feat(gateway): tier-detector with fail-fast PG/Valkey/pgvector probes (FED-M1-04) (#473)

apps/gateway/package.json

@@ -63,8 +63,10 @@
     "class-validator": "^0.15.1",
     "dotenv": "^17.3.1",
     "fastify": "^5.0.0",
+    "ioredis": "^5.10.0",
     "node-cron": "^4.2.1",
     "openai": "^6.32.0",
+    "postgres": "^3.4.8",
     "reflect-metadata": "^0.2.0",
     "rxjs": "^7.8.0",
     "socket.io": "^4.8.0",

apps/gateway/src/bootstrap/tier-detector.spec.ts

@@ -0,0 +1,344 @@
+/**
+ * Unit tests for tier-detector.ts.
+ *
+ * All external I/O (postgres, ioredis) is mocked — no live services required.
+ *
+ * Note on hoisting: vi.mock() factories are hoisted above all imports by vitest.
+ * Variables referenced inside factory callbacks must be declared via vi.hoisted()
+ * so they are available at hoist time.
+ */
+
+import { describe, it, expect, vi, beforeEach } from 'vitest';
+
+/* ------------------------------------------------------------------ */
+/*  Hoist shared mock state so factories can reference it              */
+/* ------------------------------------------------------------------ */
+
+const mocks = vi.hoisted(() => {
+  const mockSqlFn = vi.fn();
+  const mockEnd = vi.fn().mockResolvedValue(undefined);
+  const mockPostgresConstructor = vi.fn(() => {
+    const sql = mockSqlFn as ReturnType<typeof mockSqlFn>;
+    (sql as unknown as Record<string, unknown>)['end'] = mockEnd;
+    return sql;
+  });
+
+  const mockRedisConnect = vi.fn().mockResolvedValue(undefined);
+  const mockRedisPing = vi.fn().mockResolvedValue('PONG');
+  const mockRedisDisconnect = vi.fn();
+  const MockRedis = vi.fn().mockImplementation(() => ({
+    connect: mockRedisConnect,
+    ping: mockRedisPing,
+    disconnect: mockRedisDisconnect,
+  }));
+
+  return {
+    mockSqlFn,
+    mockEnd,
+    mockPostgresConstructor,
+    mockRedisConnect,
+    mockRedisPing,
+    mockRedisDisconnect,
+    MockRedis,
+  };
+});
+
+/* ------------------------------------------------------------------ */
+/*  Module mocks (registered at hoist time)                            */
+/* ------------------------------------------------------------------ */
+
+vi.mock('postgres', () => ({
+  default: mocks.mockPostgresConstructor,
+}));
+
+vi.mock('ioredis', () => ({
+  Redis: mocks.MockRedis,
+}));
+
+/* ------------------------------------------------------------------ */
+/*  Import SUT after mocks are registered                              */
+/* ------------------------------------------------------------------ */
+
+import { detectAndAssertTier, TierDetectionError } from './tier-detector.js';
+import type { MosaicConfig } from '@mosaicstack/config';
+
+/* ------------------------------------------------------------------ */
+/*  Config fixtures                                                     */
+/* ------------------------------------------------------------------ */
+
+const LOCAL_CONFIG: MosaicConfig = {
+  tier: 'local',
+  storage: { type: 'pglite', dataDir: '.mosaic/pglite' },
+  queue: { type: 'local', dataDir: '.mosaic/queue' },
+  memory: { type: 'keyword' },
+};
+
+const STANDALONE_CONFIG: MosaicConfig = {
+  tier: 'standalone',
+  storage: { type: 'postgres', url: 'postgresql://mosaic:mosaic@db-host:5432/mosaic' },
+  queue: { type: 'bullmq', url: 'redis://valkey-host:6380' },
+  memory: { type: 'keyword' },
+};
+
+const FEDERATED_CONFIG: MosaicConfig = {
+  tier: 'federated',
+  storage: {
+    type: 'postgres',
+    url: 'postgresql://mosaic:mosaic@db-host:5433/mosaic',
+    enableVector: true,
+  },
+  queue: { type: 'bullmq', url: 'redis://valkey-host:6380' },
+  memory: { type: 'pgvector' },
+};
+
+/* ------------------------------------------------------------------ */
+/*  Tests                                                               */
+/* ------------------------------------------------------------------ */
+
+describe('detectAndAssertTier', () => {
+  beforeEach(() => {
+    vi.clearAllMocks();
+
+    // Default: all probes succeed.
