Problem with locks

We're seeing a problem where occasionally a process appears to be granted a
lock but miss its semaphore signal. It eventually gets the 60s deadlock signal
and notices that it has been granted the lock and gets back to work.

I haven't gotten any further than this in the forensic analysis and it's a bit
tricky as it only happens in fairly long benchmark runs. It seems to happen
every few hours of running at 3000+ transactions per minute...

--  Gregory Stark EnterpriseDB          http://www.enterprisedb.com