yosh

Integration with Go's context library is amazing (for cancellation, deadlines, request-scoped info), but node impl is OK too

it ticks off so many of the boxes I had for "so, here's how I want to do RPC in large distributed systems"

including a couple that nobody else seemed to have considered, both on and off my own lists

you actually have a list of this? keen to see what your considerations are - doing a talk like in a couple of days hah

sorry, haven't quite got something I can readily publish

a good chunk of this is just stuff that's been brewing in my head for a long time and hasn't really been written down

canonical error codes are a must (https://godoc.org/google.golang.org/grpc/codes …), (recursive) deadlines+explicit cancellation are a must for QoS

being able to change the names of things without changing encoding (compactness is a bonus, but decoupling matters more)