[Pacemaker] NFSv4 90 sec grace period

[Florian Haas]

On 07/14/2011 09:46 PM, Michael Schwartzkopff wrote:
>> On 2011-07-13 23:08, Michael Schwartzkopff wrote:
>>> Hi,
>>>
>>> I am trying to set up a NFSv4 server cluster. After failover my client
>>> has to wait 90 sec to be able to access the data agin.
>>>
>>> I already set /proc/fs/nfsd/nfsv4leasetime to 10 but the result ist the
>>> same. The server now logs:
>>>
>>> Jul 13 22:44:27 debian1 kernel: [ 3512.102296] NFSD: starting 10-second
>>> grace period
>>>
>>> but the client waits exactly for 90 sec.
>>>
>>> Any idea what might be wrong?
>>
>> Cosmic rays.
>>
>> More seriously, post the configuration and explain _precisely_ how you
>> are testing this. May well be a bug, but no way to tell for sure (and
>> hence, no way to fix) with the scarce information given. Thanks.
> 
> </long description>
> 
> I somehow tried so set up a NFSv4 server along your SLES11 tutorial.

Can you please post your source? I'll assume for a moment that you are
referring to the NFS tech guide on our web site, found here:

http://www.linbit.com/en/education/tech-guides/highly-available-nfs-with-drbd-and-pacemaker/

> But I
> user Debian Squeeze and tried to keep it simple without the clone resources

Yeah, bad idea. The clones are in the example config for a reason.

> So here is my config:
> 
> primitive resDRBD ocf:linbit:drbd \
> 	params drbd_resource="r0"
> primitive resExportHome ocf:heartbeat:exportfs \
> 	params clientspec="192.168.87.0/24" directory="/srv/nfs/export/home" \
>                     fsid="1001" options="no_root_squash,rw"
> primitive resExportRoot ocf:heartbeat:exportfs \
> 	params clientspec="192.168.87.0/24" \
>                    options="rw,crossmnt,no_root_squash" fsid="0" \               
>                    directory="/srv/nfs/export"
> primitive resFilesystem ocf:heartbeat:Filesystem \
> 	params device="/dev/drbd0" fstype="ext4" directory="/srv/nfs"
> primitive resIP ocf:heartbeat:IPaddr2 \
> 	params ip="192.168.87.33" nic="eth0" cidr_netmask="24"
> primitive resNFScommon lsb:nfs-common \
> 	op monitor interval="60s"
> primitive resNFSserver ocf:heartbeat:nfsserver \
> 	params nfs_init_script="/etc/init.d/nfs-kernel-server" \
>                     nfs_notify_cmd="/sbin/sm-notify" \
>                     nfs_shared_infodir="/srv/nfs/infodir" \
>                     nfs_ip="192.168.87.33"

exportfs is designed such that you don't need to use nfsserver.

> group groupNFS resFilesystem resNFScommon resNFSserver \
>          resExportRoot resExportHome resIP
> ms msDRBD resDRBD \
> 	meta notify="true"
> colocation col_FS_DRBD inf: groupNFS:Started msDRBD:Master
> order ord_DRBD_FS inf: msDRBD:promote groupNFS:start
> 
> I can mount the NFS on a client
> 
> mount -t nfs4 -o udp 192.168.87.33:/home /mnt/home

NFSv4 over UDP? Why?

> accessing the NFS share every second during a failover results in:
> 
> while : ; do date; cat /mnt/home/testfile ; sleep 1 ; done
> 
> Thu Jul 14 20:59:11 CEST 2011
> Hello world
> Thu Jul 14 20:59:12 CEST 2011
> 
> cat: /mnt/home/testfile: Socket operation on non-socket
> Thu Jul 14 20:59:15 CEST 2011
> cat: /mnt/home/testfile: Socket operation on non-socket
> Thu Jul 14 20:59:17 CEST 2011
> cat: /mnt/home/testfile: Socket operation on non-socket
> Thu Jul 14 20:59:20 CEST 2011
> cat: /mnt/home/testfile: Socket operation on non-socket
> Thu Jul 14 20:59:22 CEST 2011
> (...)
> Hello world
> Thu Jul 14 21:01:03 CEST 2011
> Hello world
> 
> That means there is a 90 sec gap accesing the file. I changed nfsv4leasetime 
> already to 10 sec. But It did not help.

The NFS lease time grace period is only relevant to open file handles.
Assuming your "cat" takes a few millisecs and you are sleeping 1s, I'd
guess there's a very high chance your failover occurs while you have no
handles open on that filesystem at all. Thus the grace period shouldn't
be a factor here.

> Any other ideas? Are the LVS as in your tutorial nescessesary?

I suppose you mean LVs as in Logical Volumes, not LVS as in Linux
Virtual Server. No, the LVM based approach is not strictly necessary,
it's just practical to set things up that way.

As for other ideas, please post your "mount" output on the NFSv4 client
so we can get an idea of the default NFS mount options in effect on your
system.

Cheers,
Florian

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 262 bytes
Desc: OpenPGP digital signature
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20110715/1c466ef4/attachment-0004.sig>