Confluent Cloud for Flink

ksqlDB and Apache Flink runs Apache Flink 129 yourself: a JobManager, TaskManagers, checkpoint storage, connector JARs and SQL DDL that declares each Kafka 129 topic as a table. Confluent Cloud 17,245 for Apache Flink removes all of that. It is a regional, serverless service in which the catalog is already filled in:

So booknest.order-events, with its Avro 129 schema in Stream Governance, is immediately a table you can SELECT from, filter, window and join; and a CREATE TABLE in Flink creates a topic and registers its schema. You write Flink SQL in the console or CLI, or use the Table API from Java or Python, and can upload user-defined functions. Statements run in a compute pool, a set of resources that scales between zero and a maximum you set.

Billing counts Confluent Flink Units (CFUs) actually used per minute, not the pool's size: $0.21 per CFU-hour ($0.0035 per CFU-minute) at the base rate, with a minimum of one CFU-minute per statement. Confluent's own example: five short queries consuming five CFU-minutes cost $0.0175. A continuous streaming statement, such as live sales per genre feeding a dashboard, runs around the clock: at one CFU that is about $153 a month (730 hours times $0.21), before the Kafka network charges for what it reads and writes. Capping the pool's maximum is the main cost control, since you cannot pin the CFUs of an individual statement.

What you give up compared with ksqlDB and Apache Flink is control: no choice of Flink version or state backend, no DataStream API (only SQL, the Table API and user-defined functions), and statements whose tables are the topics of your Confluent clusters. What you gain is no checkpoints to tune, jobs to upgrade or cluster to size, usually the better trade when the processing is SQL over topics already in Confluent Cloud.