Allow users to modify the state of error datasets

[natefoo]

Sometimes it'd be useful to be able to operate on a dataset that Galaxy considers to be in error - especially if the user is confident that the error state was only due to an errant tool message or framework bug that can be worked around. This capability should probably be accompanied with a warning.

[nsoranzo]

💯

[hexylena]

xref #3779

[hexylena]

5/6 worked fine. The 6th invalidated all other uploads.

[yvanlebras]

Hi Nate, Nicola, Helena, is there some manner to fix this kind of thing now ? With @ColineRoyaux we have some datasets with an error state but the content is ok and we search a way to use it...

[hexylena]

You can always modify the database records, but, outside of that, no way for people to do it.

[mvdbeek]

due to an errant tool message or framework bug that can be worked around

I think the ambition is always to fix those and make the framework better. It's very complex to assert that a dataset is erroneously in error (is it truncated ? is there an unexpected message in the stderr ? did the tool swallow an error but set the exit code to non-zero ? did the scheduler oom message mean the output is incomplete ? other weird stuff that lead to a dataset that looks almost correct but isn't), and that overwriting the state is correct and won't lead to wrong results. I'd rather have someone waste time and compute on re-doing what they were doing than publish a wrong result. If re-doing what you were doing works, great! If not, there's a bug that needs to be fixed. @hexylena's example for instance lead to re-working the upload mechanism so that now you can have just 1 dataset fail instead of all.

My 2c is that this should be a wontfix and we should close the issue, if you are confident the data is actually fine you can change the state in the database.