Metadata: SQLite and Postgres

The stores

On Postgres

import pg from 'pg';
import { postgresMetadata } from '@rebofiles/server';

metadata: await postgresMetadata(new pg.Pool({ connectionString: process.env.DATABASE_URL })),

It takes anything with query(text, values) that answers { rows }: pg, Neon’s Pool, PGlite. Given a pool, each change holds a lock of the database while it reads and writes, so two servers never change the drive at once; without one, two folders made at the same moment could both be “(2)”. { table: 'acme_files' } for a table of another name.

What is kept

• The entries, one row each, as JSON.
• The settings of the drive and of each person, and the passwords of links as scrypt makes them.
• The activity, a row a line, found by the entry and the folders it is about.
• The text of each file: SQLite’s full-text search, or a tsvector in Postgres.
• A revision that goes up with every change, so each process knows when to read the drive again.

A store of your own

MetadataStore is load() and save(changes); revision, lock, settings, record/activity and keepText/findText are optional, and without them the server keeps those in the process. Database schema has every table.

A piece at a time

SQLite, Postgres and the store in memory answer what each request needs, and no more: an entry and the folders above it, what is in a folder, what is under one, a search by name, the trash, what is shared with someone. The drive is never loaded whole, and a change writes only what it touched. What another process changed is read from the activity they keep.

The columns and tables it is asked by are added to a database made before them the first time it is opened, and the rows filled once. A store of your own does the same with scoped, and without it, it is held whole, as before.