Project Reactor documentation suggests the following pattern for MDC logging:
.doOnEach(logOnNext(r -> LOG.debug("found restaurant {} for ${}", r.getName(), r.getPricePerPerson())))
To avoid having to wrap each logging call, would a custom subscriber, populating the MDC from the currentContext before each signal, added using Hooks.onEachOperator(Operators.lift(...)) be a good idea?
My main conerns are:
1.) The cost of populating the MDC before every signal, even if there is no logging happening.
2.) Operator fusion: Does Operators.lift(...) on each operator effectively disable operator fusion? Attempting a quick test with StepVerifier#expectFusion seems to indicate that. If this is true, how much of a performance hit is this, in practice?
Any input is appreciated!
This is the approach that was initially taken by Sleuth, via Hooks.onEachOperator. However, this is very costly and probably not worth it if you only need logging/MDC on a subset of operations in you reactive pipeline. Not to mention that this approach not only impacts reactive steps defined by you, but also any other library / framework.
The recommendation is there for a reason: better control, less impact and a more explicit approach.