// the find
polardb/PolarDB-for-PostgreSQL
A cloud-native database based on PostgreSQL developed by Alibaba Cloud.
A fork of PostgreSQL 17 that separates compute from storage. Read-only nodes share one copy of the data on shared storage, and LogIndex keeps their in-memory pages consistent with the primary's WAL. It is for teams that want that architecture, mostly on Alibaba Cloud, not for anyone who just needs a single Postgres instance.
LogIndex is a concrete design, not a slogan: the primary ships only WAL metadata to replicas, and replicas replay either lazily or in parallel. The architecture section also names the hard part, the dirty-page flush rule that stops read-only nodes from reading pages from the future, which is worth reading even if you never deploy it. It is still a PostgreSQL fork with the upstream tree layout intact, and the sync-postgres workflow suggests upstream merges are routine rather than a one-off drop. The license situation is clear: Apache 2.0 for the PolarDB changes, the PostgreSQL License for the base, and a NOTICE file for third-party code.
The quick start is a single-node Docker image on local storage, which is not the architecture the project is about. The shared-storage deployment docs live on a separate site, and this README does not say what storage layer the open-source build expects. The 'millisecond-level latency' and HTAP claims come without benchmarks, and the MPP execution for large OLAP queries gets one bullet with nothing on cost or scaling limits. Every PostgreSQL minor release becomes a merge you own, and by choosing a fork you take on the lag and divergence from upstream.