[ 
https://issues.apache.org/jira/browse/IGNITE-9558?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16681310#comment-16681310
 ] 

Alexey Goncharuk commented on IGNITE-9558:
------------------------------------------

[~ilantukh] I see the following error on TC:
{code}
[16:37:45]W:             [org.apache.ignite:ignite-core] 
java.lang.AssertionError: Invalid cached affinity: 
[cache=HistoryAffinityAssignment [topVer=AffinityTopologyVersion [topVer=251, 
minorTopVer=0], mvccCrd=MvccCoordinator 
[nodeId=1fd96fa9-55f1-46c4-854e-f78443000000, crdVer=1541684159656, 
topVer=AffinityTopologyVersion [topVer=1, minorTopVer=0]]], 
topVer=AffinityTopologyVersion [topVer=10, minorTopVer=0], 
lastAffChangedTopVer=AffinityTopologyVersion [topVer=10, minorTopVer=0]]
{code}

https://ci.ignite.apache.org/viewLog.html?buildId=2271833&buildTypeId=IgniteTests24Java8_Basic1&tab=buildResultsDiv

Please take a look.

> Avoid changing AffinityTopologyVersion on client connect when possible
> ----------------------------------------------------------------------
>
>                 Key: IGNITE-9558
>                 URL: https://issues.apache.org/jira/browse/IGNITE-9558
>             Project: Ignite
>          Issue Type: Improvement
>    Affects Versions: 2.0
>            Reporter: Alexey Goncharuk
>            Assignee: Ilya Lantukh
>            Priority: Major
>             Fix For: 2.8
>
>
> Currently a client join event changes discovery topology version which, in 
> turn, changes AffinityTopologyVersion.
> When a client maps transaction on new AffinityTopologyVersion, corresponding 
> message is not processed on remote node until remote node receives the 
> corresponding discovery event. If discovery event delivery is delayed for 
> some reason, this will result in transaction stalls on client joins.
> Since the client node does not change partition affinity, we can safely map 
> transactions on the previous topology version and do not change the affinity 
> topology version at all.
> Some cases need special care and probably do not qualify for this 
> optimization, such as when client has near cache or client hosts partition 
> for REPLICATED cache.



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

Reply via email to