Rethinking Data Ingestion as a DAG (falconer.com)

Once you decomposed the pipeline into stages, did observability become significantly easier? I imagine stage-level latency and queue depth give you a much clearer signal than profiling a single async service.