[OpenAFS-devel] Re: [OpenAFS] Re: 1.6 and post-1.6 OpenAFS branch management and schedule

[Andrew Deason]

On Thu, 17 Jun 2010 22:01:08 +0200
Christof Hanke <christof.hanke@rzg.mpg.de> wrote:

> Am Donnerstag, 17. Juni 2010 21:30:23 schrieb Andrew Deason:
>
> > And in particular, NTFS and other journalled filesystems have the
> > advantage of a journal, and probably lots of other similarly helpful
> > things. Guess what we do not have.
> 
> Right, this is actually Hartmut's point. We have user-volumes spread
> out over a few fileservers, so in each partition of such a server
> there are hundreds of user home-volumes. It is very painful for the
> admin and his phone, if these are not back up quickly. Being user-home
> volumes they are active, so DAFS does _not_ help here at all. That's
> why we take the risk of some minor corruptions rather than knowing
> that some hundred people cannot work at all (or simulation-jobs crash
> or what not).

I am not the one you need to convince about this line of thinking, but
I'm not sure I agree about DAFS not being a help here at all. When you
say the volumes are "active", how active is "active"? DAFS gives you
some configuration knobs which allow you to specify how often inactive
volumes become detached. The default is 2 hours, but you could possibly
set it much lower than that.

That has other performance implications and I'm not necessarily
recommending doing that, but it's something to think about.

> Thus, there is no real alternative to us there yet.  Unless DAFS
> salvages some hundred volumes in parallel rather than one after the
> other.  Does it do that ?  This might alleviate the problem.

You can salvage multiple volumes on a partition at once, but not
hundreds, and not with the speed you want yet. There is some
mostly-complete code to allow a much higher number of salvages to occur
at once, but I'm pretty sure that will not be merged for the 1.6
release. I'm not sure about subsequent releases in the 1.6 series, but
it will probably be in at least 1.10 the way things are going now. But
since 1.10 is the earliest possible time I've heard that fast-restart
will be removed, perhaps that's not so bad for you.

And as has been mentioned elsewhere in the thread, you need to wait for
the VG hierarchy summary scan to complete, no matter how fast salvaging
is or how many you do in parallel. That involves reading the headers of
all volumes on the partition, so it's not fast (but it is very fast if
you're comparing it to the recovery time of a 1.4 unclean shutdown)

-- 
Andrew Deason
adeason@sinenomine.net