Arvados

This might be a duplicate of #6606. That's really a call for Tom to make. The question is: do we want to fix this bug by providing #6606? Or are we willing to consider an alternative implementation?

I'd advocate for #6606 as the solution, because the logic involved in a complete copy is non-trivial, and I would rather not rewrite it.

This is actually two separate issues. (1) workbench move is incomplete, and (2) workbench copy is incomplete. In case of (1) the fix seems simpler in that we need to change the owner_uuid of all the objects belonging to the pipeline, and probably can be implemented without blocking until the issues with copy where figuring out what to do with copying of the jobs as identified in #7001 are resolved.

Radhika Chippada wrote:

This is actually two separate issues. (1) workbench move is incomplete, and (2) workbench copy is incomplete. In case of (1) the fix seems simpler in that we need to change the owner_uuid of all the objects belonging to the pipeline, and probably can be implemented without blocking until the issues with copy where figuring out what to do with copying of the jobs as identified in #7001 are resolved.

You're right that it might help to think of these as separate issues, but I'm not sure move is actually simpler. There are some corner cases where this proposal would cause different bugs. For example, imagine we're moving one pipeline that reused jobs from another pipeline that we didn't move. In that case, some users might lose permissions whether we move the jobs or not. Same goes for other dependencies like Docker images, log collections, …