Parca

What it is

Parca is an open-source continuous profiling project, originally created by Polar Signals and first introduced publicly in 2021 under the tagline “Prometheus, but for profiles.” Its core differentiator is that Parca Agent uses eBPF to sample every process running on a Linux host directly from the kernel, which means it can produce CPU (and, increasingly, other resource) profiles for an entire fleet without requiring per-language SDKs, code instrumentation, or even application restarts — a meaningfully lower-friction model than profilers that need a language-specific agent linked into each service. Profiles are stored and queried by the Parca server, a purpose-built columnar time-series-like backend for profiling data, and visualized as flame graphs (or icicle charts) that can be compared across time or between deployments.

Polar Signals, the company behind Parca, was acquired by Dash0 in August 2026. Parca itself is unaffected by that change: the project remains open source and Dash0 has stated it will continue maintaining it, including continuing Polar Signals’ upstream contributions to the OpenTelemetry eBPF profiler and to OpenTelemetry’s broader profiling signal, which reached alpha status in the OpenTelemetry specification in 2026. Over time, the columnar storage engine developed for Parca (“Great Lakes”) is expected to become part of Dash0’s own OpenTelemetry-native observability platform, correlating profiles with Dash0’s metrics, logs, and traces.

Why teams choose it

• No code changes, no per-language agents. eBPF-based sampling from the kernel means Parca can profile polyglot fleets — including languages or binaries you can’t easily instrument — uniformly, which is Parca’s single biggest differentiator versus SDK-based profilers.
• Genuinely low overhead, since sampling happens outside the target process rather than via an in-process agent, making always-on production profiling more palatable operationally.
• Purpose-built storage for profiling’s shape. Profiling data has very different cardinality and volume characteristics than metrics or traces, and Parca’s backend was designed specifically for that rather than being retrofitted from a general time-series database.
• Narrower symbolication support than some alternatives. eBPF-based profiling depends on accurate stack unwinding and symbolication, which historically has required more setup for some managed runtimes (JVM, for example) than native/compiled languages — worth checking against your actual stack before committing.

Pricing model

Parca is free, open-source software that you self-host and operate yourself, with no licensing cost. For teams that want the same eBPF-based whole-system approach without running the storage and query backend themselves, Polar Signals Cloud is the commercial managed version, priced separately under its own SaaS pricing and now operated as part of Dash0 following the 2026 acquisition.

Alternatives

Grafana Pyroscope is the most common open-source alternative, with broader SDK-based language support and tighter integration into the Grafana ecosystem, though historically less mature eBPF whole-system profiling than Parca. Datadog Continuous Profiler is a leading fully-managed commercial alternative for teams already standardized on Datadog. Pixie, an eBPF-based observability tool for Kubernetes, overlaps with Parca on the “no instrumentation required” philosophy but is scoped more broadly at auto-instrumented Kubernetes observability than at continuous profiling specifically.