[pgpool-general: 4061] Re: PgPool stability

Jose Baez pepote at gmail.com
Tue Sep 15 05:40:36 JST 2015


I have been using your scenario (2 PSQL nodes with Postgis in replication
mode). I can say pgpool is stable (it hasn't failed in 2 years). Recovery
is manual (needs to execute one command from administrator) but you don't
have to do anything extra. The dificulty laid in its initial configuration
(I had to tweak 3 scripts a lot).

Besides that, replication mode didn't work for us:

1.)  From time to time pgpool detects data mismatch between nodes (pgool
 got more results in a query from 1 node than the other), so it dettaches
the node. I guess there is some PSQL function which runs inside DDBB and
gets different results on both nodes. PostGIS maybe?

2.) When reattaching the node (doing 'online recovery') everything works
fine, but if I create new objects (DDBB, tables, etc) both nodes use
different OID's for each object. When pgpool detects this difference it
dettaches one node. I guess they are getting different OID because the
reattached node is in another PSQL "timeline" from activde node after
recovery.

3.) If I don't create new objects, after a few days there is the same
mismatch about getting different number of results from some query.

(There is a workaroung for number 2, which is to shutdown and restart the
active node. Now I can execute online recovery to reattach failed node.)

At the end I am using master-slave with PSQL's synchronus replication which
doesn't suffer from these anomalies.




On 14 September 2015 at 15:31, VILLARD, Pierre <pierre.villard at capgemini.com
> wrote:

> Hi,
>
> I am comfortable with standalone PostgreSQL but, so far, I never used
> PgPool to switch to a multi-instance installation. In order to improve the
> read-only performances on my system, I am considering to switch to a two
> (or more) nodes installation of PostgreSQL with the native replication of
> PostgreSQL and the use of PgPool. I am also using a lot PostGIS with my
> PostgreSQL installation and I am accessing this database through Java code
> using a JDBC/Hibernate layer.
>
> Unfortunately, I received a bad feedback about this approach. According to
> this feedback, PgPool would not be very stable, the load balancing would
> not be mature and the recovering on a failure would be manual and very
> complex.
>
> I have to admit that this feedback slowed things down... I need to have
> something really robust regarding my context of application.
>
> Do you have feedbacks on your side regarding my approach ? Some advice ?
>
> Pierre
> This message contains information that may be privileged or confidential
> and is the property of the Capgemini Group. It is intended only for the
> person to whom it is addressed. If you are not the intended recipient, you
> are not authorized to read, print, retain, copy, disseminate, distribute,
> or use this message or any part thereof. If you receive this message in
> error, please notify the sender immediately and delete all copies of this
> message.
>
> _______________________________________________
> pgpool-general mailing list
> pgpool-general at pgpool.net
> http://www.pgpool.net/mailman/listinfo/pgpool-general
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.sraoss.jp/pipermail/pgpool-general/attachments/20150914/74db3755/attachment.html>


More information about the pgpool-general mailing list