Got an error in the experiments table even when dvc.yaml seems to be correct

[shcheklein]

I could have had a merge conflict in dvc.lock a while ago:

[version: 1.0.14, 2023-07-08T23:11:46.912Z, pid: 9232] > /usr/local/bin/python -m dvc data status --granular --unchanged --json - FAILED with code 255 (652ms)
'./dvc.lock' is invalid (possible merge conflicts).



While scanning a simple key, in line 12, column 1

  12 <<<<<<< Updated upstream                                                   



Could not find expected ':', in line 13, column 10

  13 │     md5: acc863c0b1d50f969eba8171c96d0c0c                                

But since then, there were multiple successful dvc status calls:

[version: 1.0.14, 2023-07-08T23:50:51.459Z, pid: 24324] > /usr/local/bin/python -m dvc data status --granular --unchanged --json - COMPLETED (4282ms)

[mattseddon]

I think that message will be generated as a result of dvc stage list. What is the output of that command?

[shcheklein]

It's too late, can'd grab it now (instance reloaded).

I didn't see any errors related to it AFAIR.

In this case btw there was a dvc.yaml, I'm not sure why would we show the button at all.

After reset, renew, etc - I can't reproduce anymore. Might be related to all the updates tbh.

[mattseddon]

In this case btw there was a dvc.yaml, I'm not sure why would we show the button at all.

If the dvc.yaml is invalid then dvc stage list will return an error. Under those circumstances, we show the button because trying to run an experiment will result in an error. We decided to always show the button because we prioritized onboarding.

When the dvc.yaml becomes valid the button should disappear:

Screen.Recording.2023-07-10.at.8.56.57.am.mov

Given the feedback in #4242, the fact that we can't decide on the copy (#4225) and this issue we should consider removing the button altogether from the table. We could replace it with a Pipeline section on the Setup page. This would also give us somewhere to put a "Show DAG" button.

[shcheklein]

This was primarily about that state being stuck. I'm closing this since there were too many update at once even if we had some issue there- I think we can tolerate it, wait until we reproduce this reliably.