Files
TREK/server/scripts/generate-pg-baseline.ts
T
Isaias Tavares a77705984e feat(db): create and migrate the schema on a Postgres database
With TREK_DB_URL set, TREK now boots on Postgres; without it nothing changes.

- The database engine contract gains open, close, reopen and the schema
  bootstrap; DatabaseLifecycle delegates to the engine resolveEngine() picks.
- An empty Postgres database gets server/assets/postgres/baseline.sql
  (generated from the entities by gen:pg-baseline, with the reference rows
  the migrations up to its cut insert: tour types, provider catalogues,
  addons, app settings) and every migration up to its cut recorded as
  executed, in one transaction. A database holding tables
  without TREK's history is refused before anything is written; a migrated one
  runs the newer-database refusal and its pending migrations, which refuse
  without a copy unless TREK_DB_PRE_MIGRATE_SNAPSHOT=false.
- INTEGER columns are bigint on Postgres (SQLite's INTEGER is 64-bit: an
  epoch-millisecond column overflowed int4), and the engine's driver reads
  bigint and numeric back as numbers, refusing values past 2^53.
- The two SQLite triggers are ported to PL/pgSQL; check:pg-baseline fails
  while a SQLite trigger has no port, and while the baseline drifts from the
  entities before a later migration freezes it.
- Backups and restores on Postgres refuse until they can dump and load one.
- Migration helpers read the catalog per engine (dialect/catalog.ts), and
  the email case index no longer needs group_concat.
- CI runs check:pg-baseline and boots the image on postgres:17 twice.
2026-11-04 22:45:02 -03:00

