// the take
XXL-MQ is a DB-backed distributed message queue from the xxl-job author, aimed at teams who want Kafka/RocketMQ-style patterns (parallel, serial, sharded, broadcast, delayed, retry messages) without standing up a dedicated MQ cluster. It's a fit for small-to-mid Java shops already running MySQL or TiDB who want message queuing as an extra workload on existing infrastructure rather than a new system to operate.
Message routing is genuinely varied and well thought out — sharded messages route by hash to the same consumer for FIFO-per-key while still parallelizing across shards, which is a real pattern (e.g. per-phone-number SMS ordering) that most lightweight queues don't bother supporting. The broker is stateless with storage separated out, so horizontal scaling and failover don't require coordinating cluster state yourself. It exposes a RESTful OpenAPI, so producers/consumers aren't locked to the JVM. Comes from the same author as xxl-job, so the registration/discovery and failover mechanics are patterns already proven under production load elsewhere.
It's GPLv3 — copyleft licensing that will be an immediate no-go for a lot of companies that would otherwise happily adopt Kafka or RabbitMQ (Apache/MPL licensed), so check your legal team before you build on this. Message storage is DB-backed (MySQL/TiDB), meaning queue throughput is ultimately bounded by your database's write capacity and row-churn tolerance — the advertised 100k-500k TPS depends heavily on your DB tier, and this is architecturally a different beast from log-structured (Kafka) or in-memory-first (RabbitMQ) brokers at real scale. Documentation is effectively Chinese-only (the linked docs site has no English version despite the bilingual README), which will slow down adoption and troubleshooting for non-Chinese-speaking teams. Community size is modest (493 stars, no visible plugin ecosystem), so you're more on your own for edge cases than you would be with RocketMQ or Pulsar.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →