EEXIST when kubelet triggers retry of ADD request

[Michal Dulko]

On errors and timeouts kubelet is triggering a retry. Now if error happened after we've moved host vETH to host namespace all the future retries will fail due to pyroute2 raising exception EEXIST. To mitigate that we need to catch that exception and simply ignore it as vETH is already in the host namespace and we can safely proceed.

[Michal Dulko (michal-dulko-f)]

Okay, I'd say this is low priority, I think I've found the main reason we were even getting into this situation.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/518404

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/518404
Committed: https://git.openstack.org/cgit/openstack/kuryr-kubernetes/commit/?id=9ac4df32e2e2d3111ca0d4e57ba8b22bcc1974c8
Submitter: Zuul
Branch: master

commit 9ac4df32e2e2d3111ca0d4e57ba8b22bcc1974c8
Author: Michał Dulko <email address hidden>
Date: Tue Nov 7 12:03:27 2017 +0100

    Fix kubelet retries issues

    When an error happens during kubelet's ADD request kubelet will trigger
    a retry. If aforementioned failure happened after the host-side veth
    interface was moved into the host netns, all further kubelet retries of
    the request will fail with conflict, as it will be impossible to move
    another iface of that name into the host netns.

    This commit aims to fix the issue by checking for existence of
    conflicting interface in the host netns and removing it if needed.

    Change-Id: I1596585c8b076d09a7f8c854bb524c2374d804e8
    Closes-Bug: 1730644

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/kuryr-kubernetes 0.4.0 release.