While deploying a load balancer, we need some mechanism to inform loadbalancer to forward request to different set of servers based on the condition specified. Here I’m using HAProxy load balancer on CentOS 5 box. This is a very small test setup in Amazon EC2 environment having 3 small instances.

Let’s start the action. Login to server where you want to install HAProxy. Download it and extract it. You can download source and compile but as its a single executable file, I prefer to download precompiled file for being lazy

$ mkdir /usr/local/haproxy
$ cd /usr/local/haproxy
$ wget http://haproxy.1wt.eu/download/1.3/bin/haproxy-1.3.15.2-pcre-40kses-splice-linux-i586.notstripped.gz
$ gunzip haproxy-1.3.15.2-pcre-40kses-splice-linux-i586.notstripped.gz
$ mv haproxy-1.3.15.2-pcre-40kses-splice-linux-i586.notstripped haproxy
$ chmod 700 haproxy

As an example, we have two backend servers/domains, one to serve static contents and other for dynamic contents. Let’s create config file:

$ vi haproxy.cfg
 
  defaults
    balance roundrobin
    cookie SERVERID insert indirect</code>
 
  frontend www 10.252.130.162:80
    mode http
    acl dyn_content url_sub cgi-bin
    use_backend dyn_server if dyn_content
    default_backend stat_server
 
  backend dyn_server
    mode http
    server dserver1 dynamic.example.com:80 cookie A check
 
backend stat_server
    mode http
    server sserver1 static.example.com:80 cookie B check</code>

Here, 10.252.130.162 is the IP of your load balancer server. HAProxy configuration file has several sections called defaults, listen, frontend and backend. We used cookie to forward all subsequent requests from same user to same backend. The main thing here is the acl in frontend section which stats that if there’s a word “cgi-bin” in user’s url then use dyn_server as backend otherwise use default backend which is stat_server. You should refer HAProxy documentation for further information of configurations.

Save the file and check it for syntax:

$ ./haproxy -f ./haproxy.cfg -c

It will throw warnings regarding missing timeouts in sections, you can ignore these warnings. If there’s any error, check the config file again.

Run HAProxy:

$ ./haproxy3 -f ./haproxy.cfg -p /var/run/haproxy.pid

Put load balancer’s public IP/domain name in your browser and test this setup, it should go as expected.

You may also like to read:
* 5 steps to secure your Linux Server
* Ensuring secure access to production Linux Servers
* Bash script to backup essential log files in Linux
* Quickly change your ssh port from defualt 22 to something higher
* SSH port forwarding from remote to local machine
* Save root or user history to check later
* Install and configure denyhost to prevent brute force attacks

Step 1. Download perlbal OR install it via perl cpan, like this:

$ perl -MCPAN -e shell
cpan-> install perlbal

Step 2. Find out its sample config (/root/.cpan/build/Perlbal-1.60/doc/config-guide.txt) or if you downloaded and compiled it, file will be there. Put this file in /etc/perlbal as perlbal.conf.

$ mkdir /etc/perlbal
$ cp /root/.cpan/build/Perlbal-1.60/doc/config-guide.txt /etc/perlbal/perlbal.conf

Step 3. Update the perlbal.conf file as per your requirements.
for example, we are using it as load balancer, here is sample config

$ vi /etc/perlbal/perlbal.conf
CREATE POOL my_apaches
SET nodefile = conf/nodelist.dat  # IP of backend Apache servers.
 
CREATE SERVICE balancer
SET listen          = 0.0.0.0:80
SET role            = reverse_proxy
SET pool            = my_apaches
SET persist_client  = on
SET persist_backend = on
SET verify_backend  = on
ENABLE balancer
 
# Keep an internal management port open to reconfigure pool automatically via telnet
CREATE SERVICE mgmt
SET role   = management
SET listen = 127.0.0.1:60000
ENABLE mgmt

Step 4. Start perlbal module as daemon.

$ perlbal -d

Step 5. Test by connecting through management port

$ telnel 127.0.0.1 60000

Step 6. One major concern is that the backend servers will log entries with IP address of load balancer instead of actual user’s IP. To overcome this issue, install and configure mod_rpaf for apache at backend servers. Login in backend server and install the module:

$ cd /usr/src/
$ wget http://stderr.net/apache/rpaf/download/mod_rpaf-0.6.tar.gz
$ tar xzf mod_rpaf-0.6.tar.gz
$ cd mod_rpaf-0.6
$ apxs -i -c -n mod_rpaf-2.0.so mod_rpaf-2.0.c
$ vi /etc/httpd/conf/httpd.conf
LoadModule rpaf_module modules/mod_rpaf-2.0.so
RPAFenable On
RPAFsethostname On
RPAFproxy_ips 127.0.0.1 192.168.0.1  ## replace your IP of load balancer
RPAFheader X-Forwarded-For
 
$ service httpd restart

Testing

Step 1. Create a simple perl file and put it in cgi-bin directory of all backend servers.

$ vi /var/www/cgi-bin/test.pl
#!/usr/bin/perl
print "Content-Type: text/html\n\n";
print " Web server load balance testing ";
print "Web Server running fine.";
 
# Fetch IP of server
my $ifout =`ifconfig eth0 | grep inet`;
$ifout =~ m/\s+inet\saddr:([\d\.]+)\s.+/g;
 
print "This page is fetched from: $1 Server.";

Step 2. Start testing from browser and try to access the perl script from your load balancer server. Add more backends (by updating perlbal.conf) and test again. As per the load you will notice that perl script will be fetched from different back ends.

Adding / removing backends on the fly

Step 1. To maximum utilize the load balanced environment, there should be some technique by which backend servers can be added or removed on fly in the pool of perlbal as per the load. The topic of measuring load of backend servers and then making right decision is beyond the scope of this post. I developed some perl scripts to achieve the same in Amazon EC2[aws.amazon.com/ec2] environment, where we can create/remove servers on fly. If you are also on EC2, just post a comment or send mail to me and I will happily give scripts to you. Updating perlbal using a perl script with the help of Net::Telnet module is very easy. Here’s sample code:

use Net::Telnet;
$telnet=new Net::Telnet(Host=>$loadbalancerIP,Port=>60000,Timeout=>20, Errmode=>'Die');
$telnet->print("pool my_apaches add $newserverIP");
$telnet->waitfor('/OK/i');
$telnet->close;

Its Done. Play with it. One of the noticed minus point (as of ver 1.60), perlbal is not so efficient/capable while handling https connections. In case you have website which doesnt require https connections, perlbal should be given preferrence.

An alternate to perlbal is haproxy load balancer, I’ve covered it here, here and here as well.

You may also like to read:

* Install and configure haproxy load balancer, lightweight and fast alternative of perlbal/apache proxy.
* Enable or fix logging for Haproxy or perlbal load balancer.
* Install and setup haproxy load balancer for content switching.
* 5 steps to secure your Linux Server
* Ensuring secure access to production Linux Servers
* Bash script to backup essential log files in Linux