[Linux-cluster] GFS 6.0 node without quorum tries to fence

[Amir Guindehi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi all,

|>But if you do it that way, and you really have a netsplit, won't you get
|>into a "quickdraw" situation where each of the newly formed clusters are
|>trying to fence out the others?  In the worst case, all the nodes get
|>reset and nobody is happy.  But maybe the worst case happens so
|>infrequently that it is better than always losing the cluster whenever
|>quorum is lost.
|
| yes, worst case everyone reboots.  BUT! the data on disc is safe.  There
| is probbly a better way to go about this, but we've currently kept the
| idea that an unexpected reboot is better than looking for backup tapes.

I think you would need a form of /atomic fencing/ )which is not really
possible afaik) to resolve this problem. It's a race, and as long as you
can't do it atomically you can always reach above worst case state.

- - Amir
- --
Amir Guindehi, nospam.amir at datacore.ch
DataCore GmbH, Witikonerstrasse 289, 8053 Zurich, Switzerland

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2-nr1 (Windows 2000)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBD/1PbycOjskSVCwRAoRcAJ9cq5+FiQnIx817IdEthaB6HTgPTQCg9G3n
MLA+ulKC4Jh3BQZLbPq59/0=
=Gy9x
-----END PGP SIGNATURE-----