Merge pull request 'feat(federation): outbound mTLS FederationClient (FED-M3-08)' (#508) from feat/federation-m3-client into main

- HIGH-A: resolveEntry now uses promise-cache pattern so concurrent
  callers serialize on a single in-flight build, eliminating duplicate
  key material in heap and duplicate DB round-trips
- HIGH-B: flushPeer destroys the evicted undici Agent so stale TLS
  connections close on cert rotation
- MED-C: add regression test for PEER_MISCONFIGURED when
  STEP_CA_ROOT_CERT_PATH is unset

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

apps/gateway/package.json

@@ -73,6 +73,7 @@
     "rxjs": "^7.8.0",
     "socket.io": "^4.8.0",
     "uuid": "^11.0.0",
+    "undici": "^7.24.6",
     "zod": "^4.3.6"
   },
   "devDependencies": {

apps/gateway/src/federation/client/__tests__/federation-client.service.spec.ts

@@ -0,0 +1,553 @@
+/**
+ * Unit tests for FederationClientService (FED-M3-08).
+ *
+ * HTTP mocking strategy:
+ *   undici MockAgent is used to intercept outbound HTTP requests.  The service
+ *   uses `undici.fetch` with a `dispatcher` option, so MockAgent is set as the
+ *   global dispatcher and all requests flow through it.
+ *
+ *   Because the service builds one `undici.Agent` per peer and passes it as
+ *   the dispatcher on every fetch call, we cannot intercept at the Agent level
+ *   in unit tests without significant refactoring.  Instead, we set the global
+ *   dispatcher to a MockAgent and override the service's `doRequest` indirection
+ *   by spying on the internal fetch call.
+ *
+ *   For the cert/key wiring, we use the real `sealClientKey` function from
+ *   peer-key.util.ts with a test secret — no stubs.
+ *
+ * Sealed-key setup:
+ *   Each test (or beforeAll) calls `sealClientKey(TEST_PRIVATE_KEY_PEM)` with
+ *   BETTER_AUTH_SECRET set to a deterministic test value so that
+ *   `unsealClientKey` in the service recovers the original PEM.
+ */
+
+import 'reflect-metadata';
+import { describe, it, expect, vi, beforeEach, afterEach, beforeAll, afterAll } from 'vitest';
+import { MockAgent, setGlobalDispatcher, getGlobalDispatcher } from 'undici';
+import type { Dispatcher } from 'undici';
+import { writeFileSync, unlinkSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import type { Db } from '@mosaicstack/db';
+import { FederationClientService, FederationClientError } from '../federation-client.service.js';
+import { sealClientKey } from '../../peer-key.util.js';
+
+// ---------------------------------------------------------------------------
+// Test constants
+// ---------------------------------------------------------------------------
+
+const TEST_SECRET = 'test-secret-for-federation-client-spec-only';
+const PEER_ID = 'aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa';
+const ENDPOINT = 'https://peer.example.com';
+
+// Minimal valid RSA/EC private key PEM — does NOT need to be a real key for
+// unit tests because we only verify it round-trips through seal/unseal, not
+// that it actually negotiates TLS (MockAgent handles that).
+const TEST_PRIVATE_KEY_PEM = `-----BEGIN PRIVATE KEY-----
+MIIEvQIBADANBgkqhkiG9w0BAQEFAASCBKcwggSjAgEAAoIBAQDummyKeyForTests
+-----END PRIVATE KEY-----`;
+
+// Minimal self-signed cert PEM (dummy — only used for mTLS Agent construction)
+const TEST_CERT_PEM = `-----BEGIN CERTIFICATE-----
+MIIBdummyCertForFederationClientTests==
+-----END CERTIFICATE-----`;
+
+const TEST_CERT_SERIAL = 'ABCDEF1234567890';
+
+// ---------------------------------------------------------------------------
+// Sealed key (computed once in beforeAll)
+// ---------------------------------------------------------------------------
+
+let SEALED_KEY: string;
+
+// Path to a stub Step-CA root cert file written in beforeAll. The cert is never
+// actually used to negotiate TLS in unit tests (MockAgent + spy on resolveEntry
+// short-circuit the network), but loadStepCaRoot() requires the file to exist.
