For review: NAND out of space patch.

[Erik Garrison]

On Tue, Jul 22, 2008 at 12:53:37PM -0300, John Watlington wrote:
> 
> On Jul 22, 2008, at 12:06 PM, Chris Ball wrote:
> 
> > Hi,
> >
> >> Can you walk me through the exact steps that the user would
> >> experience if this script was installed?
> >
> > They wouldn't see anything different, but Journal entries  
> > corresponding
> > to files we chose to delete wouldn't resume properly.
> >
> >> In terms of which files, I think the oldest (or maybe LRU as they
> >> say in caches) would be better than the largest. Can we do that
> >> (e.g. delete oldest then iterate until x MBs is free)?
> >
> > I disagree; I don't think we're filling up with small Write or Paint
> > documents, my intuition is that we're filling up with recent large
> > downloads and movies.  In the case where the problem is a huge  
> > download
> > the user just made, your scheme results in deleting *everything*.
> >
> > Since we disagree, maybe best to wait until we have some disk-full
> > images back from the field so that we can see what used up all the
> > space, before deciding the algorithm.
> 
> I'm getting three images right now.
> 
> One of the machines booted, but wouldn't allow any activities to launch
> (which since you can't log in on vttys kinda locks down the machine).
> But I did notice a large number of non-standard activities (e.g. Doom).

This sounds familiar.  I think several teachers from Uruguay have
mentioned on the Sur list that their students love to download software
and have filled up their storage space.  I'll try to find the reference.
I have also heard the same from a contractor in Uruguay who has been
involved in distribution (via #olpc-ayuda).

Today I am going to test a solution in which we union-mount a tmpfs over
top of a full root filesystem (which is effectively read-only).  This
should allow us to boot, but obviously any changes made to the tmpfs
during the session will be lost.  Provided we can boot in this scheme,
we should immediately open a dialogue which asks the user to select
Activities to delete.

I think that such a 'recovery-mode' is ultimately the best we're going
to do to help resolve this issue.  We must provide students a way to
manage their systems, and to do so even in a NAND-full state, or the
solution to NAND-full will continue to be centralized and costly.  If it
is not something that we ship immediately to help resolve the issue in
Uruguay, the current situation demonstrates that it is a worthwhile
target for future releasese.

Erik