183 lines
8.7 KiB
TypeScript

/**
* Generates the Postgres baseline: the schema a fresh Postgres database starts
* from (src/db/engine/postgres/baseline.ts says how a boot uses it).
*
* The entities describe the migrated SQLite schema and are pinned to it
* (`check:entities`), so the baseline is MikroORM's schema generator over the
* entities, for the Postgres platform, after the adjustments every Postgres
* schema in this repository shares (src/db/engine/postgres/schema-adjust.ts):
* TEXT timestamps, portable defaults, the `nocase` collation. Unlike the
* probe's schema it keeps the foreign keys. No database is needed: the
* generator reads only the metadata and the platform.
*
* The baseline stands for every migration up to the newest one when it was
* generated (`through`). Once a migration lands after it, the baseline is
* frozen like a shipped migration: the newer migration runs on Postgres
* through the migrator, and `--check` only verifies the manifest.
*
* Usage:
* node --import tsx scripts/generate-pg-baseline.ts # rewrite server/assets/postgres/
* node --import tsx scripts/generate-pg-baseline.ts --check # exit 1 when it drifted (CI)
*/
import { baselineMatches, readPostgresBaseline, writePostgresBaseline, type PostgresBaseline } from '../src/db/engine/postgres/baseline';
import {
adjustColumns,
NOCASE_COLLATION_SQL,
POSTGRES_TRIGGERS,
splitStatements,
type ColumnMeta,
} from '../src/db/engine/postgres/schema-adjust';
import { ALL_ENTITIES } from '../src/db/entities';
import { knownMigrationNames } from '../src/db/known-migrations';
import type { EntityProperty } from '@mikro-orm/core';
import { MikroORM } from '@mikro-orm/postgresql';
import { migrateTempDb, tempDbPath } from './generate-entities';
import Database from 'better-sqlite3';
import fs from 'node:fs';
/** The baseline the entities and the migrations on disk describe now. */
export async function generatePostgresBaseline(known: Set<string> = knownMigrationNames()): Promise<PostgresBaseline> {
const migrations = [...known].sort();
const through = migrations.at(-1);
if (!through) throw new Error('no migrations found');
const orm = await MikroORM.init({
// Never contacted: init() only discovers metadata, and the schema generator needs nothing else.
clientUrl: 'postgresql://generator@localhost/trek',
entities: [...ALL_ENTITIES],
discovery: { warnWhenNoEntities: false },
});
try {
for (const meta of orm.getMetadata().getAll().values()) adjustColumns(meta.props as EntityProperty[] as ColumnMeta[]);
const migrated = await readMigratedSqlite();
const statements = [
NOCASE_COLLATION_SQL,
...splitStatements(await orm.schema.getCreateSchemaSQL({ wrap: false })),
...migrated.tables.flatMap((table) => rowStatements(table.name, table.rows)),
...Object.values(POSTGRES_TRIGGERS).flat(),
];
return { through, migrations, statements };
} finally {
await orm.close();
}
}
/**
* What `--check` holds. While no migration is newer than the committed
* baseline, the baseline must be exactly what the entities generate. Once one
* is, the baseline is frozen and only its manifest is checked: every migration
* it names exists, and it names every migration up to `through`.
*/
export function checkBaseline(committed: PostgresBaseline, generated: PostgresBaseline, matches: boolean): string[] {
const known = new Set(generated.migrations);
const problems: string[] = [];
const missing = committed.migrations.filter((name) => !known.has(name));
if (missing.length > 0) problems.push(`the baseline names migrations that do not exist: ${missing.join(', ')}`);
const unlisted = generated.migrations.filter((name) => name <= committed.through && !committed.migrations.includes(name));
if (unlisted.length > 0) problems.push(`migrations up to ${committed.through} are missing from it: ${unlisted.join(', ')}`);
if (committed.through === generated.through && !matches) {
problems.push('the entities generate a different schema; run npm run gen:pg-baseline');
}
return problems;
}
/** Migration bookkeeping: the baseline's manifest stands for it on Postgres. */
const BOOKKEEPING_TABLES = new Set(['mikro_orm_migrations', 'migrations', 'schema_version']);
type Row = Record<string, unknown>;
/** What the migrations leave in a SQLite database: its triggers, and the rows they insert, table by table in creation order. */
async function readMigratedSqlite(): Promise<{ triggers: string[]; tables: Array<{ name: string; rows: Row[] }> }> {
const dbPath = tempDbPath();
try {
await migrateTempDb(dbPath);
const conn = new Database(dbPath, { readonly: true });
try {
const objects = conn
.prepare("select type, name from sqlite_master where type in ('table', 'trigger') and name not like 'sqlite_%' order by rowid")
.all() as Array<{ type: string; name: string }>;
const tables = objects
.filter((object) => object.type === 'table' && !BOOKKEEPING_TABLES.has(object.name))
.map((object) => ({ name: object.name, rows: conn.prepare(`select * from "${object.name}" order by rowid`).all() as Row[] }))
.filter((table) => table.rows.length > 0);
const triggers = objects.filter((object) => object.type === 'trigger').map((object) => object.name).sort();
return { triggers, tables };
} finally {
conn.close();
}
} finally {
fs.rmSync(dbPath, { force: true });
}
}
/** A value as a Postgres literal the column's own type takes in: quoted text, which Postgres casts on insert. */
function literal(value: unknown): string {
if (value === null || value === undefined) return 'NULL';
if (Buffer.isBuffer(value)) throw new Error('a migration inserted binary data, which the baseline cannot carry');
return `'${String(value).replace(/'/g, "''")}'`;
}
/**
* The rows a migration inserted into `table` (reference data such as the tour
* types and the provider catalogues), as the statements that put them in a
* Postgres database the baseline creates, then move the table's id sequence,
* if it has one, past them. A migration up to the cut never runs on Postgres,
* so without these the database would start without what every SQLite
* install has before its first seeder.
*/
export function rowStatements(table: string, rows: readonly Row[]): string[] {
if (rows.length === 0) return [];
const columns = Object.keys(rows[0]);
const values = rows.map((row) => `(${columns.map((column) => literal(row[column])).join(', ')})`);
const statements = [`insert into "${table}" (${columns.map((c) => `"${c}"`).join(', ')}) values ${values.join(', ')}`];
// A text key (addons.id) has no sequence; a numeric one may.
if (rows.every((row) => typeof row.id === 'number')) {
statements.push(
`select setval(seq, greatest(max_id, 1)) from (select pg_get_serial_sequence('"${table}"', 'id') as seq, ` +
`max("id") as max_id from "${table}") ids where seq is not null`,
);
}
return statements;
}
/**
* The SQLite triggers without a Postgres port in `POSTGRES_TRIGGERS`, and the
* ports whose trigger is gone. The entities carry no triggers, so this is
* what keeps the hand-written ports level with the migrations.
*/
export function triggerPortProblems(sqliteTriggers: readonly string[], ported: readonly string[]): string[] {
const problems: string[] = [];
const missing = sqliteTriggers.filter((name) => !ported.includes(name));
if (missing.length > 0) problems.push(`SQLite triggers without a Postgres port in POSTGRES_TRIGGERS: ${missing.join(', ')}`);
const stale = ported.filter((name) => !sqliteTriggers.includes(name));
if (stale.length > 0) problems.push(`POSTGRES_TRIGGERS ports triggers the migrations no longer create: ${stale.join(', ')}`);
return problems;
}
async function main(): Promise<void> {
const generated = await generatePostgresBaseline();
if (!process.argv.slice(2).includes('--check')) {
writePostgresBaseline(generated);
console.log(`[generate-pg-baseline] wrote ${generated.statements.length} statements through ${generated.through}`);
return;
}
const committed = readPostgresBaseline();
const problems = checkBaseline(committed, generated, baselineMatches(generated));
// Once frozen, a later migration's trigger reaches Postgres through the migrator, not this list.
if (committed.through === generated.through) {
problems.push(...triggerPortProblems((await readMigratedSqlite()).triggers, Object.keys(POSTGRES_TRIGGERS)));
}
if (problems.length > 0) {
for (const problem of problems) console.error(`[generate-pg-baseline] ${problem}`);
process.exit(1);
}
console.log('[generate-pg-baseline] --check: the Postgres baseline matches the entities and the migrations');
}
if (require.main === module) {
main().catch((error: unknown) => {
console.error(error);
process.exit(1);
});
}