Garbage Collection and Latency Tuning Review
Diagnose pauses and tail latency in managed runtimes.
The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.
Help me diagnose pauses and tail latency in a managed runtime by interviewing me first. Ask one question at a time and wait: (1) runtime, version, and collector settings, (2) container or machine memory and CPU limits, (3) heap size and live data size, (4) what the symptoms look like (p99 spikes, periodic stalls, out-of-memory kills), (5) allocation-heavy parts of the code, (6) what metrics and GC logs I have, and (7) the latency target. What I know now: [SYMPTOMS_AND_METRICS]. After the interview, give a diagnosis plan: how to enable and read GC logs for my runtime; the metrics to chart (pause time, frequency, heap after collection, allocation rate, promotion rate); a table of likely causes (heap too small or too large, container limits misread by the runtime, large object allocation, excessive short-lived garbage, finalizers, memory leaks) with the evidence for each; three experiments with one change each (heap size, collector choice, allocation reduction in the hot path); and how to compare results with a repeatable load test. Warn about tuning flags copied from the internet. Say clearly where you are guessing.
Fill in before use
Optional: type your details and the prompt above updates. Nothing is saved or sent.
Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.