OpenTelemetry at Home: Traces for a Two-Service App

Thu, 25 Jul 2024 16:02:00 +0000

Metrics tell you a request was slow. Logs tell you what each service said while it was being slow. Neither tells you where the time went when a request passes through more than one service, and that gap is the whole reason distributed tracing exists.

You hit this the moment you split an app in two. Say you have a small front-end API that takes a request and calls a back-end service to do the real work. A user reports the thing is slow. Your metrics show the front-end’s p95 latency climbing. Your logs show the front-end received the request and, some time later, returned a response. Was the delay in the front-end’s own code? In the network hop? In the back-end? In something the back-end called? You are back to correlating timestamps across two log streams by eye, which is exactly the archaeology tracing abolishes.

Tracing - vo.rs

OpenTelemetry at Home: Traces for a Two-Service App