You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Sergey Antonov (JIRA)" <ji...@apache.org> on 2019/02/19 10:34:00 UTC
[jira] [Assigned] (IGNITE-11058) Possible OOM due to large discard
queue in TcpDiscoverySpi
[ https://issues.apache.org/jira/browse/IGNITE-11058?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Sergey Antonov reassigned IGNITE-11058:
---------------------------------------
Assignee: Sergey Antonov
> Possible OOM due to large discard queue in TcpDiscoverySpi
> ----------------------------------------------------------
>
> Key: IGNITE-11058
> URL: https://issues.apache.org/jira/browse/IGNITE-11058
> Project: Ignite
> Issue Type: Improvement
> Reporter: Alexei Scherbakov
> Assignee: Sergey Antonov
> Priority: Major
> Fix For: 2.8
>
>
> Currently it's necessary to store every ensured (marked with TcpDiscoveryEnsureDelivery annotation) in pending message queue until it's discarded from coordinator for implementing guaranteed delivery, otherwise if subsequent nodes will fail while forwarding message the guarantee couldn't be fulfilled.
> On large topologies with active changes the queue may contain many very large messages causing heap usage bursts and possible OOM.
> Possible solution:
> # off-load pending messages payload to off-heap or even on disk.
> # store messages in serialized form for avoiding JVM Object overhead.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)