Zimbra Cluster + NFS

[KenVM]

Hello,

I'm trying to setup a Zimbra Cluster on a pair of VM's for testing purposes. I'm using CentOS 4.4 and the accompanying CSGFS.

Initially, I setup one of the vm's (zimcluster1) with an extra drive that was mounted to the zimbra cluster mountpoint. This worked great for getting Zimbra running so that it worked on the one machine.

Next, I used the system-config-cluster tool to modify the cluster configuration so that it used an NFS mount instead of the Filesystem mount and copied the contents of the drive to the NFS mount.

When I start the service, the log reports that everything starts up correctly and that its running and then immediately shuts down after the following message:

May 8 10:43:45 zimcluster1 clurgmgrd[2958]: status on netfs "NFS Mount: zimcluster" returned 1 (generic error)

The NFS mount is there and working as expected, even after the shutdown.

Anyone have a suggestion on whats going on?

Thanks,

-Ken

Edit: Oh, and for reference this is for a single server cluster installation.

[KenVM]

I found another way to solve the problem, so this is no longer an issue.

If anyone does have any insights, I'd be very curious to know why the NFS didn't work, though.

-Ken

[jhahm]

You might want to check mysql error log. I have seen on test setups in the past that mysql won't run on NFS mounted volume due to file locking issues. I don't remember exactly what the error log said, but if you are indeed having the lock issue the error message will be unmistakable. I don't believe this happens with all NFS server, but it happened for me using nfs daemon on a Linux box.

You could instead go with software iSCSI. Run iSCSI server on a machine, and run iscsi client on your clustered nodes. It's not a deployable configuration from performance standpoint, but it works fine for cluster testing.

[freefall]

KenVM, I'd be interested to know what you did to solve this problem.

Thanks.

[KenVM]

I ended up creating a raw lun on the SAN and mapping that to a drive. Configure the drive on a separate raid controller and then change its settings to be shared across multiple VM's.

I haven't tried jhahm's suggestion yet, though.