+    mocks.mockSqlFn.mockResolvedValue([]);
+    mocks.mockEnd.mockResolvedValue(undefined);
+    mocks.mockRedisConnect.mockResolvedValue(undefined);
+    mocks.mockRedisPing.mockResolvedValue('PONG');
+
+    // Re-wire constructor to return a fresh sql-like object each time.
+    mocks.mockPostgresConstructor.mockImplementation(() => {
+      const sql = mocks.mockSqlFn as ReturnType<typeof mocks.mockSqlFn>;
+      (sql as unknown as Record<string, unknown>)['end'] = mocks.mockEnd;
+      return sql;
+    });
+    mocks.MockRedis.mockImplementation(() => ({
+      connect: mocks.mockRedisConnect,
+      ping: mocks.mockRedisPing,
+      disconnect: mocks.mockRedisDisconnect,
+    }));
+  });
+
+  /* ---------------------------------------------------------------- */
+  /*  1. local — no-op                                                 */
+  /* ---------------------------------------------------------------- */
+
+  it('resolves immediately for tier=local without touching postgres or ioredis', async () => {
+    await expect(detectAndAssertTier(LOCAL_CONFIG)).resolves.toBeUndefined();
+    expect(mocks.mockPostgresConstructor).not.toHaveBeenCalled();
+    expect(mocks.MockRedis).not.toHaveBeenCalled();
+  });
+
+  /* ---------------------------------------------------------------- */
+  /*  2. standalone — happy path                                       */
+  /* ---------------------------------------------------------------- */
+
+  it('resolves for tier=standalone when postgres and valkey are reachable', async () => {
+    await expect(detectAndAssertTier(STANDALONE_CONFIG)).resolves.toBeUndefined();
+
+    // Postgres was probed (SELECT 1 only — no pgvector check).
+    expect(mocks.mockPostgresConstructor).toHaveBeenCalledTimes(1);
+    expect(mocks.mockSqlFn).toHaveBeenCalledTimes(1);
+    // Valkey was probed.
+    expect(mocks.MockRedis).toHaveBeenCalledTimes(1);
+    expect(mocks.mockRedisPing).toHaveBeenCalledTimes(1);
+  });
+
+  /* ---------------------------------------------------------------- */
+  /*  3. standalone — postgres unreachable                             */
+  /* ---------------------------------------------------------------- */
+
+  it('throws TierDetectionError with service=postgres when postgres query rejects', async () => {
+    mocks.mockSqlFn.mockRejectedValueOnce(new Error('connection refused'));
+
+    const promise = detectAndAssertTier(STANDALONE_CONFIG);
+    await expect(promise).rejects.toBeInstanceOf(TierDetectionError);
+
+    // Confirm no valkey probe happened (fail fast on first error).
+    expect(mocks.MockRedis).not.toHaveBeenCalled();
+  });
+
+  it('sets service=postgres on the error when postgres fails', async () => {
+    mocks.mockSqlFn.mockRejectedValue(new Error('connection refused'));
+
+    try {
+      await detectAndAssertTier(STANDALONE_CONFIG);
+      expect.fail('should have thrown');
+    } catch (err) {
+      expect(err).toBeInstanceOf(TierDetectionError);
+      const typed = err as TierDetectionError;
+      expect(typed.service).toBe('postgres');
+      expect(typed.remediation).toContain('docker compose');
+    }
+  });
+
+  /* ---------------------------------------------------------------- */
+  /*  4. standalone — valkey unreachable                               */
+  /* ---------------------------------------------------------------- */
+
+  it('throws TierDetectionError with service=valkey when ping fails', async () => {
+    // Postgres probe succeeds; valkey connect fails.
