Helm charts, pod storage issues

Hello,
I’m using datalore helm installation version on k8s.
I selected a agent (=machine type, pod) and used well.
But everytime i changed a agent, i lost all packaged that i downloaded using pip.
How can i store my downloaded packages even though pods are changed.

Please give me any advice, thanks.

Hello June,

I’ve already answered you in person, and just duplicating the answer publicly to help anyone else who encounters the same issue.

For the sake of reproducibility, Datalore uses predefined isolated environments for notebooks. Thus, when you restart a notebook, either explicitly or by changing the underlying machine/pod, its state is reset to the defaults, including the python environment. With this in mind, there are two ways to persist installed libraries through instance restarts: either make Datalore aware of additional libraries, or incorporate these libraries into the environment itself (the latter option is only for on-premises). Let’s look at each of the options in more detail.

To notify Datalore about the libraries you want to install, use Environment tool. In this case, the notebook will install the libraries you added at startup. Depending on the list of libraries, it may take some time. This suits better for libraries that are not crucial for you daily work, and you only want to have them in a small portion of your notebooks.

Alternatively, for on-premises installations, you may change the default environment itself. This requires to build a custom Docker image on top of Datalore one, but allows to customize the environment for your specific needs, or even make multiple envs and switch between them later.

1 Like