// the find
redpanda-data/redpanda
Redpanda is a streaming data platform for developers. Kafka API compatible. 10x faster. No ZooKeeper. No JVM!
Redpanda is a Kafka-compatible event streaming broker written in C++ on top of the Seastar framework — the same thread-per-core async engine behind ScyllaDB. It drops ZooKeeper and the JVM entirely, shipping as a single binary with Raft-based consensus. Aimed at teams who want Kafka semantics but are tired of the operational overhead of the JVM ecosystem.
1. The Seastar foundation is the real story: thread-per-core with no shared state between shards means no GC pauses, no contention, and predictable tail latency at high throughput — not a rewrite-from-scratch gamble but a proven architecture. 2. No ZooKeeper is a genuine ops win: single binary, single process to monitor, Raft consensus built-in — cluster management goes from two systems to one. 3. Tiered storage offloads cold data to object storage transparently, which lets you keep infinite retention without proportionally infinite disk cost. 4. The rpk CLI (Go) is well-scoped and batteries-included: it handles tuning, topic management, WASM transforms, and cloud auth without forcing you into a separate web UI.
1. It ships under BSL (Business Source License), not Apache 2.0 — production use in a competing streaming service requires a commercial license, which is a real constraint many teams discover late. 2. Kafka compatibility is close but not complete: specific producer/consumer configs, transactional edge cases, and some admin API behaviors differ, and you will hit these eventually if you're porting a non-trivial Kafka workload. 3. Building from source is heavy: Bazel plus a wall of third-party deps plus a custom LLVM toolchain — plan on a fast machine and 20+ minutes; the prebuilt packages are fine but you're locked into their release cadence if you need a patch. 4. Seastar's share-nothing architecture makes profiling and debugging non-obvious if something goes wrong at the shard level — the mental model is different enough from standard multithreaded C++ that your team will need ramp-up time.