+const STUB_CA_PEM_PATH = join(tmpdir(), 'federation-client-spec-ca.pem');
+const STUB_CA_PEM = `-----BEGIN CERTIFICATE-----
+MIIBdummyCAforFederationClientSpecOnly==
+-----END CERTIFICATE-----
+`;
+
+// ---------------------------------------------------------------------------
+// Peer row factory
+// ---------------------------------------------------------------------------
+
+function makePeerRow(overrides: Partial<Record<string, unknown>> = {}) {
+  return {
+    id: PEER_ID,
+    commonName: 'peer-example-com',
+    displayName: 'Test Peer',
+    certPem: TEST_CERT_PEM,
+    certSerial: TEST_CERT_SERIAL,
+    certNotAfter: new Date('2030-01-01T00:00:00Z'),
+    clientKeyPem: SEALED_KEY,
+    state: 'active' as const,
+    endpointUrl: ENDPOINT,
+    lastSeenAt: null,
+    createdAt: new Date('2026-01-01T00:00:00Z'),
+    revokedAt: null,
+    ...overrides,
+  };
+}
+
+// ---------------------------------------------------------------------------
+// Mock DB builder
+// ---------------------------------------------------------------------------
+
+function makeDb(selectRows: unknown[] = [makePeerRow()]): Db {
+  const limitSelect = vi.fn().mockResolvedValue(selectRows);
+  const whereSelect = vi.fn().mockReturnValue({ limit: limitSelect });
+  const fromSelect = vi.fn().mockReturnValue({ where: whereSelect });
+  const selectMock = vi.fn().mockReturnValue({ from: fromSelect });
+
+  return {
+    select: selectMock,
+    insert: vi.fn(),
+    update: vi.fn(),
+    delete: vi.fn(),
+    transaction: vi.fn(),
+  } as unknown as Db;
+}
+
+// ---------------------------------------------------------------------------
+// Helpers for MockAgent HTTP interception
+// ---------------------------------------------------------------------------
+
+/**
+ * Create a MockAgent + MockPool for the peer endpoint, set it as the global
+ * dispatcher, and return both for per-test configuration.
+ */
+function makeMockAgent() {
+  const mockAgent = new MockAgent({ connections: 1 });
+  mockAgent.disableNetConnect();
+  setGlobalDispatcher(mockAgent);
+  const pool = mockAgent.get(ENDPOINT);
+  return { mockAgent, pool };
+}
+
+/**
+ * Build a FederationClientService with a mock DB and a spy on the internal
+ * fetch so we can intercept at the HTTP layer via MockAgent.
+ *
+ * The service calls `fetch(url, { dispatcher: agent })` where `agent` is the
+ * mTLS undici.Agent built from the peer's cert+key.  To make MockAgent work,
+ * we need the fetch dispatcher to be the MockAgent, not the per-peer Agent.
+ *
+ * Strategy: we replace the private `resolveEntry` result's `agent` field with
+ * the MockAgent's pool, so fetch uses our interceptor.  We do this by spying
+ * on `resolveEntry` and returning a controlled entry.
+ */
+function makeService(db: Db, mockPool: Dispatcher): FederationClientService {
+  const svc = new FederationClientService(db);
+
+  // Override resolveEntry to inject MockAgent pool as the dispatcher
+  vi.spyOn(
+    svc as unknown as { resolveEntry: (peerId: string) => Promise<unknown> },
+    'resolveEntry',
+  ).mockImplementation(async (_peerId: string) => {
+    // Still call DB (via the real logic) to exercise peer validation,
+    // but return mock pool as the agent.
+    // For simplicity in unit tests, directly return a controlled entry.
+    return {
+      agent: mockPool,
+      endpointUrl: ENDPOINT,
+      certPem: TEST_CERT_PEM,
+      certSerial: TEST_CERT_SERIAL,
+    };
+  });
+
+  return svc;
+}
+
+// ---------------------------------------------------------------------------
+// Test setup
+// ---------------------------------------------------------------------------
+
+let originalDispatcher: Dispatcher;
+
+beforeAll(() => {
+  // Seal the test key once — requires BETTER_AUTH_SECRET
+  const saved = process.env['BETTER_AUTH_SECRET'];
+  process.env['BETTER_AUTH_SECRET'] = TEST_SECRET;
+  try {
+    SEALED_KEY = sealClientKey(TEST_PRIVATE_KEY_PEM);
+  } finally {
+    if (saved === undefined) {
+      delete process.env['BETTER_AUTH_SECRET'];
+    } else {
+      process.env['BETTER_AUTH_SECRET'] = saved;
+    }
+  }
+  writeFileSync(STUB_CA_PEM_PATH, STUB_CA_PEM, 'utf8');
+});
+
+afterAll(() => {
+  try {
+    unlinkSync(STUB_CA_PEM_PATH);
+  } catch {
+    // best-effort cleanup
+  }
+});
+
+beforeEach(() => {
+  originalDispatcher = getGlobalDispatcher();
+  process.env['BETTER_AUTH_SECRET'] = TEST_SECRET;
+  process.env['STEP_CA_ROOT_CERT_PATH'] = STUB_CA_PEM_PATH;
+});
+
+afterEach(() => {
+  setGlobalDispatcher(originalDispatcher);
+  vi.restoreAllMocks();
+  delete process.env['BETTER_AUTH_SECRET'];
+  delete process.env['STEP_CA_ROOT_CERT_PATH'];
+});
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+/** Successful list response body */
+const LIST_BODY = {
+  items: [{ id: '1', title: 'Task One' }],
+  nextCursor: undefined,
+  _partial: false,
+};
+
+/** Successful get response body */
+const GET_BODY = {
+  item: { id: '1', title: 'Task One' },
+  _partial: false,
+};
+
+/** Successful capabilities response body */
+const CAP_BODY = {
+  resources: ['tasks'],
+  excluded_resources: [],
+  max_rows_per_query: 100,
+  supported_verbs: ['list', 'get', 'capabilities'] as const,
+};
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+describe('FederationClientService', () => {
