We use Tomcat(4.1) DBCP Connection pooling to access Postgres(7.2.1) on
Windows.
What we notice is that the number of handles used by the postmaster process
on the server
gradually increases if the same connection is used for update operations for
a extended period of time.
In a couple of days of high activity the number of handles is in the range
of 2 million and
eventually crashes the server.
You can see the number of handles in Windows Task manager. With the Process
Explorer tool
(www.sysinternals.com) you could see that these handles are all Semaphore
handles.
Running Vacuum at regular intervals helps to prolong the crash by some more
days.
Any suggestions,
Thanks,