We recently ran into this as well, because in testing we had people leaving an application running against the database while they uninstalled slony. I'm curious, what OIDs would be missing exactly, since the application does not refer directly to any slony objects? Does the cached plan "know" about the slony trigger on a given table? I don't know the extent of information stored in plans. Thanks, because I was just beginning to be puzzled by this! - DAP >-----Original Message----- >From: slony1-general-bounces@xxxxxxxxxxxxxxxxxxxx >[mailto:slony1-general-bounces@xxxxxxxxxxxxxxxxxxxx] On Behalf >Of John Sidney-Woollett >Sent: Tuesday, February 15, 2005 7:03 AM >To: Richard Huxton >Cc: postgres general; slony1-general@xxxxxxxxxxxxxxxxxxxx >Subject: [Slony1-general] Re: Slony uninstall info/warning > >Thanks for the info, Richard. > >I didn't think that it was a slony issue per se, but that a >note should be added to the slony docs warning to recycle >connections after making substantive changes to the schema. > >You're right, we use both (java) prepared statements and >pl/pgsql functions. > >The data loss aspect is not so clear cut (for us). We >definitely got records that failed to insert (missing sequence >numbers in tables) while others did insert correctly into the >same tables (using the same pl/pgsql functions). So we can't >figure out what the pattern is. > >Although we do have our connection pool recycle idle >connections - it could be that things worked when a new >recycled connection was used by the web app, and failed when >one of the "old/original" connections was used. This is >probably what masked the errors for us... > >It would be great if postgres could "recompile" pl/pgsql >functions whenever it found a missing object referenced within >the function - chances are that it would compile cleanly (in >this case) and then could be executed without error. > >something along the lines of > >execute function >OK -> return result >ERROR - OID -> recompile function, and re-execute > OK -> return result > ERROR - OID -> report error > >This would help get rid of the temp table in functions work >around having to use an execute statement. > >Thanks for your help and feedback. > >John Sidney-Woollett > >Richard Huxton wrote: > >> John Sidney-Woollett wrote: >> >>> Hopefully this will prevent data loss or problems for others using >>> slony 1.0.5 and pg 7.4.6... >>> >>> We just got bitten by something we didn't foresee when completely >>> uninstalling a slony replication cluster from the master >and slave... >>> >>> MAKE SURE YOU STOP YOUR APPLICATION RUNNING AGAINST YOUR MASTER >>> DATABASE WHEN REMOVING THE WHOLE SLONY CLUSTER, or at least >re-cycle >>> all your open connections after the event! >>> >>> The connections appear to "remember" or refer to objects which are >>> removed by the uninstall node script. And you get lots of >errors as a >>> result (and possible data loss??)... >>> >>> Question: Why do our database objects still refer to removed slony >>> objects after they are removed? >> >> >> Well, there are two areas I know cache plans/OIDs: >> 1. Prepared statements >> 2. plpgsql functions >> >> At a guess, since you mention Java the first almost >certainly applies >> to you. >> >> This isn't a slony issue so much as a cached plan issue. I'm >guessing >> the same problems would occur if you were manually changing the >> database schema. >> >> Don't think you could get data loss (unless the application ignores >> errors). You will however get to see a wide range of OID >related errors. >> >> -- >> Richard Huxton >> Archonet Ltd >_______________________________________________ >Slony1-general mailing list >Slony1-general@xxxxxxxxxxxxxxxxxxxx >http://gborg.postgresql.org/mailman/listinfo/slony1-general > ---------------------------(end of broadcast)--------------------------- TIP 8: explain analyze is your friend
