Better Error Handling for Serverless

[timid leaf]

Serverless error handling is not good enough:

1. If the function fails during the importing of the file rather than the function call
2. if the file doesn't call the runpod.start() it just creates more workers to try to respond, instead of just crashing.
3. Just in general, sometimes I've seen it rarely try to spin up 2 workers for a single request, even if I call runpod.start(); maybe something begins to hang so it tries a new worker?

But all of these, I have no way to check / capture the error easily.

[copper wave]

(3) should be fixed

[opal cipher]

@plain shore I have seen similar feedback from others

[copper wave]

fix for (3) was pushed on 1 Feb and I have not seen it since, I was seeing it a lot before that