[jira] [Commented] (QPID-7763) [Java Broker] Flow to disk if allocated direct memory exceeds broker wide broker.flowToDiskThreshold

["Lorenz Quack (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/QPID-7763?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15988814#comment-15988814 ] 

Lorenz Quack commented on QPID-7763:
------------------------------------

Also we should try to use pooled QpidByteBuffers as much as possible.
Currently there are a couple of places that can allocate non-pooled direct byte buffers.
Some offenders:
* WebSocketProvider if the incoming message is larger than QpidByteBuffer.getPooledBufferSize()
* QpidByteBuffer#inflate if the decompressed message is larger than QpidByteBuffer.getPooledBufferSize()
* MemoryMessageStore does not preserve the chunks but puts everything into a single QBB


> [Java Broker] Flow to disk if allocated direct memory exceeds broker wide broker.flowToDiskThreshold
> ----------------------------------------------------------------------------------------------------
>
>                 Key: QPID-7763
>                 URL: https://issues.apache.org/jira/browse/QPID-7763
>             Project: Qpid
>          Issue Type: Bug
>          Components: Java Broker
>            Reporter: Lorenz Quack
>
> Currently the broker.flowToDiskThreshold is compared against the size used by messages on queues.
> However, the intent was to have a mechanism for preventing the broker going out of direct memory. For this it is much more interesting to observe how much direct memory the QpidByteBuffers use. The two numbers can easily differ since the QBB can be sparsely populated. 



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@qpid.apache.org
For additional commands, e-mail: dev-help@qpid.apache.org