Builders need to run a kernel >= 2.6.32

Jarod Wilson jarod at wilsonet.com
Tue Jul 27 20:48:04 CEST 2010


On Tue, Jul 27, 2010 at 1:48 AM, Jarod Wilson <jarod at wilsonet.com> wrote:
> On Tue, Jul 27, 2010 at 12:57 AM, Jarod Wilson <jarod at wilsonet.com> wrote:
> ...
>>>> I saw a similar glibc bomb running a yum remove while cleaning up some
>>>> detritus earlier tonight too. Looks like something bugly somewhere in
>>>> the rhel6 stack. Will look into it more tomorrow, its past my bed
>>>> time. :\
>>>
>>> Red Hat Bugzilla #608710 and friends. I've applied a patch attached to
>>> that bug (patches an rpm lib) that is thought to possibly fix the
>>> problem and installed the resulting build on my builder, and just
>>> threw a previously failed build at it... Quite a few rhel6 kvm guest
>>> installs are apparently hitting this one, so I expect if this patch
>>> isn't the fix, someone is going to get to the bottom of this
>>> post-haste. :) (Fingers crossed the one-line patch is indeed the fix
>>> though).
>>
>> Sigh. Never mind. Hadn't seen the latest few comments in the bug until
>> now. However, I have a new idea for something that may help (based on
>> some comments in bug 607650)... May well be an issue with transparent
>> hugepage support in kvm guests. I've turned that feature off now.
>
> Managed to make it all the way through populating two mock chroots
> simultaneously with transparent hugepage support disabled, pretty sure
> that's the answer, so please go ahead with any recent failed builds (I
> think all I saw was ffmpeg builds from Nicolas and
> xine-lib-extras-freeworld from Rex).

So apparently, the necessary fix is actually a host-side kernel
update, which is in the latest EL6 kernel build (2.6.32-52.el6) and
the latest F13 kernel in updates-testing. Not sure yet about Fedora
12, which is what the host here is running. Will get a patched kernel
on there shortly, one way or another.

(My build croaked with a "g++: Internal error: Killed (program
cc1plus)", which may or may not be ongoing random corruption that a
patched host kernel might fix).

-- 
Jarod Wilson
jarod at wilsonet.com


More information about the rpmfusion-developers mailing list