[jira] [Commented] (AMQ-2922) Add ability for KahaDB log files to be created on a per-destination basis.

["Gary Tully (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/AMQ-2922?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13084089#comment-13084089 ] 

Gary Tully commented on AMQ-2922:
---------------------------------

Having reviewed these changes, they are quite ingrained but also some more work needs to be done to separate out the metadata into shared and per journal bits and rework recovery and journal tracking for transactions. The producer audit is one example of metadata that can be shared, but lastAppendLocation and inprogress transactions need to be separate. I think it makes sense to pull multiple journal support in to a separate persistence adapter such that we don't destabilise kahadb with lots of optional code paths.
 

> Add ability for KahaDB log files to be created on a per-destination basis.
> --------------------------------------------------------------------------
>
>                 Key: AMQ-2922
>                 URL: https://issues.apache.org/jira/browse/AMQ-2922
>             Project: ActiveMQ
>          Issue Type: New Feature
>          Components: Message Store
>    Affects Versions: 5.4.0
>         Environment: All
>            Reporter: Peter Blackburn
>            Assignee: Gary Tully
>             Fix For: 5.6.0
>
>         Attachments: activemq.xml, multistore.diff
>
>
> KahaDB persistence uses rolling log files to store unconsumed messages. These are named db-1.log, db-2.log ... db3.log. At present these files contain messages for all destinations managed by the broker.
> A configurable option could be added that would allow these files to be created on a per-destination basis. For example, if the broker contained two queue destinations, queue_1 and queue_2, the log files would become:
> queue_1-1.log, queue_1-2.log....etc
> queue_2-1.log, queue_2-2.log....etc
> Each set of log files would only contain messages for the relevant destination. 
> This would help in the following situation:
> queue_1: receives one message every 15 seconds or so. These messages remain unconsumed for several hours.
> queue_2: receives thousands of messages per second. These messages are consumed as they arrive.
> At present, the scenario above leads to each log files containing a few messages that have yet to be consumed and thousands of messages that have been consumed, but the log file cannot be deleted until all messages logged in it have been consumed (which may be hours later).
> Logging on a per-destination basis would allow the log files for queue_2 in the above example to be deleted, meaning the unconsumed messages on queue_1 take up far less disk space. This would also reduce the number of file handles required.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira