[ovs-dev] [PATCH v2 0/3] GitHub Actions integration and Travis CI update.

[Ilya Maximets]

On 11/26/20 9:31 AM, David Marchand wrote:
> On Wed, Nov 25, 2020 at 4:05 PM Ilya Maximets <i.maximets at ovn.org> wrote:
>>
>> On 11/25/20 3:57 PM, Ilya Maximets wrote:
>>> Version 2:
>>>   - Added fix for dpdk-dir cache.  We should not use 'native' machine.
> 
> Interesting.
> This explains a crash I got once on GHA dpdk side but did not
> reproduce: testpmd test failing with an Illegal instruction with a x86
> ccache in place...

I caught it only 2 times while actively developing and testing all day.

> dpdk-latest will need to force -Dmachine=default.
> 
> 
> I tested your series on dpdk-latest to check against 20.11-rc5, lgtm.
> For 20.11 support, we will need some small adjustments:
> - add -Dmachine=default in .ci/linux-build.sh,
> - install ninja, python3-setuptools packages
> - install the python3-wheel package, I suppose it is implicitly
> installed in Travis CI envs but not in GHA envs
> - with meson pip-installed only for user, PATH must be appended a
> $HOME/.local/bin

Thanks for the investigation and putting this together!
That is very helpful.

> 
> 
> As far as my own DPDK patch is concerned, your key generation is
> interesting in that a change in configuration triggers a cache
> invalidation.
> OTOH we have a ccache whose benefits could be seen with ovsrobot.
> In Travis CI, ovsrobot/dpdk has a weekly job to maintain a fresh
> ccache in its main branch.
> The "secrets" trick you mentioned offlist could be used for it for
> both use cases.

I hope, that we will not need to invalidate caches after all.
We'll enable the trick if we'll need to.  I still have a little bit
of faith in this cache action.

Here is the trick (for consistency of this mail thread):
https://github.com/actions/cache/issues/2#issuecomment-673493515

For the ccache, it might be beneficial for OVS too, so that is the
feature for ci we could consider later.

--
I found few issues with upload-artifact at v2 action.  It crashes trying
to upload socket files and also, for some reason, it's not able to
find testsuite files with wildcarded path.
I'll send v3 of this patch-set workarounding these issues soon.

Also, I prepared backports for all other branches and fixed specific
issues on them.  I'll send these patches today too.

Best regards, Ilya Maximets.