Kumar, Amit H. wrote:
Our PVFS 2.7.1 installation has started behaving strangely. One of
the 3 I/O Servers died and had to restart. Along with restarting all
of the I/O servers had to restart all of the pvfs2 clients. I noticed
the following warning in dmesg. Does this mean our PVFS2 FS is
corrupted? Because, until this error as I understand we were able to
submit jobs through SGE and they ran okay. Now none of the jobs get
started through our SGE. SGE shows they are running bit none of the
job process is running. SGE is not a problem because if the users
chose to run their job on an NFS scratch it works fine, hassle free.
Any thoughts or feedback please?
***************************************************** pvfs2_kill_sb:
(WARNING) number of inode allocs (145476) != number of inode deallocs
(145475)
Hi Amit,
This error message was broken in 2.7.1- it does not mean that your fs is
corrupted. One way to trigger it was to mount the same PVFS volume in
two places on one client node, which is perfectly valid. All but the
last one would then display this error message when unmounting.
Though this message by itself doesn't indicate a problem, you might want
to double check if PVFS is mounted in the right location everywhere.
2) What is the best way to upgrade pvfs2 from 2.7.1 to 2.8.1 on a
production machine? Can I just install the pvfs2 server first and
then mount all the clients with the old version of pvfs2-client. This
way I can bring down one compute node at a time and upgrade it to the
latest version.
Unfortunately 2.7.1 and 2.8.1 do not have compatible protocols. If you
leave any old clients mounted while you upgrade the servers, then they
will generate errors until they are also upgraded. It would be best if
you could at least unmount PVFS on your clients during before upgrading
the servers.
-Phil
_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
