go-torch

go-torch is deprecated. As of Go 1.11, flamegraph visualizations are available directly in go tool pprof.

To listen on a specific port and open a browser for visualization, use:

$ go tool pprof -http=":8081" [binary] [profile]

If you are on an older version of Go, you can install the standalone pprof tool:

$ go get -u github.com/google/pprof
$ pprof -http=":8081" [binary] [profile]

To allow go-torch to profile your application, you must expose the net/http/pprof endpoints.

If your application uses the DefaultServeMux, simply add this blank import:

import _ "net/http/pprof"

If you are using a custom multiplexer, you must manually register the net/http/pprof handlers.

You can install go-torch using go get or run it via Docker.

Using Go:

$ go get github.com/uber/go-torch

Using Docker: To avoid mounting volumes, use the -p flag to print the SVG to stdout and redirect it to a file:

$ docker run uber/go-torch -u http://[address-of-host] -p > torch.svg

The go-torch CLI is a stochastic flame graph profiler for Go programs. It takes a binary and a profile source (typically from pprof) and generates an SVG flame graph.

Usage Syntax: [options] [binary] <profile source>

Example: To generate a flame graph from a profile source using a specific binary:

go-torch ./my-binary profile.pb.gz

By default, go-torch hits http://localhost:8080/debug/pprof/profile for a 30-second CPU profile and writes the result to torch.svg.

Basic usage (default settings):

$ go-torch

Customize the base URL:

$ go-torch -u http://my-service:8080/

Customize the profiling duration:

$ go-torch --seconds 5

$ go-torch -u http://my-service:8080/
INFO[19:10:58] Run pprof command: go tool pprof -raw -seconds 30 http://my-service:8080/debug/pprof/profile
INFO[19:11:03] Writing svg to torch.svg

go-torch supports passing arguments directly to go tool pprof. This is useful if you have already generated a profile file (e.g., from a benchmark).

Example using a benchmark profile:

# 1. Create the profile
$ go test -bench . -cpuprofile=cpu.prof

# 2. Generate flame graph using go-torch
$ go-torch main.test cpu.prof

Passing extra pprof arguments: Arguments not handled by go-torch are passed through to pprof:

$ go-torch --alloc_objects main.test mem.prof

$ go-torch main.test cpu.prof
INFO[19:00:29] Run pprof command: go tool pprof -raw -seconds 30 main.test cpu.prof
INFO[19:00:29] Writing svg to torch.svg