[ovs-dev] [ovs-dev, PATCHv8, 1/3] Improved Packet Drop Statistics in OVS

Aaron Conole aconole at redhat.com
Tue Feb 12 16:22:12 UTC 2019


Ilya Maximets <i.maximets at samsung.com> writes:

> On 12.02.2019 18:55, Aaron Conole wrote:
>> Ilya Maximets <i.maximets at samsung.com> writes:
>> 
>>> Aaron, ovsrobot doesn't check this patch for some reason.
>>> I suspect, it's because "PATCH" and "v8" are glued together.
>>> Could you, please, take a look.
>> 
>> It won't.  But not for that reason.  The patch info says that this is
>> patch 1 of 3.  But 2/3 and 3/3 haven't been posted.  So doing a pull
>> from patchwork states:
>> 
>>    "total":3,"received_total":1,"received_all":false
>> 
>> Because the patchwork thinks it hasn't received all the patches, it
>> defers trying to process the series.  In this case, the submitter may
>> have intended to post 2 more patches that never hit the list.  Maybe we
>> can revisit the way the robot works now that it creates a special branch
>> for each series that can be modified.
>
> Thanks for checking. Did you thought about checking patches in a series one
> by one ?

We do.  But we don't start processing the series until all the patches
are received.  The reason is for ordering.  It's possible that we get
1/3 and 3/3, and if we start processing we will apply 1/3, but then 3/3
won't apply.  There could be more logic added to address this, but then
we'll need to re-start the whole monitor mechanism and tell it to run
all over again - and what happens if the patches *never* arrive... the
system will be spinning in loops.

> I mean that we definitely want all the patches to not break the
> build. For this purpose ovsrobot needs to apply patches one by one and
> check.

For each patch, we only check that the build doesn't break.  The full
make-check does take a long time to run, so I pushed it until the end of
the series.

> This way, I guess, there could be possible to avoid this kind of
> misunderstanding with subject prefixes as patch 1/3 should be checked first
> anyway. OTOH, this will increase testing time significantly.

See above for why it's more complex to do partial series.  Maybe there's
a better design?  Pull requests to https://github.com/orgcandman/pw-ci
are always welcome :)

>> 
>> Perhaps this should have been posted without the '1/3' series counter?
>
> Yes. I guess, Anju wanted this patch to be treated as a single patch.

I can manually kick it off if one is ever missing.  OTOH, since you've
already noted lots of problems, maybe it's best that Anju submit a v9
with the correct metadata. :)


More information about the dev mailing list