Parca

Parca is an open-source, self-hosted continuous profiler for teams that can run Linux infrastructure and want long-term control of performance data. It is most convincing when always-on sampling, multi-language coverage, and queryable history matter more than a managed service.

Overview

Parca gathers CPU, memory, I/O, and other statistical profiles, retains them in built-in storage, and makes them available for analysis over time. Its sampling model is designed for continuous production profiling rather than occasional diagnostics. The project reports Agent usage below 1% CPU and below 200 MB of memory, with actual overhead dependent on how many targets are profiled.

Collection is not tied to one ingestion path. Profiles can be sent over gRPC, retrieved from targets over HTTP, or supplied in any valid pprof format. That makes Parca practical for consolidating profiles from existing producers as well as from its own Agent.

Key features

eBPF whole-system profiling

Parca Agent uses eBPF to capture both user-space and kernel-space stack traces. Automatic profiling covers C, C++, Go, Rust, .NET, Deno, Erlang, Java, Julia, Node.js, Wasmtime, PHP, Ruby, and Python. This breadth suits mixed-language services, while the Agent requires Linux kernel 4.18 or newer plus kernel access and privileged permissions to load eBPF programs. Restricted containers and hosts that cannot grant those permissions are poor candidates.

Queries and comparisons

Label selectors and a profiling-specific query engine let teams filter profile data and compare dimensions such as processes or code versions. The multi-dimensional model is useful for finding regressions across deployments instead of inspecting one capture at a time.

Visualization and integrations

The web interface renders profiles with Icicle Graphs, an upside-down Flame Graph presentation, and supports interactive querying. A Grafana plugin can display profiles stored in Parca, which helps teams place profiling alongside existing dashboards. Parca receives profiling statistics and function-name metadata rather than application binaries or source code.

Deployment and licensing

Installation is documented for binaries, containers, Kubernetes, and Snaps. Components are Apache 2-licensed, and Polar Signals offers paid support. The project is self-hosted, so the team running it remains responsible for deployment, storage, upgrades, and access controls.

Pricing

Parca Open Source — free. This is the open-source edition, with components under the Apache 2 License and a self-hosted deployment model. It is the natural fit for engineering teams that can operate the service themselves and want unrestricted control of their profiling stack. There are no paid tiers, seat charges, published quotas, or trial terms in this plan; commercial support is available separately from Polar Signals.

Platforms

Parca is offered through an API and web interface, runs self-hosted, and targets Linux for its Agent. The documented deployment options include binaries, containers, Kubernetes, and Snaps. The Agent’s kernel-version and privilege requirements are material platform constraints, not optional configuration.

Who it's for

Choose Parca if your team needs continuous CPU and memory profiling across production services, operates Linux hosts, and values keeping profile data in its own environment. It is particularly suitable for polyglot systems and organizations that already use Grafana or can feed pprof profiles from other tools.

Look elsewhere if you need a hosted service with no infrastructure ownership, cannot provide privileged eBPF access, or primarily profile a single local runtime. Parca’s flexibility rewards platform expertise but does not remove the operational work of running a self-hosted system.

Pros and cons

Pros

• Continuous, low-overhead sampling: reported usage below 1% CPU and 200 MB memory can make production-wide profiling viable.
• Broad runtime coverage: automatic profiling spans native, managed, and scripting languages.
• Flexible ingestion: gRPC, HTTP pull, and pprof support accommodate existing profile pipelines.
• Useful comparative analysis: labels and cross-dimension queries help isolate process and version differences.
• Open deployment: Apache 2 licensing and self-hosting avoid mandatory vendor storage.

Cons

• Privileged Linux agent: kernel access, Linux 4.18+, and permissions to load eBPF programs exclude some locked-down environments.
• Operational responsibility: self-hosting leaves installation, capacity, maintenance, and security to your team.
• Support costs extra: paid Polar Signals support is separate from the free software.

Alternatives

Grafana Cloud Application Observability is the better choice when you want a hosted, cross-platform service with an always-free allowance of 2,232 host hours and a usage-based Pro tier.

Scalene fits developers seeking a free, open-source Python profiler installed through pip or conda rather than a multi-service profiling backend.

YourKit .NET Profiler is aimed at teams needing a paid .NET-focused profiler with desktop platform support and licensing options.

Javadoc is a documentation generator included with JDK 25, not a substitute for continuous application profiling.

Node.js --prof is the focused option for generating V8 profiler output from Node.js applications.

py-spy suits Python-only sampling, with an MIT-licensed command-line profiler for Linux, macOS, and Windows.

Pyroscope is preferable when you want a managed freemium service, including a free tier with 50 GB monthly ingestion and 14-day retention.

Valgrind is a free GPL tool for Linux, macOS, and Android when deep instrumentation is worth its documented slowdown and higher memory use.

For broader comparisons, browse Profiling Software, Continuous Profiling Tools, and eBPF Observability Tools.

Verdict

Parca is a strong recommendation for technically capable teams that want continuous, low-overhead, self-hosted profiling with rich labels, comparisons, and Grafana access. Choose another tool when Linux privileges, infrastructure ownership, or a managed commercial experience are non-negotiable.