I am trying to export-import mlflow objects from a deprecated databricks workspace to a new one. Thereby, I figured out following miss behavior when using local setup:
- export-model (from deprecated ws) works without any warning or error
- import-model (to new ws) works without any warning or error
- I see the model in the registry of the the new ws
- I see the experiment the model stems from in the new ws
- clicking the migrated experiment results in a null pointer
- clicking the run of the migrated model version in registry results in a null pointer
See also below screenshots.
Do you have an idea why that happens? To me it seems that the imported (meta-) data are not complete and the run might be not available in the migrated object.
Doing the same with objects only involving the same (new) ws seems to work fine. I cannot test it with the notebooks since I don't have a mounted storage that is shared by the two workspaces.
All mlflow objects that I used for my tests were generated with mlflow version 1.24.0 and on a databricks cluster with runtime 10.4 LTS ML.
I am trying to export-import mlflow objects from a deprecated databricks workspace to a new one. Thereby, I figured out following miss behavior when using local setup:
See also below screenshots.
Do you have an idea why that happens? To me it seems that the imported (meta-) data are not complete and the run might be not available in the migrated object.
Doing the same with objects only involving the same (new) ws seems to work fine. I cannot test it with the notebooks since I don't have a mounted storage that is shared by the two workspaces.
All mlflow objects that I used for my tests were generated with mlflow version 1.24.0 and on a databricks cluster with runtime 10.4 LTS ML.