It got messy after ~1,000 tables.
Schema changes every week. Connectors restarting more often than expected. Backlog recovery during peak traffic could take hours.
CDC wasn’t the hard part. Managing the system around it was.
Curious how others handle this at scale.
Details here: https://medium.com/@guichenchen/why-we-replaced-debezium-kafka-in-our-large-scale-real-time-pipeline-84a06a07e53b