0

I have bare-metal deployed Kubernetes cluster on VMware vSphere private cloud.

Spec:

  • OS: Ubuntu 20.04.3 Cloud Image (Also trying 18.04 and 21.04)
  • Kubernetes: 1.23.1 (Also trying 1.21.8) - depolyed with kubeadm and kube-proxy (with strictARP and ipvs mode)
  • CRI: Docker 20.10.12
  • MetalLB v0.11.0 (Layer 2 mode)
  • Cilium 1.11.1
  • All nodes in 192.168.50.0/24 subnet

When I simply add deployment:
kubectl create deployment nginx --image k8s.gcr.io/nginx:latest

And exposed it:
kubectl expose deployment nginx --type=LoadBalancer --port=80

All seems good. Service got ExternalIP from MetalLB pool:
nginx LoadBalancer 10.96.56.118 198.168.50.152 80:32325/TCP 4s

New ip address published on kube-ipvs0 interface:

kube-ipvs0: <BROADCAST,NOARP> mtu 1500 qdisc noop state DOWN group default
...
inet 198.168.50.152/32 scope global kube-ipvs0
valid_lft forever preferred_lft forever
...

But I cant get access to service from all K8s nodes trought ExternalIP, and also from other LAN, curl returns:
curl: (7) Failed to connect to 192.168.50.152 port 80: No route to host

ClusterIP works good, I can access it from node.

old_timer
  • 69,149
  • 8
  • 89
  • 168
Ilya Pozdnov
  • 11
  • 2

1 Answers1

0

Solution is: Your Virtual Machines and services must be in different subnets. In my case I have 192.168.50.0/24 network and I split it to 2 subnets 192.168.50.0/25 - for virtual machines and 192.168.50.128/25 for MetalLB services.

Ilya Pozdnov
  • 11
  • 2