Fencing

Main Page :: Fencing

In clustering, fence (also called 'stonith') refers to the action of removing a node from the cluster.

A fence is carried out when the cluster software determines a node is faulty. Once this decision is made, the cluster software consults it's configuration for information on how to carry out the fence. The fence action is in turn carried out by a software or hardware action. The details of which depend on the fence method(s) configured for the node being fenced. Multiple nodes may be fenced at a time, if necessary.

There are three main types of fencing:

Power Fencing

By far the most popular method of fencing; This involves forcing a power off or reboot of a node. This is often implemented using an IPMI BMC (baseboard management controller). Many OEMs have their own incarnations of IPMI like Dell's DRAC, HP's "light's out management", etc. Power fencing can also be accomplished using an addressable power distribution unit (like APC's addressable PDU). In all cases, the goal is to force the target to be powered off by some method.

Network Fencing

In some clusters using a SAN to access a shared disk, a fence could be accomplished by logically disconnecting a node from the SAN array. This is generally discouraged as a stand-alone fence solution with power fencing being more highly recommended as rebooting a node is more likely to resolve the problem and allow the effected node to rejoin the cluster when it comes back online.

Virtual Fencing

When a cluster is built on a collection of virtual servers, fencing may be accomplished by telling the virtual server hypervisor to force down an affected virtual server.

 

Any questions, feedback, advice, complaints or meanderings are welcome.
Us: Alteeve's Niche!	Support: Mailing List	IRC: #clusterlabs on Libera Chat
© Alteeve's Niche! Inc. 1997-2023		Anvil! "Intelligent Availability™" Platform
legal stuff: All info is provided "As-Is". Do not use anything here unless you are willing and able to take responsibility for your own actions.