zfs discuss - Aug 2008 - NFS shares don''t come back online with pool

I''ve just gotten a pool back online after the server booted with it
unavailable, but found that NFS shares were not automatically restarted when the
pool came online.

Although the pool was online and sharenfs was set, sharemgr was showing no pools
shared, and the NFS server service was disabled.  If I tried to manually enable
the NFS service, it would automatically disable itself, and the logs revealed
that it was stopping because "no NFS filesystems are shared".

I found that manually resetting the sharenfs property brought the share back
online, but surely this should be an automatic action as any pool changes from
an offline to online status?

I haven''t tested this with sharesmb, but I would think the same
behaviour change will be needed there too.
 
 
This message posted from opensolaris.org

>>>>> "r" == Ross  <myxiplx at hotmail.com>
writes:
     r> I''ve just gotten a pool back online after the server booted
     r> with it unavailable, but found that NFS shares were not
     r> automatically restarted when the pool came online.

``me, too.''''  in b44, in b71.

for workarounds, export/import can probably fix it iirc.  You might
try ''zfs share -a''---I saw that in another email and will try
it the
next time my box panics.

I also have problems, if I do as you did---boot with a pool
unavailable, then get it online somehow without exporting it---that
filesystems mount in the wrong order.  I have a lot of datasets nested
inside each other.  Some of the parent datasets will just silently not
mount, but their children will.  Then after I notice it, issuing ''zfs
mount -a'' will say something like can''t mount blah/blah:
directory not
empty.  The workaround is to manually unmount all the child datasets,
rmdir the mountpoints, and then mount them by hand in the right
tree-traversal order.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 304 bytes
Desc: not available
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20080826/407a5dbb/attachment.bin>

I might have found an even bigger problem.

It appears that the pool wasn''t even mounted correctly.  My
''rc-pool'' must have crashed in the past, leaving the
mountpoint /rc-pool behind.

On reboot the pool hadn''t mounted because the /rc-pool folder was
already there, and it appears that is why the NFS shares weren''t up.

However, it appears that setting sharenfs=on brought the NFS shares up, but
pointing to the wrong location.  /rc-pool was stored on the root volume, and it
looks like I''ve created 30GB or so of virtual machines on my root hard
disk.  Fortunately I''ve got enough free space to get away with it, but
that could have been a major problem.

This is only my first impressions, I''m going away now to double check
and see exactly what happened, and see if I can replicate it, but if setting
sharenfs=on doesn''t check that the pool is mounted when creating the
share that could also be a problem.
 
 
This message posted from opensolaris.org