+  // ─── Successful verb calls ─────────────────────────────────────────────────
+
+  describe('list()', () => {
+    it('returns parsed typed response on success', async () => {
+      const db = makeDb();
+      const { mockAgent, pool } = makeMockAgent();
+      const svc = makeService(db, pool);
+
+      pool
+        .intercept({
+          path: '/api/federation/v1/list/tasks',
+          method: 'POST',
+        })
+        .reply(200, LIST_BODY, { headers: { 'content-type': 'application/json' } });
+
+      const result = await svc.list(PEER_ID, 'tasks', {});
+
+      expect(result.items).toHaveLength(1);
+      expect(result.items[0]).toMatchObject({ id: '1', title: 'Task One' });
+
+      await mockAgent.close();
+    });
+  });
+
+  describe('get()', () => {
+    it('returns parsed typed response on success', async () => {
+      const db = makeDb();
+      const { mockAgent, pool } = makeMockAgent();
+      const svc = makeService(db, pool);
+
+      pool
+        .intercept({
+          path: '/api/federation/v1/get/tasks/1',
+          method: 'POST',
+        })
+        .reply(200, GET_BODY, { headers: { 'content-type': 'application/json' } });
+
+      const result = await svc.get(PEER_ID, 'tasks', '1', {});
+
+      expect(result.item).toMatchObject({ id: '1', title: 'Task One' });
+
+      await mockAgent.close();
+    });
+  });
+
+  describe('capabilities()', () => {
+    it('returns parsed capabilities response on success', async () => {
+      const db = makeDb();
+      const { mockAgent, pool } = makeMockAgent();
+      const svc = makeService(db, pool);
+
+      pool
+        .intercept({
+          path: '/api/federation/v1/capabilities',
+          method: 'GET',
+        })
+        .reply(200, CAP_BODY, { headers: { 'content-type': 'application/json' } });
+
+      const result = await svc.capabilities(PEER_ID);
+
+      expect(result.resources).toContain('tasks');
+      expect(result.max_rows_per_query).toBe(100);
+
+      await mockAgent.close();
+    });
+  });
+
+  // ─── HTTP error surfaces ──────────────────────────────────────────────────
+
+  describe('non-2xx responses', () => {
+    it('surfaces 403 as FederationClientError({ status: 403, code: "FORBIDDEN" })', async () => {
+      const db = makeDb();
+      const { mockAgent, pool } = makeMockAgent();
+      const svc = makeService(db, pool);
+
+      pool.intercept({ path: '/api/federation/v1/list/tasks', method: 'POST' }).reply(
+        403,
+        { error: { code: 'forbidden', message: 'Access denied' } },
+        {
+          headers: { 'content-type': 'application/json' },
+        },
+      );
+
+      await expect(svc.list(PEER_ID, 'tasks', {})).rejects.toMatchObject({
+        status: 403,
+        code: 'FORBIDDEN',
+        peerId: PEER_ID,
+      });
+
+      await mockAgent.close();
+    });
+
+    it('surfaces 404 as FederationClientError({ status: 404, code: "HTTP_404" })', async () => {
+      const db = makeDb();
+      const { mockAgent, pool } = makeMockAgent();
+      const svc = makeService(db, pool);
+
+      pool.intercept({ path: '/api/federation/v1/get/tasks/999', method: 'POST' }).reply(
+        404,
+        { error: { code: 'not_found', message: 'Not found' } },
+        {
+          headers: { 'content-type': 'application/json' },
+        },
+      );
+
+      await expect(svc.get(PEER_ID, 'tasks', '999', {})).rejects.toMatchObject({
+        status: 404,
+        code: 'HTTP_404',
+        peerId: PEER_ID,
+      });
+
+      await mockAgent.close();
+    });
+  });
+
+  // ─── Network error ─────────────────────────────────────────────────────────
+
+  describe('network errors', () => {
+    it('surfaces network error as FederationClientError({ code: "NETWORK" })', async () => {
+      const db = makeDb();
+      const { mockAgent, pool } = makeMockAgent();
+      const svc = makeService(db, pool);
+
+      pool
+        .intercept({ path: '/api/federation/v1/capabilities', method: 'GET' })
+        .replyWithError(new Error('ECONNREFUSED'));
+
+      await expect(svc.capabilities(PEER_ID)).rejects.toMatchObject({
+        code: 'NETWORK',
+        peerId: PEER_ID,
+      });
+
+      await mockAgent.close();
+    });
+  });
+
+  // ─── Invalid response body ─────────────────────────────────────────────────
+
+  describe('invalid response body', () => {
+    it('surfaces as FederationClientError({ code: "INVALID_RESPONSE" }) when body shape is wrong', async () => {
+      const db = makeDb();
+      const { mockAgent, pool } = makeMockAgent();
+      const svc = makeService(db, pool);
+
+      // capabilities returns wrong shape (missing required fields)
+      pool
+        .intercept({ path: '/api/federation/v1/capabilities', method: 'GET' })
+        .reply(200, { totally: 'wrong' }, { headers: { 'content-type': 'application/json' } });
+
+      await expect(svc.capabilities(PEER_ID)).rejects.toMatchObject({
+        code: 'INVALID_RESPONSE',
+        peerId: PEER_ID,
+      });
+
+      await mockAgent.close();
+    });
+  });
+
+  // ─── Peer DB validation ────────────────────────────────────────────────────
+
+  describe('peer validation (without resolveEntry spy)', () => {
+    /**
+     * These tests exercise the real `resolveEntry` path — no spy on resolveEntry.
