NSD vs TypeDB

TypeDB is releasing frequently — ten versions between late April and mid-July 2026 — with release notes that explain the reasoning behind each change rather than just listing it. The recent work concentrates on transaction machinery and server operability: a deadlock in commits exceeding the RocksDB write buffer limit, transaction open latency, guarantees that the transaction operator actually closes what it was asked to close, exposed RocksDB memory tuning, and separated listen and advertise addresses. A clustered alpha appeared alongside the community line in May, and 3.11.5 raised the driver compatibility floor, rejecting connections from anything older than 3.11.0.

The distributed story explains most of what looks like unrelated plumbing. Letting callers supply UUIDs for users and credentials exists so cluster members can share identical state instead of each generating its own; splitting listen from advertise address exists so a node can be reachable at an address it does not bind; making transaction close synchronous rather than fire-and-forget matters once another node depends on the result. Single-node correctness fixes are being made in the shape that a clustered deployment will need.

Expect the clustered line to keep pulling extension points into the community server, and the optional-return warnings introduced in 3.10.3 to escalate into compilation errors as the notes promise. Whether clustering lands in the community edition or only in the enterprise distribution is not stated here.