[Gllug] NFS problem

Nix nix at esperi.org.uk
Tue Oct 23 18:15:58 UTC 2007


On 23 Oct 2007, Kostas Georgiou verbalised:

> On Mon, Oct 22, 2007 at 08:18:48PM +0100, Nix wrote:
>
>> On 22 Oct 2007, Kostas Georgiou told this:
>> > You also risk data corruption with a soft mount
>> > http://nfs.sourceforge.net/#faq_e4
>> 
>> Of course, take that with a pinch of salt, because you *also* risk data
>> corruption with any disk medium which isn't at least battery-backed
>> RAID, even if the machine as a whole is on a UPS, and how many of us
>> have one of those?
>
> For soft mounts I follow Neil's advice which is "Don't ever use soft
> mounts" :) It's not hard to get in trouble with soft mounts, every time
> the nfs server is slow (doesn't have enough free threads, is under heavy
> load, you are rebooting it) for example instead of getting "nfs: server
> foobar not responding, still trying" you'll end up with an I/O error.

Yeah, but with no umount -f on NFS (as was true from 2.6.15, I think it
was, to 2.6.23), if your NFS server goes away for a long time without
warning you've pretty much lost the clients too: they'll need a reboot,
at least.

> The only time you don't risk data corruption with soft mounts is when
> you export a read only fs read only to the clients. Keep in mind that
> data corruption doesn't necessarily mean that you corrupted the copy
> on the disk, it could just be that the nfs clients got the wrong data.

If they're getting rebooted anyway, that's not often important.

> What is the point of using soft mounts anyway? Mounting with hard,intr
> is a much saner option, I've never seen an application that can reliably
> recover from an EIO yet.

Those I write do. It's not that hard, I mean you have to handle EINTR
anyway, handling EIO as well isn't much harder. Besides, you can get EIO
from physical media too, it's just rarer.

-- 
`Some people don't think performance issues are "real bugs", and I think 
such people shouldn't be allowed to program.' --- Linus Torvalds
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list