[
https://issues.apache.org/jira/browse/IGNITE-19234?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17846835#comment-17846835
]
Roman Puchkovskiy commented on IGNITE-19234:
--------------------------------------------
It turned out that the 'anti-double-vote' algorithm was not working for about
half a year (it was erroneously disabled). So the following was done in the PR:
# It was enabled back
# Its implementation was fixed (as the common rebalancing code switched from
storing serialized Set<Assignment> to storing serialized Assignments; this had
to be changed for the algorithm as well; also, storages had to be accessed in
correct threads, otherwise thread assertions were triggered)
# StorageRebalanceException and StorageClosedException are handled
> Race between additional volatile storage assignment update and installation
> of RAFT snapshot
> --------------------------------------------------------------------------------------------
>
> Key: IGNITE-19234
> URL: https://issues.apache.org/jira/browse/IGNITE-19234
> Project: Ignite
> Issue Type: Improvement
> Reporter: Roman Puchkovskiy
> Assignee: Roman Puchkovskiy
> Priority: Major
> Labels: ignite-3
> Fix For: 3.0.0-beta2
>
> Time Spent: 1.5h
> Remaining Estimate: 0h
>
> Code in TableManager#updateAssignmentInternal() has a specific case for
> volatile storages, see IGNITE-16668. For them, an additional rebalance might
> be triggered. This additional rebalance causes data races with
> IncomingSnapshotCopier.
> This is only about in-memory storages (namely, in-memory pagemem), it does
> not harm persistent storages.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
