Re: WAPBL patch for testing

[Thor Lancelot Simon <tls%rek.tjls.com@localhost>]

On Sat, Nov 01, 2008 at 10:42:25AM +0100, Joerg Sonnenberger wrote:
> On Sat, Nov 01, 2008 at 10:36:12AM +0100, Juergen Hannken-Illjes wrote:
> > > For snapshots, the real problem is that the strategy of
> > > fss doesn't handle WAPBL as it should.
> > 
> > And how should fss handle wapbl?
> 
> It should know by itself if it has to initiate a transaction or not
> because it is called from within or just via VOP_PUTPAGE.

It already has a perfectly good notion of a "transaction".  WAPBL added
a second one.  This is not fss's fault, it's hasty coding on the part of
the person who ported WAPBL to modern NetBSD (which may have been the
right compromise but it's still not a good situation, in the abstract).

> Given that fss is a reliable way to crash the machine anyway as soon
> as trying to remove the snapshot, I'm not sure I care about it right
> now.

I do, and so do a lot of other people.  We cannot presently take consistent
backups of several of the NetBSD.org servers because fss is broken -- which
is a severe problem, much worse than a performance issue we could address
by adding hardware.  A number of our servers are still on 4.0 because we
need fss.  I think there are probably a lot of people in this situation,
because they care more about correctness than speed.

Thor