Hi John,
1. If I attempt to cleanly shut down the cluster, I presume I am
likely to hit the ‘last man standing’ restart order problem as I
cannot easily control the shutdown order
Yes. Sequoia 4 will include a 'shutdown cluster' command that will allow
restart in any order.
2. Would it be easier to stagger shutdown so that at least one cluster
member is active at all times, effectively keeping the cluster up?
Yes, actually this is even the recommended way. Sequoia is not really
designed to be shutdown but to always be fully available so restarting
nodes 1 by 1 prevent interruption in service. Cluster maintenance
procedures are described at
http://sequoia.continuent.org/doc/infocenter/topic/org.continuent.sequoia.doc/html/Cluster_maintenance.html
3. If I go for 2, what happens if there is a catastrophic failure (eg
power out) – presumably I would need to detect this and restart the
controllers with force/init?
If you lose all your cluster, you need to use the procedure described at
http://sequoia.continuent.org/doc/infocenter/topic/org.continuent.sequoia.doc/html/Recover_from_a_controller_node_failure.html
4. There will be minimal writes happening at the time of shutdown, so
can I avoid all of the restart synchronisation by restarting with
force/init?
No because if you miss even a single write, that will lead to
inconsistency between the various backends. If you stop the application
first to make sure that all connections are closed before shutting down
Sequoia then it is fine to do an init/force (but at your own risk!).
Hope this helps,
Emmanuel
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
