There is a bit of configuration "magic" around this case and also some
First the 2 nodes when they try to perform a system restart they will
not constitute a
majority and the partitioning service hasn't got all the bells and
whistles yet during
system restart so there is a configuration parameter stating how long
to wait until it
starts in a partitioned state
StartPartitionedTimeout which by default is 60 seconds (= 60000
In addition there is another configuration parameter that ensures that
as often as
possible a full cluster is restarted. This sets the time that the
cluster waits for processes
to join before starting the system restart.
StartPartialTimeout which by default is set to 30 seconds.
So by default no system restarts are initiated until 30 seconds have
starting the DB nodes and if in partitioned state one waits for 60
attempting the system restart.
2004-09-14 kl. 01.49 skrev Jim Hoadley:
> Using a 4-node cluster with 2 replicas, my testing has shown that a
> cluster will continue to run with a subset of DB nodes, as long as all
> data is
> present. This is as expected :)
> But testing also show it is not possible to bring up just a subset of
> In practice, this means that if I'm lucky enough to have the "right" 2
> crash in a 4-node cluster, I'd best not shut them down for even a few
> for if I do the cluster won't be able to restart until all 4 nodes can
> brought back online.
> Is this observation accurate? Comments?
> -- Jim
> Do you Yahoo!?
> Take Yahoo! Mail with you! Get it on your mobile phone.
> MySQL Cluster Mailing List
> For list archives: http://lists.mysql.com/cluster
> To unsubscribe:
Mikael Ronström, Senior Software Architect
MySQL AB, www.mysql.com