Description of problem:
In a new installation or when a new node is added if the clusternetwork is configured with 2 subnets pods in one subnet cannot communicate with pods in another subnet until the ovnkube-controller container is restarted.

Version-Release number of selected component (if applicable):
reproduced in 4.14.31 and 4.14.42

How reproducible:
100% always

Steps to Reproduce:

1. Install a cluster with 2 subnets in the clusternetwork

2. Try the connectivity between pods residing on different CIDRs

Actual results:
When a cluster is installed or a new node is created, pods on different CIDRs cannot communicate until ovnkube-controller is manually restarted.

Expected results:
The clusternetwork should work without manually restart the ovnkube-controller container.

Additional info:

The workaround identified is just restarting the ovnkube-controller but for customers using clusterautoscaler this workaround is not sustainable. 

The only relevant error visible in the ovnkube-controller logs when the clusternetwork doesn't work is the following:

W1217 09:13:09.431454    3934 node_tracker.go:233] Failed to get node host CIDRs for [worker-0: k8s.ovn.org/host-cidrs annotation not found for node "worker-0"

This error is not visible after manually restarting the container.