Performance profiling

Note

The recipe section of the documentation focuses on the objectives with minimal explanations about the basics.

Objective

I want to profile sbt to understand what it spent time on.

Solution: traces

  1. Shutdown an existing server if any: sbt shutdown
  2. Run a command with --server --traces
$ sbt --server --traces compile
....
$ ls target/traces/build.trace
target/traces/build.trace

This produces a Chrome tracing file target/traces/build.trace, which can be viewed using Perfetto, shows how long each task took.

Solution: timings

  1. Shutdown an existing server if any: sbt shutdown
  2. Run a command with --server --timings
$ sbt --server --timings compile
....
Total time: 4116 ms
  aaa-build / Compile / compileIncremental      : 731 ms
....

This outputs task timing information on the terminal.

Solution: flame graph

Install async-profiler by putting asprof and jfrconv on your PATH:

$ ln -s $HOME/Applications/async-profiler-4.2/bin/asprof $HOME/bin/asprof
$ ln -s $HOME/Applications/async-profiler-4.2/bin/jfrconv $HOME/bin/jfrconv

  1. Close anything that may affect the profiling, and run sbt in one terminal

  2. In another terminal run jps to identify the process ID of sbt

    $ jps
92746 sbt-launch.jar
92780 Jps

  3. Run asprof with the duration flag. -d 60 means 60s:

    $ asprof -d 60 -f /tmp/flamegraph.jfr <process id>

    This produces a JFR (Java Flight Recorder) file.

  4. Run jfrconv as follows to convert the *.jfr file to a flame graph:

    $ jfrconv --lines /tmp/flamegraph.jfr /tmp/flamegraph.html

Alternative: JVM profiling

You can use a profiler like VisualVM to identify implementation-level timings.

  1. Shutdown all existing servers if any: sbt shutdownall
  2. Start an sbt session with --server
  3. Launch VisualVM, and double-click on xsbt.boot.Boot
  4. Go to Sampler tab, and click CPU
  5. Run a command inside the sbt shell
  6. Click on the Stop button in Sampler tab

The threads named pool-n-thread-m are the task threads.