Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
A DSL for reasoning about and manipulating groups of HA Proxy instances.
Ruby
branch: master

Fetching latest commit…

Cannot retrieve the latest commit at this time

Failed to load latest commit information.
bin
lib
.gitignore
Gemfile
Gemfile.lock
README.md
example.csv
haproxy-cluster.gemspec

README.md

haproxy-cluster

"Can we survive a rolling restart?"

"How many transactions per second am I seeing?"

"What's my session backlog?"

While there are already a handfull of HA Proxy abstraction layers on RubyGems, I wanted to be able to answer questions like those above and more, quickly, accurately, and easily. So here's one more for the pile.

HAProxyCluster::Member provides an ORM for HA Proxy's status page.

HAProxyCluster provides a simple map/reduce-inspired framework on top of HAProxyCluster::Member.

haproxy_cluster provides a shell scripting interface for HAProxyCluster. Exit codes are meaningful and intended to be useful from Nagios.

Do you deploy new code using a sequential restart of application servers? Using this common pattern carelessly can result in too many servers being down at the same time, and cutomers seeing errors. haproxy_cluster can prevent this by ensuring that every load balancer agrees that all (or enough) servers are up at each stage in the deployment. In the example below, we will deploy a new WAR to three Tomcat instances which are fronted by two HA Proxy instances. HA Proxy has been configured with option httpchk /check, a path which only returns an affirmative status code when the application is ready to serve requests.

#!bin/bash
set -o errexit
servers="server1.example.com server2.example.com server3.example.com"
load_balancers="https://lb1.example.com:8888 http://lb2.example.com:8888"

for server in $servers ; do
    haproxy_cluster --eval "wait_until(:condition => true){ myapp.rolling_restartable? }" $load_balancers
    scp myapp.war $server:/opt/tomcat/webapps
done

The code block passed to --eval will not return until every load balancer reports that at least 80% of the backend servers defined for "myapp" are ready to serve requests (or all of them are down). This condition must pass 3 times in a row for it to be considered valid. If this takes more than 5 minutes (300 seconds), the whole deployment is halted.

Maybe you'd like to know how many transactions per second your whole cluster is processing.

haproxy_cluster --eval 'poll{ puts members.map{|m|m.myapp.rate}.inject(:+) }' $load_balancers

Perhaps there's a particular server which needs to be removed from the cluster quickly and temporarily.

haproxy_cluster --eval 'each_member{ myapp.servers.find("server3").disable! }' $load_balancers

Read the full documentation.

Installation

gem install haproxy-cluster

Requires Ruby 1.9.3 and depends on RestClient.

Non-Features

  • Doesn't try to modify configuration files. Use haproxy-tools, rhaproxy, haproxy_join, or better yet, Chef for that.
  • Doesn't talk to sockets, yet. Use haproxy-ruby for now if you need this. I intend to add support for this using Net::SSH and socat(1) but for now HTTP is enough for my needs. Please comment on Issue #6 if you care about this feature.

ProTip

HA Proxy's awesome creator Willy Tarrreau loves big text files and big, flat web pages. If smaller, hyperlinked documents are more your style, you should know about the two alternative documentation sources:


Build Status

Something went wrong with that request. Please try again.