[Linux-cluster] CS4 U4 /// HA NFS / problem
Chris Simmons
csim at ices.utexas.edu
Wed Nov 28 12:48:29 UTC 2007
In 2.6 kernels, NFS clients default to mounting via TCP. Switch to UDP
using the 'udp' mount option and you'll be able to switch around your
HANFS share with only a few seconds of stalled access.
If you do keep it as the TCP, NFS clients will keep the connection in a
TIME_WAIT state for quite some time (up to 20 minutes)
Thanks,
Chris
* Alain Moulle <Alain.Moulle at bull.net> [2007-11-28 11:22:59 +0100]:
> Hi
>
> We are currently trying to set up a HA NFS Sever (CS4u4, active/active). As
> decribed in Cookbook we use "managed nfs service" fonctionnality with ext3 FS
> type. During failover process (relocate command) we have the following problem :
> On server side everything seems to be OK (exports, IP, mount points), but on
> the client side time before ressource is back may last from 3 and 20 min
> (apparently stalled in "rpc_execute"). Are they any specifics options to use in
> that case (NFS,ext3)?, or a well-known issue on that kernel ?
>
> cluster.conf
> <resources>
> <fs device="/dev/sdd" force_unmount="1" fstype="ext3"
> mountpoint="/tstha" name="nfsha" options=""/>
> <nfsexport name="nfsexport1"/>
> <nfsclient name="nfsclient1" target="182.20.10.68"
> options="rw,no_root_squash,fsid=7"/>
> <fs device="/dev/sde" force_unmount="1" fstype="ext3"
> mountpoint="/tstha1" name="nfsha1" options=""/>
> <nfsexport name="nfsexport2"/>
> <nfsclient name="nfsclient2" target="182.20.10.68"
> options="rw,no_root_squash,fsid=8"/>
> </resources>
> <service domain="NFSHA1" name="nfs_service1" autostart="1"
> checkinterval="60">
> <ip address="182.20.10.76" monitor_link="1"/>
> <fs ref="nfsha">
> <nfsexport name="nfsha">
> <nfsclient ref="nfsclient1"/>
> </nfsexport>
> </fs>
> </service>
> <service domain="NFSHA2" name="nfs_service2" autostart="1"
> checkinterval="60">
> <ip address="182.20.10.77" monitor_link="1"/>
> <fs ref="nfsha1">
> <nfsexport name="nfsha1">
> <nfsclient ref="nfsclient2"/>
> </nfsexport>
> </fs>
> </service>
>
> on the client side:
>
> [root at xena8 ~]# ps -ef |grep df
> root 85 10 0 Nov27 ? 00:00:00 [pdflush]
> root 86 10 0 Nov27 ? 00:00:00 [pdflush]
> root 3723 1 0 Nov27 ? 00:00:00 xinetd -stayalive -pidfile
> /var/run/xinetd.pid
> root 31614 31262 0 10:55 pts/4 00:00:00 df
> root 31648 28448 0 10:56 pts/2 00:00:00 grep df
> [root at xena8 ~]# cat /proc/31614/wchan
> __rpc_execute[root at xena8 ~]# umount /tstha
> umount: /tstha: device is busy
> umount: /tstha: device is busy
>
> Thanks for help.
> Regards
> Alain Moullé
>
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster
More information about the Linux-cluster
mailing list