Clustering and High Availability

Introduction

Clustering is the concept of using a load balancer to distribute connections to multiple destinations. Three of the most common load balancing methods used in clustering are:

  • Round robin = Send each request to the next server in the pool.

  • Least connections = Send requests to the server with the least amount of connections.

  • Source IP = Send requests from a source to the same destination server.

Ideally, high availability of services should also be of high importance to keep services working 100% of the time. [1]

IP Virtual Server

The IP Virtual Server (IPVS) service utilizes the Linux kernel directly for load balancing. It is designed to be a simple load balancer for internal/private networks. [2]

The only prerequisite for IPVS is that it requires the Linux system to be configured as a router (masquerading). This is an example of using the internal network 10.0.0.0/24 on the interface eth1 and forward requests to/from the public interface eth0.

$ sudo iptables -F
$ sudo iptables -t nat -F
$ sudo iptables -P INPUT ACCEPT
$ sudo iptables -P OUTPUT ACCEPT
$ sudo iptables -P FORWARD ACCEPT
$ sudo iptables -A FORWARD -i eth1 -s 10.0.0.0/255.255.255.0 -j ACCEPT
$ sudo iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE

IPVS is easily managed via the “ipvsadm” command.

  • View the current configuration.

    $ sudo ipvsadm -l

  • View more verbose information about current connections. [3]

    $ sudo ipvsadm -l -n --stats

  • Define the public IP address of the IPVS server (ex., 192.168.1.10), port (ex., :80 for HTTP), and then the type of load balancing (ex., “rr”).

    $ sudo ipvsadm -A -t <BALANCERIP>:<PORT> -s <SCHEDULER>

    $ sudo ipvsadm -A -t 192.168.1.10:80 -s rr

    • Types of clustering schedulers for “-s”:

      • rr = Round robin.

      • lc = Least connections.

      • sh = Source hash (source IP address).

  • Add a back-end server (-a), serving TCP connections (-t), using the specified public balancer IP and port, sending requests to the real back-end server’s IP address (-r), and masquerade/NAT the requests (-m).

    $ sudo ipvsadm -a -t <BALANCERIP>:<PORT> -r <DESTINATIONIP> -m

    $ sudo ipvsadm -a -t 192.168.1.10:80 -r 10.0.0.11 -m
$ sudo ipvsadm -a -t 192.168.1.10:80 -r 10.0.0.12 -m

  • The configuration rules are automatically saved, but they can viewed in standard output. These rules can then be migrated or restored to an IPVS server.

    $ sudo ipvsadm {-S|--save}

    $ sudo ipvsadm {-R|--restore}

  • The entire configuration can be cleared at any time. [2]

    $ sudo ipvsadm {-C|--clear}

History

Bibliography

  1. “Nginx Load Balancing.” Nginx. Accessed July 9, 2016. https://www.nginx.com/resources/admin-guide/load-balancer/

  2. “LVS-mini-HOWTO.” Austintek. March, 2012. Accessed July 9, 2016. http://www.austintek.com/LVS/LVS-HOWTO/mini-HOWTO/LVS-mini-HOWTO.html

  3. “Building a Load Balancer with LVS - Linux Virtual Server.” Linux Admins. January, 2013. Accessed July 9, 2016. http://www.linux-admins.net/2013/01/building-load-balancer-with-lvs-linux.html