Bug 1309881

Description of problem:

In deploying a pod we would get failures.  Looking in the deploy pod docker log:

docker log <deployer container ID>

would show the the error:

================================
 1 deployer.go:65] couldn't get deployment <namespace>/cakephp-example-6: Get https://internal.api.clustername.openshift.com:443/api/v1/namespaces/<namespace>/replicationcontrollers/cakephp-example-6: net/http: TLS handshake timeout
================================

This ultimately turned out to be an MTU issue.  The mtu size of eth0 was 9000 (jumbo frames) while the mtu of tun0 (ovs) was 1500.  We noticed through a tcpdump that anything bigger than 1500 was coming in on eth0 but was not going across tun0.  It was being dropped and caused the error shown above.

When speaking with Clayton, he suggested that this error may be able to be bubbled up and show the user that there may be an MTU issue present.  Finding this MTU bug was very involved.  Having the user alerted to this earlier may help.


Version-Release number of selected component (if applicable):

atomic-openshift-node-3.1.1.6-2.git.10.15b47fc.el7aos.x86_64
openvswitch-2.4.0-1.el7.x86_64


Steps to Reproduce:
1.  configure openshift node with eth0 with mtu size of 9000.
2.  configure tun0 to have size of 1500
3.  attempt to deploy pod to the node.
4.  if the deploy fails, on the node where the deploy failed, run "docker logs <id of deploy container>" 

Actual results:

TLS handshake timeout

Expected results:

The build should communicate with the master api without errors.

Additional info:

There is probably an underlying bug that doesn't allow tun0 to pass traffic bigger than the MTU that it has set.  Frame fragmentation should happen and traffic should pass.