Memory Leak & Resource Testing

Symptom: memory/handles/latency grow with time or request count and never plateau; ends with OOM-kill, EMFILE, or timeouts.

Detect (black-box first)

1. Run a soak test at steady load 1–24 h (performance-testing).
2. Chart RSS, heap-used, open FDs, thread count, DB connections, GC pause, queue depth. Healthy = sawtooth around a plateau; leak = rising baseline after each GC.
3. Force GC (where possible) and compare post-GC baselines at t0 vs t+N.

# Linux / container quick checks
ls /proc/$PID/fd | wc -l                     # FD count over time
ps -o rss,vsz,nlwp -p $PID                   # RSS, VSZ, threads
lsof -p $PID | awk '{print $5}' | sort | uniq -c | sort -rn | head
kubectl top pod --containers; kubectl describe pod | grep -i oomkilled

Find the owner (white-box)

Typical root causes

Unbounded caches/maps · listeners/subscriptions never removed · timers/intervals not cleared · closures holding large objects · detached DOM nodes · unclosed streams/cursors/connections · thread/goroutine leaks (blocked channel) · growing logs/metrics label cardinality · ThreadLocal misuse · session stores without TTL.

Regression test ideas

// Jest + node: loop operation and assert bounded growth
global.gc(); const a = process.memoryUsage().heapUsed;
for (let i=0;i<10000;i++) handle(sampleReq());
global.gc(); const b = process.memoryUsage().heapUsed;
expect(b - a).toBeLessThan(5 * 1024 * 1024);   // run with node --expose-gc

Also: repeat-navigation loop in Playwright then compare heap; goroutine count before/after (goleak.VerifyNone(t)); connection-pool active == 0 after test.

Report

Growth rate (MB/h), load profile, first-seen build, retained-object class + retainer path, minimal repro, suggested fix (bound/evict/close/unsubscribe).

Related

performance-testing, observability-testing, chaos-resilience-testing, mobile-testing