pkg: Do not set MTU greater than 1450 for VXLAN

[yoheiueda]

MTU of a VXLAN interface in a pod VM becomes greater than 1450,
which is the maximum allowed size for VxLAN packets, when a CNI
plugin such as Calico IPIP uses a MTU greater than 1450. This may
cause undesirable packet fragmentation. This patch fixes the issue.

Fixes #67

Signed-off-by: Yohei Ueda yohei@jp.ibm.com

[stevenhorsman]

The code changes look good to me, and I've tested it out with the ibmcloud provider and was able to successfully create an nginx peer pod

[bpradipt]

/lgtm

[huoqifeng]

@yoheiueda is it possible to accept the MTU as a parameter rather than a hard-coded const? i think Calico MTU is configurable.

[yoheiueda]

@huoqifeng MTU is not hard-coded.

MTU of the interface created by a CNI plugin is detected at a worker node. Then, the detected MTU is set to the interface of the pod VM.

When a detected MTU is greater than 1450, which is the maximum allowed MTU for VXLAN, MTU of the interface of the pod VM is set to 1450.

cloud-api-adaptor/pkg/podnetwork/workernode.go

Lines 114 to 118 in 41b75cc

	mtu, err := podNS.GetMTU(podInterface)
	if err != nil {
	return nil, fmt.Errorf("failed to get MTU size of %s: %w", podInterface, err)
	}
	config.MTU = mtu

cloud-api-adaptor/pkg/podnetwork/tunneler/vxlan/podnode.go

Lines 67 to 71 in 41b75cc

	mtu := int(config.MTU)
	if mtu > maxMTU {
	mtu = maxMTU
	}
	if err := podNS.SetMTU(podVxlanInterface, mtu); err != nil {