[jira] [Created] (IGNITE-10088) Partition can be restored in moving state instead of owning if node crashed before first checkpoint.

["Alexei Scherbakov (JIRA)" <ji...@apache.org>]

Alexei Scherbakov created IGNITE-10088:
------------------------------------------

             Summary: Partition can be restored in moving state instead of owning if node crashed before first checkpoint.
                 Key: IGNITE-10088
                 URL: https://issues.apache.org/jira/browse/IGNITE-10088
             Project: Ignite
          Issue Type: Bug
            Reporter: Alexei Scherbakov


Scenario:

1. Start grid with large enough checkpoint freq, wait for rebalance, put some data.
2. Observe all partitions in OWNING state.
3. Kill of trigger FH for node before checkpoint is started.
4. Return node to grid, observe all partitions created with moving state and unnecessary rebalanced.

Problem in org.apache.ignite.internal.processors.cache.persistence.GridCacheDatabaseSharedManager#restorePartitionStates which doesn't apply owning partition state.

{noformat}
public void testMoving() throws Exception {
        IgniteEx crd = startGrid(0);

        startGrid(1);

        crd.cluster().active(true);

        awaitPartitionMapExchange();

        stopGrid(1);

        awaitPartitionMapExchange();

        startGrid(1);

        awaitPartitionMapExchange();
}
{noformat}



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)