Setting Up A Highly Available NFS Server

[piotr]

I have read the tutorial with great interest and I found it very informative. The tutorial assumes a creation of a new (clean) NFS cluster. However, I have an active (production) NFS server that is a SPOF. I need to cluster it, as I cannot afford any downtime on it without bringing the company's productivity to a halt. The export in question is a independently mounted RAID 1 array, i.e., doesn't hold any system files. Could someone suggest how should I go about adding a duplicate machine to mirror the this array with minimal downtime? Any pointers would be greatly appreciated.

Thanks in advance,

Piotr

[falko]

You'll always have some downtime, at least when syncing the volumes (the bigger the shares, the longer it will take). So maybe this should be done at night when noone else uses the system (of course, you should test drive this before!).

But do you really need DRBD if you use RAID1? Assuming that only hard disks fail, but not processors/memory/..., RAID1 should be fine.

[piotr]

It is not the RAID1 that is the issue; it works just fine _if_ the machine is up and running. I want to deal with a situation when that particular machine is down, i.e., I would want a standby machine to take over. To that end I envision building a new machine, with the same functionality as the original, adding heartbeat to both, and then somehow (perhaps over the weekend, synchronize the RAID 1 array to a similar array on the new machine.) Can you offer any suggestions on how to do that ant what potential pitfalls to be aware of? Thanks.

Piotr

[falko]

Originally Posted by piotr

It is not the RAID1 that is the issue; it works just fine

Yes, I understood that.

Originally Posted by piotr

Can you offer any suggestions on how to do that ant what potential pitfalls to be aware of? Thanks.

Piotr

I think you can use the tutorial, but instead of /dev/sda1 or /dev/hda1 you use your RAID devices (e.g. /dev/md1). If it is possible you should test-drive it with two spare machines.