Resources limits

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Resources limits

Lev Lamberov-3
Hi,

recently I've discussed with upstream the build problems regarding the
swi-prolog package on arm{el,hf} and mipsel [0], which are also
highlighted in #887155 [1].

We're wondering whether or not some of the tests are simply too
demanding for the build machines that are probably configured with low
resource limits. Do we have the ulimit settings (ulimit -a) of the build
machines? This time it were henze both for arm{el,hf} and mipsel-aql-01
for mipsel.

On amd64 using /usr/bin/time -v, the max RSS is 170Mb. There will also
be a quite high amount of reserved virtual memory, notably for the C
stacks of the threads. We do not know the max number of threads, but
assume about a dozen.

Maybe it's better to blacklist swi-prolog for the weaker buildd
candidates?

Regards,
Lev Lamberov

P. S. Please, CC me since I'm not subscribed to debian-{arm,mips}.

[0] https://buildd.debian.org/status/package.php?p=swi-prolog

[1] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=887155

Reply | Threaded
Open this post in threaded view
|

Re: Resources limits

Adrian Bunk-3
On Tue, Jan 16, 2018 at 12:38:25AM +0500, Lev Lamberov wrote:

> Hi,
>
> recently I've discussed with upstream the build problems regarding the
> swi-prolog package on arm{el,hf} and mipsel [0], which are also
> highlighted in #887155 [1].
>
> We're wondering whether or not some of the tests are simply too
> demanding for the build machines that are probably configured with low
> resource limits. Do we have the ulimit settings (ulimit -a) of the build
> machines? This time it were henze both for arm{el,hf} and mipsel-aql-01
> for mipsel.

You've already added that to debian/rules, so that's in the build logs.

> On amd64 using /usr/bin/time -v, the max RSS is 170Mb. There will also
> be a quite high amount of reserved virtual memory, notably for the C
> stacks of the threads. We do not know the max number of threads, but
> assume about a dozen.
>
> Maybe it's better to blacklist swi-prolog for the weaker buildd
> candidates?

The weakest buildds have 4 GB RAM.

Some mips* buildds don't have an FPU, but all arm* ones have.

On armel/armhf all 6 buildds are exactly the same hardware.

A hard limit are 2 GB userspace address space on mips/mipsel
and 3 GB userspace address space on armel/armhf.

The build failed on armel/armhf but passed on mips,
that makes it less likely that it is just running
out of address space.

What strikes me is that 7.6.3+dfsg-1 built everywhere,[1]
but 4 weeks later 7.6.4+dfsg-1 had problems.

Porterbox abel is the same hardware as the armel/armhf buildds.

In the same armel unstable chroot on abel, 7.6.4+dfsg-1 did FTBFS
and 7.6.3+dfsg-1 did build successfully for me.

I suspect there is some (potentially bogus) change on 7.6.4
that triggers the problem, and it might be possible to bisect.

> Regards,
> Lev Lamberov
>...

cu
Adrian

[1] except s390x, but that failure looks unrelated

--

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed