Blackfire Continuous Profiler
Low-overhead continuous profiling for app performance optimization.
Parca
Continuous eBPF profiling for cost-effective performance insights
Parca provides zero-instrumentation, continuous profiling across Kubernetes and systemd environments, delivering pprof-compatible data with low overhead to help reduce waste, boost performance, and troubleshoot incidents.
Overview
Overview
Parca delivers continuous, zero‑instrumentation profiling for modern cloud workloads. By leveraging eBPF, it automatically discovers targets in Kubernetes clusters or systemd services, collecting CPU and memory usage down to the line number with minimal overhead.
Capabilities
The profiler outputs standard pprof profiles, enabling seamless integration with existing tooling, while also ingesting external pprof data. Profiles are stored in an optimized, label‑driven store that supports powerful queries across dimensions such as version, region, or deployment.
Deployment
Deploy the Parca agent alongside your services; it runs as a single binary with configurable storage options and optional persistence. The web UI, accessible on port 7070 by default, visualizes trends over time, helping teams identify hot paths, reduce resource waste, and accelerate incident resolution.
Highlights
eBPF‑based auto‑discovery for Kubernetes and systemd targets
Standard pprof output and ingestion for broad language support
Label‑driven storage with efficient slicing and aggregation
Very low runtime overhead across supported languages
Pros
- Zero instrumentation required
- Low overhead profiling across the entire infrastructure
- Works with multiple native languages (C, C++, Rust, Go, etc.)
- Integrates with existing pprof tooling
Considerations
- Requires Linux kernel with eBPF support
- May need root privileges for eBPF programs
- Profile storage can grow without retention policies
- Learning curve for label‑based query language
Managed products teams compare with
When teams consider Parca, these hosted platforms usually appear on the same shortlist.
Looking for a hosted option? These are the services engineering teams benchmark against before choosing open source.
Fit guide
Great for
- Teams needing continuous performance visibility across microservices
- Organizations aiming to cut cloud costs through profiling
- SREs troubleshooting CPU, memory, or I/O spikes
- Developers of native applications seeking statistical hot‑path analysis
Not ideal when
- Environments lacking eBPF support (old kernels, non‑Linux OS)
- Purely interpreted language stacks without native binaries
- Small single‑node deployments where overhead outweighs benefit
- Teams without resources to manage profiling data retention
How teams use it
Identify CPU hot paths in a Kubernetes service
Pinpoint functions consuming the majority of CPU, enabling targeted optimizations that reduce resource usage.
Detect memory leaks in a Go application
Continuous memory profiling reveals growing allocations, allowing developers to fix leaks before they impact stability.
Compare performance across deployment versions
Label‑based queries show how code changes affect latency and CPU, supporting data‑driven rollbacks.
Correlate incidents with resource spikes
During outages, profiling data provides deep insight into execution paths, speeding root‑cause analysis.
Tech snapshot
TypeScript66%
Go32%
Shell1%
Jsonnet1%
JavaScript1%
CSS1%
Frequently asked questions
Do I need to modify my code to use Parca?
No. Parca uses eBPF to profile processes without any code changes.
Which programming languages are supported?
Parca works with any language that can produce pprof profiles and has native binaries, including C, C++, Rust, Go, and more.
How does Parca store profiling data?
Profiles are stored in an optimized columnar format, indexed by user‑defined labels for fast querying.
Can I ingest existing pprof files?
Yes, Parca can ingest any pprof‑formatted profile alongside data collected by its eBPF agent.
What are the system requirements?
A Linux kernel with eBPF support (typically 4.14 or newer) and sufficient storage for profile data.
Project at a glance
Active- Stars
- 4,746
- Watchers
- 4,746
- Forks
- 243
LicenseApache-2.0
Repo age7 years old
Last commityesterday
Primary languageTypeScript
Last synced yesterday