if a node in my cluster goes down, some requests fail

[chadkouse]

should sherlock automatically use another node in this case?

[polyfractal]

It should, in the sense that removing dead nodes and retrying requests is the desired behavior. Unfortunately, Sherlock currently does not remove dead nodes, and there is no built-in retrying mechanism. It doesn't properly round-robin either...just picks random nodes.

Leave this issue open, since its a pretty important feature that is missing.