[opensuse-kernel] Kernel:HEAD armv7l kernel-lpae OOM
Hi, kernel-lpae 4.11.rc7 failed to build for armv7hl: [33246s] calling /usr/lib/rpm/brp-suse.d/brp-99-pesign [33246s] Using signing certificate /home/abuild/rpmbuild/SOURCES/_projectcert.crt [33247s] Creating /home/abuild/rpmbuild/OTHER/kernel-lpae.cpio.rsasign [33282s] [33220.985937] Out of memory: Kill process 1408 (rpmbuild) score 2 or sacrifice child [33282s] [33220.988302] Killed process 25612 (bash) total-vm:3516kB, anon-rss:160kB, file-rss:1980kB, shmem-rss:0kB [33282s] error: Bad exit status from /var/tmp/rpm-tmp.2fR0OE (%install) [33282s] [33282s] [33282s] RPM build errors: [33282s] Bad exit status from /var/tmp/rpm-tmp.2fR0OE (%install) [33285s] [33285s] armbuild17 failed "build kernel-lpae.spec" at Wed Apr 19 15:28:44 UTC 2017. Do we need to tweak rpm/constaints.in? If I'm reading correctly, it has a constraint of 2 GB RAM for binary packages, with no override for ARM. Regards, Andreas -- SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany GF: Felix Imendörffer, Jane Smithard, Graham Norton HRB 21284 (AG Nürnberg) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
Dne 19.4.2017 v 21:14 Andreas Färber napsal(a):
Hi,
kernel-lpae 4.11.rc7 failed to build for armv7hl:
[33246s] calling /usr/lib/rpm/brp-suse.d/brp-99-pesign [33246s] Using signing certificate /home/abuild/rpmbuild/SOURCES/_projectcert.crt [33247s] Creating /home/abuild/rpmbuild/OTHER/kernel-lpae.cpio.rsasign [33282s] [33220.985937] Out of memory: Kill process 1408 (rpmbuild) score 2 or sacrifice child [33282s] [33220.988302] Killed process 25612 (bash) total-vm:3516kB, anon-rss:160kB, file-rss:1980kB, shmem-rss:0kB [33282s] error: Bad exit status from /var/tmp/rpm-tmp.2fR0OE (%install) [33282s] [33282s] [33282s] RPM build errors: [33282s] Bad exit status from /var/tmp/rpm-tmp.2fR0OE (%install) [33285s] [33285s] armbuild17 failed "build kernel-lpae.spec" at Wed Apr 19 15:28:44 UTC 2017.
Do we need to tweak rpm/constaints.in? If I'm reading correctly, it has a constraint of 2 GB RAM for binary packages, with no override for ARM.
I agree that 2GB is too little, but the scheduler does not fulfill this request even: $ osc rbl --last Kernel:HEAD kernel-default standard x86_64 | grep /qemu- | grep -o ' -m [0-9]*' -m 1500 -m 1500 $ osc rbl --last Kernel:HEAD kernel-lpae ARM armv7l | grep /qemu- | grep -o ' -m [0-9]*' -m 1020 Michal -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
HI,
[33246s] calling /usr/lib/rpm/brp-suse.d/brp-99-pesign [33246s] Using signing certificate /home/abuild/rpmbuild/SOURCES/_projectcert.crt
we're using pesign on armv7 nowadays? why?
Do we need to tweak rpm/constaints.in? If I'm reading correctly, it has a constraint of 2 GB RAM for binary packages, with no override for ARM. I agree that 2GB is too little
what changed? why was it not a problem in the past?
, but the scheduler does not fulfill this request even:
$ osc rbl --last Kernel:HEAD kernel-default standard x86_64 | grep /qemu- | grep -o ' -m [0-9]*' -m 1500 -m 1500 $ osc rbl --last Kernel:HEAD kernel-lpae ARM armv7l | grep /qemu- | grep -o ' -m [0-9]*' -m 1020
we do have a few workers with 3 and 4 GB of RAM for armv7, but they're not many and always busy (we have a ton of forks of libreoffice and ceph that are basically building all the time on those).. I would prefer to be able to run the jobs with less memory, is there a way to achieve that? if it is really pesign, why are we using that on armv7 and why is it eating that much memory? Thanks, Dirk -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
Dne 19.4.2017 v 23:08 Dirk Müller napsal(a):
HI,
[33246s] calling /usr/lib/rpm/brp-suse.d/brp-99-pesign [33246s] Using signing certificate /home/abuild/rpmbuild/SOURCES/_projectcert.crt
we're using pesign on armv7 nowadays? why?
God question.
Do we need to tweak rpm/constaints.in? If I'm reading correctly, it has a constraint of 2 GB RAM for binary packages, with no override for ARM. I agree that 2GB is too little
what changed? why was it not a problem in the past?
For a single-cpu build on armv7 it's probably OK, if we get rid of pesig. I was surprised to see the x86_64 build using 1.5GB. But then the scheduler ignores the <processors> constraint as well and uses a worker with 4 CPUs, so we fit into the RAM fine...
we do have a few workers with 3 and 4 GB of RAM for armv7, but they're not many and always busy (we have a ton of forks of
So is the _constraints file actually used on build.opensuse.org?
libreoffice and ceph that are basically building all the time on those).. I would prefer to be able to run the jobs with less memory, is there a way to achieve that? if it is really pesign, why are we using that on armv7 and why is it eating that much memory?