+    mocks.mockSqlFn.mockResolvedValue([]);
+    mocks.mockRedisConnect.mockRejectedValue(new Error('ECONNREFUSED'));
+
+    try {
+      await detectAndAssertTier(STANDALONE_CONFIG);
+      expect.fail('should have thrown');
+    } catch (err) {
+      expect(err).toBeInstanceOf(TierDetectionError);
+      const typed = err as TierDetectionError;
+      expect(typed.service).toBe('valkey');
+      expect(typed.message).toContain('valkey');
+      expect(typed.remediation).toContain('valkey-federated');
+    }
+  });
+
+  /* ---------------------------------------------------------------- */
+  /*  5. federated — happy path                                        */
+  /* ---------------------------------------------------------------- */
+
+  it('resolves for tier=federated when all three checks pass', async () => {
+    // SELECT 1 and CREATE EXTENSION both succeed.
+    mocks.mockSqlFn.mockResolvedValue([]);
+
+    await expect(detectAndAssertTier(FEDERATED_CONFIG)).resolves.toBeUndefined();
+
+    // postgres probe (SELECT 1) + pgvector probe (CREATE EXTENSION) = 2 postgres constructors.
+    expect(mocks.mockPostgresConstructor).toHaveBeenCalledTimes(2);
+    expect(mocks.mockSqlFn).toHaveBeenCalledTimes(2);
+    // Valkey probed once.
+    expect(mocks.MockRedis).toHaveBeenCalledTimes(1);
+  });
+
+  /* ---------------------------------------------------------------- */
+  /*  6. federated — pgvector not installable                          */
+  /* ---------------------------------------------------------------- */
+
+  it('throws TierDetectionError with service=pgvector when CREATE EXTENSION fails', async () => {
+    // SELECT 1 succeeds (first call), CREATE EXTENSION fails (second call).
+    mocks.mockSqlFn
+      .mockResolvedValueOnce([]) // SELECT 1
+      .mockRejectedValueOnce(new Error('extension "vector" is not available'));
+
+    try {
+      await detectAndAssertTier(FEDERATED_CONFIG);
+      expect.fail('should have thrown');
+    } catch (err) {
+      expect(err).toBeInstanceOf(TierDetectionError);
+      const typed = err as TierDetectionError;
+      expect(typed.service).toBe('pgvector');
+      expect(typed.message).toContain('pgvector');
+      expect(typed.remediation).toContain('pgvector/pgvector');
+    }
+  });
+
+  /* ---------------------------------------------------------------- */
+  /*  7. probeValkey honors connectTimeout                             */
+  /* ---------------------------------------------------------------- */
+
+  it('constructs the ioredis Redis client with connectTimeout: 5000', async () => {
+    await detectAndAssertTier(STANDALONE_CONFIG);
+
+    expect(mocks.MockRedis).toHaveBeenCalledOnce();
+    const [, options] = mocks.MockRedis.mock.calls[0] as [string, Record<string, unknown>];
+    expect(options).toMatchObject({ connectTimeout: 5000 });
+  });
+
+  /* ---------------------------------------------------------------- */
+  /*  8. probePgvector — library-not-installed remediation             */
+  /* ---------------------------------------------------------------- */
+
+  it('includes pgvector/pgvector:pg17 in remediation when pgvector library is missing', async () => {
+    // SELECT 1 succeeds; CREATE EXTENSION fails with the canonical library-missing message.
+    mocks.mockSqlFn
+      .mockResolvedValueOnce([]) // SELECT 1 (probePostgres)
+      .mockRejectedValueOnce(new Error('extension "vector" is not available')); // probePgvector
+
+    try {
+      await detectAndAssertTier(FEDERATED_CONFIG);
+      expect.fail('should have thrown');
+    } catch (err) {
+      expect(err).toBeInstanceOf(TierDetectionError);
+      const typed = err as TierDetectionError;
+      expect(typed.service).toBe('pgvector');
+      expect(typed.remediation).toContain('pgvector/pgvector:pg17');
+    }
+  });
+
+  /* ---------------------------------------------------------------- */
+  /*  9. probePgvector — permission / other error remediation          */
+  /* ---------------------------------------------------------------- */
+
+  it('mentions CREATE permission or superuser in remediation for a generic pgvector error', async () => {
+    // SELECT 1 succeeds; CREATE EXTENSION fails with a permission error (not the library-missing message).
