On Thu, Apr 26, 2012 at 3:07 PM, dennis jenkins <dennis.jenkins...@gmail.com > wrote:
> On Thu, Apr 26, 2012 at 1:59 PM, Kenneth Tilton <ktil...@mcna.net> wrote: > >> On Thu, Apr 26, 2012 at 2:24 PM, Tom Lane <t...@sss.pgh.pa.us> wrote: >> >> >>> Can you produce a self-contained test case? >>> >> >> I doubt it. Every test iteration I run includes a lot of redefining of >> functions and triggers all over the map, and it works fine thru dozens of >> iterations. The wheels seem to come off after a serious refactoring. >> Nothing is needed to clear the problem except (it seems) close any >> connections to the DB, so I guess some optimization does not work in the >> face of sufficient dynamic redefinition. >> >> >> > Suggestion: create a program which attempts to stress-test Postgresql by > doing similar things. Even if a single test case does not reliably > reproduce the issue, a stress-test might trigger the condition after some > time. > > I am starting to suspect this does it (having reproduced the problem with vastly less effort than suspected): with function A calling function B, and function B coming in several variants (diff params), drop then define A and then drop/define only one variant of B. Order there may not matter. I plan anyway to enhance our builder so it knows about function overloading and rebuilds all variants when it rebuilds one. If the problem goes away I'll take a crack at creating a reproducible. Thx for the feedback. -ken