Yeah, this particular failure is due to pesign, which should only run if CONFIG_MODULE_SIG=y. Michal -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
we're using pesign on armv7 nowadays? why? God question.
hmm, ok, so it is worth investigating that part then if there is no good reason for that.
we do have a few workers with 3 and 4 GB of RAM for armv7, but they're not many and always busy (we have a ton of forks of So is the _constraints file actually used on build.opensuse.org?
yeah, but your constraint specifies <memory> (which is physicalmemory+swap). The workers with 1GB of physical RAM still have 2 (4?) G of swap, so they are elligible choices. the oom killer might kill this sooner though. is this maybe another OOM regression? the other variable that could have changed is kernel-obs-build's oom behavior. Thanks, Dirk-- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
Dne 19.4.2017 v 23:56 Dirk Müller napsal(a):
we're using pesign on armv7 nowadays? why? God question.
hmm, ok, so it is worth investigating that part then if there is no good reason for that.
I fixed it now as bsc#1035053.
we do have a few workers with 3 and 4 GB of RAM for armv7, but they're not many and always busy (we have a ton of forks of So is the _constraints file actually used on build.opensuse.org?
yeah, but your constraint specifies <memory> (which is physicalmemory+swap). The workers with 1GB of physical RAM still have 2 (4?) G of swap, so they are elligible choices. the oom killer might kill this sooner though.
OK.
is this maybe another OOM regression? the other variable that could have changed is kernel-obs-build's oom behavior.
The only mm changes after v4.12-rc6 were some THP fixes, so that's rather unlikely. It's possible that we were just unlucky this time. I triggered a rebuild of Kernel:HEAD/kernel-lpae with the original spec file, let's see if it fails again or not. Michal -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
Am 20.04.2017 um 00:21 schrieb Michal Marek:
Dne 19.4.2017 v 23:56 Dirk Müller napsal(a):
we're using pesign on armv7 nowadays? why? God question.
hmm, ok, so it is worth investigating that part then if there is no good reason for that.
I fixed it now as bsc#1035053.
To me it looks as if aarch64 is still doing some weird signing despite the above fix being merged. My kernel-default package has been waiting
10 hours today to get scheduled the second magic time...
aarch64 must be running into the %if %CONFIG_EFI_STUB == "y" path. Regards, Andreas http://kernel.opensuse.org/cgit/kernel-source/commit/?id=50db9ef9aeb9e47e03f... -- SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany GF: Felix Imendörffer, Jane Smithard, Graham Norton HRB 21284 (AG Nürnberg) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
On Mai 24 2017, Andreas Färber <afaerber@suse.de> wrote:
To me it looks as if aarch64 is still doing some weird signing despite the above fix being merged. My kernel-default package has been waiting
10 hours today to get scheduled the second magic time...
kernel-source has only been updated 1.5 hours ago.... Andreas. -- Andreas Schwab, SUSE Labs, schwab@suse.de GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE 1748 E4D4 88E3 0EEA B9D7 "And now for something completely different." -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
Am 24.05.2017 um 18:24 schrieb Andreas Schwab:
On Mai 24 2017, Andreas Färber <afaerber@suse.de> wrote:
To me it looks as if aarch64 is still doing some weird signing despite the above fix being merged. My kernel-default package has been waiting
10 hours today to get scheduled the second magic time...
kernel-source has only been updated 1.5 hours ago....
I used scripts/osc_wrapper upload, as suggested by Michal. The second build printed pesign stuff, it did not do a full multi-hour rebuild of the sources. Cheers, Andreas -- SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany GF: Felix Imendörffer, Jane Smithard, Graham Norton HRB 21284 (AG Nürnberg) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
Dne 24.5.2017 v 17:57 Andreas Färber napsal(a):
To me it looks as if aarch64 is still doing some weird signing despite the above fix being merged. My kernel-default package has been waiting
10 hours today to get scheduled the second magic time...
aarch64 must be running into the %if %CONFIG_EFI_STUB == "y" path.
Right, because it has CONFIG_EFI_STUB=y and we sign the kernel image if it's an EFI binary. I can of course change the condition to depend on CONFIG_EFI_STUB=y && CONFIG_MODULE_SIG=y, but to be hones, I do not know what the requirements are in TW with respect to secure boot. Michal -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
Hi Dirk, Am 19.04.2017 um 23:08 schrieb Dirk Müller:
Do we need to tweak rpm/constaints.in? If I'm reading correctly, it has a constraint of 2 GB RAM for binary packages, with no override for ARM. I agree that 2GB is too little
what changed? why was it not a problem in the past?
The last successful build was for -rc6 (2nd revision, with serdev), so the tarball changed but no armv7hl Kconfig options for -rc7: http://kernel.opensuse.org/cgit/kernel-source/commit/?id=6e80a142558610d3515... In general of course the number of modules (and of built-in options) grows over time. Regards, Andreas -- SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany GF: Felix Imendörffer, Jane Smithard, Graham Norton HRB 21284 (AG Nürnberg) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
participants (4)
-
Andreas Färber -
Andreas Schwab -
Dirk Müller -
Michal Marek