ADR-0001: The Services Are Go, and the Inventory Is Kept in SQLite, with No cgo

Context

The inventory needs a store from its first slice, the owner’s iTunes library. It holds one owner’s music: a few thousand albums, not millions. It runs in EdS’s tenant, where EdS’s back end may run on one Pi or several (EdS ADR-0033), so whatever it is made of must run there.

The matching the first slice needs already exists, in Go, in EdS (internal/library, EdS ADR-0046), measured against the owner’s library and spoken requests.

Options considered

Go, with SQLite through a driver with no cgo (modernc.org/sqlite). One static binary for each service, built for a Pi from the Mac as EdS’s are. The database is one file, backed up by copying it. EdS’s matching comes across unchanged.

Go, with Postgres. A server to run, upgrade and back up beside the service, for a collection one file holds easily. Better when several writers work at once, which nothing here does yet.

Python, as EdS’s palette and scribe are. Good for work that leans on its libraries, as theirs does; nothing here does, and EdS’s matching would have to be ported.

Decision

The services are Go, one module for each, and the inventory is kept in SQLite, through modernc.org/sqlite, with no cgo anywhere.

• The schema is SQL in the store package, applied in steps and numbered in SQLite’s user_version, so a database is brought up to date when it is opened.
• Everything that touches the database is in internal/store. Nothing else writes SQL, so moving to Postgres, if it is ever needed, changes one package.
• Searching is done on a copy of the inventory held in memory, read again when a collection changes it.

Consequences

• The module needs Go 1.26, which the driver requires; EdS’s are on 1.25.
• One writer at a time. SQLite allows many readers and one writer; a collection is one transaction, and the server only reads. Several collectors writing at once would wait on one another.
• The database is a file on one machine. Where that file lives, and how it is backed up, is a question for where the inventory runs.
• Postgres is the step if SQLite is outgrown: several writers, or the inventory needed from more than one machine.

What would settle it

• The owner’s real iTunes library collected, and the inventory asked, on the machine it will run on, with times that are no concern.

Settled, 2026-10-24 (CHG-0001): the owner’s library, 3,665 albums by 1,912 artists, collected on EdS’s services workload in one transaction, and asked in 10 to 17 ms. The database is 1.4 MB.