activemq-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Gary Tully (JIRA)" <j...@apache.org>
Subject [jira] Reopened: (AMQ-2103) Memory leak when marshaling ActiveMQTextMessage to persistent store
Date Mon, 11 Oct 2010 16:24:40 GMT

     [ https://issues.apache.org/activemq/browse/AMQ-2103?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Gary Tully reopened AMQ-2103:
-----------------------------

      Assignee: Gary Tully  (was: Rob Davies)

This fix breaks the topic case with multiple consumers and a vm transport producer. There
is contention over marshaling so whacking the marshaled state is not possible, results in
missing content for some of the consumers. https://issues.apache.org/activemq/browse/AMQ-2966

Think the fix is to have an reduceMemoryFootprint policy option that when enabled for queues,
will clear the marshaled state (can work for map and object messages also) when the message
is first persisted. This is a natural sync point that is contention free provided the concurrentStoreAndDispatchQueue
KahaDB option that tries to optimize out persistence, is not enabled.



> Memory leak when marshaling ActiveMQTextMessage to persistent store
> -------------------------------------------------------------------
>
>                 Key: AMQ-2103
>                 URL: https://issues.apache.org/activemq/browse/AMQ-2103
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker
>    Affects Versions: 5.0.0
>         Environment: ActiveMQ 5.0.0.20-fuse
>            Reporter: Trevor Pounds
>            Assignee: Gary Tully
>             Fix For: 5.3.0
>
>         Attachments: AMQ-2103.diff, Duplicate Message Data (Internal Marshalling).png,
heap_100_1MB_messages.png, jhat_ActiveMQTextMessage_0xe837a478.htm, jhat_ByteSequence_0xe837a5c0.htm,
jhat_ByteSequence_data_0xe837adb0.htm
>
>
> When an org.apache.activemq.command.ActiveMQTextMessage is marshaled into the persistence
store some portion of the messages are stored in memory (i.e. pending cursor/consumer dispatch
queue).  The messages stored in memory have the potential to cause the broker to run out of
memory because org.apache.activemq.command.ActiveMQTextMessage objects can store the data
twice, once in the 'text' field and once in the 'content' field.  Normally this isn't a problem
since the 'content' field is cleared when the message is being used in a client application
(i.e. by calling getText() clears content).  The problem occurs when a consumer is slow and
a large number of messages are sitting around on the broker in pending/dispatch memory space.
 The message is marshaled for the store and then persisted to disk and copied to pending memory
when space is available.
> This bug affects any ActiveMQ*Message object that does not clear its temporary data (i.e.
'text' field) once it has been marshaled.  When a message is marshaled we should null the
derived objects memory space once the data has been written to the parent object's 'content'
field.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message