daily_ci_report.yml

name: Daily CI Failure Report

on:
  schedule:
    - cron: "0 12 * * *"
  workflow_dispatch:

permissions:
  issues: write

jobs:
  check-and-report:
    name: Check daily CI and report failures
    runs-on: ubuntu-latest
    steps:
      - name: Check daily CI status and open/update issue on failure
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          REPO: ${{ github.repository }}
          DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
        run: |
          python3 << 'PYEOF'
          import os
          import sys
          import requests
          from datetime import datetime, timezone, timedelta

          TOKEN = os.environ["GITHUB_TOKEN"]
          REPO = os.environ["REPO"]
          DEFAULT_BRANCH = os.environ.get("DEFAULT_BRANCH") or "master"

          HEADERS = {
              "Authorization": f"Bearer {TOKEN}",
              "Accept": "application/vnd.github+json",
              "X-GitHub-Api-Version": "2022-11-28",
          }
          BASE = f"https://api.github.com/repos/{REPO}"

          # Daily workflows to monitor
          WORKFLOWS = [
              ("Daily Nimbus", "daily_nimbus.yml"),
              ("Daily test all (latests dependnecies)", "daily_test_all_latest_deps.yml"),
              ("Daily test all (without feature flags)", "daily_test_all_no_flags.yml"),
              ("Daily runnable examples", "daily_runnable_examples.yml"),
              ("Daily interop", "daily_interop.yml"),
          ]

          LABEL_NAME = "daily-ci-failure"
          LABEL_COLOR = "d93f0b"
          LABEL_DESC = "Tracks daily CI job failures"

          now = datetime.now(timezone.utc)
          # Allow up to 26 hours so we always catch the 6:30 UTC scheduled run
          cutoff = now - timedelta(hours=26)

          # ── 1. Find workflows that failed in their most recent run ────────────
          failed = []
          for wf_name, wf_file in WORKFLOWS:
              url = f"{BASE}/actions/workflows/{wf_file}/runs"
              params = {
                  "status": "completed",
                  "branch": DEFAULT_BRANCH,
                  "per_page": 1,
              }
              r = requests.get(url, headers=HEADERS, params=params)
              r.raise_for_status()
              runs = r.json().get("workflow_runs", [])

              if not runs:
                  print(f"{wf_name}: no completed runs found, skipping")
                  continue

              run = runs[0]
              created_at = datetime.fromisoformat(run["created_at"].replace("Z", "+00:00"))
              if created_at < cutoff:
                  print(
                      f"{wf_name}: most recent completed run ({run['created_at']}) "
                      f"is older than 26 h, skipping"
                  )
                  continue

              if run["conclusion"] == "failure":
                  print(f"{wf_name}: FAILED — {run['html_url']}")
                  failed.append(
                      {"name": wf_name, "url": run["html_url"], "run_id": run["id"]}
                  )
              else:
                  print(f"{wf_name}: {run['conclusion']}")

          if not failed:
              print("All monitored daily workflows passed (or no recent runs). Nothing to report.")
              sys.exit(0)

          # ── 2. Ensure the tracking label exists ───────────────────────────────
          r = requests.get(f"{BASE}/labels/{LABEL_NAME}", headers=HEADERS)
          if r.status_code == 404:
              payload = {
                  "name": LABEL_NAME,
                  "color": LABEL_COLOR,
                  "description": LABEL_DESC,
              }
              r = requests.post(f"{BASE}/labels", headers=HEADERS, json=payload)
              r.raise_for_status()
              print(f"Created label '{LABEL_NAME}'")
          else:
              r.raise_for_status()
              print(f"Label '{LABEL_NAME}' already exists")

          # ── 3. Compute ISO week identifier ────────────────────────────────────
          year, week, _ = now.isocalendar()
          week_id = f"{year}-W{week:02d}"
          issue_title = f"[Daily CI] Failures - Week {week_id}"

          # ── 4. Find an existing issue (open or closed) for this week ────────
          # Searching both states prevents creating duplicate issues when
          # someone manually closes the weekly tracking issue mid-week.
          existing_issue = None
          for state in ("open", "closed"):
              if existing_issue:
                  break
              page = 1
              while True:
                  r = requests.get(
                      f"{BASE}/issues",
                      headers=HEADERS,
                      params={
                          "state": state,
                          "labels": LABEL_NAME,
                          "per_page": 50,
                          "page": page,
                      },
                  )
                  r.raise_for_status()
                  issues = r.json()
                  if not issues:
                      break
                  for issue in issues:
                      # Exclude pull requests (GitHub returns them in /issues)
                      if issue.get("pull_request"):
                          continue
                      if issue["title"] == issue_title:
                          existing_issue = issue
                          break
                  if existing_issue or len(issues) < 50:
                      break
                  page += 1

          # ── 5. Build the failure report content ───────────────────────────────
          date_str = now.strftime("%Y-%m-%d")
          lines = [f"### Daily CI failures detected on {date_str}\n"]
          for f in failed:
              lines.append(f"- **{f['name']}** failed → [View run]({f['url']})")
          content = "\n".join(lines)

          # ── 6. Comment on existing issue or create a new one ─────────────────
          ASSIGNEES = ["gmelodie", "richard-ramos", "vladopajic"]
          assignee = ASSIGNEES[(week - 1) % len(ASSIGNEES)]

          if existing_issue:
              # Reopen the issue if it was closed, then add a comment
              if existing_issue["state"] == "closed":
                  r = requests.patch(
                      f"{BASE}/issues/{existing_issue['number']}",
                      headers=HEADERS,
                      json={"state": "open"},
                  )
                  r.raise_for_status()
                  print(f"Reopened issue #{existing_issue['number']}")
              r = requests.post(
                  f"{BASE}/issues/{existing_issue['number']}/comments",
                  headers=HEADERS,
                  json={"body": content},
              )
              r.raise_for_status()
              print(
                  f"Added comment to issue #{existing_issue['number']}: "
                  f"{existing_issue['html_url']}"
              )
          else:
              body = (
                  f"This issue tracks daily CI failures for week **{week_id}**.\n\n"
                  f"---\n\n{content}"
              )
              r = requests.post(
                  f"{BASE}/issues",
                  headers=HEADERS,
                  json={"title": issue_title, "body": body, "labels": [LABEL_NAME], "assignees": [assignee]},
              )
              r.raise_for_status()
              new_issue = r.json()
              print(f"Created issue #{new_issue['number']}: {new_issue['html_url']}")
          PYEOF