Description of problem:
Running v3_4/upgrade.yml on a 3.4 system to bring it to the latest version within 3.4. The upgrade fails with:
TASK [Generate etcd backup] ****************************************************
fatal: [master.ose33.org]: FAILED! => {
    "changed": true,
    "cmd": [
        "etcdctl",
        "backup",
        "--data-dir=/var/lib/etcd/default.etcd",
        "--backup-dir=/var/lib/origin/etcd-backup-pre-upgrade-20170620143759"
    ],
    "delta": "0:00:00.016078",
    "end": "2017-06-20 14:37:59.698638",
    "failed": true,
    "rc": 1,
    "start": "2017-06-20 14:37:59.682560",
    "warnings": []
}

STDERR:

2017-06-20 14:37:59.698101 I | wal: file not found
        to retry, use: --limit @/home/ec2-user/.ansible-retry/upgrade.retry


Version-Release number of selected component (if applicable):
$ oc version
oc v3.4.1.24
kubernetes v1.4.0+776c994

How reproducible:
Working cluster - not at the latest dot release. Run:

ansible-playbook -i ose34.inventory /usr/share/ansible/openshift-ansible/playbooks/byo/openshift-cluster/upgrades/v3_4/upgrade.yml

From IRC chat, it may be the data-dir location that's wrong: --data-dir=/var/lib/etcd/default.etcd". There was no default.etcd directory. Creating this directory by hand did not resolve the problem.