CentOS - Aug 2011 - Multipath w/ iscsi

I have several CentOS 6 boxes that mount iscsi based luns and use mpath.
They all had problems shutting down as a result of unused maps not getting
flushed as the system halted.

After examining the init scripts, netfs, iscsi and multipathd all had the
correct
order but mpath failed to flush these maps and the system waited indefinitely.

In the meantime I hacked this by adding a `/sbin/multipath -F` at the end of the
stop clause in the init script.

I seriously doubt this problems exists w/o being the result of my error in
configuration.
Anyone know what the required mpath config might be in this scenario where
the block devices all disappear once netfs unmounts and iscsi stops?

Thanks!
jlc

Am 21.08.2011 21:49, schrieb Joseph L. Casale:
> I have several CentOS 6 boxes that mount iscsi based luns and use mpath.
> They all had problems shutting down as a result of unused maps not getting
> flushed as the system halted.
> 
> After examining the init scripts, netfs, iscsi and multipathd all had the
correct
> order but mpath failed to flush these maps and the system waited
indefinitely.
That sounds as if the paths (SCSI block devices) where removed before
multipath had a chance to flush its map(s).
> In the meantime I hacked this by adding a `/sbin/multipath -F` at the end
of the
> stop clause in the init script.
> 
> I seriously doubt this problems exists w/o being the result of my error in
configuration.
> Anyone know what the required mpath config might be in this scenario where
> the block devices all disappear once netfs unmounts and iscsi stops?
You are sure about the order of the service stops? If you stop iscsi and
remove the devices before multipath flushes the maps, you will end up in
the situation described.

1) umount
2) vgchange -an if LVM is used on LUNs
3) flush multipaths
4) stop iscsi
> Thanks!
> jlc
Alexander