We are using the Databricks managed MLflow to log experiment runs for quite some time already and never experienced issues. However, now we seem to have encountered a bug in the associated Databricks UI.

[332588]

We observe the following behavior when we keep adding new runs to an experiment:

- In the beginning, the runs are still displayed correctly in the UI.

- After a certain number of total runs, the following bug occurs in the UI:

   - In the UI, there are no longer any runs in the experiment. ("No runs yet.")

   - If you change the way of sorting the runs, some of the runs will be displayed again.

   - Then when you scroll down to look at the runs below, the UI breaks down. Instead of the experiment, you see the message: "Something went wrong".

- We have not yet been able to clearly determine the number of runs above which the UI breaks, but it seems to be in the low three-digit range.

- Regardless of the number of runs, they are correctly stored in the underlying storage, in our case an S3 bucket. Thus it seams to be an UI bug.

- Furthermore, you can still get the meta information from the last 100 runs using the "Download CSV" Button.

Is this behavior already known? Is a fix planned for this issue?

[Debayan]

Hi, It will be helpful if you make a video on it and post it here. Thanks in advance!

[332588]

Hi, please find the video ...

Thanks in advance!

[Debayan]

Hi @Timo Burmeister​ Apologies for the delay! I went through the video, does it happen all the time? I see after sorting it with different filter the list appears.