On Sun, Aug 23, 2009 at 10:59:13PM -0700, roctaiwan wrote:
> 
> Hi Andrew,
> 
> If I understand the slony replication correct, is that after I do switchover
> or failover what it can do is promote another node to replace its origin
> node. 

Hold on: switchover and failover are completely different.  Failover
abandons the former origin.  Switchover does not.  Switchover is
intelligent about this, but failover just takes over at the new node
at whatever stage that node is in.  This works on a _set_, and not on
the node as such.

> If lets say failover, after origin node failed and another node
> replaced it, all other slave nodes in group will seek for sets from new
> master, basically failover script will change the "provider" from subscribe
> set command to the new master node. Therefore, new master should first sync
> with old master with all the data and having all sets isn't it? 

Not for failover, no.  Switchover does this in a controlled manner,
and therefore moves the set at a logical point in time for that set.
 
> Let's say Old master(node 1) has five tables (table 1,2,3,4,5) and new
> master (node 2) has only (table 1 and 2). After switch over or failover to
> the new master (node 2), will node 2 sync with old master and has all 5
> tables? 

I don't think so, no.

A

-- 
Andrew Sullivan
ajs at crankycanuck.ca