// the find
linkedin/brooklin
An extensible distributed system for reliable nearline data streaming at scale
Brooklin is LinkedIn's internal Kafka mirroring and CDC system, open-sourced. It sits between data sources (Kafka clusters, databases) and destinations, with a pluggable connector/transport-provider model. The primary audience is platform engineers running multi-datacenter Kafka replication at scale — not teams evaluating it for a greenfield project.
The partition-level fault isolation for Kafka mirroring is genuinely useful: one bad partition doesn't stall the whole topic, which Kafka MirrorMaker 1 notoriously couldn't handle. The REST management API covering pause/resume at the individual partition level is operationally practical, not just CRUD. The connector abstraction is well-factored — source and destination are orthogonal, so you can wire MySQL CDC to Kafka or Kafka to a custom transport without touching the core. The repo is actively maintained (last push July 2026) and has real production mileage at LinkedIn scale.
The connector ecosystem outside Kafka is thin: MySQL CDC was listed as 'under development' in the README and there's no evidence it shipped; if you need CDC from Postgres or Oracle, you're writing a connector from scratch. Deployment story is LinkedIn-internal Zookeeper-based coordination, which means significant operational overhead to adopt outside their stack — no Kubernetes operator, no Helm chart, no managed hosting path. At under 1000 stars with a narrow contributor base, this is effectively LinkedIn infrastructure with a public license, not a community project; your bug reports will compete with LinkedIn's internal priorities. Alternatives like Kafka MirrorMaker 2 (which fixed the partition isolation problem) and Debezium for CDC are better-maintained with larger ecosystems.