On Fri, Sep 14, 2012 at 6:15 PM, David Roid <dataroid(at)gmail(dot)com> wrote:
> No long-running trac, commit is done fair often. I also tried "one conn for
> all threads" vs "one conn for one thread", and "one cursor all the way" vs
> "use and throw away cursor", issue remains.
David has provided me further information about his setup and a gcc
traceback, from which it was obvious that he was performing database
queries from a signal handler. It seems he is not experiencing any
more problems since he got rid of this unsafe practice.
-- Daniele