[csw-maintainers] there is something rotten around checkpkg

[Peter FELECAN]

"Maciej (Matchek) Bliziński" <maciej at opencsw.org> writes:

> No dia 16/01/2013 19:02, "Peter FELECAN" <pfelecan at opencsw.org> escreveu:
>>
>> "Maciej (Matchek) Bliziński" <maciej at opencsw.org> writes:
>>
>> > 2013/1/16 Dagobert Michelsen <dam at opencsw.org>:
>> >> The new link symbol inspection is in place for some time now, maybe it
> has
>> >> some serious drawbacks on database access?
>> >
>> > Some packages produce metadata objects serializing to more than 32MB.
>> > The out of space error could be a memory leak which I have been
>> > fighting over the weekend, but did not manage to find the cause. In my
>> > case, I was trying to run a full catalog import.
>> >
>> > Peter, what operation is causing the out of space error? How much RAM
>> > does the machine have?
>>
>> The "out of space" issue happened when packaging the *huge* TeXLive
>> hundred packages on the Baltic Online build farm; I think that you know
>> better than me how much RAM unstable10s and its companions have. I'm
>> trying again but one cycle merge-package is long: I start the process in
>> the morning and have the results the other day. You can see the log of
>> the action in progress in ~pfelecan/logs/texlive
>
> Maybe Yann can look at the logs and figure out what was the immediate
> problem. The main issue is that we art collecting way more date than
> previously, and that puts more pressure on the database. Do you think they
> the problem you had I'd transient? If it's persistent, we can back out the
> changes and stop collecting the additional data. I would prefer to
> gradually solve the performance problems without having to roll back, but
> if people can't build packages, we will have to.

Th issue is not transient as I repeated 3 times the operation. However,
having a more important disk quota will delay it.

> I did rebuild GCC with the new code and GCC is a pretty hefty build. But
> maybe TeX live is so large that it pushes the system off the cliff, while
> GCC is still small enough.

The difference is in the number of files and, among them, the number of
binary executable files which are all candidates for the new checks.

> Yann, maybe we can selectively disable the ldd and elfdump data collection
> for packages that wish to have these checks disabled? The checking code
> would need to allow for missing ldd and elfdump data. Since all that
> checkpkg sees is the package file, there would have to be a specific bit of
> information in the package itself that would instruct checkpkg to skip ldd
> end elfdump. It could be in pkginfo for example.

This is an interesting option but I suggest to make it temporary, i.e.,
until solving the issue. How do you think to set this option, obviously
in the recipe.
-- 
Peter