Change default namespace logic

[anishasthana]

• Update developer instructions
• Change Default namespace logic to use user's current namespace

[MichaelClifford]

Ok, kind of annoying, but I think we need to change this PR a bit to account for the fact that using the MCAD scheduler also needs a get_namespace call to make sure it doesn't just use the "default" namespace.

codeflare-sdk/src/codeflare_sdk/job/jobs.py

Lines 146 to 149 in f529763

	scheduler="kubernetes_mcad",
	cfg=self.scheduler_args if self.scheduler_args is not None else None,
	workspace="",
	)

We need to add something here like:

self.scheduler_args["namespace"] = openshift.get_project()

We also need to deal with the fact that we already have a get_current_namespace() function within the cluster.py file. (This is an artifact from when we assumed a cluster object would always be required).

My suggestion is that we delete get_current_namespace() as its not used by any other function, and then rely on the openshift package to call openshift.get_project() wherever we need the project name (basically your idea from yesterday 😄). This avoids the need for a higher level utils.py type file to house a shared function like get_current_namespace.

Let me know what you think.

[anishasthana]

My suggestion is that we delete get_current_namespace() as its not used by any other function, and then rely on the openshift package to call openshift.get_project() wherever we need the project name (basically your idea from yesterday 😄). This avoids the need for a higher level utils.py type file to house a shared function like get_current_namespace.

Question about that -- yesterday we also said that there is a need to allow users to determine what namespace they are in from the API. How would we structure things for that?

[MichaelClifford]

Question about that -- yesterday we also said that there is a need to allow users to determine what namespace they are in from the API. How would we structure things for that?

I did say that 😄, But on reflection, not sure how critical that really is. Especially if we are outputting it somewhere when we define the job or cluster objects.

If this becomes an issue we could add it in in a later PR, as a utils function like I suggested. But I think correctly auto-setting the namespace is a much more important feature than just getting it.

That said, if you disagree, by all means go the utils route :)

[MichaelClifford]

thanks @anishasthana
LGTM!