On Thu, Nov 04, 2004 at 10:57:06AM +0100, Mark Jaroski wrote:
> 
> But, it just occurred to me that if we're replicating the db
> anyhow we could possibly turn fsync off for the master, but
> leave it on on the slave(s), thereby having our fsynced logs
> on the slave, while also enjoying the performance benifit on
> the master.
> 
> Is this completely nuts?

Yes.  The problem with turning off fsync isn't that the writes on the
disk aren't _ever_ going to happen, but that they aren't guaranteed
at COMMIT time.  Moreover, the very recently COMMITed transactions
are also the ones not yet replicated.  So Slony-I will not save you
at all if you turn off fsync.

We've found that fsync is close to zero additional cost if you have a
big enough battery-backed write cache on your disk subsystem.   I'd
have a look at that sort of thing.

A

-- 
Andrew Sullivan  | ajs at crankycanuck.ca
The fact that technology doesn't work is no bar to success in the marketplace.
		--Philip Greenspun