Hi,
I have jira ticket https://issues.apache.org/jira/browse/IGNITE-5061 with
subj and now I'm trying to understand partition exchange mechanism.
As far as I understand, task point is: it need when we have to grow cluster
(start few new nodes without partition exchange and enable exchange after
we completely get desired topology).
All nodes know actual partition topology, so partitions can stay unbalanced
for long time.
How I see solution:
1) Add in ignite MBean some flag enablePartitionExchange
2) Test it in GridDhtPartitionDemander.requestPartitions and complete
RebalanceFuture if partitionExchange is disabled
3) Somehow fire up force rebalance when flag change to true.
Does anybody see problems with this solution? Maybe we need to add
per-cache granularity or let system caches rebalance even if
enablePartitionExchange=false? Does we have some huge changes in progress,
which can conflict with it?
