On Feb 9, 2008, at 4:35 AM, Alan D. Cabrera wrote:
I thought that I knew JCA pretty well. What is a permit leak?
It's not an external JCA concept, but an internal implementation
mechanism. A "permit" is an acquired semaphore used to grant
permission to obtain a connection. In some exceptional cases, looks
like we weren't properly releasing the "permit". If you lose enough
permits, you stop handing out connections. So, things slow down a
bit... ;-) There might have been another problem adjusting pool size,
but I'm not sure. A bit hard to tease out from some code re-factoring
occurred...
Regards,
Alan
On Feb 9, 2008, at 1:22 AM, David Jencks wrote:
Working with Tomasz Mazan we discovered that there was a permit
leak in the connector pooling code (GERONIMO-3834). While fixing
this I revamped quite a bit of the pooling code and added several
permit count and resizing tests, hopefully making it more reliable
and reducing the bug count.
I think we should get this bug fix in g. 2.1 which will require
releasing the "components" jars also. I'm OK with a vote on the
component jars + geronimo 2.1.
Agreed. I'll start pulling things together...
We might consider releasing 2.0.3 with this fix as well.
Possible. What do others think?
--kevan
