[jira] [Closed] (GEODE-9607) radish publish or subscribe operations could run the server out of memory

["Owen Nichols (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/GEODE-9607?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Owen Nichols closed GEODE-9607.
-------------------------------

> radish publish or subscribe operations could run the server out of memory
> -------------------------------------------------------------------------
>
>                 Key: GEODE-9607
>                 URL: https://issues.apache.org/jira/browse/GEODE-9607
>             Project: Geode
>          Issue Type: Bug
>          Components: redis
>            Reporter: Darrel Schneider
>            Assignee: Darrel Schneider
>            Priority: Major
>              Labels: pull-request-available, unreleased
>             Fix For: 1.15.0
>
>
> Each subscribe op stores some data in the server memory. If you keep doing this the server will eventually run out of memory. Operations that store data are supposed to honor the geode critical memory threshold and fail with a LowMemory exception but subscribe does no check for critical but instead just uses more memory.
> Each publish op is added to an unbound queue that can take a while to process (longer when more than one server is running). If enough publish ops are received in a burst they can also cause the server to run out of memory. Before adding the op to the queue geode's critical memory threshold should be checked.
> Since the server queues publish ops and needs to send them to remote servers using a geode function, the implementation should be enhanced to do this with a batch of publish ops instead  of doing them one at a time. This will improve the performance of publish which will allow it to free up memory faster.



--
This message was sent by Atlassian Jira
(v8.20.7#820007)