Does ext3 prevent partial page writes?

[Bruno Wolff III]

On Wed, Dec 20, 2006 at 06:42:20 +0000,
  Christian Kujau <lists at nerdbynature.de> wrote:
> On Tue, 19 Dec 2006, ext3 at jks.tupari.net wrote:
> >Basically I want to know if I can turn off full_page_writes in my postgres 
> >config.
> 
> if your devices support write barriers, you can turn off this option, 
> methinks (mount -o barrier=1). Of course, you'll run a few tests before 
> going live, right?

I have tested write barriers in FC5 using ext3 on top of sofware raid (raid
1 is the only type of raid that supports write barriers) and it seems to
be working correctly. If you use this with the data=journal option I would
expect you would be OK. It might be a good idea to check on the postgres
list about the exact semantics you need for this.

I also asked about write barriers on the dm-crypt list and they are supported
there, but there is probably a problem with it on 2.6.19 kernels on SMP
machines relating to a change to use per cpu work queues.