ActiveRecordStats

Provides Rails, Resque, and Sidekiq integrations for emitting metrics about ActiveRecord usage to StatsD using Shopify's statsd-instrument.

Installation

As usual:

gem 'active_record_stats', require: false

This library assumes that statsd has already been configured, and is available via the StatsD constant.

Usage

For Rails web request instrumentation, add the following to your config/application.rb:

require 'active_record_stats/rack_middleware'

module YourApp
  class Application < Rails::Application
    config.middleware.use 'ActiveRecordStats::RackMiddleware'
  end
end

For Resque job instrumentation, extend your job classes like so:

require 'active_record_stats/resque_plugin'

class SomeJob
  extend ActiveRecordStats::ResquePlugin
end

For Sidekiq instrumentation, enable the server middleware:

require 'active_record_stats/sidekiq_server_middleware'

Sidekiq.configure_server do |config|
  config.server_middleware do |chain|
    chain.add ActiveRecordStats::SidekiqServerMiddleware
  end
end

Metric names

Metrics will be with names in the following formats:

Gauge db.web.controller.action.statement: the number of statement queries performed for a given controller action.
Timer db.web.controller.action.runtime: the total db_runtime cost of a given controller action, in milliseconds.
Gauge db.job.class.statement: the number of statement queries performed for a given job.

As Graphite metric names may not contain a / character, they are replaced with two underscores (__). For sanity's sake, the same is done for :: delimiters.

Example metrics that might be emitted:

db.web.v1__widgets.update.UPDATE:2|g: two UPDATE statements were issued when processing V1::WidgetsController#update.
db.web.posts.create.runtime:123|g: Rails reported the total time spent in ActiveRecord for the PostsController#create action as 123 ms.
db.job.cache__prime.SELECT:10|g: ten SELECT statements were issued when processing the Cache::Prime job.

Caveats

This library does not actually parse SQL queries; it uses a very naive string munging tactic to identify the statement type involved. Because of this, some of the metrics are a bit nonsensical, such as WITH statements being reported when a query begins with a CTE.

License