Clarification Regarding RepoAction

[Sid9993]

run_args.get('item', run_args.get('artifact'))

In this line, the keys item and artifact, are these passed when we call mlcflow functions from the script automation code.

[arjunsuresh]

We normally don't call repo actions from script automation right? But, when we do we need to pass it.

[Sid9993]

We normally don't call repo actions from script automation right? But, when we do we need to pass it.

Oh ok fine.
I would like to clarify the logic behind this find function in the RepoAction :
https://github.com/mlcommons/mlcflow/blob/23897070a011328190843104159a258d78ba866b/mlc/main.py#L939C5-L944C34

While I went through the code I found out that the command for finding the repo would be like :
find repo <repo_owner>@<repos_name>,<repo_uid>

Is this the right way to find the repo?

[arjunsuresh]

Actually we should support any identifier for a repo like below.

1. <repo_owner>@<repos_name>
2. <repo_url>
3. <repo_uid>
4. <repo_alias>
5. <repo_alias>,<repo_uid>

[Sid9993]

Actually we should support any identifier for a repo like below.

1. <repo_owner>@<repos_name>
2. <repo_url>
3. <repo_uid>
4. <repo_alias>
5. <repo_alias>,<repo_uid>

Oh yes. Got it.
Thank You for the clarification.

[anandhu-eng]

Hi @arjunsuresh , is there any situation where a user tries the fifth option?

[Sid9993]

mlcflow/mlc/main.py

Line 961 in 9e9c933

elif repo_name == i.meta['alias']:

Suppose there is a scenario where I pull a forked repo and I find the same by using find repo <my_forked_repo_alias>. In that case, according to the above logic the alias in the meta.yaml file would be the original repo and not the forked one.
So I doubt whether that would result in a conflict while we use find repo command.

[arjunsuresh]

Shouldn't cause a conflict as mlc pull repo will only register one repo and if a fork is pulled, the previous fork or the origin repo automatically gets unregistered.

[anandhu-eng]

Hey folks,

I'm reopening this issue as we might have missed a tricky part while merging PR for mlc find repo.

here we have used repos_list which loads data from the function load_repos_and_meta. But for comparing the repo uid as well as alias, we are using repos variable that comes from Index class which only initialize at the beginning when calling mlc.

The above condition would be working without failures when we are using the pull and find as a seperate mlc command because the Index class gets initialized each time it is called, and the updation in repos.json would be reflected in repos variable.

For the situation where a person uses both pull and find action in same python script, this creates a problem as the Index class is not re-initialised. Example could be found in failed GitHub actions of this PR.

[anandhu-eng]

Closing this discussion as its been decided to not concentrate on repo modification actions through mlc access.

Feel free to reopen if you feel this feature is relevant to any of you use cases.