Re: [PATCH] Revert "tests/migration: Reduce autoconverge initial bandwidth"

[Thomas Huth <thuth@redhat.com>]

On 24/06/2020 18.26, Philippe Mathieu-Daudé wrote:
> Hi Thomas,
> 
> On 6/24/20 12:21 PM, Philippe Mathieu-Daudé wrote:
>> On 6/24/20 7:04 AM, Thomas Huth wrote:
>>> On 23/06/2020 19.35, Philippe Mathieu-Daudé wrote:
>>>> On 6/23/20 7:07 PM, Thomas Huth wrote:
>>>>> On 23/06/2020 17.39, Philippe Mathieu-Daudé wrote:
>>>>>> On 6/23/20 4:56 PM, Michael S. Tsirkin wrote:
>>>>>>> This reverts commit 6d1da867e65f ("tests/migration: Reduce autoconverge initial bandwidth")
>>>>>>> since that change makes unit tests much slower for all developers, while it's not
>>>>>>> a robust way to fix migration tests. Migration tests need to find
>>>>>>> a more robust way to discover a reasonable bandwidth without slowing
>>>>>>> things down for everyone.
>>>>>>
>>>>>> Please also mention we can do this since 1de8e4c4dcf which allow
>>>>>> marked the s390x job as "unstable" and allow it to fail.
>>>>>>
>>>>>> But if nobody is going to look at it, instead lets disable
>>>>>> it until someone figure out the issue:
>>>>>>
>>>>>> -- >8 --
>>>>>> diff --git a/.travis.yml b/.travis.yml
>>>>>> index 74158f741b..364e67b14b 100644
>>>>>> --- a/.travis.yml
>>>>>> +++ b/.travis.yml
>>>>>> @@ -507,6 +507,7 @@ jobs:
>>>>>>
>>>>>>       - name: "[s390x] Clang (disable-tcg)"
>>>>>>         arch: s390x
>>>>>> +      if: false # Temporarily disabled due to issue testing migration
>>>>>> (see commit 6d1da867e65).
>>>>>>         dist: bionic
>>>>>>         compiler: clang
>>>>>>         addons:
>>>>>
>>>>> Sorry, but that looks wrong. First, the disable-tcg test does not run
>>>>> the qtests at all. So this is certainly the wrong location here.
>>>>
>>>> Indeed, this is the previous job:
>>>>
>>>> -- >8 --
>>>> diff --git a/.travis.yml b/.travis.yml
>>>> index 74158f741b..b399e20078 100644
>>>> --- a/.travis.yml
>>>> +++ b/.travis.yml
>>>> @@ -464,6 +464,7 @@ jobs:
>>>>           - CONFIG="--disable-containers
>>>> --target-list=ppc64-softmmu,ppc64le-linux-user"
>>>>
>>>>       - name: "[s390x] GCC check-tcg"
>>>> +      if: false # Temporarily disabled due to issue testing migration
>>>> (see commit 6d1da867e65).
>>>>         arch: s390x
>>>>         dist: bionic
>>>>         addons:
>>>> ---
>>>>
>>>>> Second,
>>>>> if just one of the qtests is failing, please only disable that single
>>>>> failing qtest and not the whole test pipeline.
>>>>
>>>> Last time we talked about this Dave was against that option:
>>>>
>>>> https://www.mail-archive.com/qemu-devel@nongnu.org/msg690085.html
>>>>
>>>
>>> Was he? Citing his reply to the mail from your URL:
>>>
>>>   "Before we take the hammer to it, could you try reducing it's initial
>>> bandwidth"
>>>
>>> So all I can see is that he first wanted to try something different than
>>> disabling the test. And now,  instead of using a small hammer to disable
>>> just this test, you now even use a very *big* hammer to disable *all*
>>> tests. That's just a very bad idea. Please don't.
> 
> You asked on IRC the CI failures history:
> 
> https://travis-ci.org/github/philmd/qemu/jobs/663607963
> https://travis-ci.org/github/philmd/qemu/jobs/663622229
> https://travis-ci.org/github/philmd/qemu/jobs/663642522
> 
> "No output has been received in the last 10m0s, this potentially
> indicates a stalled build or something wrong with the build itself."

Ok, but none of these hangs seem to be related to the migration test. I 
assume that's just a generic unreliability of the builder machines, 
which should have been addresses with Alex' patch 1de8e4c4dcf2af8e1 ?

  Thomas