Release Notes for Red Hat Cluster Manager

1.2.0

Split Brain:
- In a 2-member cluster using IP-based tiebreaking when one member has
network connectivity (and is in the quorum), do not disconnect it from
the network until the other member is operational.  Doing so will result
in a split-brain.
- In a 2-member cluster using disk-based tiebreaking, do not disconnect
the network from a member and start the cluster daemons on the same member.
This results in a broken cluster (some services won't run).

Shrinking a Cluster:
- Moving from a 3-member cluster to a 2-member cluster which uses the shared
medium as a tie-breaker requires a full cluster restart.

Regarding Samba services:
- Samba share names may not contain spaces.
- Only one Samba configuration is allowed per service.  If multiple shares are
desired, the administrator must create them manually by editing
/etc/samba/smb.conf.<sharename>, where <sharename> is the name of the share
created by the cluster.

Using Real-Time Priority with the Membership & Quorum Daemons:
- The membership daemon has the ability to run in the round-robin real-time
priority scheduling queue.  To make it do this, run:

   cludb -p "clumembd%rtp" <priority>
   cludb -p "cluquorumd%rtp" <priority>

The membership daemon should be one of the most priority tasks on the 
system - a failed membership daemon will cause a STONITH event.  The quorum
daemon should never exceed the membership daemon in priority.

Stopping the cluster:
- If you cut your members down so that a majority is no longer present (IE, 
you no longer have a quorum), all the remaining members will slam to a 
grinding halt.  Services will stop cleanly - except for their state on the 
shared storage, which will look as though the members crashed.  This will get 
cleaned up when a new quorum is formed.

Maintenance:
- If you need to stop all clustered services for maintenance, run "clushutdown"
on one member first.  This command will stop all services and lock the service
managers so that member transitions do not affect service state.  At this point,
it is safe to shut down all cluster members.  Do NOT start any new cluster
members after running "clushutdown".  It is a good idea to disable automatic
starting of the cluster services, since members which do not have controlling
power switches will reboot when the quorum daemon detects a minority (ie, loss
of quorum).