[Linux-cluster] Asymmetric config for two-node cluster with qdisk?

[Jos Vos]

Hi,

Should I put some asymmetry in the cluster config (and/or the qdisk
heuristic scripts) of a two-node cluster to make (only) one node
decide to continue its services in case of a split-brain problem?

Or are heartbeat failures ignored *if* the quorum disk still shows
that the other node is running ok (or does this only affect whether
to fence or not)?

During my tests (disconnecting nodes, shared storage via working)
I got tons of messages like

  qdiskd[4012]: <crit> A master exists, but it's not me?!
  qdiskd[4012]: <crit> Critical Error: More than one master found!

and I finally had to reboot one node to make this "master race" be
solved.

I'm getting more confused now, some samples would help...  As said
yesterday, what's really missing is a clear picture of the algorithm
CMAN uses to determine membership, calculate votes and quorum, etc.

P.S.
While testing disconnectivity I got an almost synchronous (on both
nodes) kernel panic in gfs_lockd (still using the 42.0.3.EL kernel,
will upgrade soon...).

--
--    Jos Vos <jos at xos.nl>
--    X/OS Experts in Open Systems BV   |   Phone: +31 20 6938364
--    Amsterdam, The Netherlands        |     Fax: +31 20 6948204