annoying umount NFS issue

[ogniemi]

hello,

anyone knows how to help HACMP to unmount an NFS filesystem during takeover (when opposite node in a cluster serving NFSes crashed)?

'lsof' shows no NFS filesystems so sounds like filesystem is not used but see HACMP can;t unmount it.

the NFS is seen on mount output and force umount option can not unmount it (still in use error).

any workaround? only reboot can fixe such problem??

I am thinking of stopping using NFS "features" in HACMP's RGs and use for mounting/unmounting simple scripts instead.


thx in advance,

 

[frenger]

Hi

If your NFS mount is mounted with the recommended options "bg,soft,intr" if the NFS Server stays down you should be able to get rid of the mount with at least umount -f <mountpoint>. Power HA Cluster have theier own special ways on killing dead NFS mounts like /usr/es/sbin/cluster/events/utils/cl_nfskill or /usr/es/sbin/cluster/events/utils/cl_deactivate_nfs you can use them too but be careful!

Cheers frenger

 

[ogniemi]

thx.

but the filesystems are configured in cluster resource group to be exported and mounted.

doesn it mean that the recommendations "bg,soft,intr" are not HACMP recommendations? is it possiblem the HACMP recommends them and mounts by default "hard,intr" (what I have seen on mount output)

How to learn HACMP to mount with "bg,soft,intr"? There are no nfs stanzas in /etc/filesystems as HACMP takes care by itself for NFS mounts/exports. Should I add combined NFS stanzas to /etc/filesystems to learn HACMP mount with new options or there is another way to achieve this within HACMP configuration?

thx in advance,

 

[frenger]

Hi

HACMP (or Power HA) does not need /etc/filesystem entries for NFS mounting (or crossmounting) but if there is a stanza for the nfs mount in your /etc/filesystems HACMP will use it. So it will be best to add the mount options there. I made it with our Crossmounting SAP HACMP Cluster that way, but found no way to get the option bg (background mounting) to work. But most hangups in the HACP script is caused by the more problematic "hard" option. And that could be changed to soft via this method. The option intr maybe only sufficent for hardmounts, but it will not be bad anyway

Cheers frenger

 

[ogniemi]

sounds like the recommendation change:

http://www-01.ibm.com/support/docview.wss?uid=isg1IZ43725

and in /usr/es/sbin/cluster/README5.4.1.UPDATE is:


Changes in NFS mount options
=====================================================================
Default options for NFS cross mounts are changed on HACMP 5.4.1.

If there are no mount options specified for this filesystme in
/etc/filesystems,
HACMP 5.4.1 does a hard, interruptable mount
options=hard,intr,vers=$vers
HACMP 5.4.0 does a soft, interruptable mount
options=soft,intr

 

[frenger]

Well I did mean the recommendation for a non hanging NFS mount not on IBM's opinion on that.
Our problem was that we had cross mounts and if one node was down and the second node should have taken over somewhere in the swich progress he tried to acesss the dead ended hard nfs mount and hung to infinity on it, due hard mode says "wait to the end of days to get something you can't get and do not report errors on it". Soft mount does report errors after a timount periode and that error let the cluster script continue. So there is the question wy hard mount when soft is better to use? Data integrity is the key. IBM in the past was always the company that puts data integrity on first place, before it comes to performance or usability. So it's no wonder why they prefer the hard mount option.
If you find another way to get rid of hard nfs mounts you can do it, for us the easiest solution was to switch to soft NFS Crossmounts.

Cheers frenger

 

[polani]

hi

I think , soft NFS cross mounts is best solution to this problem, but i would also suggest to have a look all related documentation of NFS V3 as well....

Here comes polani Once again!!!

See my personal blogs at

http://reliablesolutions.blogspot.com

for unique solutions and tips on AIX, Linux,Storage and TSM.