As mentioned in RHODS-1981, we need to document the following (or similar) for RHODS 1.1:

While every effort is made to make Red Hat OpenShift Data Science resilient to OpenShift node failure, upgrades, and similarly disruptive operations, individual users' notebook environments can be interrupted during these events. If an OpenShift node restarts or becomes unavailable, the user's notebook environment is restarted on a different node. When this occurs, any ongoing process executing in the user's notebook environment is interrupted, and the user needs to re-execute it when their environment becomes available again.

Due to this limitation, Red Hat recommends that processes for which interruption is unacceptable are not executed in the Jupyter notebook environment on OpenShift Data Science. For example, rather than execute long-running data science model training activities within a Jupyter notebook on OpenShift Data Science, we recommend leveraging the model serving capabilities of Seldon instead.