Pipeline Error Propagation, Handling & Suppression

[JoeHegarty]

There are some complex scenarios we need to handle with the pipeline error system.

The main cases are:

• Outgoing requests implicitly have an error handler (the caller) who is expected to wait on the response, so in this case Orbit should suppress the error and just propagate to the caller. There is some additional complexity around ResponseTracking handling a deeper error (say in serialization) that does not come back as a response.

• Outgoing one way requests have an implicit handler until they reach ResponseTracking, at which point we complete the promise. Before this step Orbit should suppress and propagate the error. After that point it should treat the error as unhandled.

• Incoming requests are implicitly unhandled until they have been received by ResponseTracking which registers a listener to propagate the error back to the calling node as a response. Before ResponseTracking they should be treated as unhandled, afterwards they should be suppressed and propagated.

• Incoming one way requests are unhandled entirely and should always be treated as such.