On 11-06-01 05:58 AM, Dave Page wrote:
> Hi Guillaume
>
> What happened with this in the end? I've just run into the pg_listener
> bug again. I see from the thread you said you were going to work on
> it, but then we got side-tracked into a discussion on whether we
> should have slony support at all.
>
Since your discussing slony support I'll add my my personal thoughts (I
can't say if the other Slony developers feel the same way).
If your going to continue to have pgadmin change a slony cluster then I
think pgadmin should be issuing commands by invoking slonik as a
sub-process and not by calling the slony stored procedures directly.
My thinking is
1) The API for the stored procedures has been known to change both
between major releases and minor ones, while the syntax of slonik
commands has mostly stayed the same.
2) In a number of cases slonik does more than just call a stored
procedure (ie FAILOVER) and with 2.1 this has increased both because for
most commands need to have obtained a lock on 'sl_event_lock' as the
first command in a transaction (therefore before any stored procedures
have been obtained) and to take advantage of the 'wait for' logic in 2.1