect: 'follow', |
| 126 | + }); |
| 127 | +
|
| 128 | + if (!logsResponse.ok) { |
| 129 | + throw new Error(`Failed to download logs for job ${job.name}: ${logsResponse.status} ${logsResponse.statusText}`); |
| 130 | + } |
| 131 | +
|
| 132 | + const logText = await logsResponse.text(); |
| 133 | + const hasTransientMatch = transientPatterns.some((pattern) => pattern.test(logText)); |
| 134 | +
|
| 135 | + if (!hasTransientMatch) { |
| 136 | + const summary = buildSummary({ |
| 137 | + status: 'skipped-no-transient-match', |
| 138 | + failedJobs: failedJobs.map((failedJob) => failedJob.name), |
| 139 | + matchedJobs, |
| 140 | + }); |
| 141 | +
|
| 142 | + core.setOutput('status', 'skipped-no-transient-match'); |
| 143 | + core.setOutput('summary', summary); |
| 144 | +
|
| 145 | + return; |
| 146 | + } |
| 147 | +
|
| 148 | + matchedJobs.push(job.name); |
| 149 | + } |
| 150 | +
|
| 151 | + await github.request('POST /repos/{owner}/{repo}/actions/runs/{run_id}/rerun-failed-jobs', { |
| 152 | + owner, |
| 153 | + repo, |
| 154 | + run_id: runId, |
| 155 | + }); |
| 156 | +
|
| 157 | + const summary = buildSummary({ |
| 158 | + status: 'rerun-requested', |
| 159 | + failedJobs: failedJobs.map((job) => job.name), |
| 160 | + matchedJobs, |
| 161 | + }); |
| 162 | +
|
| 163 | + core.setOutput('status', 'rerun-requested'); |
| 164 | + core.setOutput('summary', summary); |
| 165 | +
|
| 166 | + - name: Write step summary |
| 167 | + env: |
| 168 | + RETRY_SUMMARY: ${{ steps.retry.outputs.summary }} |
| 169 | + run: printf '%s\n' "$RETRY_SUMMARY" >> "$GITHUB_STEP_SUMMARY" |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?v=4&size=40){kind=avatar}
0 commit comments