Profiling

Build the server with memory profiling enabled:

cargo build --release -p riffle-server --features memory-prof

With the HTTP service configured and reachable, use the debug endpoints:

http://<server>:<http-port>/debug/heap/profile
http://<server>:<http-port>/debug/pprof/profile

The heap endpoint produces heap profiling data; the pprof endpoint captures a CPU profile for the configured sampling window.

Safe capture workflow

1. Identify one affected server from metrics before profiling.
2. Record the server version, workload window, and active configuration.
3. Capture a short baseline outside the incident window.
4. Capture the same duration during the problem.
5. Compare profiles together with request rate, memory pressure, and storage latency.

Profiling consumes CPU, memory, and I/O. Start with a short capture on one server and observe its impact before collecting from additional processes.