continuous-profilinglisted

# Continuous Profiling ## Purpose Continuous profiling moves a bounded set of collection decisions before an incident so that historical code-resource evidence remains queryable afterward. It does not make every future question answerable: the chosen event, interval, context, retention, stack fidelity, and backend schema define the evidence envelope. Treat it as a production telemetry system. It consumes CPU, memory, network, disk, backend compute, and operational attention on every instance; it can expose sensitive code and tenant metadata; and it can fail silently while its dashboard remains available. ## Ownership boundary This skill owns collection architecture, permanent budgets, context dimensions, storage, retention, comparison protocol, security, and operational health. Delegate: - event/engine mechanics to `async-profiler-advanced` and `jfr-advanced`; - one-off collection choice to `jfr-and-async-profiler`; - statistical interpretation and differential graphs to `latency-statistics` and `flame-graph-analysis`; - metric-label economics to `metrics-and-cardinality`; - system-wide eBPF/JIT symbol mechanics to `ebpf-for-jvm`. ## Define the retroactive questions Start with decisions, not a vendor: | Question | Required evidence | Common missing dimension | | --------------------------------------------------- | -------------------------------------