Failure and input readiness entangled

[liamhuber]

When I have a failed node and try to run again, I get the error message:

The following node require inputs before you can run the graph:

Since I broke the node on purpose, I know which one it is and I can go hit the "reset" button (🚀 🚀 🚀 ), re-run, and everything is fine. I'm actually quite happy that running when there is a failed child is triggering a sort of controlled error message, but the message itself just needs to disentangle missing-input and actually-failed nodes and list them all. Direct users to the "reset" button right in the message would probably also be worthwhile.

[Tara-Lakshmipathy]

Should the user be pointed to the reset button, or should the reset be done automatically?

[liamhuber]

The user should be alerted to the failed node and informed how to resolve it; if the presence of a failure is unmissable on the graph level, I would be a bit more relaxed about automating it, but as it stands the colour coding is not working consistently enough. What I find critical to avoid is that the user keeps re-running a graph and it keeps failing and they never receive a clear message about where and why it's failing.