[jira] [Commented] (IGNITE-2951) Cluster is unstable when there are more than 500 clients

["Valentin Kulichenko (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IGNITE-2951?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15231103#comment-15231103 ] 

Valentin Kulichenko commented on IGNITE-2951:
---------------------------------------------

The following is done:

* Removed serialization buffer from discovery message worker. This fixed OOM issues.
* Fixed race in continuous processor. {{clientInfos}} map was not copied before adding to the message. This could cause concurrent serialization and modification.

Changes are committed to master: https://github.com/apache/ignite/commit/5e266153707021a8866e91dfa3f958066f80fc99

> Cluster is unstable when there are more than 500 clients
> --------------------------------------------------------
>
>                 Key: IGNITE-2951
>                 URL: https://issues.apache.org/jira/browse/IGNITE-2951
>             Project: Ignite
>          Issue Type: Bug
>          Components: general
>    Affects Versions: 1.5.0.final
>            Reporter: Valentin Kulichenko
>            Assignee: Valentin Kulichenko
>            Priority: Blocker
>             Fix For: 1.6
>
>
> To reproduce:
> # Start three servers with 4GB heap.
> # Start 500 clients on parallel.
> Server nodes will eventually run out of memory of fail with various exceptions in discovery.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)