[Pgpool-general] Kick off inconsistent node from cluster

[Jaume Sabater]

On Fri, Oct 9, 2009 at 12:56 AM, Nikolay Popov
<nikolay.popov at altima.com.ua> wrote:

> What if I need more than 2 nodes placed on remote sides with a bit
> unreliable links?
> It will be good if for example node3 lost connection to rest part of pool,
> applications that are using it on this site continue to work (as far as
> there is nobody to failover to there) but not getting any requests from
> node1 and node2 even after link is restored. After node synchronize their
> changes with rest nodes in pool (possibly in manual way to avoid conflicts)
> it becomes fully functional member again.
> It's even acceptable to lost local changes at node3 - until link to it is
> down, the rest nodes don't need it's data so we can just take info from any
> server available.
> I suppose we need pgpool to be installed on each remote node to do that, but
> this will prevent us from using online restore (as directed in
> documentation)
> Is there anything to do with pgpool-II at all or I need another solution to
> use?

I am not sure that I understand what you are asking, but here you are
some advice:

- There is pgpoll-II and there is PostgreSQL.
- There can be a number of pgpool-II instances in a master slave configuration.
- There can be a number of PostgreSQL nodes behind pgpool-II.
- When a node is lost/degenerated (e.g. link lost), the rest of the
cluster should be working fine.
- When the link to the currently active pgpool-II instance is lost,
another should take its place (via Heartbeat or similar).
- Automatic failback is, IMHO, a bad idea in this sort of setup.

-- 
Jaume Sabater
http://linuxsilo.net/

"Ubi sapientas ibi libertas"