[IGNITE-14828] No fallback to full rebalance after exception on historical

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.11

Release Note

Nodes will now properly fallback to full rebalance if historical rebalance fails

• Ignite Flags: Release Notes Required
• Assignee: Mirza Aliev
• Reporter: Mirza Aliev
• Reviewer: Vyacheslav Koptilin
• Created: 06/Jun/21 23:26
• Updated: 01/Jul/21 09:12
• Resolved: 08/Jun/21 16:38

Description

We have noticed that in some cases when we handle demand message in GridDhtPartitionSupplier.java it is possible for some reasons that

iter = grp.offheap().rebalanceIterator(demandMsg.partitions(), demandMsg.topologyVersion());

throw an exception. In that case, rebalance should switch to full, but the code has a bug and remainingParts has been filed after rebalance iterator has been created

                for (int i = 0; i < histMap.size(); i++) {
                    int p = histMap.partitionAt(i);

                    remainingParts.add(p);
                }

That means that we lost partitions that meant to be rebalanced by historical rebalance.

The solution is to create an iterator after remainingParts filling.

Issue Links

• links to: GitHub Pull Request #9155