Parca
What it is
Parca is an open-source, self-hosted system for continuously collecting and querying CPU, memory, I/O, and other profiles over time. Parca Agent uses eBPF to capture user-space and kernel-space stack traces. Profiles can be sent over gRPC or gathered from targets over HTTP, and Parca accepts valid pprof-formatted profiles.
Its profiling-specific query engine supports label selectors and comparisons across dimensions such as processes or code versions. The web interface can query and visualize profiles with Icicle Graphs, and a Grafana plugin can display profiles stored in Parca. Sampling profiling is intended for continuous use in production; the project reports observed Agent overhead below 1% CPU and below 200 Mb memory, depending on active targets. The Agent requires Linux kernel version 4.18 or newer and privileged-user permissions to load eBPF programs. Parca says it receives profiling statistics and function-name metadata, not users’ binaries or source code. Components are available under the Apache 2 License, and paid support is offered for the open-source project.
For and against
For
- Collects CPU, memory, I/O, and other profiles
- Accepts valid pprof-formatted profiles
- Web UI includes Icicle Graph visualizations
- Grafana plugin displays stored profiles
Against
- Agent requires Linux kernel 4.18 or newer
- Agent needs privileged access to load eBPF programs
- Deployment is self-hosted
The verdict
Parca provides self-hosted collection, storage, and querying for ongoing profiling, with web and Grafana visualization options. Its Agent’s Linux and privilege requirements are important deployment considerations.
Who it suits
Parca suits teams that want to run their own continuous profiler and examine profiles across processes or code versions. It requires Linux kernel 4.18 or newer and privileged access for its Agent.
Plans and prices
Parca Open Source
Open source; components available under Apache 2 License
Read 7 Oct 2026 · parca.dev
On the record
- Agent
- Parca Agent is an eBPF-based whole-system profiler.parca.dev · read 7 Oct 2026
- Analysis
- Parca supports label-selector queries, a profiling-specific query engine, and comparisons across label dimensions such as processes or code versions.parca.dev · read 7 Oct 2026
- Commercial API
- The project identifies Polar Signals as the only commercial provider of Parca-compatible APIs, through Polar Signals Cloud.parca.dev · read 7 Oct 2026
- Data collection
- Profiles can be pushed to Parca through gRPC or pulled from targets through HTTP.parca.dev · read 7 Oct 2026
- Data handling
- Parca says it receives profiling statistics and function-name metadata, not users’ binaries or source code.parca.dev · read 7 Oct 2026
- Deployment
- The project documents installation using binaries, containers, Kubernetes, and Snaps.parca.dev · read 7 Oct 2026
- Features
- Features include a multi-dimensional data model, label-selector queries, a profiling-specific query engine, built-in storage, and support for pushing and pulling profiles.parca.dev · read 7 Oct 2026
- Grafana
- Profiles stored in Parca can be visualized using the Grafana plugin.parca.dev · read 7 Oct 2026
- Integration
- A Grafana plugin can visualize profiles stored in Parca.parca.dev · read 7 Oct 2026
- Languages
- The agent’s automatic profiling supports languages including C, C++, Rust, Go, Ruby, Node.js, Python, Java, .NET, Perl, and PHP.parca.dev · read 7 Oct 2026
- Linux requirement
- The quick start says the Parca Agent binary works only on Linux.parca.dev · read 7 Oct 2026
- Maker
- Polar Signals says it was founded in 2020.polarsignals.com · read 7 Oct 2026
- Open source
- Parca’s components are available under the Apache 2 License.parca.dev · read 7 Oct 2026
- Overhead
- The FAQ reports observed Agent overhead of less than 1% CPU and less than 200 Mb memory, depending on the number of actively profiled targets.parca.dev · read 7 Oct 2026
- Product
- Parca continuously collects, stores, and makes CPU, memory, I/O, and other profiles available for querying over time.parca.dev · read 7 Oct 2026
- Profiler
- Parca Agent uses eBPF to capture user-space and kernel-space stack traces with low overhead.parca.dev · read 7 Oct 2026
- Profiling
- Parca focuses on sampling profiling, which the project says can run continuously in production with very little overhead.parca.dev · read 7 Oct 2026
- Security
- The project says the Agent source is open source, it is written in Go, and its binaries and container images are fully reproducible; the Agent needs privileged access to load eBPF programs.parca.dev · read 7 Oct 2026
