Push haproxy logs to newrelic rpm

This is useful for languages where newrelic does not provide an agent, such as erlang or node.js based applications.

It runs in syslog mode or tail a log file.

Installation

gem install haproxy2rpm or clone it

  • copy your newrelic.yml file to $HOME/.newrelic/newrelic.yml
  • or set $NRCONFIG to point to your newrelic.yml file

Tell haproxy to log to syslog. e.g:

# /etc/haproxy/haproxy.cfg
global
  log 127.0.0.1:3333   daemon

Running it

haproxy2rpm /path/to/logfile 
haproxy2rpm --syslog
haproxy2rpm --syslog --daemonize
haproxy2rpm --syslog -c /path/to/config/file.rb

Configuration

Check the examples folder

Problems with haproxy timers

Slow clients (e.g mobile clients over an edge connection) influence the measurements haproxy returns. The same happens for requests that do send a lot of data (e.g POST/PUT). In that case, the 'tr' value provided by haproxy is generally a bit higher. When you have many slow clients, it is significantely higher. For more information about the values haproxy provides, refer to the manual.

The issue of the default route

When haproxy2rpm does not find a route, it will record it under a default name so that NewRelic is not flooded with millions of routes. You can provide by setting 'config.default_route'.

Node.js support

Check the examples folder

Supported RPM features

  • Response times in application server
  • Error rate
  • Queue time
  • Apdex
  • Histogram
  • Web transactions

Supported Ruby versions

Tested in production

  • mri-1.9.2-p180
  • ree-1.8.7-2011.03
  • mri-1.8.7-p33

Passing manual integration test with test logs

  • mri-1.9.3
  • mri-1.9.2-p180
  • ree-1.8.7-2011.03
  • mri-1.8.7-p33
  • jruby 1.6.2

rbx-head fails to write to the local db file:

Error serializing data to disk: #<TypeError: incompatible marshal file format (can't be read)format version 4.8 required

Passing unit tests

  • mri-1.9.2-p180
  • ree-1.8.7-2011.03
  • mri-1.8.7-p33
  • rbx-head
  • jruby 1.6.2

Performance

No performance testing done so fare. Judging by the results of jordansissel, MRI 1.9.2 and jRuby should be the fastest.

Known issues

  • Daemonize is broken. Does not work nicely with new relic agent
  • Does not work with Rubinius

Roadmap

  • Working daemonized mode
  • remove haproxy dependency and make it a more generic rpm recorder. Maybe change the name to log2rpm or log4rpm
  • Automated benchmarks to see how fast it performs and what Ruby version might be best suited for a use case