I just realized. Even with _netdev, we have a problem on Red
Hat. Normally, we assume that netfs will start up our _netdev
filesystems and umount them on shutdown. But what about filesystems
that aren't in /etc/fstab? What about filesystems that are noauto? I'm
not sure what netfs does with noauto on shutdown, but I'm certain it has
no idea about OCFS2 filesystems that aren't in /etc/fstab. It knows
about nfs/gfs/etc, because it hardcodes those fs types. But it only
tracks OCFS2 filesystems via the /etc/fstab entries.
What does this mean? Even with /etc/init.d/netfs, any
filesystem that isn't in /etc/fstab will not be umounted before the
network is stopped. So every other node mounting that filesystem will
see:
o network connection went away
o heartbeat is still alive
=> Ergo, make a quorum decision, perhaps killing myself.
Yuk.
What should we do? Try to persuade Red Hat to add OCFS2 as an
explicit type in netfs (means we don't even need to tell our customers
_netdev anymore)? That doesn't help sles/debian. We have a script for
sles. Maybe use that everywhere?
What are your thoughts?
Joel
--
"Reality is merely an illusion, albeit a very persistent one."
- Albert Einstien
Joel Becker
Principal Software Developer
Oracle
E-mail: joel.becker@oracle.com
Phone: (650) 506-8127
umount script should look at /etc/mtab and not /etc/fstab. netfs looks at /etc/mtab or /proc/mounts depending on the version. Joel Becker wrote:> I just realized. Even with _netdev, we have a problem on Red >Hat. Normally, we assume that netfs will start up our _netdev >filesystems and umount them on shutdown. But what about filesystems >that aren't in /etc/fstab? What about filesystems that are noauto? I'm >not sure what netfs does with noauto on shutdown, but I'm certain it has >no idea about OCFS2 filesystems that aren't in /etc/fstab. It knows >about nfs/gfs/etc, because it hardcodes those fs types. But it only >tracks OCFS2 filesystems via the /etc/fstab entries. > What does this mean? Even with /etc/init.d/netfs, any >filesystem that isn't in /etc/fstab will not be umounted before the >network is stopped. So every other node mounting that filesystem will >see: > > o network connection went away > o heartbeat is still alive > => Ergo, make a quorum decision, perhaps killing myself. > >Yuk. > What should we do? Try to persuade Red Hat to add OCFS2 as an >explicit type in netfs (means we don't even need to tell our customers >_netdev anymore)? That doesn't help sles/debian. We have a script for >sles. Maybe use that everywhere? > What are your thoughts? > >Joel > >
On Wed, Sep 28, 2005 at 04:46:06PM -0700, Sunil Mushran wrote:> umount script should look at /etc/mtab and not /etc/fstab. > > netfs looks at /etc/mtab or /proc/mounts depending on the version.And how many hand-mounted OCFS2 filesystems have _netdev in the options? Oh, we force it? I guess that works. Joel -- "I always thought the hardest questions were those I could not answer. Now I know they are the ones I can never ask." - Charlie Watkins Joel Becker Principal Software Developer Oracle E-mail: joel.becker@oracle.com Phone: (650) 506-8127
and oh btw, netfs should look at /etc/mtab as _netdev is not something the kernel understands. Joel Becker wrote:>On Wed, Sep 28, 2005 at 04:46:06PM -0700, Sunil Mushran wrote: > > >>umount script should look at /etc/mtab and not /etc/fstab. >> >>netfs looks at /etc/mtab or /proc/mounts depending on the version. >> >> > > And how many hand-mounted OCFS2 filesystems have _netdev in the >options? Oh, we force it? I guess that works. > >Joel > > >