On Sat, Feb 23, 2008 at 02:37:23AM +0800, Ow Mun Heng wrote:
> When the logs start to fill up >2G, I will see the dreaded fetch 100
> from log and it takes upwards to 500++ seconds for it to get the 1st row
> and thus replication will fall behind _way_way_ much.

If you don't have a replica into which to send your data, what's the
difference how long the operation takes?

I'm supposing that in the end you're going to want to drop the failed node
and start again, because the rate you seem to be talking about sounds like
it's going to overwhelm you.  I just have a nagging suspicion that your DROP
NODE message isn't going to have any effect until you get the log switch
completed.  I may be wrong, though.

A