[
https://issues.apache.org/jira/browse/IGNITE-10755?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16760049#comment-16760049
]
Roman Kondakov commented on IGNITE-10755:
-----------------------------------------
[~gvvinblade],
# Listeners initialization (obtaining the current update counters map) was
moved under the listeners write lock. This allows to avoid collecting pending
events in the dedicated map.
# Listener registration lock has been moved from the cache context to the
group context to avoid races between adding cache to the
{{CacheGroupContext#contQryCaches}} and listeners registration. When each cache
has it's own listener lock, adding and notifying listeners are not synchronized
between different caches within same group. And there were possible situations
when uninitialized listener was supplied with a skip update event fired in the
neighbor cache.
> MVCC: Flaky continuous query tests
> ----------------------------------
>
> Key: IGNITE-10755
> URL: https://issues.apache.org/jira/browse/IGNITE-10755
> Project: Ignite
> Issue Type: Bug
> Components: mvcc
> Reporter: Roman Kondakov
> Assignee: Roman Kondakov
> Priority: Major
> Labels: CQ, MakeTeamcityGreenAgain, mvcc_stabilization_stage_1
> Fix For: 2.8
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Some continuous query tests are flaky when MVCC is enabled:
> * {{CacheContinuousQueryConcurrentPartitionUpdateTest}}
> ** {{testConcurrentUpdatesAndQueryStartMvccTxCacheGroup}}
> ** {{testConcurrentUpdatesAndQueryStartMvccTx}}
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
