Dear list,
we've had several servers block today because they all have a NFS mount point
from a linux kernel 3.7.6 server. This is the fstab entry:
1.1.1.1:/nfsshare/all /shared nfs4 defaults,tcp,intr,relatime,async,acl 0 9
The NFS server crashed and was offline, that made all clients block. Even
after the NFS server was up again, most clients were stuck to the point they
had to be rebooted. What mount options should we use so clients don't block
completely when the NFS server is gone? It's no problem to unmount the NFS
share, or get errors, but the clients must not stop their work.
--
mit freundlichen Grüssen,
Michael Monnerie, Ing. BSc | Tel: +43 660 415 6531
XING: https://www.xing.com/profile/Michael_Monnerie
Facebook: https://www.facebook.com/michael.monnerie
Twitter: @MichaelMonnerie https://twitter.com/MichaelMonnerie
LinkedIn: http://lnkd.in/uGx6ug
Google+: https://plus.google.com/u/0/100598203632716687928/
Protéger.at Internet Services Austria [gesprochen: Prot-e-schee]
http://protéger.at | http://proteger.at
Facebook: https://www.facebook.com/protegerat
Mitglied im it-management Netzwerk http://it-management.at
On Sat, Feb 16, 2013 at 12:30:07PM +0100, Michael Monnerie wrote:
> Dear list,
>
> we've had several servers block today because they all have a NFS mount point
> from a linux kernel 3.7.6 server. This is the fstab entry:
>
> 1.1.1.1:/nfsshare/all /shared nfs4 defaults,tcp,intr,relatime,async,acl 0 9
>
> The NFS server crashed and was offline, that made all clients block.
That's expected.
> Even
> after the NFS server was up again, most clients were stuck to the point they
> had to be rebooted.
But that sounds like a bug, on either client or server.
> What mount options should we use so clients don't block
> completely when the NFS server is gone?
As Boaz says, you can mount with "soft"--but there are good reasons it
isn't the default. (Worst case it can cause data corruption if your
applications aren't extremely careful.)
--b.
> It's no problem to unmount the NFS
> share, or get errors, but the clients must not stop their work.
>
> --
> mit freundlichen Grüssen,
> Michael Monnerie, Ing. BSc | Tel: +43 660 415 6531
>
> XING: https://www.xing.com/profile/Michael_Monnerie
> Facebook: https://www.facebook.com/michael.monnerie
> Twitter: @MichaelMonnerie https://twitter.com/MichaelMonnerie
> LinkedIn: http://lnkd.in/uGx6ug
> Google+: https://plus.google.com/u/0/100598203632716687928/
>
> Protéger.at Internet Services Austria [gesprochen: Prot-e-schee]
> http://protéger.at | http://proteger.at
> Facebook: https://www.facebook.com/protegerat
> Mitglied im it-management Netzwerk http://it-management.at
On 02/16/2013 01:30 PM, Michael Monnerie wrote:
> Dear list,
>
> we've had several servers block today because they all have a NFS mount point
> from a linux kernel 3.7.6 server. This is the fstab entry:
>
> 1.1.1.1:/nfsshare/all /shared nfs4 defaults,tcp,intr,relatime,async,acl 0 9
>
> The NFS server crashed and was offline, that made all clients block. Even
> after the NFS server was up again, most clients were stuck to the point they
> had to be rebooted. What mount options should we use so clients don't block
> completely when the NFS server is gone? It's no problem to unmount the NFS
> share, or get errors, but the clients must not stop their work.
>
It's the "-o soft" mount option to mount or fstab. Note that it will unstuck
you at most operations, after a timeout the Applications will receive EIO.
("most", because rarely it'll D state anyway)
Cheers
Boaz