+    mocks.mockSqlFn
+      .mockResolvedValueOnce([]) // SELECT 1 (probePostgres)
+      .mockRejectedValueOnce(new Error('permission denied to create extension'));
+
+    try {
+      await detectAndAssertTier(FEDERATED_CONFIG);
+      expect.fail('should have thrown');
+    } catch (err) {
+      expect(err).toBeInstanceOf(TierDetectionError);
+      const typed = err as TierDetectionError;
+      expect(typed.service).toBe('pgvector');
+      // Must NOT point to the image fix — that's only for the library-missing case.
+      expect(typed.remediation).not.toContain('pgvector/pgvector:pg17');
+      // Must mention permissions or superuser.
+      expect(typed.remediation).toMatch(/CREATE|superuser/i);
+    }
+  });
+
+  /* ---------------------------------------------------------------- */
+  /*  10. federated tier rejects non-bullmq queue.type                 */
+  /* ---------------------------------------------------------------- */
+
+  it('throws TierDetectionError with service=config for federated tier with queue.type !== bullmq', async () => {
+    const badConfig: MosaicConfig = {
+      tier: 'federated',
+      storage: {
+        type: 'postgres',
+        url: 'postgresql://mosaic:mosaic@db-host:5433/mosaic',
+        enableVector: true,
+      },
+      queue: { type: 'local', dataDir: '.mosaic/queue' },
+      memory: { type: 'pgvector' },
+    };
+
+    try {
+      await detectAndAssertTier(badConfig);
+      expect.fail('should have thrown');
+    } catch (err) {
+      expect(err).toBeInstanceOf(TierDetectionError);
+      const typed = err as TierDetectionError;
+      expect(typed.service).toBe('config');
+      expect(typed.remediation).toContain('bullmq');
+    }
+
+    // No network probes should have been attempted.
+    expect(mocks.mockPostgresConstructor).not.toHaveBeenCalled();
+    expect(mocks.MockRedis).not.toHaveBeenCalled();
+  });
+
+  /* ---------------------------------------------------------------- */
+  /*  11. Error fields populated                                        */
+  /* ---------------------------------------------------------------- */
+
+  it('populates host, port, and remediation on a thrown TierDetectionError', async () => {
+    mocks.mockSqlFn.mockRejectedValue(new Error('connection refused'));
+
+    let caught: TierDetectionError | undefined;
+    try {
+      await detectAndAssertTier(STANDALONE_CONFIG);
+    } catch (err) {
+      caught = err as TierDetectionError;
+    }
+
+    expect(caught).toBeInstanceOf(TierDetectionError);
+    expect(caught!.service).toBe('postgres');
+    // Host and port are extracted from the Postgres URL in STANDALONE_CONFIG.
+    expect(caught!.host).toBe('db-host');
+    expect(caught!.port).toBe(5432);
+    expect(caught!.remediation).toMatch(/docker compose/i);
+    expect(caught!.message).toContain('db-host:5432');
+  });
+});

apps/gateway/src/bootstrap/tier-detector.ts

@@ -0,0 +1,220 @@
+/**
+ * Tier Detector — pre-flight service reachability assertions.
+ *
+ * Runs BEFORE NestFactory.create() to surface actionable errors immediately
+ * rather than crashing mid-boot with an opaque stack trace.
+ *
+ * Library choices:
+ * - Postgres: `postgres` npm package (already a dep via @mosaicstack/db / drizzle-orm).
+ *   The spec mentions `pg`, but only `postgres` is installed in this monorepo.
+ * - Valkey: `ioredis` (already a dep via @mosaicstack/queue → bullmq; same URL
+ *   convention used by the bullmq adapter).
