Added docker image test script

[vfdev-5]

Addresses #1644

Description:

• Added docker image test python script

Check list:

• New tests are added (if a new feature is added)
• New doc strings: description and/or example code are in RST format
• Documentation is updated (if required)

[trsvchn]

there is python package docker? Nice!)

Since, I've started from subprocess.run :)

[vfdev-5]

Yes, I tried that from my side to see if this could work. @trsvchn if you'd like to review, feel free to comment out and give a feedback.

[vfdev-5]

And sorry for overlapping !

[trsvchn]

No problem, sorry, I was busy yesterday. and now I know about docker python so it worth it :)

[vfdev-5]

@trsvchn do you have an idea why it can't find installed docker package ?
https://github.com/pytorch/ignite/pull/1733/checks?check_run_id=2031974887#step:4:838

[trsvchn]

@vfdev-5 Oh, I think I know what is happening

[trsvchn]

$AGENT_TOOLSDIRECTORY is /opt/hostedtoolcache

and

  env:
    pythonLocation: /opt/hostedtoolcache/Python/3.7.10/x64
    LD_LIBRARY_PATH: /opt/hostedtoolcache/Python/3.7.10/x64/lib

Thus:

sudo rm -rf "/usr/local/share/boost"
sudo rm -rf "$AGENT_TOOLSDIRECTORY"

Removes python libs and python

[vfdev-5]

Oh, I see. Should we just reinstall it after building ?

[trsvchn]

But, wait, pyaml works fine

[vfdev-5]

pyyaml is OK as it is used only before the suppression

[trsvchn]

Yes, for base images
but we reuse it in later steps

[trsvchn]

Let's try

sudo rm -rf "/usr/local/share/boost"
sudo rm -rf "$AGENT_TOOLSDIRECTORY"

Just after the checkout, and before Python install actiob

[vfdev-5]

@trsvchn seems like it works !

[trsvchn]

Great!

[vfdev-5]

@trsvchn I'm struggling with test_image.py script and the way how to expose details of the error when using docker python api. For example, here https://github.com/pytorch/ignite/pull/1733/checks?check_run_id=2046167278#step:6:605
import command is failing but we have no idea what is the error message.
Basically, the following code is missing something.

def run_python_cmd(cmd):
    try:
        out = client.containers.run(args.image, f"python -c '{cmd}'", auto_remove=True, stderr=True,)
        assert isinstance(out, bytes), type(out)
        out = out.decode("utf-8").strip()
    except docker.errors.ContainerError as e:
        raise RuntimeError(e)
    return out

Any ideas how to improve it ?

[trsvchn]

@vfdev-5 sure! I'll test test_image.py in my sandbox

[vfdev-5]

I'm thinking to do something like

def run_python_cmd(cmd):

    try_except_cmd = f"""

def main():
    {cmd}

try:
    main()
except Exception as e:
    import traceback
    print(traceback.format_exc())    
    """
    try:
        out = client.containers.run(args.image, f"python -c '{try_except_cmd}'", auto_remove=True, stderr=True,)
        assert isinstance(out, bytes), type(out)
        out = out.decode("utf-8").strip()

        if any([k in out.lower() for k in ["error", "exception"]]):
            raise RuntimeError(out)

    except docker.errors.ContainerError as e:
        raise RuntimeError(e)
    return out

but this looks like a hack...

[trsvchn]

@vfdev-5 Yes!, I think the problem is in that cmd raises exceptions, since this generates exit code 1 and it forces run method to raise exception:

$ python -c 'raise RuntimeError'
Traceback (most recent call last):
  File "<string>", line 1, in <module>
RuntimeError

$ echo $?
1

    Raises:
        :py:class:`docker.errors.ContainerError`
            If the container exits with a non-zero exit code and

we should handle this quietly

[vfdev-5]

That's right. But what I'd like to have is to execute tests like import cv2 and ensure that it passes and if not say that explicitly : "ModuleNotFoundError: No module named 'cv2'".
I was just thinking if there is a proper way to get error message without 2 levels of exception trackings

[trsvchn]

What about to try detach=True, with this param, run method returns container and logs:

       If ``detach`` is ``True``, a :py:class:`Container` object is returned instead.

https://github.com/docker/docker-py/blob/55f405e04a91ddbbc26aa738bd3cb41bd28f8dbd/docker/models/containers.py#L536-L541

[vfdev-5]

detach or not detach this is the question :)
Well, actually, the context is that I do not understand why there is no cv2 module in msdp-apex-vision image and I can not rerpo that locally...

[trsvchn]

cv or not 2cv ... or cv2

[trsvchn]

It seems to work fine

[vfdev-5]

Let's see if we can finally update our docker images