This requires the backend API to be there. @IronPan to chime in, is there an issue tracking that work?
+1 to this being important.
API would provided as part of this https://github.com/kubeflow/pipelines/pull/112
More context from @IronPan: That API orphans the deleted experiment's runs though, it doesn't delete them.
This is fine for now, but if we don't also offer a delete run API, the All runs view will be cluttered with all the orphaned runs.
@ajayalfred any thoughts here? What do you think of this as a starting point?
Since runs are organized inside experiments, a user would expect deleting an experiment to delete the runs contained as well. I don't see a usecase where the user will expect this to behave otherwise and produce orphaned runs.
A delete run API is still needed for the case when a user wants to delete an individual run/set of runs inside an experiment.
Does that make sense?
Yup, that's what I'd prefer to do too.
So I guess we'll block this on support for deleting an experiment's runs from the backend.
cc/ @IronPan.
Ho folks - any progress on this one?
I don't think we have anyone working on this, feel free to pick up the discussion if you feel like it.
There is however some work I'm doing now to support archiving runs, which should help declutter the workspace. Experiments are still create-only though.
+1 on this, in multi-user settings experiments will pile up quickly
+1 on this, in multi-user settings experiments will pile up quickly
The solution to this is probably a default filter that only shows your own experiments.
@ark-kun maybe we should have labels based filter, people can add labels like owner, project,..
maybe we should have labels based filter, people can add labels like owner, project,..
I like this approach.
What if instead of deleting, we just had an option to archive, like with runs. There should also be an option to archive pipelines. That way no metadata is lost, but old experiments are out of the way
Sent with GitHawk
Why not allow archiving / deleting for experiments with no unarchived runs? Would be interested in this feature.
no metadata is lost, but old experiments are out of the wa
Would showing the experiments with recent runs first solve your issues? https://github.com/kubeflow/pipelines/issues/249
no metadata is lost, but old experiments are out of the wa
Would showing the experiments with recent runs first solve your issues? #249
Probably. I just want to hide the experiments that aren't relevant to me. The filter ideally should only show experiments you create by default too (in the situation where multiple scientists are sharing a cluster).
Closing this as a duplicate of https://github.com/kubeflow/pipelines/issues/1263
We plan to support archiving experiments in KFP 1.0
/close
@Bobgy: Closing this issue.
In response to this:
/close
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.
Most helpful comment
+1 to this being important.