+     */
+
+    it('throws PEER_NOT_FOUND when peer is not in DB', async () => {
+      // DB returns empty array (peer not found)
+      const db = makeDb([]);
+      const svc = new FederationClientService(db);
+
+      await expect(svc.capabilities(PEER_ID)).rejects.toMatchObject({
+        code: 'PEER_NOT_FOUND',
+        peerId: PEER_ID,
+      });
+    });
+
+    it('throws PEER_INACTIVE when peer state is not "active"', async () => {
+      const db = makeDb([makePeerRow({ state: 'suspended' })]);
+      const svc = new FederationClientService(db);
+
+      await expect(svc.capabilities(PEER_ID)).rejects.toMatchObject({
+        code: 'PEER_INACTIVE',
+        peerId: PEER_ID,
+      });
+    });
+  });
+
+  // ─── Cache behaviour ───────────────────────────────────────────────────────
+
+  describe('cache behaviour', () => {
+    it('hits cache on second call — only one DB lookup happens', async () => {
+      // Verify cache by calling the private resolveEntry directly twice and
+      // asserting the DB was queried only once. This avoids the HTTP layer,
+      // which would require either a real network or per-peer Agent rewiring
+      // that the cache invariant doesn't depend on.
+      const db = makeDb();
+      const selectSpy = vi.spyOn(db, 'select');
+      const svc = new FederationClientService(db);
+      const resolveEntry = (
+        svc as unknown as { resolveEntry: (peerId: string) => Promise<unknown> }
+      ).resolveEntry.bind(svc);
+
+      const first = await resolveEntry(PEER_ID);
+      const second = await resolveEntry(PEER_ID);
+
+      expect(first).toBe(second);
+      expect(selectSpy).toHaveBeenCalledTimes(1);
+    });
+
+    it('serializes concurrent resolveEntry calls — only one DB lookup', async () => {
+      const db = makeDb();
+      const selectSpy = vi.spyOn(db, 'select');
+      const svc = new FederationClientService(db);
+      const resolveEntry = (
+        svc as unknown as {
+          resolveEntry: (peerId: string) => Promise<unknown>;
+        }
+      ).resolveEntry.bind(svc);
+
+      const [a, b] = await Promise.all([resolveEntry(PEER_ID), resolveEntry(PEER_ID)]);
+      expect(a).toBe(b);
+      expect(selectSpy).toHaveBeenCalledTimes(1);
+    });
+
+    it('flushPeer destroys the evicted Agent so old TLS connections close', async () => {
+      const db = makeDb();
+      const svc = new FederationClientService(db);
+      const resolveEntry = (
+        svc as unknown as {
+          resolveEntry: (peerId: string) => Promise<{ agent: { destroy: () => Promise<void> } }>;
+        }
+      ).resolveEntry.bind(svc);
+
+      const entry = await resolveEntry(PEER_ID);
+      const destroySpy = vi.spyOn(entry.agent, 'destroy').mockResolvedValue();
+
+      svc.flushPeer(PEER_ID);
+      expect(destroySpy).toHaveBeenCalledTimes(1);
+    });
+
+    it('flushPeer() invalidates cache — next call re-reads DB', async () => {
+      const db = makeDb();
+      const { mockAgent, pool } = makeMockAgent();
+      const svc = makeService(db, pool);
+
+      pool
+        .intercept({ path: '/api/federation/v1/capabilities', method: 'GET' })
+        .reply(200, CAP_BODY, { headers: { 'content-type': 'application/json' } })
+        .times(2);
+
+      // First call — populates cache (via mock resolveEntry)
+      await svc.capabilities(PEER_ID);
+
+      // Flush the cache
+      svc.flushPeer(PEER_ID);
+
+      // The spy on resolveEntry is still active — check it's called again after flush
+      const resolveEntrySpy = vi.spyOn(
+        svc as unknown as { resolveEntry: (peerId: string) => Promise<unknown> },
+        'resolveEntry',
+      );
+
+      // Second call after flush — should call resolveEntry again
+      await svc.capabilities(PEER_ID);
+
