AnsweredAssumed Answered

Force M5 CLDB failover

Question asked by tpepper on Jul 30, 2013
Latest reply on Jul 30, 2013 by Ted Dunning
We have a test cluster of 5 nodes, all of which run CLDB and Zookeeper, and have noticed that a slow rolling restart (e.g. kernel update) across the nodes waiting 5m in between node restarts will eventually hit the CLDB master and cause MapR to wait 5-10m to nominate a new CLDB master even though a ZK quorum and plenty of CLDB slaves are still available.

The simplest means to mitigate this scenario would seem to be to force election of CLDB master onto a known-good node, so:

Is there a way to instruct MapR to designate a current valid CLDB slave as master?  Is there some other procedure we could use in this case that would prevent stalling CLDB for 10m or so?

Outcomes