regular fsck runs are too disturbing

[Bryan Haskins]

I completely like the LVM idea, as I was saying on IRC a bit ago, that 
would really be an elegant system. LVM up root, and whatever other 
chosen disks, and safely check that in the background (possibly a nice 
notification icon even?) and pop up a ping box when an error is found 
(the level of error it goes into rigorous fsck mode being user 
configurable, but shipping with a default of some sort, tbd later) this 
would REALLY cut down on issues...

The only discrepancy here is what happens when the disk is corrupted to 
a high degree and we try to boot it? Fairly simple yet also complex 
response to that. It would have to work similar to bulletproof X... 
though obviously on a lower level. We could flag to a safe location to 
fsck on boot. Or even have a special grub entry that fscks 
automatically, that would be interesting.

The first being more elegant, though rather hard... it would require us 
to have a "safe-zone" to store this sort of small information. And we 
have no idea what part of the FS/Disk could be bad.

Possibly a combination of the two might be in order.

Honestly it is a tad complex but it is REALLY a cool idea.

We should write up a formal spec and see where it goes, still needs some 
development, but it's really promising In my opinion.

John Dong wrote:
> A partial check doesn't make sense with the current fsck tools AFAIK. We
> should do a full filesystem check if anything, and if a user decides to abort
> it, it's his choice.
>
> There should be a graphical or otherwise easily accessible way of re-touching
> the /forcefsck flag so that users can choose which bootup to do a check on.
> Another "idea" is on LVM-capable systems to take a snapshot of important
> filesystems while they are unmounted or read-only then fsck the snapshot
> device as a background task. If any serious errors are detected in the
> snapshot, then schedule an uncancelable boot scan.
>
> I agree with everyone who says that the current fsck experience is a blemish
> to Ubuntu's general user-friendliness, and also that we should not be entirely
> removing the regular fsck as it catches hardware irregularities and potential
> software bugs with ext3.
>
>
> John
>
> On Wed, Oct 10, 2007 at 01:25:54PM +0200, Vincenzo Ciancia wrote:
>   
>> Problem is that users will just skip the test, and get tired of "having"
>> to skip the test each time. Perhaps an alternative would be to check
>> only a part of the filesystem (e.g. randomly choosen) each time, but I
>> don't know enough about filesystem (even though I should :) ) to say
>> it's impossible or feasible.
>>
>> Vincenzo
>>