Обсуждение: pg_listener records

- DAP
----------------------------------------------------------------------------------
David Parker    Tazz Networks    (401) 709-5130
 

On Thu, 2005-05-26 at 13:41, David Parker wrote:
> In failover testing we have been doing recently (postgres 7.4.5 w/
> slony 1.0.2) we have seen several times when the database comes back
> up after a power failure it still has old pg_listener records hanging
> around from its previous life. This causes some problems with slony,
> but of course it is easy enough to implement a procedure to clean
> those records out, which we have done.
>
> But I'm wondering - shouldn't that be part of normal server startup,
> cleaning out the pg_listener table? Or has this been addressed in
> 8.X.? Or is there a reason this isn't a good idea?

You should really be running the latest version of slony, 1.0.5.  There
were plenty of little niggling bugs in the earlier version that have
been fixed.  I'd upgrade postgresql while I was at it too, but slony
DEFINITELY needs to be the latest version.

I'm pretty sure the problem you speak of was in fact fixed in later
versions, btw.

Thanks. Yeah, I know slony 1.0.5 cleans up after itself, and is better
in general, and I want to get there, but upgrading is not an option at
the moment, unfortunately. Same for postgres 8.

But it still seems like this is something the server itself should be
taking care of, not a client process....

- DAP

>-----Original Message-----
>From: Scott Marlowe [mailto:smarlowe@g2switchworks.com]
>Sent: Thursday, May 26, 2005 2:51 PM
>To: David Parker
>Cc: postgres general
>Subject: Re: [GENERAL] pg_listener records
>
>On Thu, 2005-05-26 at 13:41, David Parker wrote:
>> In failover testing we have been doing recently (postgres 7.4.5 w/
>> slony 1.0.2) we have seen several times when the database comes back
>> up after a power failure it still has old pg_listener
>records hanging
>> around from its previous life. This causes some problems with slony,
>> but of course it is easy enough to implement a procedure to clean
>> those records out, which we have done.
>>
>> But I'm wondering - shouldn't that be part of normal server startup,
>> cleaning out the pg_listener table? Or has this been addressed in
>> 8.X.? Or is there a reason this isn't a good idea?
>
>You should really be running the latest version of slony,
>1.0.5.  There were plenty of little niggling bugs in the
>earlier version that have been fixed.  I'd upgrade postgresql
>while I was at it too, but slony DEFINITELY needs to be the
>latest version.
>
>I'm pretty sure the problem you speak of was in fact fixed in
>later versions, btw.
>

"David Parker" <dparker@tazznetworks.com> writes:
> But I'm wondering - shouldn't that be part of normal server startup,
> cleaning out the pg_listener table?

Perhaps, but the code is written such that it's unlikely to be a major
problem --- notifying processes automatically clean out entries that
don't correspond to active backend PIDs.

The long-term solution to LISTEN/NOTIFY performance issues is to get rid
of the table altogether, so I don't see a lot of point in putting effort
into band-aids like cleaning out the table during restart.

            regards, tom lane

On May 26, 2005, at 2:41 PM, David Parker wrote:

> But I'm wondering - shouldn't that be part of normal server startup,
> cleaning out the pg_listener table? Or has this been addressed in
> 8.X.?
> Or is there a reason this isn't a good idea?
>

Try slony 1.0.5, which fixed *many* issues and bugs.  I believe this
is one of them.  The upgrade should be uneventful.