Dimster is a new open-source performance benchmarking tool for Apache Kafka, built by a distributed systems engineer with a background in RabbitMQ and Pulsar tooling. Its core concept is 'dimensional testing' — systematically varying one or more configuration parameters (batch.size, linger.ms, consumer type, partition count, etc.) across a series of identical benchmark runs to isolate their performance impact. Key features include: shareable self-contained result packages (JSON, CSV, interactive charts, Grafana dashboards as HTML), four test modes (run, explore, drain-backlog, correctness), a resources command for pre-benchmark capacity planning, a compare command for cross-run analysis, and Kubernetes as the primary runtime (supporting k3d, minikube, EKS, GKE). The tool is demonstrated with real benchmark examples comparing classic consumer groups vs. Kafka 4.x share groups across varying throughput levels and CPU allocations. It is currently at version 0.x and available on GitHub.
