Files
TREK/server/scripts/generate-pg-baseline.ts
T
Isaias Tavares 2e3762afe1 feat(db): run TREK on PostgreSQL
TREK runs on PostgreSQL when TREK_DB_URL is set, and on SQLite as before
otherwise: an engine port under src/db/engine, the dialect helpers under
src/db/dialect, a Postgres baseline for the migrations up to the cut,
pg_dump/pg_restore backups, a probe of every repository method against a
real server and a Postgres test lane in CI.
2026-11-17 23:03:33 +01:00

230 lines
9.9 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 { POSTGRES_TRIGGERS } from '../src/db/engine/postgres/postgres-triggers';
import { applyPostgresColumnAdjustments, NOCASE_COLLATION_SQL } from '../src/db/engine/postgres/schema-adjust';
import { splitStatements } from '../src/db/engine/postgres/sql-split';
import { ALL_ENTITIES } from '../src/db/entities';
import { knownMigrationNames } from '../src/db/known-migrations';
import { migrateTempDb, tempDbPath } from './generate-entities';
import { MikroORM } from '@mikro-orm/postgresql';
import Database from 'better-sqlite3';
import fs from 'node:fs';
/** What one pass over the migrated SQLite database yields: the baseline, and the triggers the migrations leave there. */
interface BaselinePass {
baseline: PostgresBaseline;
sqliteTriggers: string[];
}
/** The baseline the entities and the migrations on disk describe now, from one migration of the SQLite database. */
async function generateBaseline(known: Set<string> = knownMigrationNames()): Promise<BaselinePass> {
const migrations = [...known].sort();
const through = migrations.at(-1);
if (!through) throw new Error('no migrations found');
const orm = await MikroORM.init({
clientUrl: 'postgresql://generator@localhost/trek',
entities: [...ALL_ENTITIES],
discovery: { warnWhenNoEntities: false },
});
try {
for (const meta of orm.getMetadata().getAll().values()) applyPostgresColumnAdjustments(meta.props);
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 { baseline: { through, migrations, statements }, sqliteTriggers: migrated.triggers };
} finally {
await orm.close();
}
}
/**
* What `--check` holds, as the problems it finds. Every manifest must name
* migrations that exist and every migration up to `through`. While no migration
* is newer than the committed baseline (`through` matches), the baseline is
* not frozen: it must be exactly what the entities generate, and each SQLite
* trigger needs a port. Once a newer migration lands, the baseline is frozen
* and only the manifest is checked.
*/
export function checkBaseline(
committed: PostgresBaseline,
generated: PostgresBaseline,
matches: boolean,
sqliteTriggers: readonly string[],
): 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: ${unlisted.join(', ')}`);
const frozen = committed.through !== generated.through;
if (frozen) return problems;
if (!matches) problems.push('the entities generate a different schema; run npm run gen:pg-baseline');
problems.push(...triggerPortProblems(sqliteTriggers, Object.keys(POSTGRES_TRIGGERS)));
return problems;
}
/** Migration bookkeeping: the baseline's manifest stands for it on Postgres. */
const BOOKKEEPING_TABLES = new Set(['mikro_orm_migrations', 'migrations', 'schema_version']);
type Row = Record<string, unknown>;
/** What the migrations leave in a SQLite database: its triggers, and the rows they insert, table by table in creation order. */
interface MigratedSchema {
triggers: string[];
tables: Array<{ name: string; rows: Row[] }>;
}
/** The tables and triggers of a SQLite database in creation order, each as its type and name. */
function listSchemaObjects(conn: Database.Database): Array<{ type: string; name: string }> {
return conn
.prepare(
"select type, name from sqlite_master where type in ('table', 'trigger') and name not like 'sqlite_%' order by rowid",
)
.all() as Array<{ type: string; name: string }>;
}
/** Reads a migrated SQLite database, read-only. */
function readMigratedSchema(dbPath: string): MigratedSchema {
const conn = new Database(dbPath, { readonly: true });
try {
const objects = listSchemaObjects(conn);
const tables = objects
.filter((object) => object.type === 'table' && !BOOKKEEPING_TABLES.has(object.name))
.map((object) => ({
name: object.name,
rows: conn.prepare(`select * from "${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();
}
}
/** Migrates a temporary SQLite database and reads what the migrations leave in it; the file is removed afterwards. */
async function readMigratedSqlite(): Promise<MigratedSchema> {
const dbPath = tempDbPath();
try {
await migrateTempDb(dbPath);
return readMigratedSchema(dbPath);
} 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 ${values.join(', ')}`,
];
if (rows.every((row) => typeof row.id === 'number')) {
statements.push(
`select setval(seq, greatest(max_id, 1)) from (select pg_get_serial_sequence('"${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;
}
/** The default mode: writes the baseline the entities and the migrations describe. */
async function writeMode(): Promise<void> {
const { baseline } = await generateBaseline();
writePostgresBaseline(baseline);
console.log(`[generate-pg-baseline] wrote ${baseline.statements.length} statements through ${baseline.through}`);
}
/** `--check` (CI): exits 1, naming every problem, when the committed baseline has drifted. */
async function checkMode(): Promise<void> {
const { baseline: generated, sqliteTriggers } = await generateBaseline();
const committed = readPostgresBaseline();
const problems = checkBaseline(committed, generated, baselineMatches(generated), sqliteTriggers);
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');
}
async function main(): Promise<void> {
if (process.argv.slice(2).includes('--check')) await checkMode();
else await writeMode();
}
if (require.main === module) {
main().catch((error: unknown) => {
console.error(error);
process.exit(1);
});
}