[
https://issues.apache.org/jira/browse/IGNITE-5100?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16013635#comment-16013635
]
Stanilovsky Evgeny commented on IGNITE-5100:
--------------------------------------------
tests looks like correct:
http://ci.ignite.apache.org/viewQueued.html?itemId=611081
failed :
TcpDiscoverySslSelfTest.testFailBeforeNodeAddedSent
IgniteCacheConnectionRecovery10ConnectionsTest.testConnectionRecovery - looks
like flapped, work properly well on my local.
RandomLruNearEnabledPageEvictionMultinodeTest.testPageEviction
and
IgniteCacheJoinPartitionedAndReplicatedTest.testReplicatedToPartitionedLeftJoin
probably not mine, due to fail in parent branch too.
> Exchange queue is not used properly
> -----------------------------------
>
> Key: IGNITE-5100
> URL: https://issues.apache.org/jira/browse/IGNITE-5100
> Project: Ignite
> Issue Type: Bug
> Components: general
> Affects Versions: 1.6
> Reporter: Alexei Scherbakov
> Assignee: Stanilovsky Evgeny
> Priority: Critical
> Fix For: 2.1
>
>
> Currently exchange futures share same queue for pending(incomplete) and
> completed exchanges.
> The queue has fixed hardcoded size of 1000.
> This leads to a problem when > 1000 nodes try to enter grid.
> In such case oldest exchanges will be removed by ExchangeFutureSet size
> limit, leading to whole exchange hanging.
> Solution:
> 1. Use separate queues for pending and completed exchanges.
> 2. Pending exchange queue must be unbounded.
> 3. Add system property to control exchange history.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
