Currently the architecture field does not used by Docker engine, so the
value amd64 also works on ppc64le platform.

Yes, I saw the image was created locally, but it was not used by docker
containerizer tests.

When I run docker conainerizer tests on ppc64le, it still use the alpine
image from Docker hub which is the amd64 architecture, and the tests still
failed.

So I think if the docker containerizer tests can use the image created in
runtime_isolator_tests.cpp, these test cases will not fail on ppc64le
platform. So can we also make this local docker image for Docker
containerizer testing?


Thanks,
Zhiwei


On Wed, Apr 13, 2016 at 2:02 AM, Gilbert Song <[email protected]> wrote:

> Sorry Zhiwei, just see your email. Yes, I can do that. Just for curious,
> are you using
> the manifest architecture field in ppc64le?
>
> We already have tests relying on the image created locally. Please see:
> runtime_isolator_tests.cpp
>
> What do you mean by "And could you make it create and load the image
> before
> running Docker related tests"? The local docker image is for unified
> containerizer
> testing. It is irrelevant to any test for docker containerizer.
>
> Cheers,
> Gilbert
>
> On Mon, Apr 11, 2016 at 6:27 PM, zhiwei <[email protected]> wrote:
>
>> Hi Gilbert, do you have any updates on about this issue?
>>
>> When will you use the local build Docker image for Docker related testing?
>>
>> On Mon, Mar 14, 2016 at 9:44 PM, zhiwei <[email protected]> wrote:
>>
>>> Hi Gibert,
>>>
>>> I tested your patch, it works on ppc64le, but there is an issue that the
>>> architecture in manifest file is hardcode to amd64[1].
>>>
>>> This field is currently not used by Docker engine, but I think you can
>>> enhance it to use `os.uname().machine` [2]. Or I can submit a patch for
>>> this.
>>>
>>> And could you make it create and load the image before running Docker
>>> related tests?
>>>
>>> [1]:
>>> https://github.com/apache/mesos/blob/69b2ad528dd79979a8ee113a8edbbab2669e32e6/src/tests/containerizer/docker_archive.hpp#L150
>>> [2]:
>>> https://github.com/docker/distribution/blob/master/docs/spec/manifest-v2-2.md#manifest-list-field-descriptions
>>>
>>>
>>> On Sat, Mar 12, 2016 at 8:45 PM, zhiwei <[email protected]> wrote:
>>>
>>>> this because the architecture, alpine is x86_64, the power is ppc64le.
>>>>
>>>> this may help in IBM Power.
>>>> On Mar 12, 2016 01:10, "Gilbert Song" <[email protected]> wrote:
>>>>
>>>>> Hi Zhiwei,
>>>>>
>>>>> I am trying to understand why 'alpine' is not compatible with IBM Power
>>>>> platform. Is it because of the image's rootfs?
>>>>>
>>>>> We currently have a JIRA(
>>>>> https://issues.apache.org/jira/browse/MESOS-4684)
>>>>> in progress to create a local docker image, which is used for docker
>>>>> runtime isolator local tests. This test image cp the host's linux
>>>>> rootfs.
>>>>> Does it possibly help in your case?
>>>>>
>>>>> Cheers,
>>>>> Gilbert
>>>>>
>>>>> On Fri, Mar 11, 2016 at 2:30 AM, zhiwei <[email protected]> wrote:
>>>>>
>>>>> > Yes, I prefer to create specific mesos test images and make them
>>>>> > configurable in configuration file.
>>>>> >
>>>>> > On Fri, Mar 11, 2016 at 6:20 PM, Alex Rukletsov <[email protected]
>>>>> >
>>>>> > wrote:
>>>>> >
>>>>> > > It also looks strange to me that we use "random" containers in
>>>>> Mesos
>>>>> > tests.
>>>>> > > The proper way would be to have "mesos" or "apache" account on
>>>>> docker hub
>>>>> > > managed by PMC. Do you think it's worth to set up one or it's too
>>>>> much
>>>>> > time
>>>>> > > investment?
>>>>> > >
>>>>> > > On Thu, Mar 10, 2016 at 12:19 PM, Jan Schlicht <[email protected]>
>>>>> > wrote:
>>>>> > >
>>>>> > > > Hi Zhiwei,
>>>>> > > >
>>>>> > > > I was thinking about this as well, but for different reasons:
>>>>> Pulling
>>>>> > in
>>>>> > > > Docker images for tests is not the ideal solution. Sure, testing
>>>>> a
>>>>> > `sleep
>>>>> > > > 1000` should work, but testing an executor leaves some questions
>>>>> on how
>>>>> > > to
>>>>> > > > handle changes/deprecation of the interfaces. It's not a
>>>>> pressing issue
>>>>> > > > right now, but might become one in the future.
>>>>> > > >
>>>>> > > > I think this is also what Timothy had in mind with his comment
>>>>> > (Timothy,
>>>>> > > > please correct me if I'm wrong): These problems can be resolved
>>>>> by
>>>>> > using
>>>>> > > > local Docker images, ideally ones that are created during `make
>>>>> check`.
>>>>> > > But
>>>>> > > > this would create new problems. Either we would have to build
>>>>> libmesos
>>>>> > > > inside our local container -- to be able to build test executors
>>>>> --
>>>>> > which
>>>>> > > > would take a long time, or we'd have to make sure that the
>>>>> container
>>>>> > > > environment is the same as the dev environment, to be able to
>>>>> copy test
>>>>> > > > executors into it, which isn't easy unless we'd restrict
>>>>> ourselves to
>>>>> > > only
>>>>> > > > a couple of environments.
>>>>> > > >
>>>>> > > > Cheers,
>>>>> > > > Jan
>>>>> > > >
>>>>> > > > On Thu, Mar 10, 2016 at 11:25 AM, zhiwei <[email protected]>
>>>>> wrote:
>>>>> > > >
>>>>> > > > > Hi all,
>>>>> > > > >
>>>>> > > > > The Docker related test cases that hardcoded "alpine" as the
>>>>> Docker
>>>>> > > image
>>>>> > > > > which caused test cases failed on IBM Power platform, since the
>>>>> > Docker
>>>>> > > > > image "alpine" is not compatible with IBM Power platform.
>>>>> > > > >
>>>>> > > > > And I saw an inline comment by Timothy: "// TODO(tnachen): Use
>>>>> local
>>>>> > > > image
>>>>> > > > > to test if possible."
>>>>> > > > >
>>>>> > > > > So just wonder if someone has plan to implement this, or could
>>>>> you
>>>>> > give
>>>>> > > > me
>>>>> > > > > some tips? I can implement this.
>>>>> > > > >
>>>>> > > > > Following are the images that used in Mesos test cases:
>>>>> > > > >
>>>>> > > > > 1. alpine
>>>>> > > > > 2. mesosphere/alpine-expect
>>>>> > > > > 3. mesosphere/inky
>>>>> > > > > 4. mesosphere/test-executor
>>>>> > > > > 5. tnachen/test-executor
>>>>> > > > >
>>>>> > > > >
>>>>> > > > > Thanks,
>>>>> > > > > Zhiwei
>>>>> > > > >
>>>>> > > >
>>>>> > > >
>>>>> > > >
>>>>> > > > --
>>>>> > > > *Jan Schlicht*
>>>>> > > > Distributed Systems Engineer, Mesosphere
>>>>> > > >
>>>>> > >
>>>>> >
>>>>>
>>>>
>>>
>>
>

Reply via email to