[Pgpool-general] PgPool install, Part Deux

[Tatsuo Ishii]

> > Hi, Marc!
> >
> > 2009/7/7 Marc G. Fournier <scrappy at hub.org>
> >
> >> Last time I tried, something got "out of sync", ended up breaking the 
> >> cluster and going to one DB until I had time to dive better into it, 
> >> with two brand new servers ... so, I'm going to make attempt two at 
> >> this ...
> >>
> >
> > Be sure that both your nodes are a "hard copy" one of the other. For 
> > example, use tar+gzip+cp or rsync in order to copy the whole pg-cluster 
> > from one node to another.
> 
> I'm building this from scratch, actually ... two brand new 8.4.0 installs, 
> with any interaction with them going *through* pgpool ... there is nothing 
> local to the machine(s) that are touching them ...

I think he is talking about your *data*, not program.

> >> load_balance_mode?
> >
> >
> > From the docs:
> > load_balance_mode:
> > When set to true, SELECT queries will be distributed to each backend for
> > load balance. Default is false.
> >
> > So, if you want to load balance your selects do use it! :)
> 
> Right, but does that mean that the same SELECT will be run on each node 
> simultaneously, or only passed to one nodes?  If the former, would kinds 
> defeat the point of 'load balancing', but if you read the docs, that is 
> what it implies :)

When a DB client connects pgpool and start a session, it is redirected
randomly to one of two PostgreSQL servers. If the other DB client
connects to pgpool, it will also be redirected to one of
PostgreSQL. So if there is only one DB client, actually load balancing
is meaningless.

> >> replication_stop_on_mismatch?
> >
> >
> > From the docs:
> > replication_stop_on_mismatch:
> > When set to true, pgpool-II degenerates the backends and keeps the service
> > only with the Master DB if data mismatch occurs. If false, pgpool-II just
> > terminates the query. Default is false.
> 
> But, if I'm doing replication / load balancing ... which is Master, which 
> is Slave?  Therefore, which one gets degenerated?  The Docs, again, do not 
> make this clear ...

Ok, let me explain. "master" means the first PostgreSQL appeared in
pgpool.conf. If it goes down, the second one will become new master
automatically.

If you have only two servers and reposnses from PostgreSQL do not
agree, always second one will be degenerated.

If you have more than two PostgreSQL servers, it will be a little bit
complicated. Consider if you have 3 PostgreSQL servers, A, B and C.
If the response from A and C are identical while B is different, B
will be degenrated. Because B is likely a failed server.

> > AFAIU, we could say that if this option is enabled, then in case that pgpool
> > finds data mismatch between nodes, it will degenerate 'not master nodes', in
> > other words, it deactivates replication. It could be interesting for
> > production enviroments..
> 
> 
> Well, considering that I'm running this as a backend to something that 
> will be *very* visible to the PostgreSQL community at large (ie. the 
> mailing lists), I'd rather would with facts vs "intersting for production 
> enviroments" ... your comment almost implies that pgpool isn't even ready 
> for production environments yet ... is that the case?
> 
> If pgpool-II is not ready for a production environment, please do let me 
> know ... what I'm trying to implement is something that will be visible to 
> the PostgreSQL community at large, and would hate to create a shadow 
> because I used something that wasn't production ready :(

Actually pgpool-II has been already used in production in Japan. For
example, JTB, one of the largest travel agency in Japan
(http://www.jtbcorp.jp/en/), having 28,352 employees, uses
pgpool-II+PostgreSQL for their brand new mission critical system which
provides membership services to their customers.
The web site of the system is located at:

http://travelpoint.jtb.co.jp/tabicard/

The system was bilt by NTT DATA CCS, one of child companies of NTT
DATA (the largest system integrator in Japan). SRA OSS helped them.

Also SRA OSS provides commercial support to number of customers who
seriously use pgpool-II for their business.
--
Tatsuo Ishii
SRA OSS, Inc. Japan