Google Pub/Sub vs Kafka

Google Cloud Pub/Sub and Apache Kafka both enable asynchronous messaging and event streaming, but their operational models are fundamentally different. Pub/Sub is a fully managed, serverless messaging service. Kafka is a self-managed distributed streaming platform.

Architecture Differences

Pub/Sub abstracts away all infrastructure. You create topics and subscriptions. Google handles partitioning, replication, scaling, and failover. There are no brokers to manage, no partitions to count, and no rebalancing to worry about.

Kafka gives you full control over partitions, replication factors, broker configurations, and retention policies. This control enables optimization but requires expertise.

Ordering and Delivery

Pub/Sub supports ordering via ordering keys — messages with the same key are delivered in order. Without ordering keys, messages may arrive out of order. Kafka guarantees ordering within partitions, and partition assignment is deterministic.

Pub/Sub delivers messages at least once by default. With Dataflow, you can achieve exactly-once processing. Kafka supports exactly-once semantics natively with idempotent producers and transactional writes.

Retention and Replay

Pub/Sub retains acknowledged messages for 7 days by default (up to 31 days). You can seek subscriptions to a timestamp for replay. Kafka retains messages based on time or size policies and supports indefinite retention — consumers can seek to any offset in the log.

For event sourcing patterns that require replaying years of history, Kafka's unlimited retention is essential. For shorter retention needs, Pub/Sub's managed approach is simpler.

The Verdict

Pub/Sub wins on operational simplicity and automatic scaling. Kafka wins on control, ecosystem, and long-term retention. In system design interviews, choose Pub/Sub for GCP architectures and Kafka for multi-cloud or stream-processing-heavy designs. Check our concepts guide for messaging fundamentals.