+      // resolveEntry should have been called once after we started spying (post-flush)
+      expect(resolveEntrySpy).toHaveBeenCalledTimes(1);
+
+      await mockAgent.close();
+    });
+  });
+
+  // ─── loadStepCaRoot env-var guard ─────────────────────────────────────────
+
+  describe('loadStepCaRoot() env-var guard', () => {
+    it('throws PEER_MISCONFIGURED when STEP_CA_ROOT_CERT_PATH is not set', async () => {
+      delete process.env['STEP_CA_ROOT_CERT_PATH'];
+      const db = makeDb();
+      const svc = new FederationClientService(db);
+      const resolveEntry = (
+        svc as unknown as {
+          resolveEntry: (peerId: string) => Promise<unknown>;
+        }
+      ).resolveEntry.bind(svc);
+
+      await expect(resolveEntry(PEER_ID)).rejects.toMatchObject({
+        code: 'PEER_MISCONFIGURED',
+      });
+    });
+  });
+
+  // ─── FederationClientError class ──────────────────────────────────────────
+
+  describe('FederationClientError', () => {
+    it('is instanceof Error and FederationClientError', () => {
+      const err = new FederationClientError({
+        code: 'PEER_NOT_FOUND',
+        message: 'test',
+        peerId: PEER_ID,
+      });
+      expect(err).toBeInstanceOf(Error);
+      expect(err).toBeInstanceOf(FederationClientError);
+      expect(err.name).toBe('FederationClientError');
+    });
+
+    it('carries status, code, and peerId', () => {
+      const err = new FederationClientError({
+        status: 403,
+        code: 'FORBIDDEN',
+        message: 'forbidden',
+        peerId: PEER_ID,
+      });
+      expect(err.status).toBe(403);
+      expect(err.code).toBe('FORBIDDEN');
+      expect(err.peerId).toBe(PEER_ID);
+    });
+  });
+});

apps/gateway/src/federation/client/federation-client.service.ts

@@ -0,0 +1,500 @@
+/**
+ * FederationClientService — outbound mTLS client for federation requests (FED-M3-08).
+ *
+ * Dials peer gateways over mTLS using the cert+sealed-key stored in `federation_peers`,
+ * invokes federation verbs (list / get / capabilities), and surfaces all failure modes
+ * as typed `FederationClientError` instances.
+ *
+ * ## Error code taxonomy
+ *
+ * | Code               | When                                                          |
+ * | ------------------ | ------------------------------------------------------------- |
+ * | PEER_NOT_FOUND     | No row in federation_peers for the given peerId               |
+ * | PEER_INACTIVE      | Peer row exists but state !== 'active'                        |
+ * | PEER_MISCONFIGURED | Peer row is active but missing endpointUrl or clientKeyPem    |
+ * | NETWORK            | undici threw a connection / TLS / timeout error               |
+ * | HTTP_{status}      | Peer returned a non-2xx response (e.g. HTTP_403, HTTP_404)    |
+ * | FORBIDDEN          | Peer returned 403 (convenience alias alongside HTTP_403)      |
+ * | INVALID_RESPONSE   | Response body failed Zod schema validation                    |
+ *
+ * ## Cache strategy
+ *
+ * Per-peer `undici.Agent` instances are cached in a `Map<peerId, AgentCacheEntry>` for
+ * the lifetime of the service instance.  The cache is keyed on peerId (UUID).
+ *
+ * Cache invalidation:
+ *  - `flushPeer(peerId)` — removes the entry immediately.  M5/M6 MUST call this on
+ *    cert rotation or peer revocation events so the next request re-reads the DB and
+ *    builds a fresh TLS Agent with the new cert material.
+ *  - On cache miss: re-reads the DB, checks state === 'active', rebuilds Agent.
+ *
+ * Cache does NOT auto-expire.  The service is expected to be a singleton scoped to the
+ * NestJS application lifecycle; flushing on revocation/rotation is the only invalidation
+ * path by design (avoids redundant DB round-trips on the hot path).
