parca-agent0.46.0

parca-agent continuously samples running systems and captures profiling data with very low overhead, giving operators an ongoing view of where applications and the kernel spend time. It uses eBPF from Linux kernel space to collect user-space and kernel-space stack traces 100 times per second, then turns the extracted samples into pprof-formatted profiles. Local HTTP endpoints expose the collected profiles, and the agent can be configured to send them to a Parca server for querying and analysis over time.

Profiles reveal CPU, memory, and I/O bottlenecks down to method name, class name, and line number across applications written in any language or framework. The agent supports both pushing profiles to targets and pulling them from targets, with targets identified through service discovery or static configuration. Parca supplies a profiling-focused query engine with label-selector queries for examining the resulting data.

Parca Team’s agent is aimed at engineers operating production infrastructure who need continuous visibility without instrumenting every application. Historical profiles help identify hot paths, compare software versions or other dimensions, and pinpoint performance regressions after a deployment. They also provide detail for diagnosing memory leaks and brief CPU or I/O spikes that are difficult to reproduce.

Version v0.46.0 runs on Linux x86_64 (amd64) systems and requires snapd. It is open-source software designed to work with a Parca server when long-term storage and investigation are needed.