Re: pgsql Replication Proxy (was Re: Replication for a

Then it is removed from the list of "valid" servers, and it isn't included in
updates, at least until a) a DBA fixes the issue, or b) the DBA tells the
database server to re-mirror itself from the other available servers.

Once two-phased commit support is available, this could be changed to be much
more robust.

Any alternative ideas?

On Tuesday 06 May 2003 10:54 am, Mendola Gaetano wrote:
> Postgres don't have the support ( yet ) for the 2 phase commit, so
> I think that is impossible do it now.
> What happen if the last server do an error in commit phase ?
>
> Regards
> Gaetano
>
> ----- Original Message -----
> From: "Diehl, Jeffrey" <jdiehl(at)sandia(dot)gov>
> To: "'Michael A Nachbaur'" <mike(at)nachbaur(dot)com>; <pgsql-sql(at)postgresql(dot)org>
> Sent: Tuesday, May 06, 2003 6:28 PM
> Subject: Re: pgsql Replication Proxy (was Re: [SQL] Replication for a
>
> > I love this idea. The proxy could return immediately instead of making
> > my program block on update.
> >
> > One note, though. Instead of a stack, you need a FIFO. For example:
> >
> > delete from sometable where field=value;
> > insert into sometable (field) values (value1);
> > insert into sometable (field) values (value2);
> > ....
> >
> >
> > This code breaks in a stack and only works in a fifo. Minor point,
>
> though.
>
> > So do we have a volunteer to write such a tool? <grin>
> >
> > Mike Diehl.
> >
> > -----Original Message-----
> > From: Michael A Nachbaur [mailto:mike(at)nachbaur(dot)com]
> > Sent: Monday, May 05, 2003 1:57 PM
> > To: pgsql-sql(at)postgresql(dot)org
> > Subject: pgsql Replication Proxy (was Re: [SQL] Replication for a large
> > database)
> >
> >
> > I've thought some more about this, and I want to pass this idea past you
> > guys.
> > What do you think about a replication proxy, essentially a daemon that
>
> sits
>
> > between a PostgreSQL client and server. Every single SQL query,
>
> transaction
>
> > statement, etc that the proxy recieves it repeats back to all the
> > database servers. In this way, if a back-end database server goes down
> > queries
>
> will
>
> > continue unabated (except the downed server won't recieve updates).
> >
> > Basically, the proxy server could intercept these queries and place them
>
> in
>
> > a
> > stack (on a per-database basis) and when every server in the queue
> > acknowledges the update, the query is removed from the stack. Each
>
> database
>
> > server can have their own position in the stack, so if servers A and B
> > successfully run a query, but C doesn't (e.g. it requires human
> > intervention), C is removed from the list of acceptable servers but A and
>
> B
>
> > can keep moving through the queue.
> >
> > What do you think? Also, should this discussion be moved to another
>
> mailing
>
> > list?
> >
> > On Monday 05 May 2003 12:26 pm, Michael A Nachbaur wrote:
> > > I have thought about this. The problem I come into is data
> > > consistancy.
> >
> > I
> >
> > > have at least 8 different processes that harvest data, and an intranet
> > > website that can also manipulate the database (to assign customers to
> > > different packages, re-assign modems to different customers, etc).
>
> Trying
>
> > > to maintain consistancy across the entire application would be such a
> > > nightmare, I don't want to think about it.
> > >
> > > If I go with a centralized middleware server that manages all database
> > > access, then I could perhaps do that in there...and then I could use
> > > transactions on both databases, and if either transaction fails then
>
> I'll
>
> > > roll back the other. But this would make my entire framework very
>
> rigid.
>
> > ---------------------------(end of broadcast)---------------------------
> > TIP 1: subscribe and unsubscribe commands go to majordomo(at)postgresql(dot)org
> >
> >
> > ---------------------------(end of broadcast)---------------------------
> > TIP 3: if posting/reading through Usenet, please send an appropriate
> > subscribe-nomail command to majordomo(at)postgresql(dot)org so that your
> > message can get through to the mailing list cleanly