[
https://issues.apache.org/jira/browse/IGNITE-7930?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16401953#comment-16401953
]
Ilya Lantukh commented on IGNITE-7930:
--------------------------------------
Deeper investigation uncovered the cause of problem: it is intended in test
that nodes 0 and 2 belong to "datacenter" A, and nodes 1 and 3 - to
"datacenter" B. BackupFilter implementation in reproducer ensures that primary
and backup nodes are located in different datacenters. However, after
restarting node 2, it gets reassigned to datacenter B due to flaw in
reproducer. Node 0 becomes the only node in datacenter A, and so it is forced
to keep all partitions in OWNING state.
Regarding cluster-wide hang up: I couldn't reproduce it neither in current
master nor in ignite-2.4.
> Partition map hang in incorrect state when backup filter is assigned
> --------------------------------------------------------------------
>
> Key: IGNITE-7930
> URL: https://issues.apache.org/jira/browse/IGNITE-7930
> Project: Ignite
> Issue Type: Bug
> Reporter: Vladislav Pyatkov
> Assignee: Ilya Lantukh
> Priority: Major
> Attachments: IgnitePdsRebalanceCompletionTest.java
>
>
> The test ([^IgnitePdsRebalanceCompletionTest.java]) shown, which some
> partition turn up OWNING (but this should not be so) state and whole cluster
> hangs.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