+ */
+
+import postgres from 'postgres';
+import { Redis } from 'ioredis';
+import type { MosaicConfig } from '@mosaicstack/config';
+
+/* ------------------------------------------------------------------ */
+/*  Structured error type                                               */
+/* ------------------------------------------------------------------ */
+
+export class TierDetectionError extends Error {
+  public readonly service: 'postgres' | 'valkey' | 'pgvector' | 'config';
+  public readonly host: string;
+  public readonly port: number;
+  public readonly remediation: string;
+
+  constructor(opts: {
+    service: 'postgres' | 'valkey' | 'pgvector' | 'config';
+    host: string;
+    port: number;
+    remediation: string;
+    cause?: unknown;
+  }) {
+    const message =
+      `[tier-detector] ${opts.service} unreachable or unusable at ` +
+      `${opts.host}:${opts.port} — ${opts.remediation}`;
+    super(message, { cause: opts.cause });
+    this.name = 'TierDetectionError';
+    this.service = opts.service;
+    this.host = opts.host;
+    this.port = opts.port;
+    this.remediation = opts.remediation;
+  }
+}
+
+/* ------------------------------------------------------------------ */
+/*  URL helpers                                                        */
+/* ------------------------------------------------------------------ */
+
+/** Extract host and port from a URL string, returning safe fallbacks on parse failure. */
+function parseHostPort(url: string, defaultPort: number): { host: string; port: number } {
+  try {
+    const parsed = new URL(url);
+    const host = parsed.hostname || 'unknown';
+    const port = parsed.port ? parseInt(parsed.port, 10) : defaultPort;
+    return { host, port };
+  } catch {
+    return { host: 'unknown', port: defaultPort };
+  }
+}
+
+/* ------------------------------------------------------------------ */
+/*  Postgres probe                                                      */
+/* ------------------------------------------------------------------ */
+
+async function probePostgres(url: string): Promise<void> {
+  const { host, port } = parseHostPort(url, 5432);
+  let sql: ReturnType<typeof postgres> | undefined;
+  try {
+    sql = postgres(url, {
+      max: 1,
+      connect_timeout: 5,
+      idle_timeout: 5,
+    });
+    // Run a trivial query to confirm connectivity.
+    await sql`SELECT 1`;
+  } catch (cause) {
+    throw new TierDetectionError({
+      service: 'postgres',
+      host,
+      port,
+      remediation:
+        'Start Postgres: `docker compose -f docker-compose.federated.yml --profile federated up -d postgres-federated`',
+      cause,
+    });
+  } finally {
+    if (sql) {
+      await sql.end({ timeout: 2 }).catch(() => {
+        // Ignore cleanup errors — we already have what we need.
+      });
+    }
+  }
+}
+
+/* ------------------------------------------------------------------ */
+/*  pgvector probe                                                      */
+/* ------------------------------------------------------------------ */
+
+async function probePgvector(url: string): Promise<void> {
+  const { host, port } = parseHostPort(url, 5432);
+  let sql: ReturnType<typeof postgres> | undefined;
+  try {
+    sql = postgres(url, {
+      max: 1,
+      connect_timeout: 5,
+      idle_timeout: 5,
+    });
+    // This succeeds whether the extension is already installed or freshly created.
+    // It errors only if the pgvector shared library is missing from the Postgres binary.
+    await sql`CREATE EXTENSION IF NOT EXISTS vector`;
+  } catch (cause) {
+    const causeMsg = cause instanceof Error ? cause.message.toLowerCase() : '';
+    const isLibraryMissing = causeMsg.includes('extension "vector" is not available');
+    const remediation = isLibraryMissing
+      ? 'Use the `pgvector/pgvector:pg17` image, not the stock `postgres:17` image. See `docker-compose.federated.yml`.'
+      : 'The database role lacks permission to CREATE EXTENSION. Grant `CREATE` on the database, or run as a superuser.';
+    throw new TierDetectionError({
+      service: 'pgvector',
+      host,
+      port,
+      remediation,
+      cause,
+    });
+  } finally {
+    if (sql) {
+      await sql.end({ timeout: 2 }).catch(() => {
+        // Ignore cleanup errors.
