Re: [PULL 00/11] Optional gitlab-CI and doc fixes for -rc4

[Alexander Bulekov]

On 210816 1211, Daniel P. Berrangé wrote:
> On Mon, Aug 16, 2021 at 06:22:46AM -0400, Alexander Bulekov wrote:
> > On 210816 1001, Peter Maydell wrote:
> > > On Sat, 14 Aug 2021 at 07:10, Thomas Huth <thuth@redhat.com> wrote:
> > > >
> > > >  Hi Peter,
> > > >
> > > > in case we're going to have an -rc4, here's a pull request that contains
> > > > the fixes for getting the gitlab-CI green again. I also added some doc
> > > > updates since they should be completely riskless. But if we won't have 
> > > > an
> > > > rc4 due to other reasons, this pull request here certainly also does not
> > > > justify another RC, so please ignore this PR in that case.
> > > >
> > > > The following changes since commit 
> > > > 703e8cd6189cf699c8d5c094bc68b5f3afa6ad71:
> > > >
> > > >   Update version for v6.1.0-rc3 release (2021-08-10 19:08:09 +0100)
> > > >
> > > > are available in the Git repository at:
> > > >
> > > >   https://gitlab.com/thuth/qemu.git tags/pull-request-2021-08-11
> > > >
> > > > for you to fetch changes up to 36b508993c4dcc6b3ef4b5c00e293ee9560926ee:
> > > >
> > > >   docs/about/removed-features: Document removed machines from older 
> > > > QEMU versions (2021-08-11 15:39:09 +0200)
> > > >
> > > > CI run can be seen here:
> > > >
> > > >  https://gitlab.com/thuth/qemu/-/pipelines/351602605
> > > >
> > > > ----------------------------------------------------------------
> > > > * Fixes for the gitlab-CI (fix the hanging  build-oss-fuzz pipeline)
> > > > * Add documentation about features that have been removed in older 
> > > > versions
> > > >
> > > 
> > > 
> > > Applied, thanks.
> > > 
> > > Please update the changelog at https://wiki.qemu.org/ChangeLog/6.1
> > > for any user-visible changes.
> > 
> > https://gitlab.com/qemu-project/qemu/-/jobs/1505950978
> > Looks like build-oss-fuzz is still timing out, even without the issue
> > in the vhost-usr-blk test. At this point the job should essentially just
> > build + test qemu-system-i386 with some extra time spent on linking
> > the fuzzer and briefly running through all the fuzzer configs. Maybe the
> > only way to make this work is to split the job into a build + test
> > stage?
> 
> At this point I think we should just disable the job in gitlab entirely.
> We've spent too long debugging this, while leaving CI red for everyone.
> 
> Whomever is interested in this can then work to find a way to make it
> reliable and request it be re-enabled once confident that it will work.
> 

On my mirror the job succeeded in 41 minutes... I guess you have to get
lucky with scheduling/ambient load.
https://gitlab.com/a1xndr/qemu/-/jobs/1506197531

If all we want is to check that oss-fuzz builds work while keeping green
CI, the only thing that needs to be done is to remove the "make check
check-qtest-i386 check-unit" from the job.

The problem is that QEMU has a ton of nice tests, and by simply building
with --enable-sanitizers, these tests become even more useful because we
can catch all sorts of memory-corruptions. In fact, some of the
regression tests require --enable-sanitizers to catch issues.  When I
looked at this originally, build-oss-fuzz was the only job that was
running with any sanitizers. Now, it looks like at least
ubuntu-18.04-s390x-clang and ubuntu-20.04-aarch64-clang also build with
ASAN. Is there some other i386 build (gcc or clang) where we could add
--enable-sanitizers?

-Alex