openshift/origin-ovn-kubernetes image does not contain kubectl binary

[danwinship]

As currently built, the openshift/origin-ovn-kubernetes image does not contain kubectl, meaning that even with #4 the node still can't start

[pecameron]

Can we just eliminate the need for kubectl in the image? It is quite large.

[danwinship]

Not trivially; ovnkube.sh uses it to find the address of the nbdb and sbdb to pass to ovnkube --init-node.

In theory we could change ovnkube to let us say --nb-service=openshift-ovn-kubernetes/ovn-northd --sb-service=openshift-ovn-kubernetes/ovn-southd instead, and then ovnkube would do the lookup rather than having the wrapper script do it. That might be nice.

[danwinship]

In theory we could change ovnkube to let us say --nb-service=openshift-ovn-kubernetes/ovn-northd

actually we don't even need to do that; we can just create kube Services for northd and southd and then add, eg

env:
- name: OVN_NORTH
  value: tcp://ovn-northd.openshift-ovn-kubernetes.svc:6641
- name: OVN_SOUTH
  value: tcp://ovn-southd.openshift-ovn-kubernetes.svc:6642

and it can resolve the service IP via DNS. But that still requires rewriting stuff because currently it always uses kubectl to check the ovnkube-master service to figure out when it can start the ovn node

[squeed]

Maybe, though we often come up before kube-dns, so it's probably better to add native service support directly.

[dcbw]

@squeed @danwinship isn't this (and some other PRs) diverging from the upstream script? Shouldn't we be copying upstream into our downstream repo first, and anything we need to fix, doing that upstream first and then pulling downstream?

[danwinship]

Maybe, though we often come up before kube-dns, so it's probably better to add native service support directly.

ah, indeed. meh

isn't this (and some other PRs) diverging from the upstream script? Shouldn't we be copying upstream into our downstream repo first, and anything we need to fix, doing that upstream first and then pulling downstream?

In my most recent comments I meant to imply making those changes upstream, not here.

For the moment, I am trying to get this repo to a point where we can get ovn-kube to actually come up and work. Once that happens, I'll start upstreaming fixes and rebasing. It doesn't make sense to spend time filing PRs upstream now since we can't even test them yet, and don't actually know if they're going to work.

[pecameron]

Maybe it would be good to merge upstream onto this repo and start with that?

[dcbw]

@danwinship Right, but there's a pretty big delta between upstream and what's in openshift's downstream repo. IMHO it would be better to copy over what's upstream, and start from there, otherwise you have to do some work twice.