feat: automate verified SQLite backups and safe recovery

This commit is contained in:
syntaxbullet
2026-09-04 18:28:28 +02:00
parent 32002151e0
commit 661bc6c346
10 changed files with 227 additions and 2 deletions

96
src/ops/backups.ts Normal file
View File

@@ -0,0 +1,96 @@
import { Database } from "bun:sqlite";
import { chmodSync, closeSync, constants, copyFileSync, existsSync, fsyncSync, linkSync, mkdirSync, openSync, readdirSync, renameSync, rmSync, statSync, unlinkSync } from "node:fs";
import { dirname, join, resolve } from "node:path";
export type BackupConfig = { directory: string; replica?: string; retain: number; intervalMs: number; enabled: boolean };
const filenamePattern = /^minabot-\d{4}-\d{2}-\d{2}T[\d-]+Z-[a-f0-9-]+\.sqlite$/;
export function backupConfig(database: string, env = process.env): BackupConfig {
const retain = Number(env.BACKUP_RETAIN ?? 7);
const hours = Number(env.BACKUP_INTERVAL_HOURS ?? 24);
if (!Number.isInteger(retain) || retain < 2 || retain > 365 || !Number.isFinite(hours) || hours < 1 || hours > 168)
throw new Error("BACKUP_RETAIN must be 2–365 and BACKUP_INTERVAL_HOURS must be 1–168");
const base = dirname(database);
const directory = resolve(base, env.BACKUP_DIR ?? "backups");
const replica = env.BACKUP_REPLICA_DIR ? resolve(base, env.BACKUP_REPLICA_DIR) : undefined;
if (replica === directory) throw new Error("The backup replica must use a separate directory");
return { directory, replica, retain, intervalMs: hours * 3600000, enabled: env.BACKUP_ENABLED !== "false" };
}
function durable(path: string) {
chmodSync(path, 0o600);
const fd = openSync(path, "r"); try { fsyncSync(fd); } finally { closeSync(fd); }
}
export function verifyBackup(path: string) {
const db = new Database(path, { readonly: true, strict: true });
try {
const integrity = db.query("PRAGMA integrity_check").all() as { integrity_check: string }[];
if (integrity.length !== 1 || integrity[0]?.integrity_check !== "ok" || db.query("PRAGMA foreign_key_check").all().length)
throw new Error("Backup integrity check failed");
db.query("SELECT id FROM users LIMIT 1").all();
db.query("SELECT id FROM habits LIMIT 1").all();
db.query("SELECT hash FROM __drizzle_migrations LIMIT 1").all();
} finally { db.close(); }
}
function snapshots(directory: string) {
if (!existsSync(directory)) return [];
return readdirSync(directory).filter(name => filenamePattern.test(name)).sort().reverse();
}
export function latestBackupTime(config: BackupConfig) {
const latest = snapshots(config.directory)[0];
if (!latest || (config.replica && !existsSync(join(config.replica, latest)))) return 0;
return statSync(join(config.directory, latest)).mtimeMs;
}
export function createBackup(database: Database, config: BackupConfig) {
const name = `minabot-${new Date().toISOString().replace(/[:.]/g, "-")}-${crypto.randomUUID()}.sqlite`;
mkdirSync(config.directory, { recursive: true, mode: 0o700 });
const path = join(config.directory, name);
const staging = `${path}.partial`;
mkdirSync(staging, { mode: 0o700 });
const temporary = join(staging, "snapshot.sqlite");
try {
// VACUUM INTO includes committed WAL transactions in one consistent snapshot.
// SQLite creates the output inside a private staging directory.
database.query("VACUUM INTO ?").run(temporary);
verifyBackup(temporary); durable(temporary); renameSync(temporary, path);
if (config.replica) {
// Require the configured mount to exist; never silently create an absent mount.
if (!statSync(config.replica).isDirectory()) throw new Error("Backup replica directory unavailable");
const replica = join(config.replica, name);
copyFileSync(path, `${replica}.partial`, constants.COPYFILE_EXCL);
durable(`${replica}.partial`); verifyBackup(`${replica}.partial`); renameSync(`${replica}.partial`, replica);
}
for (const directory of [config.directory, ...(config.replica ? [config.replica] : [])])
for (const old of snapshots(directory).slice(config.retain)) unlinkSync(join(directory, old));
return path;
} finally { rmSync(staging, { recursive: true, force: true }); }
}
export function restoreBackup(source: string, destination: string) {
if (existsSync(destination) || existsSync(`${destination}-wal`) || existsSync(`${destination}-shm`))
throw new Error("Restore destination must be a new path with no SQLite sidecars");
verifyBackup(source);
mkdirSync(dirname(destination), { recursive: true, mode: 0o700 });
const temporary = `${destination}.${crypto.randomUUID()}.partial`;
try {
copyFileSync(source, temporary, constants.COPYFILE_EXCL); chmodSync(temporary, 0o600);
const restored = new Database(temporary);
try { restored.exec("PRAGMA journal_mode=DELETE; DELETE FROM sessions;"); } finally { restored.close(); }
verifyBackup(temporary); durable(temporary);
// Link is exclusive: a concurrently-created destination is never overwritten.
linkSync(temporary, destination);
} finally { if (existsSync(temporary)) unlinkSync(temporary); }
return destination;
}
export function startBackups(database: Database, config: BackupConfig) {
const state = { enabled: config.enabled, lastSuccess: 0, failed: false };
if (!config.enabled) return { state, stop() {} };
const tick = () => {
try {
const latest = latestBackupTime(config);
if (latest && Date.now() - latest < config.intervalMs) { state.lastSuccess = latest; state.failed = false; return; }
createBackup(database, config); state.lastSuccess = Date.now(); state.failed = false;
console.log(JSON.stringify({ event: "backup_completed", timestamp: new Date().toISOString() }));
} catch { state.failed = true; console.error(JSON.stringify({ event: "backup_failed", timestamp: new Date().toISOString() })); }
};
tick();
const timer = setInterval(tick, 60000); timer.unref();
return { state, stop: () => clearInterval(timer) };
}