Consuming messages from multiple sources in a scalable service sharing the same Scylla cluster can be problematic, especially when those sources provide data from two queues each and cannot deliver guaranteed in-order messages. At Zillow, we provide the Scylla write timestamp and employ a couple other tricks at Zillow to provide correct and consistent data to our consuming services and avoid doing transactions. Additionally, our data model is a simple compound key that contains a large binary Avro document.
