tasks: fail to submit workflow start operation

[diegodelemos]

Currently, if there are dangling images in the Kubernetes node(s), the check_predefined_conditions call will fail when we call list_nodes because of kubernetes-client/python#895. The exact reason of this is described in kubernetes-client/python#895 (comment), there has been a fix attempt but it hasn't been tackled yet.

The traceback when this happens:

$ kubectl logs reana-server-xxxxx-yyy scheduler --previous
  File "/usr/local/lib/python3.6/site-packages/kubernetes/client/models/v1_container_image.py", line 75, in names  
    raise ValueError("Invalid value for `names`, must not be `None`")  # noqa: E501           
ValueError: Invalid value for `names`, must not be `None` 

Note the usage of --previous, because the process exits you won't see any logs if you do kubectl logs reana-server-xxxxx-yyy as you will see the logs of the new pod.

Since we cannot control this, we should be more resilient and avoid blocking the whole system for this reason.