bug#41710: Possibly found a bug while doing a "guix pull"

["Ludovic Courtès" <ludo@gnu.org>]

Hello,

Danny Milosavljevic <dannym@scratchpost.org> skribis:

> On Mon, 8 Jun 2020 01:15:46 +0200
> Léon Lain Delysid <leon.lain.delysid@gmail.com> wrote:
>
>>or
>> if my system crashing twice during a pull command somehow broke it, 
>
> Probably.
>
>>but I
>> hope this feedback helped. 
>
> It sure helped.  It's good to know that that can happen.
>
> I remember the first time I used Guix, I picked some file system that would
> keep doing that: leave empty files if the system crashed (among lots of other
> things).  And that system crashed a lot.  I had the same result as you,
> and a lot of additional problems.
>
> Back then we already improved a lot of places that were really really
> important (added fsync calls), so the remaining places should be quite
> harmless--like this one.  Because of Guix, you can always rebuild
> /gnu/store just as it was--after a long build time maybe, but it's possible
> (could be made a LOT more usable, though).
>
> (fsync degrades performance, so it makes no sense to fsync for /gnu/store)
>
> I think we can't really do more without imposing undue mainentance burden on
> us (for something the file system shouldn't be doing in the first place),
> or we could recommend another file system or different file system options
> in the manual.  What would the latter be?
>
> Also, how it the world didn't the file system checker fsck
>
> (1) automatically run and
> (2) fix this
>
> in your case?

Yeah, that’s really weird.  I never experienced it first-hand, but it’s
not the first time we have such a report.

Ext4 & co. reportedly can leave empty files upon crashes; perhaps that’s
a problem with those file systems (though I’ve always used ext2/3/4 and
never had this problem myself, but that’s not statically significant).

Anyway, closing.

Thank you, Léon!

Ludo’.