+      });
+    }
+  }
+}
+
+/* ------------------------------------------------------------------ */
+/*  Valkey probe                                                        */
+/* ------------------------------------------------------------------ */
+
+const DEFAULT_VALKEY_URL = 'redis://localhost:6380';
+
+async function probeValkey(url: string): Promise<void> {
+  const { host, port } = parseHostPort(url, 6380);
+  const client = new Redis(url, {
+    enableReadyCheck: false,
+    maxRetriesPerRequest: 0,
+    retryStrategy: () => null, // no retries — fail fast
+    lazyConnect: true,
+    connectTimeout: 5000, // fail-fast: 5-second hard cap on connection attempt
+  });
+
+  try {
+    await client.connect();
+    const pong = await client.ping();
+    if (pong !== 'PONG') {
+      throw new Error(`Unexpected PING response: ${pong}`);
+    }
+  } catch (cause) {
+    throw new TierDetectionError({
+      service: 'valkey',
+      host,
+      port,
+      remediation:
+        'Start Valkey: `docker compose -f docker-compose.federated.yml --profile federated up -d valkey-federated`',
+      cause,
+    });
+  } finally {
+    client.disconnect();
+  }
+}
+
+/* ------------------------------------------------------------------ */
+/*  Public entry point                                                  */
+/* ------------------------------------------------------------------ */
+
+/**
+ * Assert that all services required by `config.tier` are reachable.
+ *
+ * - `local`      — no-op (PGlite is in-process; no external services).
+ * - `standalone` — assert Postgres + Valkey (if queue.type === 'bullmq').
+ * - `federated`  — assert Postgres + Valkey + pgvector installability.
+ *
+ * Throws `TierDetectionError` on the first failure with host:port and
+ * a remediation hint.
+ */
+export async function detectAndAssertTier(config: MosaicConfig): Promise<void> {
+  if (config.tier === 'local') {
+    // PGlite runs in-process — nothing to probe.
+    return;
+  }
+
+  const pgUrl =
+    config.storage.type === 'postgres' ? config.storage.url : 'postgresql://localhost:5432/mosaic';
+
+  const valkeyUrl =
+    config.queue.type === 'bullmq' ? (config.queue.url ?? DEFAULT_VALKEY_URL) : null;
+
+  if (config.tier === 'standalone') {
+    await probePostgres(pgUrl);
+    if (valkeyUrl) {
+      await probeValkey(valkeyUrl);
+    }
+    return;
+  }
+
+  // tier === 'federated'
+  // Reject misconfigured queue upfront — federated requires bullmq + a Valkey URL.
+  if (config.queue.type !== 'bullmq') {
+    throw new TierDetectionError({
+      service: 'config',
+      host: 'localhost',
+      port: 0,
+      remediation:
+        "Federated tier requires queue.type === 'bullmq'. " +
+        "Set queue: { type: 'bullmq', url: 'redis://...' } in your mosaic.config.json.",
+    });
+  }
+  const federatedValkeyUrl = config.queue.url ?? DEFAULT_VALKEY_URL;
+  await probePostgres(pgUrl);
+  await probeValkey(federatedValkeyUrl);
+  await probePgvector(pgUrl);
+}

apps/gateway/src/main.ts

@@ -20,10 +20,12 @@ import { Logger, ValidationPipe } from '@nestjs/common';
 import { FastifyAdapter, type NestFastifyApplication } from '@nestjs/platform-fastify';
 import helmet from '@fastify/helmet';
 import { listSsoStartupWarnings } from '@mosaicstack/auth';
+import { loadConfig } from '@mosaicstack/config';
 import { AppModule } from './app.module.js';
 import { mountAuthHandler } from './auth/auth.controller.js';
 import { mountMcpHandler } from './mcp/mcp.controller.js';
 import { McpService } from './mcp/mcp.service.js';
+import { detectAndAssertTier, TierDetectionError } from './bootstrap/tier-detector.js';
 
 async function bootstrap(): Promise<void> {
   const logger = new Logger('Bootstrap');
@@ -32,6 +34,20 @@ async function bootstrap(): Promise<void> {
     throw new Error('BETTER_AUTH_SECRET is required');
   }
 
+  // Pre-flight: assert all external services required by the configured tier
+  // are reachable. Runs before NestFactory.create() so failures are visible
+  // immediately with actionable remediation hints.
+  const mosaicConfig = loadConfig();
+  try {
+    await detectAndAssertTier(mosaicConfig);
+  } catch (err) {
+    if (err instanceof TierDetectionError) {
+      logger.error(`Tier detection failed: ${err.message}`);
+      logger.error(`Remediation: ${err.remediation}`);
+    }
+    throw err;
+  }
+
   for (const warning of listSsoStartupWarnings()) {
     logger.warn(warning);
   }