Observability Wiki
Open source

Grafana Pyroscope

Grafana Labs' open-source continuous profiling backend, storing and querying flame-graph data alongside metrics, logs, and traces.

Last updated

What it is

Grafana Pyroscope is an open-source continuous profiling database and backend, now maintained by Grafana Labs as part of its observability stack alongside Prometheus/Mimir (metrics), Loki (logs), and Tempo (traces). The project originated as the independent open-source Pyroscope project, which Grafana Labs acquired in 2023 and rebranded as Grafana Phlare before ultimately merging the two projects and settling on the Grafana Pyroscope name. It ingests profiling data — CPU, memory allocation, goroutine/thread state, and similar resource profiles — from language-specific SDKs, eBPF-based agents, or the OpenTelemetry Collector, stores it efficiently at scale, and renders it as flame graphs inside Grafana, correlated against the same labels used by the rest of the LGTM stack.

In 2026 Grafana Labs shipped Pyroscope 2.0, a rearchitected storage engine aimed at reducing storage cost and improving query performance and operational simplicity at scale. Pyroscope has also become one of the primary reference backends for OpenTelemetry’s profiling signal, which reached alpha status in the OpenTelemetry specification in 2026 — Pyroscope supports ingesting profiles natively over OTLP, positioning it as a first-class destination as profiling standardizes alongside the three established OpenTelemetry signals.

Why teams choose it

  • Native correlation with the rest of the Grafana stack. Because Pyroscope shares Grafana’s label model, teams can pivot directly from a slow trace span or a CPU metric spike into the exact flame graph for that service and time window, inside one UI.
  • OpenTelemetry alignment. As OTel profiling matures from alpha toward stability, Pyroscope’s early and direct OTLP support reduces the risk of picking a profiling backend that has to be replaced later.
  • Open-source with a real-cost story. Pyroscope 2.0’s storage redesign specifically targets the cost problem that has historically made continuous profiling (which generates much higher data volumes than metrics) expensive to retain at scale.
  • Younger and narrower than some alternatives. Pyroscope is less battle-tested at extreme scale than, say, Prometheus is for metrics, and eBPF-based whole-system profiling (no code changes required) is comparatively newer in Pyroscope than in eBPF-native tools like Parca.

Pricing model

Pyroscope is free, open-source software that can be self-hosted, with cost limited to your own storage and compute. Grafana Labs also offers it as a managed component of Grafana Cloud, billed under Grafana Cloud’s usage-based pricing (metered on ingested profiling data volume), with a free tier available as part of Grafana Cloud’s broader free allowance.

Alternatives

Parca, the eBPF-based open-source profiler from Polar Signals (now part of Dash0), takes a whole-system, no-instrumentation approach and is a common alternative or complement, particularly for infrastructure-level profiling. Datadog Continuous Profiler and Polar Signals Cloud (Parca’s commercial managed counterpart) are the leading fully-managed commercial alternatives for teams that don’t want to run profiling storage themselves.

Visit official site →