When we upgrade the IDH deployment of RHODS from 1.0.3 to 1.0.4 we found that despite the jupyterhub image stream being pointed at the correct image (`quay.io/odh-jupyterhub/jupyterhub-img:v0.2.2`), the pod wasn't actually using the correct image, as evidenced by

sh-4.2$ cat /opt/app-root/bin/start-jupyterhub.sh 
 #!/bin/bash 
 set -x 
 set -eo pipefail
. Add scripts directory to program search path.
PATH=$PATH:/opt/app-root/scripts # Read any custom environment variables. 
 . /opt/app-root/etc/jupyterhub_config.sh
. Start the JupyterHub instance.
 exec jupyterhub -f /opt/app-root/etc/jupyterhub_config.py

The manual fix involved the following:

1. Run `oc import-image jupyterhub-img` in the ods applications namespace
2. Start a rollout in the Jupyterhub DC