Grafana Pyroscope — Continuous Profiling Backend

What it is

Grafana Pyroscope is a continuous profiling backend that records CPU, memory, goroutine, and lock profiles from production services. After joining Grafana Labs, it was merged with Phlare into a single Grafana-integrated backend that supports both pull (pprof-scrape) and push (agent SDK) ingestion.

It is designed for backend engineers and SREs who need to understand where their services spend time and memory in production, not just in local benchmarks.

How it saves time or tokens

Continuous profiling eliminates the guesswork of performance debugging. Instead of adding ad-hoc profiling code and redeploying, Pyroscope records profiles continuously with low overhead. The diff mode lets you compare two time ranges or deployments instantly, pinpointing which function calls regressed.

How to use

1. Run the all-in-one binary: docker run -d --name pyroscope -p 4040:4040 grafana/pyroscope:latest
2. Instrument your service with the Pyroscope SDK (Go, Java, Python, Ruby, .NET, Node.js, Rust)
3. View flamegraphs at http://localhost:4040 or in Grafana with the Pyroscope data source

Example

package main

import (
    "github.com/grafana/pyroscope-go"
)

func main() {
    pyroscope.Start(pyroscope.Config{
        ApplicationName: "my.service",
        ServerAddress:   "http://pyroscope:4040",
        ProfileTypes: []pyroscope.ProfileType{
            pyroscope.ProfileCPU,
            pyroscope.ProfileAllocObjects,
            pyroscope.ProfileAllocSpace,
            pyroscope.ProfileInuseObjects,
            pyroscope.ProfileInuseSpace,
        },
    })
    defer pyroscope.Stop()

    // Your application code here
}

# Run Pyroscope locally
docker run -d --name pyroscope -p 4040:4040 grafana/pyroscope:latest

# Open the UI
open http://localhost:4040

Related on TokRepo

• AI Tools for Monitoring -- Observability and monitoring tools
• AI Tools for DevOps -- Infrastructure and operations tooling

Common pitfalls

• CPU profiling overhead is typically under 2%, but memory profiling with high allocation rates can add noticeable overhead
• The pull mode requires your service to expose a pprof endpoint, which may need firewall rules in production
• Grafana integration requires configuring the Pyroscope data source; it does not appear automatically