Zipkin-style shared span vs two spans per RPC · Issue #102 · opentracing/opentracing.io · GitHub
There are two popular ways of modelling RPC calls in tracing: Zipkin's
model, where both the client and the server share one span, and the
model in which a span is exclusive to a single tracer, causing client
and server to have two separate spans.
OpenTracing, as far as I can tell, has no opinion on which of the two
models should be used. However, OT's goal is to allow swapping out
OpenTracing implementations with a simple configuration change and to
have standardized instrumentation for popular libraries like gRPC. The
two models, however, require different use of the OT interfaces and
thus different instrumentation, which means that implementations
aren't truly interchangeable.
Has this topic come up before? Is this something that OT feels
responsible for and that should be addressed? This issue further
extends to things like logs, e.g. Zipkin's cs, cr, sr and sr, that
would need to be set for some tracers but not others.
Read full article from Zipkin-style shared span vs two spans per RPC · Issue #102 · opentracing/opentracing.io · GitHub
No comments:
Post a Comment