Crate postrust_worker 

Postrust’s REST API in a Cloudflare Worker, reaching PostgreSQL through Hyperdrive.

The request pipeline is postrust-engine, the same one the server runs, and the database driver is postrust_core::db::PgSocketDb: the PostgreSQL wire protocol over a worker::Socket to Hyperdrive. That driver is held to answering exactly as the server’s sqlx driver does. Every request runs in a transaction that sets the token’s role and claims, as on the server, so row-level security applies the same way.

What is different here, and why:

• One connection per request. A Worker may not share an I/O object between requests. Hyperdrive pools connections on its side, so opening one costs little.
• The schema cache is kept in Workers KV (POSTRUST_SCHEMA_CACHE), so a cold isolate reads it rather than querying the catalog. It is loaded live on a miss, refreshed by a Cron Trigger or by an authenticated POST /_/reload, and re-read from KV once PGRST_WORKER_SCHEMA_TTL seconds (default 60) have passed. The KV copy expires after the TTL (at least 60 seconds), so a migration is seen without a reload too.
• Rows are capped at PGRST_DB_MAX_ROWS, 1000 unless set, because an isolate has 128 MB. unlimited, under that name or PGRST_MAX_ROWS, lifts the cap.
• REST only. No GraphQL, no subscriptions, no LISTEN.

The REST API answers at the root, as PostgREST does: /todos, /rpc/f. Configuration is the server’s own variables (PGRST_DB_SCHEMAS, PGRST_DB_ANON_ROLE, PGRST_JWT_SECRET, …), set as Worker variables or secrets. See docs/workers.md.

Its semver covers what a deployment sees, not Rust items: see docs/stability.md.