Metriks Client

This is an experiment in making a threadsafe, low impact library to measure aspects of your ruby.

The library is very much a work-in-progress. It is being developed as I find needs while developing Papertrail.

Installing

The API is still in flux, but you can add this to your project by installing the gem.

To install, add this to your Gemfile:

gem 'metriks'

and re-run bundle.

Metric API Overview

Counters

Basic atomic counter. Used as an underlying metric for many of the other more advanced metrics.

increment(incr = 1)

Increment the counter. Without an argument it will increment by 1.

  counter = Metriks.counter('calls')
  counter.increment

decrement(decr = 1)

Decrement the counter. Without an argument it will decrement by 1.

  counter = Metriks.counter('calls')
  counter.decrement

count()

Return the current value of the counter.

  counter = Metriks.counter('calls')
  puts "counter: #{counter.count}"

Meters

A meter that measures the mean throughput and the one-, five-, and fifteen-minute exponentially-weighted moving average throughputs.

mark(val = 1)

Record an event with the meter. Without an argument it will record one event.

  meter = Metriks.meter('requests')
  meter.mark

count()

Returns the total number of events that have been recorded.

  meter = Metriks.meter('requests')
  puts "total: #{meter.count}"

one_minute_rate()

Returns the one-minute average rate.

  meter = Metriks.meter('requests')
  puts "rate: #{meter.one_minute_rate}/sec"

five_minute_rate()

Returns the five-minute average rate.

  meter = Metriks.meter('requests')
  puts "rate: #{meter.five_minute_rate}/sec"

fifteen_minute_rate()

Returns the fifteen-minute average rate.

  meter = Metriks.meter('requests')
  puts "rate: #{meter.fifteen_minute_rate}/sec"

mean_rate()

Returns the mean (average) rate of the events since the start of the process.

  meter = Metriks.meter('requests')
  puts "rate: #{meter.mean_rate}/sec"

Timers

A timer that measures the average time as well as throughput metrics via a meter.

update(duration)

Records the duration of an operation. This normally wouldn't need to be called — the #time method is provided to simplify recording a duration.

  timer = Metriks.timer('requests')
  t0 = Time.now
  work
  timer.update(Time.now - t0)

time(callable = nil, &block)

Measure the amount of time a proc takes to execute. Takes either a block or an object responding to #call (normally a proc or lambda).

  timer = Metriks.timer('requests')
  timer.time do
    work
  end

If neither a block or an object is passed to the method, an object that responds to #stop will be returned. When #stop is called, the time will be recorded.

  timer = Metriks.timer('requests')
  t = timer.time
  work
  t.stop

count()

Returns the number of measurements that have been made.

  timer = Metriks.timer('requests')
  puts "calls: #{timer.count}"

one_minute_rate()

Returns the one-minute average rate.

  meter = Metriks.timer('requests')
  puts "rate: #{meter.one_minute_rate}/sec"

five_minute_rate()

Returns the five-minute average rate.

  meter = Metriks.timer('requests')
  puts "rate: #{meter.five_minute_rate}/sec"

fifteen_minute_rate()

Returns the fifteen-minute average rate.

  meter = Metriks.timer('requests')
  puts "rate: #{meter.fifteen_minute_rate}/sec"

mean_rate()

Returns the mean (average) rate of the events since the start of the process.

  meter = Metriks.timer('requests')
  puts "rate: #{meter.mean_rate}/sec"

min()

Returns the minimum amount of time spent in the operation.

  meter = Metriks.timer('requests')
  puts "time: #{meter.min} seconds"

max()

Returns the maximum time spent in the operation.

  meter = Metriks.timer('requests')
  puts "time: #{meter.max} seconds"

mean()

Returns the mean (average) time spent in the operation.

  meter = Metriks.timer('requests')
  puts "time: #{meter.mean} seconds"

stddev()

Returns the standard deviation of the mean spent in the operation.

  meter = Metriks.timer('requests')
  puts "time: #{meter.stddev} seconds"

Utilization Timer

A specialized Timer that calculates the percentage (between 0.0 and 1.0) of wall-clock time that was spent. It includes all of the methods of Timer.

one_minute_utilization()

Returns the one-minute average utilization as a percentage between 0.0 and 1.0.

  meter = Metriks.utilization_timer('requests')
  puts "utilization: #{meter.one_minute_utilization * 100}%"

five_minute_utilization()

Returns the five-minute average utilization as a percentage between 0.0 and 1.0.

  meter = Metriks.utilization_timer('requests')
  puts "utilization: #{meter.five_minute_utilization * 100}%"

fifteen_minute_utilization()

Returns the fifteen-minute average utilization as a percentage between 0.0 and 1.0.

  meter = Metriks.utilization_timer('requests')
  puts "utilization: #{meter.fifteen_minute_utilization * 100}%"

mean_utilization()

Returns the mean (average) utilization as a percentage between 0.0 and 1.0 since the process started.

  meter = Metriks.utilization_timer('requests')
  puts "utilization: #{meter.mean_utilization * 100}%"

Reporter Overview

How to get metrics out of the process.

Graphite Reporter

Sends metrics to Graphite every 60 seconds.

  reporter = Metriks::Reporter::Graphite.new 'localhost', 3004
  reporter.start

Logger Reporter

Send metrics to a logger every 60 seconds.

  reporter = Metriks::Reporter::Logger.new(:logger => Logger.new('log/metrics.log'))
  reporter.start

Librato Metrics Reporter

Send metrics to Librato Metrics every 60 seconds.

  reporter = Metriks::Reporter::LibratoMetrics.new('email', 'token')
  reporter.start

Proc Title Reporter

Provides a simple way to get up-to-date statistics from a process by updating the proctitle every 5 seconds (default).

  reporter = Metriks::Reporter::ProcTitle.new :interval => 5
  reporter.add 'reqs', 'sec' do
    Metriks.meter('rack.requests').one_minute_rate
  end
  reporter.start

will display:

501      17015 26.0  1.9 416976 246956 ?       Ss   18:54  11:43 thin reqs: 273.3/sec

Application Server Configuration

Depending on how your application server operates, you may need to configure how reporters are created. Please look at Troubleshooting for more information.

Plans

An incomplete list of things I would like to see added:

• Rack middleware to measure utilization, throughput and worker time
• Basic reporters:
  • Rack endpoint returning JSON
  • Statsd reporter
• Metaprogramming instrumentation hooks like Shopify's statsd-instrument

Credits

Most of the inspiration for this project comes from Coda Hale's amazing Metrics, Metrics Everywhere talk at CodeConf and his sweet Metrics Java Library.

License

Copyright (c) 2012 Eric Lindvall

Published under the MIT License, see LICENSE