Symlink network namespace failed

[Pengfei Ni]

When restarting a pod, sometimes the pod failed to start because of failure symlink network namespace, e.g.

Error while adding to cni network: error of symlink "/var/run/netns/kube-dns-1476438210-3x9gt": symlink /var/run/netns/cni-28c96265-8a69-a09e-266d-edb8437ae4d2 /var/run/netns/kube-dns-1476438210-3x9gt: file exists

In this case, the original netns has already been removed by container runtime, but the symlink hasn't been removed. There are two problems here:

1. cni plugin should clean up the symlink on cases of failed operations in cmdAdd
2. the podName is used in symlink, but it may be conflict in different kubernetes namespaces, we should concat kubernetes namespace with podName.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/496173

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/496173
Committed: https://git.openstack.org/cgit/openstack/stackube/commit/?id=7e2fc21c2c79f6fb54eb4940f9810e2db5a9c53d
Submitter: Jenkins
Branch: master

commit 7e2fc21c2c79f6fb54eb4940f9810e2db5a9c53d
Author: Pengfei Ni <email address hidden>
Date: Tue Aug 22 17:19:02 2017 +0800

    Clean up netns symlink of failure

    * change the netns symlink name to pod full name (concating the
      namespace and name)
    * clean up the symlink on failure

    Change-Id: Ic80b07e2e3b6c3a312d95af4068e69f4740e44eb
    Closes-Bug: 1712273
    Signed-off-by: Pengfei Ni <email address hidden>