[Pgcluster-general] PGcluster recovery error with -R option

[Rick Vernam]

On Thursday 02 July 2009 3:17:24 am Pedro Delgado wrote:
> Hi, I configured pgcluster with 2 data base servers.
> All it works fine, but when I start a database in recovery mode con -R
> option and I make a lot of insert/update in the other database, I lost some
> rows in the process. This problem it's so stragne because a big number of
> rows are correctly replicated when the data base it's in recovery mode, but
> a few number no. I tryed tu insert 100 rows in the started database when
> one data base it's recoveryng with -R. When the process ended i checked the
> content of two databases. The first (active) have the 100 rows, the second
> (recover) have 96 rows, it lost four rows. This fail it's critical because
> the database cluster it's inconsistent. Some one know about this problem?
Yes.  It's been a while, but I recall having come up with two possible ways 
for this to happen..

1 - It could be that PGCluster's queue of queries received at first (active) 
server during replication becomes full (I know for sure this was a bug at one 
point - not sure if it got fixed & released).

2 - It could be that the four missing queries came to the first (active) server 
while it was executing the queued queries on the second (recovery) server.  (I 
don't recall if I ever proved this was happening...)

> it's that a problem o a configuration error?
not a configuration problem...

> In my configuration all the
> features works fine (cluster, load balancer, replication server) and i
> think that it's a bug of pgcluster, but I'm not sure. If some one can help
> me I'll be very agreed.
> Greetings.
> _______________________________________________
> Pgcluster-general mailing list
> Pgcluster-general at pgfoundry.org
> http://pgfoundry.org/mailman/listinfo/pgcluster-general