I have been able to reproduce this twice.

Steps to reproduce:

• Download kubernetes 1.5.5 tar, extract and cd into the directory
• run cluster/get-kube-binaries.sh
• run cluster/kube-up.sh
• Run the steps defined here here
  • You don’t need to wait for the HPA to react, just verify that the script says “OK!” a bunch of times to ensure it can find the servier
• Download the v1.6.0-rc.1 release tar, extract, and cd into the directory
• run cluster/get-kube-binaries.sh
• Warning: Set the following env or you will not be able to downgrade later

export TARGET_STORAGE=etcd3
export ETCD_IMAGE=3.0.17
export TARGET_VERSION=3.0.17
export STORAGE_MEDIA_TYPE=application/json

• run ./cluster/gce/upgrade.sh -M v1.6.0-rc.1
• Run the steps defined here here to verify the Pod can see the service
• run ./cluster/gce/upgrade.sh -N -o v1.6.0-rc.1 to upgrade the nodes
• Run the steps defined here here to verify the Pod can see the service
• run ./cluster/gce/upgrade.sh -N -o v1.5.5 to downgrade the nodes
• Run the steps defined here here to verify the Pod can see the service
• Warning: Set the following env to downgrade

export TARGET_STORAGE=etcd2
export ETCD_IMAGE=3.0.17
export TARGET_VERSION=2.2.1
export STORAGE_MEDIA_TYPE=application/json

• run ./cluster/gce/upgrade.sh -M v1.5.5 to downgrade the master
• Run the steps defined here here to verify the Pod can see the service
  • This is where things fail for me. I am not longer able to connect to the service through the DNS address.

wget php-apache.default.svc.cluster.local -q -O -
wget: bad address 'php-apache.default.svc.cluster.local'

• I am able to connect to the service using its IP address

kubectl get services
NAME         CLUSTER-IP    EXTERNAL-IP   PORT(S)   AGE
kubernetes   10.0.0.1      <none>        443/TCP   1h
php-apache   10.0.52.190   <none>        80/TCP    1h
wget -q -O - 10.0.52.190
OK!