+ */
+
+import { Injectable, Inject, Logger } from '@nestjs/common';
+import { readFileSync } from 'node:fs';
+import { Agent, fetch as undiciFetch } from 'undici';
+import type { Dispatcher } from 'undici';
+import { z } from 'zod';
+import { type Db, eq, federationPeers } from '@mosaicstack/db';
+import {
+  FederationListResponseSchema,
+  FederationGetResponseSchema,
+  FederationCapabilitiesResponseSchema,
+  FederationErrorEnvelopeSchema,
+  type FederationListResponse,
+  type FederationGetResponse,
+  type FederationCapabilitiesResponse,
+} from '@mosaicstack/types';
+import { DB } from '../../database/database.module.js';
+import { unsealClientKey } from '../peer-key.util.js';
+
+// ---------------------------------------------------------------------------
+// Error taxonomy
+// ---------------------------------------------------------------------------
+
+/**
+ * Client-side error code set.  Distinct from the server-side `FederationErrorCode`
+ * (which lives in `@mosaicstack/types`) because the client has additional failure
+ * modes (PEER_NOT_FOUND, PEER_INACTIVE, PEER_MISCONFIGURED, NETWORK) that the
+ * server never emits.
+ */
+export type FederationClientErrorCode =
+  | 'PEER_NOT_FOUND'
+  | 'PEER_INACTIVE'
+  | 'PEER_MISCONFIGURED'
+  | 'NETWORK'
+  | 'FORBIDDEN'
+  | 'INVALID_RESPONSE'
+  | `HTTP_${number}`;
+
+export interface FederationClientErrorOptions {
+  status?: number;
+  code: FederationClientErrorCode;
+  message: string;
+  peerId: string;
+  cause?: unknown;
+}
+
+/**
+ * Thrown by FederationClientService on every failure path.
+ * Callers can dispatch on `error.code` for programmatic handling.
+ */
+export class FederationClientError extends Error {
+  readonly status?: number;
+  readonly code: FederationClientErrorCode;
+  readonly peerId: string;
+  readonly cause?: unknown;
+
+  constructor(opts: FederationClientErrorOptions) {
+    super(opts.message);
+    this.name = 'FederationClientError';
+    this.status = opts.status;
+    this.code = opts.code;
+    this.peerId = opts.peerId;
+    this.cause = opts.cause;
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Internal cache types
+// ---------------------------------------------------------------------------
+
+interface AgentCacheEntry {
+  agent: Agent;
+  endpointUrl: string;
+  certPem: string;
+  certSerial: string;
+}
+
+// ---------------------------------------------------------------------------
+// Service
+// ---------------------------------------------------------------------------
+
+@Injectable()
+export class FederationClientService {
+  private readonly logger = new Logger(FederationClientService.name);
+
+  /**
+   * Per-peer undici Agent cache.
+   * Key = peerId (UUID string).
+   *
+   * Values are either a resolved `AgentCacheEntry` or an in-flight
+   * `Promise<AgentCacheEntry>` (promise-cache pattern).  Storing the promise
+   * prevents duplicate DB lookups and duplicate key-unseal operations when two
+   * requests for the same peer arrive before the first build completes.
+   *
+   * Flush via `flushPeer(peerId)` on cert rotation / peer revocation (M5/M6).
+   */
+  private readonly cache = new Map<string, AgentCacheEntry | Promise<AgentCacheEntry>>();
+
+  /**
+   * Step-CA root cert PEM, loaded once from `STEP_CA_ROOT_CERT_PATH`.
+   * Used as the trust anchor for peer server certificates so federation TLS is
+   * pinned to our PKI, not the public trust store. Lazily loaded on first use
+   * so unit tests that don't exercise the agent path can run without the env var.
+   */
+  private cachedCaPem: string | null = null;
+
+  constructor(@Inject(DB) private readonly db: Db) {}
+
+  // -------------------------------------------------------------------------
+  // Public verb API
+  // -------------------------------------------------------------------------
+
+  /**
+   * Invoke the `list` verb on a remote peer.
+   *
+   * @param peerId   UUID of the peer row in `federation_peers`.
+   * @param resource Resource path, e.g. "tasks".
+   * @param request  Free-form body sent as JSON in the POST body.
+   * @returns Parsed `FederationListResponse<T>`.
+   */
+  async list<T>(
+    peerId: string,
+    resource: string,
+    request: Record<string, unknown>,
+  ): Promise<FederationListResponse<T>> {
+    const { endpointUrl, agent } = await this.resolveEntry(peerId);
+    const url = `${endpointUrl}/api/federation/v1/list/${encodeURIComponent(resource)}`;
+    const body = await this.doPost(peerId, url, agent, request);
+    return this.parseWith<FederationListResponse<T>>(
+      peerId,
+      body,
+      FederationListResponseSchema(z.unknown()),
+    );
+  }
+
+  /**
+   * Invoke the `get` verb on a remote peer.
+   *
+   * @param peerId   UUID of the peer row in `federation_peers`.
+   * @param resource Resource path, e.g. "tasks".
+   * @param id       Resource identifier.
+   * @param request  Free-form body sent as JSON in the POST body.
+   * @returns Parsed `FederationGetResponse<T>`.
