Thanks! I'm glad to know I'm not going crazy - ungraceful slon exits are
definitely part of the scenario in our testing. The postgres
documentation on NOTIFY seemed to indicate that these records got
cleaned out automatically when a process exits, so I was confused.

Do you mean RESTART NODE? I will take a look. Certainly easier to
provide operations with an interface to that command rather than telling
them to start deleting records from postgres system tables ;-)

- DAP 

>-----Original Message-----
>From: cbbrowne at ca.afilias.info [mailto:cbbrowne at ca.afilias.info] 
>Sent: Sunday, May 01, 2005 9:40 PM
>To: David Parker
>Cc: slony1-general at gborg.postgresql.org
>Subject: Re: [Slony1-general] slon and pg_listener
>
>> So I thought I would stick my head out of the funhouse for a moment 
>> and ask people out in the world if they have ever seen this behavior 
>> with slony, or if any of the postgres gurus on the list know of any 
>> situation in which a pg_listener record could get left lying around.
>
>You're still running 1.0.2, right?
>
>If either a postmaster or a slon goes down ungracefully, that 
>will leave notifications sitting around that would need to be 
>cleaned out.
>
>As of 1.0.5, that is done automatically when each slon starts 
>up, but it wasn't done in earlier versions.  With earlier 
>versions it was pretty easy for failures of any sort to leave 
>notifications lying around such that you'd need to run the 
>slonik command RESET NODE to clean up.
>
>This was something we found massively useful at Afilias; we 
>had quite a lot of cases of replication falling somewhat 
>behind because a node got deranged where the report was "Will 
>be addressed when we get to version 1.0.5".  Those cases 
>certainly did go away...
>
>