[OE-core] [PATCH 1/3] dbus-test: Fix QA host-contamination errors

Richard Purdie richard.purdie at linuxfoundation.org
Mon Mar 2 21:43:56 UTC 2020


On Mon, 2020-03-02 at 13:33 -0800, Khem Raj wrote:
> 
> On 3/2/20 1:08 PM, Richard Purdie wrote:
> > On Mon, 2020-03-02 at 10:49 -0800, Khem Raj wrote:
> > > On 3/1/20 9:05 AM, Richard Purdie wrote:
> > > > On Sun, 2020-03-01 at 08:20 +0000, Richard Purdie wrote:
> > > > > On Sun, 2020-03-01 at 00:17 -0800, Khem Raj wrote:
> > > > > I understand the need for the fixes, I'm just very concerned
> > > > > we
> > > > > have
> > > > > what amounts to undetected non-determinism in the build :(
> > > > > 
> > > > > I'm more concerned about fixing that (and ensuring we can
> > > > > detect/fix
> > > > > all cases) than I am about the individual errors.
> > > > 
> > > > I did a bit more thinking/checking on this.
> > > > 
> > > > An interesting command to experiment with is:
> > > > 
> > > > $ touch /tmp/test; ls -la /tmp/test; ./tmp/sysroots-
> > > > components/x86_64/pseudo-native/usr/bin/pseudo sh -c "ls -la
> > > > /tmp/test*; cp /tmp/test /tmp/test2; ls -la /tmp/test*; rm
> > > > /tmp/test*"
> > > > 
> > > > which for me shows:
> > > > 
> > > > -rw-rw-r-- 1 richard richard 0 Mar  1 17:03 /tmp/test
> > > > Warning: PSEUDO_PREFIX unset, defaulting to XXX./tmp/sysroots-
> > > > components/x86_64/pseudo-native/usr.
> > > > -rw-rw-r-- 1 1000 1000 0 Mar  1 17:03 /tmp/test
> > > > -rw-rw-r-- 1 1000 1000 0 Mar  1 17:03 /tmp/test
> > > > -rw-rw-r-- 1    0    0 0 Mar  1 17:03 /tmp/test2
> > > > 
> > > > Can you see if that is different on your two machines?
> > > 
> > > above cmd output is exactly same as yours.
> > > 
> > > -rw-r--r-- 1 build build 0 Mar  2 18:49 /tmp/test
> > > Warning: PSEUDO_PREFIX unset, defaulting to
> > > /mnt/b/yoe/build/./tmp/sysroots-components/x86_64/pseudo-
> > > native/usr.
> > > -rw-r--r-- 1 1000 1000 0 Mar  2 18:49 /tmp/test
> > > -rw-r--r-- 1 1000 1000 0 Mar  2 18:49 /tmp/test
> > > -rw-r--r-- 1    0    0 0 Mar  2 18:49 /tmp/test2
> > 
> > Hmm, this means the test is flawed as we need to find out how
> > /tmp/test2 becomes owned by 1000.1000.
> > 
> > Any ideas how we can simplify this down to reproduce that?
> > 
> 
> Even on baremetal ubuntu 18.04 I am seeing
> 
> dlm-4.0.9: dlm: /usr/lib/libdlmcontrol.so is owned by uid 3004, which is 
> the same as the user running bitbake. This may be due to host contamination
> dlm: /usr/lib/libdlm_lt.so is owned by uid 3004, which is the same as 
> the user running bitbake. This may be due to host contamination
> dlm: /usr/lib/libdlm.so is owned by uid 3004, which is the same as the 
> user running bitbake. This may be due to host contamination 
> [host-user-contaminated]
> dlm-4.0.9: dlm: /usr/lib/libdlmcontrol.so.3 is owned by uid 3004, which 
> is the same as the user running bitbake. This may be due to host 
> contamination
> dlm: /usr/lib/libdlm_lt.so.3 is owned by uid 3004, which is the same as 
> the user running bitbake. This may be due to host contamination
> dlm: /usr/lib/libdlm.so.3 is owned by uid 3004, which is the same as the 
> user running bitbake. This may be due to host contamination 
> [host-user-contaminated]
> 
> this might be again a dlm issue to fix.  but perhaps a good one to
> try  if AB system can build it without these QA diagnostics.
> 
> dlm is in meta-networking.

That could well be a dlm recipe bug. I'm more interested in the ones
which reproducibly break on one machine but work on another.

> > I did wonder if its coreutils-native was somehow creeping into
> > DEPENDS
> > and had a different config between your two hosts which was causing
> > different behaviour but I'd need to check into whether that
> > happens.
> > 
> > Any other ideas why the behaviour difference or how to reproduce
> > it?
> > 
> 
> can you promote host-user-contaminated form QA warning to QA error
> in poky and see if builds fail ?

As its already a warning, we'd see it on the autobuilder. *Any* warning
turns the build orange rather than green. We don't any warnings in core
builds, we fixed them all. This is why I know its not happening on the
autobuilder...

Cheers,

Richard



More information about the Openembedded-core mailing list