Tracing works by inspecting and pattern matching on all of the messages being sent around the system, to inspect a subset of them.
The current stringly-typed messages are both wonderful and terrible to inspect, since they may contain serialized information that we don't know how to parse.
This is wonderful because tracers can be built very naively to match on very specific successful parsers, and a failure should just indicate that a message should be dropped, but it poses a problem when we want to discover what messages are really occupying the most of our bandwidth and causing trouble.
Tracing works by inspecting and pattern matching on all of the messages being sent around the system, to inspect a subset of them.
The current stringly-typed messages are both wonderful and terrible to inspect, since they may contain serialized information that we don't know how to parse.
This is wonderful because tracers can be built very naively to match on very specific successful parsers, and a failure should just indicate that a message should be dropped, but it poses a problem when we want to discover what messages are really occupying the most of our bandwidth and causing trouble.