LibreChat/.github/workflows/retry-docker-builds.yml

name: Retry Failed Docker Builds

on:
  workflow_run:
    workflows:
      - Docker Build Smoke Tests
      - Docker Compose Build Latest Main Image Tag (Manual Dispatch)
      - Docker Dev Branch Images Build
      - Docker Dev Images Build
      - Docker Dev Staging Images Build
      - Docker Images Build on Tag
    types:
      - completed

permissions:
  actions: write
  contents: read

jobs:
  retry-failed-jobs:
    name: Re-run failed jobs
    if: >
      (github.event.workflow_run.conclusion == 'failure' ||
       github.event.workflow_run.conclusion == 'timed_out') &&
      github.event.workflow_run.run_attempt < 3
    runs-on: ubuntu-latest
    steps:
      - name: Check failed run is still current
        id: current-run
        env:
          GH_TOKEN: ${{ github.token }}
          HEAD_BRANCH: ${{ github.event.workflow_run.head_branch }}
          HEAD_SHA: ${{ github.event.workflow_run.head_sha }}
          RUN_EVENT: ${{ github.event.workflow_run.event }}
          RUN_ID: ${{ github.event.workflow_run.id }}
          WORKFLOW_ID: ${{ github.event.workflow_run.workflow_id }}
          WORKFLOW_NAME: ${{ github.event.workflow_run.name }}
        run: |
          set -euo pipefail

          encode_uri() {
            jq -nr --arg value "$1" '$value | @uri'
          }

          workflow_runs_path="repos/${GITHUB_REPOSITORY}/actions/workflows/${WORKFLOW_ID}/runs"
          query="per_page=1&exclude_pull_requests=false"

          if [ "$WORKFLOW_NAME" = "Docker Images Build on Tag" ]; then
            if [ -z "$HEAD_BRANCH" ]; then
              echo "No tag name found for ${WORKFLOW_NAME}; skipping retry."
              echo "should_retry=false" >> "$GITHUB_OUTPUT"
              exit 0
            fi

            if ! tag_ref=$(gh api "repos/${GITHUB_REPOSITORY}/git/ref/tags/$(encode_uri "$HEAD_BRANCH")" --jq '.object'); then
              echo "Tag ${HEAD_BRANCH} no longer exists; skipping retry."
              echo "should_retry=false" >> "$GITHUB_OUTPUT"
              exit 0
            fi

            tag_object_type=$(jq -r '.type' <<< "$tag_ref")
            tag_object_sha=$(jq -r '.sha' <<< "$tag_ref")

            if [ "$tag_object_type" = "tag" ]; then
              tag_head_sha=$(gh api "repos/${GITHUB_REPOSITORY}/git/tags/${tag_object_sha}" --jq '.object.sha')
            else
              tag_head_sha="$tag_object_sha"
            fi

            if [ "$tag_head_sha" = "$HEAD_SHA" ]; then
              echo "should_retry=true" >> "$GITHUB_OUTPUT"
              exit 0
            fi

            echo "Skipping retry for stale ${WORKFLOW_NAME} run ${RUN_ID}; tag ${HEAD_BRANCH} now points to ${tag_head_sha}."
            echo "should_retry=false" >> "$GITHUB_OUTPUT"
            exit 0
          fi

          if [ -n "$RUN_EVENT" ]; then
            query="${query}&event=$(encode_uri "$RUN_EVENT")"
          fi

          if [ -n "$HEAD_BRANCH" ]; then
            query_with_ref="${query}&branch=$(encode_uri "$HEAD_BRANCH")"
            latest_run=$(gh api "${workflow_runs_path}?${query_with_ref}" --jq '.workflow_runs[0] // empty')
          else
            latest_run=""
          fi

          if [ -z "$latest_run" ]; then
            latest_run=$(gh api "${workflow_runs_path}?${query}" --jq '.workflow_runs[0] // empty')
          fi

          if [ -z "$latest_run" ]; then
            echo "No matching workflow runs found for ${WORKFLOW_NAME}; skipping retry."
            echo "should_retry=false" >> "$GITHUB_OUTPUT"
            exit 0
          fi

          latest_run_id=$(jq -r '.id' <<< "$latest_run")
          latest_head_sha=$(jq -r '.head_sha' <<< "$latest_run")

          if [ "$latest_run_id" = "$RUN_ID" ] || [ "$latest_head_sha" = "$HEAD_SHA" ]; then
            echo "should_retry=true" >> "$GITHUB_OUTPUT"
            exit 0
          fi

          echo "Skipping retry for stale ${WORKFLOW_NAME} run ${RUN_ID}; newer run ${latest_run_id} is at ${latest_head_sha}."
          echo "should_retry=false" >> "$GITHUB_OUTPUT"

      - name: Re-run failed Docker jobs
        if: steps.current-run.outputs.should_retry == 'true'
        env:
          GH_TOKEN: ${{ github.token }}
          RUN_ATTEMPT: ${{ github.event.workflow_run.run_attempt }}
          RUN_ID: ${{ github.event.workflow_run.id }}
          WORKFLOW_NAME: ${{ github.event.workflow_run.name }}
        run: |
          set -euo pipefail

          cancelled_jobs=$(gh api --paginate "repos/${GITHUB_REPOSITORY}/actions/runs/${RUN_ID}/jobs?per_page=100" \
            | jq -s '[.[].jobs[]? | select(.conclusion == "cancelled")] | length')

          if [ "$cancelled_jobs" -gt 0 ]; then
            endpoint="rerun"
            echo "Found ${cancelled_jobs} cancelled job(s); re-running the full workflow run."
          else
            endpoint="rerun-failed-jobs"
            echo "Re-running failed jobs only."
          fi

          echo "Retrying ${WORKFLOW_NAME} (run ${RUN_ID}, attempt ${RUN_ATTEMPT})."
          gh api \
            --method POST \
            "repos/${GITHUB_REPOSITORY}/actions/runs/${RUN_ID}/${endpoint}"