[Pgpool-hackers] [Pgpool-general] pgpool limitations

[Gurjeet Singh]

On Thu, Jul 28, 2011 at 6:05 PM, Tatsuo Ishii <ishii at sraoss.co.jp> wrote:

> >     I have configured pgpool 3.0.3 in master/slave mode and setup
> streaming
> Note that pgpool-II 3.0.4 is the latest stable release.
>
>
I am using the version that EnterpriseDB bundles with PPAS (Postgres Plus
Advanced Server)


>  > replication between the master and the slave systems. I want to be able
> to
> > add more slaves to the setup whenever read-only load increases beyond the
> > current capacity, and remove slave databases when the load subsides.
> >
> >     I have found the following limitations, and wanted to check with you
> > guys if these are really true or is it just me that I misconfigured
> pgpool.
>
> They are all expected behavior.
>

Sorry to hear that; doesn't really help in seamlessly scaling up/down the
read-only behaviour.


>
> > .) Killing one slave Postgres instance (in immediate mode) kills all
> > connections to the pool, even those connections whose queries are
> currently
> > running on a different Postgres instance which is still available.
> >
> >     In essence a connection loss from a slave backend causes pgpool to
> > reinitialize itself, hence breaking all existing connections.
> >
> > .) The commands pcp_attach_node and pcp_detach node cause the pool to
> > disconnect all connections and reinitialize the pool.
> >
> > .) Adding a new entries backend_hostname<n> and backend_port<n> to the
> end
> > of the current node list in pgpool.conf and reloading the config does add
> > the new node to pgpool (as documented)l, but the pcp_node_info command
> shows
> > that this new node is in 'down' state (code 3) and the docs don't say how
> to
> > bring a node from 'down' state to 'up' state (code 1 or 2). Doing
> > pcp_node_attach does bring it in 'up' mode but as noted above, this
> command
> > causes all connections to be disconnected.
>
> Because newly added node is likely to have out of sync databases. Thus
> bringing it online immediately will allow user to use obsolete
> databases. You should do online recovery first.
>

Since the new slave being added is configured using streaming replication,
it is lagging behind the master just as much as any other currently
configured slave. So from that standpoint the new slave should be allowed to
serve queries.


> >     How can I add/remove slave databases to pgpool cleanly without
> affecting
> > the connections that are working on other databases.
>
> It's impossible with pgpool-II 3.0.x.
>
> Upcoming pgpool-II 3.1 will not disconnect connections after online
> recovery.
>

I guess we should allow for the case where we want to add an existing slave,
which was configured independent of pgpool.

Thanks,
-- 
Gurjeet Singh
EnterpriseDB Corporation
The Enterprise PostgreSQL Company
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://pgfoundry.org/pipermail/pgpool-hackers/attachments/20110728/c1b7fedd/attachment.html>