debian-mirror-gitlab/doc/development/profiling.md
2021-02-22 17:27:13 +05:30

5.7 KiB

stage group info
none unassigned To determine the technical writer assigned to the Stage/Group associated with this page, see https://about.gitlab.com/handbook/engineering/ux/technical-writing/#assignments

Profiling

To make it easier to track down performance problems GitLab comes with a set of profiling tools, some of these are available by default while others need to be explicitly enabled.

Profiling a URL

There is a Gitlab::Profiler.profile method, and corresponding bin/profile-url script, that enable profiling a GET or POST request to a specific URL, either as an anonymous user (the default) or as a specific user.

The first argument to the profiler is either a full URL (including the instance hostname) or an absolute path, including the leading slash.

When using the script, command-line documentation is available by passing no arguments.

When using the method in an interactive console session, any changes to the application code within that console session is reflected in the profiler output.

For example:

Gitlab::Profiler.profile('/my-user')
# Returns a RubyProf::Profile for the regular operation of this request
class UsersController; def show; sleep 100; end; end
Gitlab::Profiler.profile('/my-user')
# Returns a RubyProf::Profile where 100 seconds is spent in UsersController#show

For routes that require authorization you must provide a user to Gitlab::Profiler. You can do this like so:

Gitlab::Profiler.profile('/gitlab-org/gitlab-test', user: User.first)

Passing a logger: keyword argument to Gitlab::Profiler.profile sends ActiveRecord and ActionController log output to that logger. Further options are documented with the method source.

Gitlab::Profiler.profile('/gitlab-org/gitlab-test', user: User.first, logger: Logger.new(STDOUT))

There is also a RubyProf printer available: Gitlab::Profiler::TotalTimeFlatPrinter. This acts like RubyProf::FlatPrinter, but its min_percent option works on the method's total time, not its self time. (This is because we often spend most of our time in library code, but this comes from calls in our application.) It also offers a max_percent option to help filter out outer calls that aren't useful (like ActionDispatch::Integration::Session#process).

There is a convenience method for using this, Gitlab::Profiler.print_by_total_time:

result = Gitlab::Profiler.profile('/my-user')
Gitlab::Profiler.print_by_total_time(result, max_percent: 60, min_percent: 2)
# Measure Mode: wall_time
# Thread ID: 70005223698240
# Fiber ID: 70004894952580
# Total: 1.768912
# Sort by: total_time
#
#  %self      total      self      wait     child     calls  name
#   0.00      1.017     0.000     0.000     1.017       14  *ActionView::Helpers::RenderingHelper#render
#   0.00      1.017     0.000     0.000     1.017       14  *ActionView::Renderer#render_partial
#   0.00      1.017     0.000     0.000     1.017       14  *ActionView::PartialRenderer#render
#   0.00      1.007     0.000     0.000     1.007       14  *ActionView::PartialRenderer#render_partial
#   0.00      0.930     0.000     0.000     0.930       14   Hamlit::TemplateHandler#call
#   0.00      0.928     0.000     0.000     0.928       14   Temple::Engine#call
#   0.02      0.865     0.000     0.000     0.864      638  *Enumerable#inject

To print the profile in HTML format, use the following example:

result = Gitlab::Profiler.profile('/my-user')

printer = RubyProf::CallStackPrinter.new(result)
printer.print(File.open('/tmp/profile.html', 'w'))

GitLab-Profiler is a project that builds on this to add some additional niceties, such as allowing configuration with a single YAML file for multiple URLs, and uploading of the profile and log output to S3.

Sherlock

Sherlock is a custom profiling tool built into GitLab. Sherlock is only available when running GitLab in development mode and when setting the environment variable ENABLE_SHERLOCK to a non empty value. For example:

ENABLE_SHERLOCK=1 bundle exec rails s

Recorded transactions can be found by navigating to /sherlock/transactions.

Bullet

Bullet is a Gem that can be used to track down N+1 query problems. Bullet section is displayed on the performance-bar.

Bullet

Because Bullet adds quite a bit of logging noise the logging is disabled by default. To enable the logging, set the environment variable ENABLE_BULLET to a non-empty value before starting GitLab. For example:

ENABLE_BULLET=true bundle exec rails s

Bullet logs query problems to both the Rails log as well as the Chrome console.

As a follow up to finding N+1 queries with Bullet, consider writing a QueryRecoder test to prevent a regression.

Settings that impact performance

  1. development environment by default works with hot-reloading enabled, this makes Rails to check file changes every request, and create a potential contention lock, as hot reload is single threaded.
  2. development environment can load code lazily once the request is fired which results in first request to always be slow.

To disable those features for profiling/benchmarking set the RAILS_PROFILE environment variable to true before starting GitLab. For example when using GDK:

  • create a file env.runit in the root GDK directory
  • add export RAILS_PROFILE=true to your env.runit file
  • restart GDK with gdk restart

This environment variable is only applicable for the development mode.