Uniq Advances Pty Ltd

UNIQfail - Unix Failover Management Software

UNIQfail provides a mechanism for managing and controlling the switchover of users and applications from one system to a second system after a failure.

The diagram above shows a typical configuration; in normal operation one system runs as the master, and has full control of the shared data disks and applications. Disk failures are masked by the use of complete mirroring of all disks, including all partitions on the system disks. The other system is configured as a hot standby or slave system, monitoring the master system and ready to take over in the event the master system fails.

There are two states that each system may be in at any given time, master or standby. A system in master mode has access to all the shared data disks and is running the user applications. In standby mode the system is running a process to monitor the master system for failure and is ready to become master if required. Only one system can be in master mode at any time.

It is possible to make use of the standby system when it is idle to run other applications, however, it must be remembered that if the master system fails the primary function of the standby is to run the master's applications. This could impact the performance of standby applications. Alternatively standby applications could be shut down when the system switches to master mode.

Failover and Switchback

Failover describes the actions taken when the standby machine notices that the other machine has failed the server availability test. It performs the necessary tasks to bring the standby system into the master role in an orderly fashion without corrupting any of the data and starts specific applications that were running on the other system when it failed.

Switchback describes the action taken to release the shared disks and transfer a machine from master to standby mode. The action is performed on the master machine. Switchback releases all of the shared disks so that the standby system can become the master in an orderly manner. Switchback is a manual procedure as it requires a scheduled outage for users. This outage is of similar duration to that required for failover.

When failover is initiated, the actions described below take place automatically. Failover can be initiated manually by executing a command, or automatically when a machine detects that its partner has failed.

Some of the steps performed during a failover are:

• Check if systemB is running as the master, if so then ask for it to be halted.
• Configure ethernet interfaces with the IP addresses of the virtual machine or the failed system if the new master is to mimic the failed system on the network.
• Start applications, typically database and related user processes.
• User processes which do not cater for their server restarting will themselves need to be restarted. Current database transactions on the failed machine may be lost.
• systemB now only has access to its system disks, and can be booted for diagnostic or repair purposes, or to go into standby mode.

The switchback procedure reverses the failover steps.

Contact Details

© Copyright 2001 Uniq Advances Pty Ltd ABN 60 091 811 915