And to also correct something Brent said earlier... MDS _and_ OST device
startup has a built in delay (the recovery period) to allow clients to
reconnect and recover/replay incomplete operations.
Fergal.
--
Fergal.McCarthy@HP.com
(The contents of this message and any attachments to it are confidential
and may be legally privileged. If you have received this message in
error you should delete it from your system immediately and advise the
sender. To any recipient of this message within HP, unless otherwise
stated, you should consider this message and attachments as "HP
CONFIDENTIAL".)
-----Original Message-----
From: lustre-discuss-bounces@clusterfs.com
[mailto:lustre-discuss-bounces@clusterfs.com] On Behalf Of Andreas
Dilger
Sent: 09 March 2006 23:36
To: Brent A Nelson
Cc: lustre-discuss@clusterfs.com
Subject: Re: [Lustre-discuss] Faster failover?
On Mar 09, 2006 17:02 -0500, Brent A Nelson wrote:> What are the ways to perform quicker Lustre failovers? The OSTs are a
> little slow to mount, although I''m not sure much can be done about
that. > I remember there was a kernel patch for doing much faster ext2/3
mounts a > long time ago (mounts were doing a lot of unnecessary work which the
patch > removed), but I think that was incorporated into the kernel a long
time > ago (and would certainly be in 2.6.13). A smaller journal, perhaps...
>
> However, the MDS failover has that built-in delay waiting for clients
to > reconnect; how do you shorten that, and what are the consequences of
doing > so?
If you know you do not have any clients that need recovery, you can run
"lctl --device {mds_device} abort_recovery" on the MDS, and similar
for
the OSTs. With 1.4.6 the OST mount time (when using mballoc) has been
reduced noticably.
The whole point of the failover delay is to allow clients to detect the
server is down, reconnect and recover any shared state, and continue on
without any noticable problem for the application. If the clients all
reconnect after server failover there shouldn''t be any extra delay.
Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.
_______________________________________________
Lustre-discuss mailing list
Lustre-discuss@clusterfs.com
https://mail.clusterfs.com/mailman/listinfo/lustre-discuss