You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@activemq.apache.org by "Izzy Alanis (JIRA)" <ji...@apache.org> on 2011/05/18 19:54:47 UTC
[jira] [Created] (AMQ-3326) Synchronization point in RegionBroker
adding and removing producers and consumers
Synchronization point in RegionBroker adding and removing producers and consumers
---------------------------------------------------------------------------------
Key: AMQ-3326
URL: https://issues.apache.org/jira/browse/AMQ-3326
Project: ActiveMQ
Issue Type: Improvement
Components: Broker
Affects Versions: 5.5.0, 5.4.2
Reporter: Izzy Alanis
Synchronization blocks were added to the RegionBroker as a result of AMQ-2821 (automatic removal of inactive destinations). I believe the author was trying to prevent adding and removing producers while the inactive destination removal process was running.
This is a major performance bottle neck in use case scenarios where many clients are subscribing and unsubscribing simultaneously -- specifically, in order to retrieve one and only one message from a queue using the STOMP protocol (without holding onto a prefetch message) you must subscribe, retrieve a message and then unsubscribe from the destination. This activity causes lots of adding and removing of producers and consumers.
I'm still not entirely convinced that the destination removal process needs to be protected in that manner, but the attached patch uses a ReentrantReadWriteLock to continue to protect the destination removal process, while still allowing clients to add and remove producers and consumers concurrently when the destination process is not running.
This patch also relates to AMQ-3070 (where the user was complaining about this same synchronization block).
I'm attaching patches both for the trunk and for 5.4.2 (which is what I'm still using).
There are other synchronization blocks further down in the code (inside AbstractRegion and Queue implementations), but in my particular use case, the 2-3x performance enhancement I get of this patch alone are enough to satisfy my immediate concerns.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Resolved] (AMQ-3326) Synchronization point in RegionBroker
adding and removing producers and consumers
Posted by "Timothy Bish (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/AMQ-3326?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Timothy Bish resolved AMQ-3326.
-------------------------------
Resolution: Fixed
Fix Version/s: 5.6.0
Assignee: Timothy Bish
Confirmed the serialization of calls into those methods is not needed when the inactivity purge is not running so a reader / writer lock is the correct choice.
> Synchronization point in RegionBroker adding and removing producers and consumers
> ---------------------------------------------------------------------------------
>
> Key: AMQ-3326
> URL: https://issues.apache.org/jira/browse/AMQ-3326
> Project: ActiveMQ
> Issue Type: Improvement
> Components: Broker
> Affects Versions: 5.4.2, 5.5.0
> Reporter: Izzy Alanis
> Assignee: Timothy Bish
> Fix For: 5.6.0
>
> Attachments: RegionBroker-542.patch, RegionBroker-trunk.patch
>
>
> Synchronization blocks were added to the RegionBroker as a result of AMQ-2821 (automatic removal of inactive destinations). I believe the author was trying to prevent adding and removing producers while the inactive destination removal process was running.
> This is a major performance bottle neck in use case scenarios where many clients are subscribing and unsubscribing simultaneously -- specifically, in order to retrieve one and only one message from a queue using the STOMP protocol (without holding onto a prefetch message) you must subscribe, retrieve a message and then unsubscribe from the destination. This activity causes lots of adding and removing of producers and consumers.
> I'm still not entirely convinced that the destination removal process needs to be protected in that manner, but the attached patch uses a ReentrantReadWriteLock to continue to protect the destination removal process, while still allowing clients to add and remove producers and consumers concurrently when the destination process is not running.
> This patch also relates to AMQ-3070 (where the user was complaining about this same synchronization block).
> I'm attaching patches both for the trunk and for 5.4.2 (which is what I'm still using).
> There are other synchronization blocks further down in the code (inside AbstractRegion and Queue implementations), but in my particular use case, the 2-3x performance enhancement I get of this patch alone are enough to satisfy my immediate concerns.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Updated] (AMQ-3326) Synchronization point in RegionBroker
adding and removing producers and consumers
Posted by "Izzy Alanis (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/AMQ-3326?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Izzy Alanis updated AMQ-3326:
-----------------------------
Attachment: RegionBroker-542.patch
RegionBroker-trunk.patch
> Synchronization point in RegionBroker adding and removing producers and consumers
> ---------------------------------------------------------------------------------
>
> Key: AMQ-3326
> URL: https://issues.apache.org/jira/browse/AMQ-3326
> Project: ActiveMQ
> Issue Type: Improvement
> Components: Broker
> Affects Versions: 5.4.2, 5.5.0
> Reporter: Izzy Alanis
> Attachments: RegionBroker-542.patch, RegionBroker-trunk.patch
>
>
> Synchronization blocks were added to the RegionBroker as a result of AMQ-2821 (automatic removal of inactive destinations). I believe the author was trying to prevent adding and removing producers while the inactive destination removal process was running.
> This is a major performance bottle neck in use case scenarios where many clients are subscribing and unsubscribing simultaneously -- specifically, in order to retrieve one and only one message from a queue using the STOMP protocol (without holding onto a prefetch message) you must subscribe, retrieve a message and then unsubscribe from the destination. This activity causes lots of adding and removing of producers and consumers.
> I'm still not entirely convinced that the destination removal process needs to be protected in that manner, but the attached patch uses a ReentrantReadWriteLock to continue to protect the destination removal process, while still allowing clients to add and remove producers and consumers concurrently when the destination process is not running.
> This patch also relates to AMQ-3070 (where the user was complaining about this same synchronization block).
> I'm attaching patches both for the trunk and for 5.4.2 (which is what I'm still using).
> There are other synchronization blocks further down in the code (inside AbstractRegion and Queue implementations), but in my particular use case, the 2-3x performance enhancement I get of this patch alone are enough to satisfy my immediate concerns.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira