Trashed

Keep an eye on resource usage.

• Sends per-request object counts, heap growth, GC time, and more to StatsD.
• Sends snapshots of resource usage, e.g. live String objects, to StatsD.
• Supports new stuff: Rails 5.1 and latest Ruby 2.x features.
• Supports old stuff: Rails 2/3/4, Ruby 1.9+, REE, Ruby 1.8 with RubyBench patches.

Setup

Rails 5

On Rails 5 (and Rails 3 and 4), add this to the top of config/application.rb:

require 'trashed/railtie'

And in the body of your app config:

module YourApp
  class Application < Rails::Application
    config.trashed.statsd = YourApp.statsd

Rails 2

On Rails 2, add the middleware to config/environment.rb:

Rails::Initializer.run do |config|
  reporter = Trashed::Reporter.new
  reporter.logger = Rails.logger
  reporter.statsd = YourApp.statsd

  config.middleware.use Trashed::Rack, reporter
end

Custom dimensions

You probably want stats per controller, action, right?

Set a #timing_dimensions lambda to return a list of dimensions to qualify per-request measurements like time elapsed, GC time, objects allocated, etc.

For example:

config.trashed.timing_dimensions = ->(env) do
  # Rails 3, 4, and 5, set this. Other Rack endpoints won't have it.
  if controller = env['action_controller.instance']
    name    = controller.controller_name
    action  = controller.action_name
    format  = controller.rendered_format || :none
    variant = controller.request.variant || :none  # Rails 4.1+ only!

    [ :All,
      :"Controllers.#{name}",
      :"Actions.#{name}.#{action}.#{format}+#{variant}" ]
  end
end

Results in metrics like:

YourNamespace.All.Time.wall
YourNamespace.Controllers.SessionsController.Time.wall
YourNamespace.Actions.SessionsController.index.json+phone.Time.wall

Similarly, set a #gauge_dimensions lambda to return a list of dimensions to qualify measurements which gauge current state, like heap slots used or total number of live String objects.

For example:

config.trashed.gauge_dimensions = ->(env) {
  [ :All,
    :"Stage.#{Rails.env}",
    :"Hosts.#{`hostname -s`.chomp}" ]
}

Results in metrics like:

YourNamespace.All.Objects.T_STRING
YourNamespace.Stage.production.Objects.T_STRING
YourNamespace.Hosts.host-001.Objects.T_STRING

Version history

3.2.8 (January 31, 2022)

• REE: Fix that GC.time is reported in microseconds instead of milliseconds

3.2.7 (November 8, 2017)

• Ruby 1.8.7 compatibility

3.2.6 (June 21, 2017)

• Mention Rails 5 support

3.2.5 (Feb 26, 2015)

• Support Ruby 2.2 GC.stat naming, avoiding 2.1 warnings

3.2.4 (July 25, 2014)

• Fix compatibility with Rails 3.x tagged logging - @calavera

3.2.3 (June 23, 2014)

• Report CPU/Idle time in tenths of a percent

3.2.2 (March 31, 2014)

• Reduce default sampling rates.
• Stop gauging all GC::Profiler data. Too noisy.
• Report gauge readings as StatsD timings.
• Support providing a Statsd::Batch since using Statsd#batch results in underfilled packets at low sample rates.
• Fix bug with sending arrays of timings to StatsD.
• Record GC timings in milliseconds.

3.1.0 (March 30, 2014)

• Report percent CPU/idle time: Time.pct.cpu and Time.pct.idle.
• Measure out-of-band GC count, time, and stats. Only meaningful for single-threaded servers like Unicorn. But then again so is per-request GC monitoring.
• Support @tmm1's GC::OOB (https://github.com/tmm1/gctools).
• Measure time between GCs.
• Spiff up logger reports with more timings.
• Support Rails log tags on logged reports.
• Allow instruments' #start to set timings/gauges.

3.0.1 (March 30, 2014)

• Sample requests to instrument based on StatsD sample rate.

3.0.0 (March 29, 2014)

• Support new Ruby 2.0 and 2.1 GC stats.
• Gauge GC details with GC::Profiler.
• Performance rework. Faster, fewer allocations.
• Rework counters and gauges as instruments.
• Batch StatsD messages to decrease overhead on the server.
• Drop NewRelic samplers.

2.0.5 (December 15, 2012)

• Relax outdated statsd-ruby dependency.

2.0.0 (December 1, 2011)

• Rails 3 support.
• NewRelic samplers.

1.0.0 (August 24, 2009)

• Initial release.