+   */
+  async get<T>(
+    peerId: string,
+    resource: string,
+    id: string,
+    request: Record<string, unknown>,
+  ): Promise<FederationGetResponse<T>> {
+    const { endpointUrl, agent } = await this.resolveEntry(peerId);
+    const url = `${endpointUrl}/api/federation/v1/get/${encodeURIComponent(resource)}/${encodeURIComponent(id)}`;
+    const body = await this.doPost(peerId, url, agent, request);
+    return this.parseWith<FederationGetResponse<T>>(
+      peerId,
+      body,
+      FederationGetResponseSchema(z.unknown()),
+    );
+  }
+
+  /**
+   * Invoke the `capabilities` verb on a remote peer.
+   *
+   * @param peerId UUID of the peer row in `federation_peers`.
+   * @returns Parsed `FederationCapabilitiesResponse`.
+   */
+  async capabilities(peerId: string): Promise<FederationCapabilitiesResponse> {
+    const { endpointUrl, agent } = await this.resolveEntry(peerId);
+    const url = `${endpointUrl}/api/federation/v1/capabilities`;
+    const body = await this.doGet(peerId, url, agent);
+    return this.parseWith<FederationCapabilitiesResponse>(
+      peerId,
+      body,
+      FederationCapabilitiesResponseSchema,
+    );
+  }
+
+  // -------------------------------------------------------------------------
+  // Cache management
+  // -------------------------------------------------------------------------
+
+  /**
+   * Flush the cached Agent for a specific peer.
+   *
+   * M5/M6 MUST call this on:
+   *  - cert rotation events (so new cert material is picked up)
+   *  - peer revocation events (so future requests fail at PEER_INACTIVE)
+   *
+   * After flushing, the next call to `list`, `get`, or `capabilities` for
+   * this peer will re-read the DB and rebuild the Agent.
+   */
+  flushPeer(peerId: string): void {
+    const entry = this.cache.get(peerId);
+    if (entry === undefined) {
+      return;
+    }
+    this.cache.delete(peerId);
+    if (!(entry instanceof Promise)) {
+      // best-effort destroy; promise-cached entries skip destroy because
+      // the in-flight build owns its own Agent which will be GC'd when the
+      // owning request handles the rejection from the cache miss
+      entry.agent.destroy().catch(() => {
+        // intentionally ignored — destroy errors are not actionable
+      });
+    }
+    this.logger.log(`Cache flushed for peer ${peerId}`);
+  }
+
+  // -------------------------------------------------------------------------
+  // Internal helpers
+  // -------------------------------------------------------------------------
+
+  /**
+   * Load and cache the Step-CA root cert PEM from `STEP_CA_ROOT_CERT_PATH`.
+   * Throws `FederationClientError` if the env var is unset or the file cannot
+   * be read — mTLS to a peer without a pinned trust anchor would silently
+   * fall back to the public trust store.
+   */
+  private loadStepCaRoot(): string {
+    if (this.cachedCaPem !== null) {
+      return this.cachedCaPem;
+    }
+    const path = process.env['STEP_CA_ROOT_CERT_PATH'];
+    if (!path) {
+      throw new FederationClientError({
+        code: 'PEER_MISCONFIGURED',
+        message: 'STEP_CA_ROOT_CERT_PATH is not set; refusing to dial peer without pinned CA trust',
+        peerId: '',
+      });
+    }
+    try {
+      const pem = readFileSync(path, 'utf8');
+      this.cachedCaPem = pem;
+      return pem;
+    } catch (err) {
+      throw new FederationClientError({
+        code: 'PEER_MISCONFIGURED',
+        message: `Failed to read STEP_CA_ROOT_CERT_PATH (${path})`,
+        peerId: '',
+        cause: err,
+      });
+    }
+  }
+
+  /**
+   * Resolve the cache entry for a peer, reading DB on miss.
+   *
+   * Uses a promise-cache pattern: concurrent callers for the same uncached
+   * `peerId` all `await` the same in-flight `Promise<AgentCacheEntry>` so
+   * only one DB lookup and one key-unseal ever runs per peer per cache miss.
+   * The promise is replaced with the concrete entry on success, or deleted on
+   * rejection so a transient error does not poison the cache permanently.
+   *
+   * Throws `FederationClientError` with appropriate code if the peer is not
+   * found, is inactive, or is missing required fields.
+   */
+  private async resolveEntry(peerId: string): Promise<AgentCacheEntry> {
+    const cached = this.cache.get(peerId);
+    if (cached) {
+      return cached; // Promise or concrete entry — both are awaitable
+    }
+
+    const inflight = this.buildEntry(peerId).then(
+      (entry) => {
+        this.cache.set(peerId, entry); // replace promise with concrete value
+        return entry;
+      },
+      (err: unknown) => {
+        this.cache.delete(peerId); // don't poison the cache with a rejected promise
+        throw err;
+      },
+    );
+
+    this.cache.set(peerId, inflight);
+    return inflight;
+  }
+
+  /**
+   * Build the `AgentCacheEntry` for a peer by reading the DB, validating the
+   * peer's state, unsealing the private key, and constructing the mTLS Agent.
