Docker Torch builds flame graphs from pprof output from a docker daemon.
It uses github.com/uber/go-torch to read pprof data into a flame graph and
github.com/crosbymichael/docker-stress to provide workload against the Docker
daemon to measure.
$ cat stress-config.json | docker run -i --rm cpuguy83/docker-torch --host <host>:<port> --concurrency 8 --containers 2000 --time 60 > torch.svg
That will create 2000 containers with a concurrency of 8 while reading the
pprof output for 60 seconds output an SVG to stdout. If
-t is reached it will
stop creating containers.
Usage of /opt/flame-on/flame-on: -concurrency string max number of concurrent operations to Docker (default "1") -containers string number of containers to run (default "1000") -dockerversion string specify the version of Docker client to use (default "latest") -host string host:port to use to connect to docker -kill string time to kill a container after it is executed (default "10s") -time string amount of time to collect data (default "30")
Docker Torch does not spin up a Docker daemon for you, you must do this and make
sure to listen on a TCP port without TLS.