Autoperf (w/ HTTPerf.rb)

Autoperf is a ruby driver for httperf, designed to help you automate load and performance testing of any web application - for a single end point, or through log replay.

This has been refactored from the original -- github.com/igrigorik/autoperf -- to include HTTPerf.rb as a simplification and to include additional features. This has been refactored and released with permission from Ilya Grigorik, the original author.

More info on the original Autoperf: http://www.igvita.com/2008/09/30/load-testing-with-log-replay/.

Requirements

To get started, first download & install httperf: http://www.hpl.hp.com/research/linux/httperf/.

Or for HTTPerf.rb percentile support, install this fork: http://github.com/jmervine/httperf.

Rubies

It's important to note that tests pass for this on the following Ruby versions, but I've only used this at length on 1.9.2p290.

• 1.9.3p392
• 2.0.0p0

Note: I was also able to get tests to pass on 1.8.7 and 1.9.2, but not on Traivs and I would consider it unstable on both.

Installation

:::shell
gem install bundler
echo 'gem "autoperf"' >> Gemfile
bundle install

Without bundler:

:::shell
gem install autoperf

From source:

:::shell
git clone git://github.com/rubyops/autoperf.git
cd autoperf
gem build autoperf.gemspec
gem install ./autoperf-VERSION.gem

Usage

Next, either run a simple test straight from the command line (man httperf), or create an execution plan for autoperf. If you want to replay an access log from your production environment, follow these steps:

:::shell
# grab last 10000 lines from nginx log, and extract a certain pattern (if needed)
tail -n 10000 nginx.log | bundle exec make_replay_log > replay_log

Next, configure your execution plan (see sample.yml), and run autoperf:

:::shell
bundle exec autoperf -c sample.yml

Sample output:

+-------------------------------------------------------------------------------------------------------------------------------+
| rate | connection_rate_per_sec | request_rate_per_sec | connection_time_avg | errors_total | reply_status_5xx | net_io_kb_sec |
+-------------------------------------------------------------------------------------------------------------------------------+
|    5 | 154.4                   | 154.4                | 6.5                 | 0            | 0                | 1176.9        |
|   10 | 153.4                   | 153.4                | 6.5                 | 0            | 0                | 1169.5        |
|   15 | 153.0                   | 153.0                | 6.5                 | 0            | 0                | 1166.6        |
|   20 | 147.4                   | 147.4                | 6.8                 | 0            | 0                | 1123.4        |
+-------------------------------------------------------------------------------------------------------------------------------+

If your server uses caching, making it pointless to run the same requests over and over, you can use different requests for each run.

:::shell
# Create 10 1000-line files (xa, xb, xc etc)
split -a 1 requests_path

# Convert to null-terminated strings
for x in x?; do tr "\n" "\0" < $x > $x.nul; done

# run as before, but use the `wlog` line instead of `httperf_wlog` in the conf file
autoperf -c sample.yml