Did a new grizzly rev get put into glassfish b73?
We're seeing an odd thing in build 73 where tasks get rejected by the
SyncThreadPool because it is full. Actually, it's not that odd, because
the size of that pool is only 1000 in our configuration, and we have
27200 connections, so the odds of there being 1000 tasks actually seems
fairly big to me. There's an 8 second wait time between requests that
average 3 seconds to execute, so we'd really expect to average 1133
tasks in the queue.
But we had the same configuration in build 72 and didn't see that error,
which really makes me think maybe the grizzly build in glassfish 72 had
an error in the task pool sizing. In v2, we would size the task pool to
310000, and we thought we were doing that in v3, but of course our
configuration scripts are still buggy. But also we changed versions of
the faban driver, and it is possible that is sending the load
differently (and more in line with expectations).
-Scott