Database · C++
SubsecondDb
Postgres scaffoldSubsecondDb is a hot-to-archive database project built around SubsecondId. The Postgres extension is currently a scaffold. GPU scheduler source and seam tests are available; a separate hot KV/auth tier and integrated GPU vector search are later phases of the architecture.
github.com/AStarStarship/… →Specification
| Components | Postgres C-extension scaffold · C++ GPU scheduler |
|---|---|
| Identity | 64-bit SubsecondId as the primary key |
| Target tiers | Hot in-memory KV · GPU vector · Postgres archive |
Design & implementation
Why hot-to-archive
The architecture is intended to serve recent content from a hot tier and keep older records in a durable archive. Scheduled movement between tiers is a design goal, not a deployed storage pipeline claimed here.
Phase 1 — Postgres + SubsecondId
Pg/ contains the build scaffold and seam harness. A native SubsecondId SQL type, compaction function, and collision ledger are extension build targets; the public README does not mark the extension implementation as complete.
Phase 2 — Auth container + IPC
A separate in-memory auth and KV module is planned, with IPC to the Postgres tier. This phase is intentionally compartmentalized from the archive extension and is not yet built.
Phase 3 — CUDA-accelerated vector DB
Gpu/ contains scheduler source and seam tests. Integrating that component with a vector index and CUDA runners is a later phase; the scheduler's presence does not establish a complete GPU database.
Related