[Sugar-devel] Reducing Stick Failures - Was Re: [Marketing] press release opportunity...

[Martin Langhoff]

On Fri, Jul 31, 2009 at 6:34 AM, James Cameron<quozl at laptop.org> wrote:
> A general comment ... if any state is preserved by the children on the
> USB sticks, and there is no copy of the state kept elsewhere, and there
> is a possibility of power failure, premature removal, or other
> interruptions, then every software component that uses the saved state
> must be either capable of detecting corruption of the saved state, or
> graceful recovery from apparently invalid state.

This is a lot more basic.

The current SoaS uses

 - FAT16
 - An overlay FS in a file on FAT16

The most trivial "non-100% kosher" shutdown completely cooks the
overlay. One serious oddity is that the overlay is mounted by
something that depends on X or the gnome / sugar session, so an X
crash unmounts the overlay, somtimes corrupting it.

And more serious hangs or "oops, did I just pull the USB stick"
moments have a very high chance of cooking the FAT partition itself.

We have 2 brittle filesystems, one on top of eachother. The proposals
if using a "fully installed USB", probably using a journalled FS are
definitely a step in the good direction.

Once that is fixed, then it makes sense to look at the app layer that
you mention. But that's the same app layer we have on the XO, and it's
been handling unclean shutdowns pretty well for a while (on 0.82 at
least).

cheers,


m
-- 
 martin.langhoff at gmail.com
 martin at laptop.org -- School Server Architect
 - ask interesting questions
 - don't get distracted with shiny stuff  - working code first
 - http://wiki.laptop.org/go/User:Martinlanghoff