Description of problem: After upgrading 3.9 to 3.10 applications are getting 'No route to host' exceptions. Deleting the SDN pod on the node seems to resolve the issue temporarily. We also see that curling the endpoint IP from the router shows 'No Route to host'. This env is using ovs-networkpolicy plugin. Version-Release number of selected component (if applicable): 3.10.83-1 How reproducible: Steps to Reproduce: 1. 2. 3. Actual results: Applications intermittently get 'No route to host' exceptions Expected results: Application able to resolve a .svc and access that endpoint. Additional info:
At the time of any issue where a pod gets "getsockopt: no route to host" when connecting to the Kubernetes service IP. The output of iptables nat table is needed. The likely reasons for this issue are either the NAT rules are not updated on the nodes or the masters api endpoints are not healthy.
*** This bug has been marked as a duplicate of bug 1689690 ***