Field notes · 6 min read

Event-loop lag symptoms that look like database faults

Your APM shows database query duration climbing. DBAs insist indexes are fine. Meanwhile p99 latency on unrelated read-only endpoints climbs in sync. That pattern often means the event loop is starved, and every I/O callback waits longer in queue.

Symptoms that overlap with DB slowness

• Uniform latency increase across routes that share no tables
• Health checks timing out while CPU sits below 40%
• Spiky GC pause metrics adjacent to user-facing errors

Quick checks in production

Expose eventLoopUtilization() or use your APM's Node runtime tab if available. Compare lag against deploy timestamps and cron schedules—month-end report generation running inside the API process is a frequent culprit.

Isolate CPU work

Move PDF generation, large JSON serialisation, and synchronous bcrypt rounds off the request thread. Worker threads or a dedicated job consumer are boring solutions that actually stick.

Need a second opinion while production is noisy? Describe your graphs and we will suggest the next measurement before you scale database hardware.