On 12/20/2006 11:32 AM, Andreas Kostyrka wrote:
> * Norman Yamada <nyamada at millburncorp.com> [061219 22:22]:
>>    ?? Sorry -- what's the right node in that case? The master node?
>>    In my case I have a master node and two slaves and one is definitely
>>    lagging. So can I drop the slave node and as long as the master node sees
>>    the event, I'm okay?
> 
> Basically yes. All nodes that are still part of the cluster after the
> event, should see it. It should probably probagate from the master
> node without troubles.
> 
> The only case where there might be troubles would be:
> 
> MASTER <-> SLAVE1 <-> SLAVE2
> 
> If you drop SLAVE1, you've got a problem that SLAVE2 has no path to
> master.

No, this wouldn't be a problem either. To kill SLAVE1 you would

     1)  STORE PATH (SLAVE2 -> MASTER)
     2)  STORE PATH (MASTER -> SLAVE2)
     3)  SUBSCRIBE SET (receiver=SLAVE2, provider=MASTER)
     4)  wait until SLAVE2 is replicating from MASTER
     5)  DROP NODE SLAVE1

Note that step 4) will _NOT_ cause SLAVE2 to actually recopy the entire 
data. Since MASTER still has all unconfirmed changes that SLAVE2 has to 
replicate, it will simply pick up from there and continue.


Jan

-- 
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck at Yahoo.com #