Event loop behavior, runtime performance, and resilient service design.
I capture runtime telemetry like event loop delay, CPU profiles, and async call stacks to separate blocking CPU work from I/O stalls. Once identified, I either optimize hot code paths, batch expensive operations, or move compute-heavy tasks to worker threads and queues.
Worker threads are best for CPU-heavy work that must stay close to request lifecycle and shared memory constraints. Queue workers are better for durable asynchronous jobs, retries, and horizontal scaling across machines.
I use timeouts, exponential backoff with jitter, circuit breakers for unstable dependencies, and request hedging only where idempotency is guaranteed. I also apply structured logging and tracing so failures are diagnosable across service boundaries.