Redpanda

Use it when

• You want the Kafka API with less to operate: one C++ binary with built-in Raft, no JVM, no ZooKeeper, no separate coordination layer.
• Tail latency matters; the thread-per-core design targets consistent low latency under load.
• Existing Kafka clients, Connect pipelines, and stream processors must work unchanged.
• Hardware efficiency is a goal: fewer, smaller nodes for the same throughput, with tiered storage keeping local disks small.

Think twice when

• OSI-approved open source is a requirement; the core is source-available under BUSL, not open source in the OSI sense.
• You depend on the newest Kafka broker features immediately; compatibility tracks the protocol, not every capability the day it ships.
• Your platform team already runs Kafka well at scale; the migration must earn more than familiarity costs.

How it runs

A single binary per broker on local NVMe with Raft replication; add brokers and rebalance to scale, offload history to object storage with tiered storage. Redpanda Connect provides YAML-defined pipelines with hundreds of connectors, and WebAssembly transforms run inside the broker. Free to self-host; Redpanda Cloud bills data, partitions, and uptime.