[
https://issues.apache.org/jira/browse/AMQ-4311?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13577334#comment-13577334
]
Christian Posta commented on AMQ-4311:
--------------------------------------
What store are you assuming for this? Kahadb? Jdbc? Memory?
Can you put together a test that shows PFC won't block? message size for
pending messages in a tx should be counted against memory usage... any time we
do a message.incrementReferenceCount(), the usage memory will be updated (if
it's the first reference).
In org.apache.activemq.broker.region.Queue#doMessageSend we check if
transaction (context.isInTransaction) and then we do
message.incrementReferenceCount()...
In org.apache.activemq.broker.region.Topic#doMessageSend we increment the
reference regardless of whether it's in a tx.
You might be on to something, but can you give more details?
> Spooling pending messages in the large transaction to the disk
> --------------------------------------------------------------
>
> Key: AMQ-4311
> URL: https://issues.apache.org/jira/browse/AMQ-4311
> Project: ActiveMQ
> Issue Type: Improvement
> Components: Broker, Message Store
> Affects Versions: 5.x
> Reporter: SuoNayi
> Fix For: 5.9.0
>
>
> Currently MemoryTransactionStore is used to hold pending messages in a
> transaction.
> More pending messages in a transaction more memory is consumed.
> Because the message size of pending messages is not counted in usage limit
> before they are committed so PFC wont block producers as well and large
> transaction will cause broker throws OOM at last.
> Spooling pending messages in the transaction to the disk is one of solution.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
