Skip to main content

Continuous CPU profiling: flame graphs, always on

avuru obs team
Maintainers

The Profiling screen is live: cluster-wide CPU flame graphs, sampled by eBPF with zero app changes and stored in the same ClickHouse as your traces, logs and metrics. Experimental and opt-in for v0.1 (sensor.profiler.enabled=true) — the upstream alpha loader does not support every kernel yet.

  • Cluster-wide sampling. The sensor's profiler container (OpenTelemetry eBPF profiler, version-pinned) samples every process on every node at ~20 Hz — tune with sensor.profiler.samplesPerSecond.
  • Per-service flame graphs. Pick a profiled service, read its aggregated icicle for the selected window, click any frame to zoom into that subtree.
  • Efficient storage. Stacks are deduplicated (stored once, referenced by hash) with a dedicated retention knob (retention.profiles, default 3 days). The alpha OTLP Profiles format is isolated behind a single adapter seam in the hub, so upstream evolution stays contained.