An endpoint has high CPU usage and slow responses. How would you profile it in Node?

First, reproduce the load, then capture a CPU profile. Run the load locally or on a staging server with a load tool such as autocannon.

The simplest ways to capture a profile:

• Start Node with --cpu-prof. It writes a .cpuprofile file when the process exits. You open that file in Chrome DevTools.
• Start Node with --inspect, attach DevTools, and record in the Performance or Profiler tab.
• --prof uses V8's tick profiler. You process its output with node --prof-process.

Read the result as a flame graph. Look for wide frames: these are functions where a lot of time is spent. Typical findings:

• Synchronous JSON work on big payloads.
• Expensive regexes.
• Synchronous crypto.
• Repeated work that should be cached.

Also watch the event loop delay with perf_hooks.monitorEventLoopDelay. This lets you connect high CPU use with slow responses.