Justin, all
during my last weeks of arcsde gce stress testing, I found a potential concurrency penalty in ResourcePool.getGridCoverageReader. If hit with various concurrent requests at it may instantiate lots of readers to ultimately leave a single one cached.
The following patch applies the same trick than for ResourcePool.getDataStore, please let me know if it seems appropriate to you: <http://pastebin.com/m32ebb87b>
It just double checks for the cached reader after acquiring the lock on the cache.
Cheers,
Gabriel
--
Gabriel Roldan
OpenGeo - http://opengeo.org
Expert service straight from the developers.
Gabriel Roldan ha scritto:
Justin, all
during my last weeks of arcsde gce stress testing, I found a potential concurrency penalty in ResourcePool.getGridCoverageReader. If hit with various concurrent requests at it may instantiate lots of readers to ultimately leave a single one cached.
The following patch applies the same trick than for ResourcePool.getDataStore, please let me know if it seems appropriate to you: <http://pastebin.com/m32ebb87b>
It just double checks for the cached reader after acquiring the lock on the cache.
The patch looks good to me.
Maybe remove those capital letters comments 
(I noticed they are also in the original code, not sure why...)
Cheers
Andrea
--
Andrea Aime
OpenGeo - http://opengeo.org
Expert service straight from the developers.
Thanks for the quick review Andrea,
I will commit with a related jira then just in case we want to roll it back later.
Cheers,
Gabriel
Andrea Aime wrote:
Gabriel Roldan ha scritto:
Justin, all
during my last weeks of arcsde gce stress testing, I found a potential concurrency penalty in ResourcePool.getGridCoverageReader. If hit with various concurrent requests at it may instantiate lots of readers to ultimately leave a single one cached.
The following patch applies the same trick than for ResourcePool.getDataStore, please let me know if it seems appropriate to you: <http://pastebin.com/m32ebb87b>
It just double checks for the cached reader after acquiring the lock on the cache.
The patch looks good to me.
Maybe remove those capital letters comments
(I noticed they are also in the original code, not sure why...)
Cheers
Andrea
--
Gabriel Roldan
OpenGeo - http://opengeo.org
Expert service straight from the developers.