networking: bump to CNI v0.3.0, and adapt & refactor code #2696
Conversation
| @@ -406,6 +426,11 @@ | |||
| "Rev": "a1b8ba5163b7f041b22761461eabd02b70d1f824" | |||
| }, | |||
| { | |||
| "ImportPath": "github.com/gogo/protobuf/proto", | |||
| "Comment": "v0.1-125-g82d16f7", | |||
| "Rev": "82d16f734d6d871204a3feb1a73cb220cc92574c" | |||
There was a problem hiding this comment.
I don't know if gogo/protobuf#182 would be done in time.
Does it actually fail with the old tag v0.2?
#2673 (comment)
| if err != nil { | ||
| return nil, err | ||
| func podNSerrorOK(podNSPath string, err error) bool { | ||
|
|
|
Additionally intelsdi-x@706e8b2 provides changes included in containernetworking/cni#149 to kvm flavor. I'll look on this deeper on Monday, but it seems that this PR breaks kvm network teardown. tap interfaces and iptables rules are preserved after gc. |
|
@jellonek can you point us to a direction why the teardown wouldn't work for KVM? I changed the handling of netns creation and switching, but I thought KVM doesn't involve that. |
|
@jellonek also, It seems to be likely that the disabled/failing tests are dependent on the cleanup of these interfaces. can you confirm that? |
| @@ -215,12 +200,20 @@ func Load(podRoot string, podID *types.UUID) (*Networking, error) { | |||
| nets = append(nets, *n) | |||
| } | |||
|
|
|||
| // TODO load podNS | |||
|
LGTM |
* Fixes KVM flavor, which doesn't use the NS file
This will implicitly disable these tests for the KVM flavor on which they are currently failing.
| @@ -1,4 +1,5 @@ | |||
| #!/bin/bash -e | |||
| #!/usr/bin/env bash | |||
|
@steveej: i'll try to confirm that on monday. |
TODO