It is important to make regular backups of your GFS2 file system in case of emergency, regardless of the size of your file system. Many system administrators feel safe because they are protected by RAID, multipath, mirroring, snapshots, and other forms of redundancy, but there is no such thing as safe enough.
It can be a problem to create a backup since the process of backing up a node or set of nodes usually involves reading the entire file system in sequence. If this is done from a single node, that node will retain all the information in cache until other nodes in the cluster start requesting locks. Running this type of backup program while the cluster is in operation will negatively impact performance.
Dropping the caches once the backup is complete reduces the time required by other nodes to regain ownership of their cluster locks/caches. This is still not ideal, however, because the other nodes will have stopped caching the data that they were caching before the backup process began. You can drop caches using the following command after the backup is complete:
echo -n 3 > /proc/sys/vm/drop_caches
It is faster if each node in the cluster backs up its own files so that the task is split between the nodes. You might be able to accomplish this with a script that uses the
rsync command on node-specific directories.
The best way to make a GFS2 backup is to create a hardware snapshot on the SAN, present the snapshot to another system, and back it up there. The backup system should mount the snapshot with
-o lockproto=lock_nolock since it will not be in a cluster.