kubernetes: HTTP readiness/liveness probes always fail from Windows node against container running on same node

What happened:

Readiness/liveness probes from a Windows node against container running on same node using httpGet always seem to fail causing a never-ending loop of pods being terminated and recreated.

What you expected to happen:

If I configure a readiness/liveness probe for a Windows container in a pod and that container is functioning as expected, I expect the probe to succeed and the pod to continue running.

How to reproduce it (as minimally and precisely as possible):

1. Add Windows node to a K8s cluster running in overlay mode
2. Create a pod with a Windows container which exposes an HTTP endpoint and a liveness/readiness httpGet probe that queries that endpoint
3. Note the pod getting terminated due to the failing probe
4. Update/patch the pod to remove the probe
5. Grab the pod IP
6. Query the probe target using the pod IP from a Linux node and confirm it succeeds
7. Query the probe target using the pod IP from the Windows node and note that it times out

Anything else we need to know?:

I’ve uploaded the route table for the Windows node just in case that helps. I assume there should be routes configured for the pod network CIDR but I don’t see any?

Environment:

• Kubernetes version: v1.15.1
• Cloud provider or hardware configuration: Local Hyper-V VMs
• OS: Windows Server 2019 1809 (17763)
• Network plugin and version: flannel 0.11.0

Attachments:

• route-table.txt
• net-conf.json
• cni.conf