Files
Charon/docs/implementation/ci_remediation_summary.md
GitHub Actions 21b52959f5 chore: e3e triage
2026-02-05 11:00:56 +00:00

1.9 KiB

CI Remediation Summary

Date: February 5, 2026 Task: Stabilize E2E testing pipeline and fix workflow timeouts.

Problem

The end-to-end (E2E) testing pipeline was experiencing significant instability, characterized by:

  1. Workflow Timeouts: Shard 4 was consistently timing out (>20 minutes), obstructing the CI process.
  2. Missing Dependencies: Security jobs for Firefox and WebKit were failing because they lacked the required Chromium dependency.
  3. Flaky Tests:
    • certificates.spec.ts failed intermittently due to race conditions when ensuring either an empty state or a table was visible.
    • crowdsec-import.spec.ts failed due to transient locks on the backend API.

Solution

Workflow Optimization

  • Shard Rebalancing: Reduced the number of shards from 4 to 3. This seemingly counter-intuitive move rebalanced the test load, preventing the specific bottlenecks that were causing Shard 4 to hang.
  • Dependency Fix: Explicitly added the Chromium installation step to Firefox and WebKit security jobs to ensure all shared test utilities function correctly.

Test Logic Improvements

  • Robust Empty State Detection: Replaced fragile boolean checks with Playwright's .or() locator pattern.
    • Old: isVisible().catch() (Bypassed auto-waits, led to race conditions)
    • New: expect(locatorA.or(locatorB)).toBeVisible() (Leverages built-in retry logic)
  • Resilient API Retries: Implemented .toPass() for the CrowdSec import test.
    • This allows the test to automatically retry the import request with exponential backoff if the backend is temporarily locked or busy, significantly reducing flakes.

Results

  • Stability: The "Empty State OR Table" flake in certificates is resolved.
  • Reliability: CrowdSec import tests now handle transient backend states gracefully.
  • Performance: CI jobs now complete within the allocated time budget with balanced shards.