[
https://issues.apache.org/activemq/browse/AMQ-1926?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Gary Tully resolved AMQ-1926.
-----------------------------
Fix Version/s: 5.2.0
Resolution: Fixed
fixed in revision 691621
problem: allocated data locations not used because ack received before async
write can complete were remaining in memory in the inProgressDataFile map,
hence preventing cleanup. Fix is to remove the reference when message is
removed if message still exists when ack is received. At that stage it should
normally be stored.
> Persistent store journal files (occasionally) not being cleaned up at runtime
> for queues with active consumers
> --------------------------------------------------------------------------------------------------------------
>
> Key: AMQ-1926
> URL: https://issues.apache.org/activemq/browse/AMQ-1926
> Project: ActiveMQ
> Issue Type: Bug
> Components: Message Store
> Affects Versions: 5.1.0
> Environment: AMQPersistenceAdapter with syncOnWrite=false
> Reporter: Gary Tully
> Assignee: Gary Tully
> Fix For: 5.2.0
>
>
> Over a long period data files fill up in the
> activemq-data/borker1/data/journal directory. Scenario is single queue,
> persistent producer and consumer, consumer periodically stalls to allow
> messages to build up in broker.
> Problem appears to be related to message ack before async message store gets
> a chance to persist messages. Reference to active data file remain which
> stops them being cleaned up by the periodic checkpoint.
> This limits the available up time of the broker with large volumes of
> messages as excessive amounts of disk space are consumed.
> Using syncOnWrite=true or setting a
> AMQPersistenceAdapter.maxCheckpointMessageAddSize=0 will help avoid the
> problem, both of which will effect performance.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
