// the find
AutoMQ/automq
Diskless Kafka® on S3. 10x Cost-Effective. No Cross-AZ Traffic Cost. Autoscale in seconds. Single-digit ms latency. Multi-AZ Availability.
AutoMQ is a Kafka fork that replaces local disk storage with S3 (or S3-compatible object storage), making brokers stateless. The value proposition is real: no cross-AZ replication traffic, auto-scaling in seconds instead of hours, and dramatically lower storage costs versus EBS-backed Kafka. It's for teams running Kafka on AWS/GCP who are tired of paying cross-AZ bills and hand-tuning partition rebalancing.
The architecture is genuinely clever — by pushing durability to S3 and using a WAL for low-latency writes, they get the cost benefits of object storage without catastrophic tail latency on every produce. The built-in Auto Balancer replacing Cruise Control is a meaningful operational improvement; partition reassignment in vanilla Kafka is a pain most teams underestimate. Native OpenTelemetry/Prometheus metrics export (instead of JMX) is a quiet upgrade that makes it drop-in compatible with modern observability stacks. 100% Kafka protocol compatibility means zero client-side changes — you swap the cluster, not your code.
Single-digit millisecond latency is enterprise-edition only; the open-source version defaults to S3 directly, which means hundreds-of-milliseconds p99 on produces — fine for batch, problematic for latency-sensitive workloads. The README buries this distinction, which will bite someone benchmarking it. Production deployment is explicitly discouraged in the quick-start, and the 'contact us' CTA for production guidance is a red flag: docs for self-hosted production clusters are thin. The Table Topic / Iceberg integration is promising but clearly very new, and mixing streaming and table semantics in one system adds operational complexity that the documentation doesn't adequately address.