Tobias Brox wrote:
I think those pages probably should be merged ... hmm ... if
I manage to solve my locking issues I should probably try and
contribute to the wiki.
Certainly the 3rd one could be merged with one of the other two, and maybe all merged into one. I haven't cleaned up that whole area in a whole, it's due for a round of it soon. You're welcome to take a shot at it. We can always revert any mistakes made.
reducing deadlock_timeout.
It's set to one second, and some of the jams we have been experiencing
has lasted for several minutes. I also think it should say in the pg
log if there is a deadlock situation?
deadlock_timeout is how long a process trying to acquire a lock waits before it a) looks for a deadlock, and b) prints a report that it's waiting into the logs when log_lock_waits is on. So if you're looking for slow lock acquisition that's in the sub-second range, it can be useful to reduce regardless of whether deadlock ever happens. That doesn't sound like your situation though.
--
Greg Smith, 2ndQuadrant US greg@2ndQuadrant.com Baltimore, MD
PostgreSQL Training, Services and Support www.2ndQuadrant.us
Author, "PostgreSQL 9.0 High Performance" Pre-ordering at:
https://www.packtpub.com/postgresql-9-0-high-performance/book