[Lustre-discuss] Shutdown procedure

[Brian J. Murrell]

On Wed, 2009-06-10 at 15:25 +0200, Enrico Morelli wrote:
> 
> which is the correct sequence for a MDT/MGS server shutdown?

Well, if you want to take an entire filesystem down, technically and
ideally (there is no strict requirement, just optimum), you unmount in
this order: clients, MDT, OSTs.

> Because, when I have to shutdown the server, first I try to umount the
> lustreFS on all client and after I reboot the server. But the client
> keeps a lot of time to umount the filesystem, more than 20 minutes.

If all you want to do is simply reboot a server, you can do that without
unmounting the clients.  When the server comes back, the clients will
just resume where they left off.  This is called recovery/failover.
Lustre was designed to allow servers to reboot while clients are still
using them.

> Moreover, if during the client umounting  I have to poweroff the server,
> after the server power up, I'm unable to remount lustreFS on the
> clients without reboot each client, because if I try to enter in the
> lustreFS directory I receive 
> 
> lustre cannot change directory to /lustre_homes/: Cannot send
> after transport endpoint shutdown

Hrm.  This might be related to rebooting the server during the client
unmount, although that should allowable as well.  I know of no specific
bugs in this area.

Also, I should ask, do you have your targets (OSTs, MDT) configured for
failover?  The above scenario where you reboot a server and clients
simply resume working when it comes back requires that failover be
configured for the targets.

b.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20090610/60cb9a7b/attachment.pgp>