On Sat, Jul 26, 2008 at 3:24 AM, chris <cbbrowne at ca.afilias.info> wrote:

> "Gurjeet Singh" <singh.gurjeet at gmail.com> writes:
> >                                                                    Hi
> All,
> >       The docs for CREATE SET say 'No application-visible locking should
> take place'. But we saw that it was hung, and noticed that there was a
> >          VACUUM running on master node. We waited for quite a while, and
> after we killed that VACUUM, the CREATE SET moved forward.
> >                                                 PG is 8.1.11, and Slony
> is 1.2.14, if that helps.
>
> The only thing that CREATE SET does, initially, that involves locking
> of *anything* is that it takes out a lock out on sl_config_lock, a
> table internal to Slony-I.
>
> The only way for that to lock things that are application-visible is
> if you have some application that's vacuuming *everything*, and which
> therefore takes out a lock on sl_config_lock that prevents it from
> being granted to the CREATE SET request.
>
> The locking that is done should indeed not be visible to applications.
> It only became visible because you had a VACUUM that was working on
> the Slony-I schema.
>
> I'm trying to think of what more to say, in the documentation; nothing
> is really coming to me.  I don't think that the documentation is
> misleading.
>

I was not implying that docs are insufficient in any way; just trying to get
attention to a problem I faced in production environment. Maybe it's only an
issue with the old 8.1.11!

One more little thingy, I prefer to call the product simply Slony, rather
than Slony-I. I'd love to call it Slony-I, but only if Slony-II was any
nearer than 'beyond' the horizon; I don't see Slony-II progressing, and even
if it does get implemented, it'd be radically different technology than the
current implementation. So, lets spare all of us some (mild-) pain by making
'Slony' the official name; at least like PostgreSQL has accepted Postgres as
an alternate correct name.

On topic, the VACUUM i saw running was on one table, which was a user table,
not a Slony table. We do not have DB wide vacuuming policy (yet) (I'll
re-confirm if you wish), so I don't think it was a DB-wide vacuum; and even
if it was, IMHO vacuum takes locks on one table at a time.


Best regards and kudos to a great product,
-- =

gurjeet[.singh]@EnterpriseDB.com
singh.gurjeet@{ gmail | hotmail | indiatimes | yahoo }.com

EnterpriseDB http://www.enterprisedb.com

Mail sent from my BlackLaptop device
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.slony.info/pipermail/slony1-general/attachments/20080726/=
0cca2395/attachment-0001.htm