When conducting startup-tests with multiple nodes (11) starting up at the same time, I sometimes end up in the situation where the replicated cache is not completely funtional: when adding an entry to the cache, not all nodes receive an event.
This only happens when starting multiple nodes at the same time and I want to understand what the problem is.
There are no errors in the logs.
I already started a thead on the jgroups-mailinglist:
Google Groupes
But I want to post it here as well.
I have these questions basically:
- When a node want to add something to the (replicated) cache (UDP stack), is it expected that this node only sends it to 1 other member, and that this other member should replicated to all the rest?
- If so, what can be the cause if that doesn't seem to happen, even if the view is consistent and stable? (both infinispan and jgroups have the same idea on who's in the cluster)
- Based on that, is there a way this situation could be detected at runtime? Because it looks that the situation can be resolved by triggering a 'rebalance'. But this requires that the situation needs to be detected first.
This is on wildfly-9.0.2, infinispan-7.2.3
For more details about the situation (such as configuration) I refer to the jgroups-mailinglist:https://groups.google.com/forum/#!topic/jgroups-dev/b9B6SAQfCko
Thanks for any info
