Root cause: With BROWSER_COUNT=1, the hourly browser restart set
restarting=true, drained all pages, closed the browser, THEN launched
a new one. During that window (seconds), all requests queued and
timed out after 30s with QUEUE_FULL errors.
Fix: Launch the new browser BEFORE closing the old one (hot-swap).
This ensures zero downtime during browser recycling, even with a
single browser instance.
Performance fixes:
- Default waitUntil changed from networkidle2 to domcontentloaded (saves ~500ms+)
- Add waitUntil parameter so users can choose (load/domcontentloaded/networkidle0/networkidle2)
- Optimize page recycle: use DOM reset instead of about:blank navigation
- Add Chromium flags to disable unnecessary features (background networking, extensions, sync, etc.)
Swagger fixes:
- Fix apis glob to include dist/*.js (was only matching src/*.ts, empty at runtime)
- Document new waitUntil parameter on POST /v1/screenshot
- Add OpenAPI docs for /status endpoint
- Extract complete codebase from running staging pod
- Add Dockerfile with multi-stage build for Node.js + Puppeteer
- Configure CI/CD workflows for staging and production deployment
- Include all source files, configs, and public assets
