Master is always switching between the same two members.

Question

Hi all,
We have observed that our replicated environment, is always switching between the same two members even though there are other members with more cpu power.
As said on message {message:id=4169549}: "+In the event of a tie, an arbitrary, but predictable attribute is used to break the tie. In the current release, that tie breaker is an implementation artifact, and is not available through the public API.+"
I know the availability of "je.rep.node.priority" property but, is it possible to set a custom tie breaker or add some king of randomness to the default tie breaker?
BR,
/César.
Edited by: Cesar Alvarez on 04-jul-2011 18:05

Answer 1

Cesar,
No, currently there is no way for the application to influence the election, except through the use of je.rep.node.priority. However, is it possible to use node priority to achieve your goals? If use a positive priority value for all nodes (and avoid using zero), elections would make this choice:
- first consider nodes with most recent log files as master candidates
- node priority used as a tie for those most-recent nodes.
Since node priority is mutable, you could even change the node priority randomly on each node, which is rather awkward, but would work.
Out of curiosity, could you tell us your motivation for injecting some randomness into the master selection? We understand that some nodes have more resources and should be higher priority, but have not thought about why randomness would be useful.
Linda