For RHODS updates of to software versions that could potentially break existing notebook code (updates to Python, TensorFlow, PyTorch, any package included in supported notebook images. What is the support lifecycle for older versions? Is that n and n-1?
The planned model is to support n (latest - recommended) and n-1 versions of notebook images. When we release a new notebook image version, it will be the recommended version. To help mitigate concerns about breaking existing notebooks, we will also support the previous version (n-1). We expect to release new notebook images ~ 1/quarter, so that will give users some time to adopt the new notebook image before the previous one is deprecated. As a future roadmap item, we also plan to support custom notebook images, which will allow customers to "freeze" specific versions in a custom-defined notebook image.