[Sugar-devel] GPA School Visit report for Tuesday July 14

[Tomeu Vizoso]

On Wed, Jul 15, 2009 at 04:14, Martin Langhoff<martin.langhoff at gmail.com> wrote:
> On Wed, Jul 15, 2009 at 8:35 AM, Greg Smith<gregsmithpm at gmail.com> wrote:
>> As I got there I noticed a kid had unplugged their USB before full
>> shut down and they were frantically trying to plug it back in while
>> errors scrolled on the screen. Still not sure if they lost data (will
>> check tomorrow on kid who made the same mistake last week). Is there
>> anything we can do to ensure that all data is written/saved ASAP on
>> clicking Shut down from the menu?
>
> This is a very valid concern, and very tricky. I've seen lots of
> problems with LiveUSB sticks and unclean shutdown.
>
> The short of it is that the scheme behind "LiveUSB", specifically the
> "persistent storage" is very brittle, to the point that it can hose
> the entire disk badly. The USB stick will only be usable again after
> re-formatting and re-installing the SoaS software (losing user data,
> etc).
>
> It is a bad combination of
>
>  - Using "rw" overlay mountpoints that only sync every 5s. If the
> overlay partition is corrupted, it will prevent the SoaS from booting.
>
>  - Writing on a FAT partition which is notoriously fragile... and
> mounted async. Mounting it sync slows things tremendously and wears
> the USB stick out much faster.
>
> IIRC, recent kernels have a mount flag for vfat (and maybe other FSs?)
> that makes the kernel more eager to flush things to disk, so you can
> keep the mountpoint async but reduce the 5s window significantly.
>
> Unfortunately, I can't find it now. It was well reported on LWN and
> kernelnewbies a few months ago.

I agree with the conclusion that Linux support for live usb sticks
still has a long way to go. See
https://bugzilla.redhat.com/show_bug.cgi?id=465725 for more details
about the overlay issue, though note that in Fedora the whole stick is
not thrashed, you can discard any changed you did to / by passing the
kernel the option reset_overlay.

If deployers of SoaS keep in mind this limitation and users mostly
write inside /home, this limitation is much less important.

This really has very little to do with Sugar itself, like hardware
support. I think we need to find the resources to make the needed work
in the kernel and in the distro level before we can do anything in
SoaS itself.

Regards,

Tomeu

> 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
> _______________________________________________
> Sugar-devel mailing list
> Sugar-devel at lists.sugarlabs.org
> http://lists.sugarlabs.org/listinfo/sugar-devel
>