TLS handshake fails due to large packets discarded for OpenShift 4 on Azure

Issue

• TLS handshake errors occur although TCP communication is possible. A traffic capture shows large packets being discarded (see "Diagnostic Steps").
• Unexpected ICMP fragmentation needed messages are received for direct communications happening between OpenShift nodes but without vxlan encapsulation. Requested MTU is lower than the one set in both ends and/or required by any intermediate element.
• Routing cache shows bad entries as described in "Diagnostic Steps".
• After some time, the OpenShift Cluster becomes very slow and many operators start to become unhealthy (degraded state).