show sidebar
HA-CFS Filesystem Failover

Root filesystem failover is one of the features of OpenSSI.
You either need to use DRBD or shared disk hardware for this to work.
If you choose to use DRBD, you should also refer to README.drbd. If you 
use shared disk hardware, then the root filesystem must be installed on 
the shared disk.

During installation you will be asked if you want a failover root 
filesystem.  It is recommended that the root filesystem be ext3.  
While adding new nodes you will also be asked if the new node has 
access to the shared root.  At ssi-addnode time, if you elect to 
enable root filesystem failover for that node, you need to set
up a local boot device, so that it can boot independently of the 
originally installed node.

If all goes well the first line of /etc/fstab looks something like this:

UUID=60926ab6-5992-4f05-907c-23d4c719e2c7 / ext3 chard,defaults,node=1:2 1 1

The /etc/clustertab looks something like this:

<<<SNIP COMMENTS>>>
# Node  IP address   Hardware address   Boot Prot  Initnode?   Boot device
# ----  -----------  -----------------  ---------  ---------   -----------
1       192.168.0.63    00:04:5A:4E:3C:72       E       1       /dev/sda1
2       192.168.0.64    00:04:5A:4E:F7:BB       E       1       /dev/sda1


NON-ROOT FILESYSTEM FAILOVER

To set-up for failover CFS hard mounts, you modify the fstab file.  For example,
if you had an entry like this (must have LABEL or UUID as device):

LABEL=/opt      /opt    ext3    rw,node=3  1 2

you can change it as follows:

LABEL=/opt      /opt    ext3    rw,chard,node=3:4  1       2

Assuming that both nodes 3 and 4 have direct access to the disk partition, this filesystem will be available as long as nodes 3 or 4 is up.  Assuming that it
is already mounted, you'll want to umount the filesystem and mount it again,
so that the chard option is set.

onnode 3 umount /opt
onnode 3 mount /opt

The node= of the fstab entry above dictates the preferred node order.  In this
case, it says that node 3 is preferred over node 4 for this filesystem, but if
node 4 boots first, then node 4 will mount the filesystem.  If the filesystem
is initially mounted on node 3, and node 3 fails, the system will automatically
fsck and remount the filesystem on node 4.  When there are more than
two nodes, if the node mounting the filesystem fails, the earliest node in
the list which is UP will mount the filesystem.

It is recommended that ext3 filesystem be used because of its journalling 
feature.

Although you can manually mount filesystems with failover options using 
"mount -o chard,node=...", the cluster will not automatically fail it over, 
unless the node information is in the /etc/fstab file.

To test, perform filesystem activity in a loop; then crash the node that
did the mount.  The Ongoing operations should hang for a short period.  During
that time, an fsck followed by a remount will automatically occur if another
node with access to the disk is specified in /etc/fstab.  If no node is
currently available, access to that filesystem will hang until a node joins 
the cluster which is specified in /etc/fstab as having access to the 
filesystem.

This page last updated on Tue Apr 12 03:39:53 2005 GMT
privacy and legal statement