Introduction
UltraFabric is a new distributed database: PostgreSQL-compatible on the outside, with its own storage engine, query plan language and security verifier on the inside.
UltraFabric is in development. A first SQL subset runs over the PostgreSQL protocol in our tests, but there is no public build or hosted database yet. These docs describe what exists today and are marked wherever they describe what is planned. Status and roadmap has the details.
What it is for
UltraFabric is designed for multi-tenant applications: one database serving many workspaces, teams or customers, where a row leaking from one tenant to another is the worst bug there is. Three ideas shape it:
- Isolation is proved, not assumed. Every query becomes a plan, and every plan must show the data node that it isolates the tenant, applies the access policy and hides restricted fields before it can run.
- Low, predictable resource use. The engine keeps memory inside an explicit budget and pushes back on writers instead of stalling or running out of memory.
- Correctness first. Every component is tested against a simple reference model, with crashes injected, before it is made fast.
How these docs are organised
- Status and roadmap: what works, what is next, and when hosted databases open.
- Architecture: the layers from disk to SQL.
- Storage engine: how data is written, kept and read.
- Ultra IR: the query plan language every query passes through.
- Security model: what the verifier checks and why.
- SQL and psql: connecting with a token and the SQL that works today.
- Node protocol v0: the local binary protocol of the
ufdserver. - How we test: crash campaigns, differential tests, model checking and benchmarks.