[
https://issues.apache.org/jira/browse/IGNITE-22806?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Mirza Aliev reassigned IGNITE-22806:
------------------------------------
Assignee: Mirza Aliev
> Investigate table counters problems
> -----------------------------------
>
> Key: IGNITE-22806
> URL: https://issues.apache.org/jira/browse/IGNITE-22806
> Project: Ignite
> Issue Type: Bug
> Reporter: Mirza Aliev
> Assignee: Mirza Aliev
> Priority: Major
> Labels: ignite-3
>
> For the purposes of the colocation feature, we have implemented counters for
> table rebalances from a zone
> https://issues.apache.org/jira/browse/IGNITE-21253.
>
> This solution has a lot of restrictions, like we cannot start a table in a
> zone if there is ongoing rebalance in this zone, or we cannot remove a table,
> etc.
>
> In general, solution was implemented as a workaround in we had a plan to
> remove it once colocation feature is implemented.
>
> Now we have several flaky tests because of some bugs in the counters
> implementation.
> 1) When we receive call back from raft that configuration of a group was
> applied, we count down counter, but we don't have a revision of MS and we
> cannot take actual counter for this event. It could lead to problems for some
> recovery scenarios or disaster recovery.
>
> 2) Also there is a NPE problem when we count down, because counter could be
> null, as we can see from
> `org.apache.ignite.internal.disaster.ItDisasterRecoveryReconfigurationTest#testManualRebalanceIfPartitionIsLost`
>
> *Definition of done*
> * We must decide, should we rollback this feature, as long as corresponding
> logic for zone-based rebalance has already been implemented
> * We must invent some solution for Raft configuration applying and reading
> MS in that callbacks
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
