contrail with Kubernetes issue with master node

40 views

Skip to first unread message

Juraj Markotic

unread,

Aug 14, 2018, 6:31:39 AM8/14/18

to Tungsten Fabric Users

Hi,

tried adding TF CNI to Kubernetes cluster.

Followed this installation links.

https://tungstenfabric.github.io/website/Tungsten-Fabric-one-line-install-on-k8s.html

https://github.com/Juniper/contrail-controller/wiki/Install-K8s-using-Kubeadm

During master node install, one should disable networking on master node.

Problem is that in latest versions of K8s (1.11), this env variable is no longer present in 10-kubeadm.conf file).

vi /etc/systemd/system/kubelet.service.d/10-kubeadm.conf

Comment out "#Environment="KUBELET_NETWORK_ARGS=--network-plugin=cni --cni-conf-dir=/etc/cni/net.d --cni-bin-dir=/opt/cni/bin"

so, i proceeded without it. Kube master node install and I deployed TF CNI with kubeclt one-liner.

I assigned Master-node VM with 8 vcpu and 16Gb RAM.

After vrouter is spinned up, it effectively kills master node - taking all cpu resources (I can access instance only via console, ssh timeouts).

When checking contrail-status, I can see some errors (some pods not up, some images in pullback status etc).

I assume I would need to stop vrouter (and maybe some oter modules) not to be spinned up.

Any idea how would I recover from this issue ? Either by stopping certain TF pods/modules or I can reinstall but would need to know how to stop networking to be brought up on master node.