Contributor Guide
Performance Investigation

Performance Investigation

This document provides information on how to obtain performance insight on Orchestrion, when trying to identify performance optimization opportunities.

Built-in Profiling

Orchestrion has built-in support for profiling; which can be turned on by specifying command line arguments to an orchestrion command. These flags are not show by the output of orchestrion --help as they are only intended for deep troubleshooting:

  • -profile-path=DIR specifies the path to a directory where profiler outputs will be written. The directory will be created if it does not exist.
  • -profile=[cpu|heap|trace] determines which standard Go profilers are enabled during the execution. This can be specified multiple times to enable multiple profilers.

Note that enabling the profiles may affect the performance of builds (how exactly depends on the peorifler being used). It may be a good idea to slightly increase any build timeout that may prevent the build from completing fully.

The main use-case for these involve using the cpu or heap profilers, then using the go tool pprof tool to combine all generated profiles (a build involves a great many orchestrion processes) into a single one, then investigating it in some graphical user interface:

$ orchestrion --profile-path="$PWD/profiles" --profile=cpu go build .
$ go tool pprof -proto $PWD/profiles/*.pprof > profile.pprof
$ go tool pprof -http=localhost:6060 profile.pprof

The generated profile does not contain direct information about the code being compiled, and should typically be safe for users to share with maintainers. When investigating customer performance issues that cannot be reproduced (e.g, because there is no simple reproduction and their code is private), consider asking customers to submit both a cpu and heap profile for investigation.

go tool pprof

For more information on how to use go tool pprof, you may refer to the following resources:

Last updated on
Introduction