A pipeline consists of a sequence of structured operations. Multiple operations can sit in a pipeline, and individual items can pass through the pipeline. Understanding a pipeline, then, implies understanding both every step of the way, and understanding the end-to-end throughput. Setting up observability for a pipeline, then, implies observing the end-to-end progress of items; it also implies observing each service individually (see Service Calls, above).
The top of the stack are the entry point and exit point from the pipeline, as well as any service boundaries in between.
The request, as it moves through the pipeline.
Now that the outer stages of the pipeline are instrumented, try instrumenting individual services within it. Start with more complex services or stages that are more likely to fail. If the observation can maintain a request ID through the pipeline ,then this is an ideal candidate for the Tracing API, because it can be possible to trace a request all the way from high level to low-level internal functions.