[CRIU] Fwd: CRIU » CRIU-iter » criu-dev - Build # 1209 - Failure! (origin/criu-dev)

[Pavel Emelyanov]

On 10/26/2016 07:03 PM, Tycho Andersen wrote:
> On Wed, Oct 26, 2016 at 11:24:31AM +0300, Pavel Emelyanov wrote:
>> On 10/24/2016 10:32 PM, Andrei Vagin wrote:
>>> On Mon, Oct 24, 2016 at 06:18:56PM +0300, Pavel Emelyanov wrote:
>>>> On 10/24/2016 05:10 PM, Tycho Andersen wrote:
>>>>> On Mon, Oct 24, 2016 at 02:52:38PM +0300, Pavel Emelyanov wrote:
>>>>>> Hi, Tycho.
>>>>>>
>>>>>> The macvlan test fails in uns flavor. Would you tale a look, please?
>>>>>
>>>>> Hmm. Is it possible to get the output of the hook somehow? Seems like
>>>>> that's what really failed.
>>>>
>>>> Well, it ... sometimes passes actually :\
>>>>
>>>>> One thing I thought of is that using pidof to get the pid of the test
>>>>> probably isn't very robust, since there might be more than one test
>>>>> running at the same time. But it would be good to know what really
>>>>> failed :)
>>>>
>>>> Andrey, would you suggest something?
>>>
>>> 17386 pts/0    T      0:00              |               \_ make --no-print-directory -C zdtm/static macvlan.pid
>>> 17400 pts/0    T      0:00              |               |   \_ ./macvlan --pidfile=macvlan.pid --outfile=macvlan.out
>>> 17401 ?        Ss     0:00              |               |       \_ ./macvlan --pidfile=macvlan.pid --outfile=macvlan.out
>>> 17403 ?        S      0:00              |               |           \_ ./macvlan --pidfile=macvlan.pid --outfile=macvlan.out
>>> 17404 ?        Ss     0:00              |               |               \_ ./macvlan --pidfile=macvlan.pid --outfile=macvlan.out
>>> 17387 pts/0    T      0:00              |               \_ /bin/bash zdtm/static/macvlan.hook --post-start
>>> 17388 pts/0    T      0:00              |                   \_ sleep 1000
>>>
>>> When you call pidof there is four processes with the macvlan name
>>> and the first one is running in the current namespaces, it will
>>> exit when the test calls test_daemon().
>>>
>>> I think you need to write a pid in a file from test instead of using "pidof"
>>
>> Test also lives in another pid namespace, so value from the pidfile won't help hook script :)
> 
> I think we might be able to use /proc/<pid>/task/<pid>/children and walk
> down the stack until we find one with no children. I'll look into it.

Tycho, look at my set here:
https://lists.openvz.org/pipermail/criu/2016-October/032498.html

I have the same problem and solved this by generating $testname.pid.inprogress
file from inside the zdtm/lib/ and use it in .hook :)

-- Pavel