Charon/docs/reports/archive/phase1_analysis.md

Phase 1.1: Test Execution Order Analysis

Date: February 2, 2026 Phase: Analyze Test Execution Order Duration: 30 minutes

Current Configuration Analysis

Project Dependency Chain (playwright.config.js:195-223)

setup (auth)
   ↓
security-tests (sequential, 1 worker, headless chromium)
   ↓
security-teardown (cleanup)
   ↓
┌──────────┬──────────┬──────────┐
│ chromium │ firefox  │ webkit   │  ← Parallel execution (no inter-dependencies)
└──────────┴──────────┴──────────┘

Configuration Details:

• Workers (CI): workers: 1 (Line 116) - Forces sequential execution
• Retries (CI): retries: 2 (Line 114) - Tests retry twice on failure
• Timeout: 90s per test (Line 108)
• Dependencies: Browser projects depend on setup and security-tests, NOT on each other

Why Sequential Execution Amplifies Failure

The Problem:

With workers: 1 in CI, Playwright runs ALL projects sequentially in a single worker:

Worker 1: [setup] → [security-tests] → [security-teardown] → [chromium] → [firefox] → [webkit]

When Chromium encounters an interruption (not a normal failure):

1. Error: Target page, context or browser has been closed at test #263
2. This is an INTERRUPTION, not a normal test failure
3. The worker encounters an unrecoverable error (browser context closed unexpectedly)
4. Playwright terminates the worker to prevent cascading failures
5. Since there's only 1 worker, the entire test run terminates
6. Firefox and WebKit never start - marked as "did not run"

Root Cause: The interruption is treated as a fatal worker error, not a test failure.

Interruption vs Failure

Type	Behavior	Impact
Normal Failure	Test fails assertion, runner continues	Next test runs
Interruption	Browser/context closed unexpectedly	Worker terminates
Timeout	Test exceeds 90s, marked as timeout	Next test runs
Error	Uncaught exception, test marked as error	Next test runs

Interruptions are non-recoverable - they indicate the test environment is in an inconsistent state.

Current GitHub Actions Architecture

Current workflow uses matrix sharding:

strategy:
  matrix:
    shard: [1, 2, 3, 4]
    browser: [chromium, firefox, webkit]

This creates 12 jobs:

• chromium-shard-1, chromium-shard-2, chromium-shard-3, chromium-shard-4
• firefox-shard-1, firefox-shard-2, firefox-shard-3, firefox-shard-4
• webkit-shard-1, webkit-shard-2, webkit-shard-3, webkit-shard-4

BUT: All jobs run in the same e2e-tests job definition. If one browser has issues, it affects that browser's shards only.

The issue: The sharding is already browser-isolated at the GitHub Actions level. The problem is likely in local testing or in how the interruption is being reported.

Analysis Conclusion

Finding: The GitHub Actions workflow is ALREADY browser-isolated via matrix strategy. Each browser runs in separate jobs.

The Real Problem:

1. The diagnostic report shows Chromium interrupted at test #263
2. Firefox and WebKit show "did not run" (0 tests executed)
3. This suggests the issue is in the Playwright CLI command or local testing, NOT GitHub Actions

Next Steps:

1. Verify if the issue is in local testing vs CI
2. Check if there's a project dependency issue in playwright.config.js
3. Implement Phase 1.2 hotfix to ensure complete browser isolation
4. Add diagnostic logging to capture the actual interruption error

Recommendation: Proceed with Phase 1.2 to add explicit browser job separation and enhanced logging.