+   *
+   * Throws `FederationClientError` with appropriate code if the peer is not
+   * found, is inactive, or is missing required fields.
+   */
+  private async buildEntry(peerId: string): Promise<AgentCacheEntry> {
+    // DB lookup
+    const [peer] = await this.db
+      .select()
+      .from(federationPeers)
+      .where(eq(federationPeers.id, peerId))
+      .limit(1);
+
+    if (!peer) {
+      throw new FederationClientError({
+        code: 'PEER_NOT_FOUND',
+        message: `Federation peer ${peerId} not found`,
+        peerId,
+      });
+    }
+
+    if (peer.state !== 'active') {
+      throw new FederationClientError({
+        code: 'PEER_INACTIVE',
+        message: `Federation peer ${peerId} is not active (state: ${peer.state})`,
+        peerId,
+      });
+    }
+
+    if (!peer.endpointUrl || !peer.clientKeyPem) {
+      throw new FederationClientError({
+        code: 'PEER_MISCONFIGURED',
+        message: `Federation peer ${peerId} is missing endpointUrl or clientKeyPem`,
+        peerId,
+      });
+    }
+
+    // Unseal the private key
+    let privateKeyPem: string;
+    try {
+      privateKeyPem = unsealClientKey(peer.clientKeyPem);
+    } catch (err) {
+      throw new FederationClientError({
+        code: 'PEER_MISCONFIGURED',
+        message: `Failed to unseal client key for peer ${peerId}`,
+        peerId,
+        cause: err,
+      });
+    }
+
+    // Build mTLS agent — pin trust to Step-CA root so we never accept
+    // a peer cert signed by a public CA (defense against MITM with a
+    // publicly-trusted DV cert for the peer's hostname).
+    const agent = new Agent({
+      connect: {
+        cert: peer.certPem,
+        key: privateKeyPem,
+        ca: this.loadStepCaRoot(),
+        // rejectUnauthorized: true is the undici default for HTTPS
+      },
+    });
+
+    const entry: AgentCacheEntry = {
+      agent,
+      endpointUrl: peer.endpointUrl,
+      certPem: peer.certPem,
+      certSerial: peer.certSerial,
+    };
+
+    this.logger.log(`Agent cached for peer ${peerId} (serial: ${peer.certSerial})`);
+
+    return entry;
+  }
+
+  /**
+   * Execute a POST request with a JSON body.
+   * Returns the parsed response body as an unknown value.
+   * Throws `FederationClientError` on network errors and non-2xx responses.
+   */
+  private async doPost(
+    peerId: string,
+    url: string,
+    agent: Dispatcher,
+    body: Record<string, unknown>,
+  ): Promise<unknown> {
+    return this.doRequest(peerId, url, agent, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify(body),
+    });
+  }
+
+  /**
+   * Execute a GET request.
+   * Returns the parsed response body as an unknown value.
+   * Throws `FederationClientError` on network errors and non-2xx responses.
+   */
+  private async doGet(peerId: string, url: string, agent: Dispatcher): Promise<unknown> {
+    return this.doRequest(peerId, url, agent, { method: 'GET' });
+  }
+
+  private async doRequest(
+    peerId: string,
+    url: string,
+    agent: Dispatcher,
+    init: { method: string; headers?: Record<string, string>; body?: string },
+  ): Promise<unknown> {
+    let response: Awaited<ReturnType<typeof undiciFetch>>;
+
+    try {
+      response = await undiciFetch(url, {
+        ...init,
+        dispatcher: agent,
+      });
+    } catch (err) {
+      throw new FederationClientError({
+        code: 'NETWORK',
+        message: `Network error calling peer ${peerId} at ${url}: ${err instanceof Error ? err.message : String(err)}`,
+        peerId,
+        cause: err,
+      });
+    }
+
+    const rawBody = await response.text().catch(() => '');
+
+    if (!response.ok) {
+      const status = response.status;
+
+      // Attempt to parse as federation error envelope
+      let serverMessage = `HTTP ${status}`;
+      try {
+        const json: unknown = JSON.parse(rawBody);
+        const result = FederationErrorEnvelopeSchema.safeParse(json);
+        if (result.success) {
+          serverMessage = result.data.error.message;
+        }
+      } catch {
+        // Not valid JSON or not a federation envelope — use generic message
+      }
+
+      // Specific code for 403 (most actionable for callers); generic HTTP_{n} for others
+      const code: FederationClientErrorCode = status === 403 ? 'FORBIDDEN' : `HTTP_${status}`;
+
+      throw new FederationClientError({
+        status,
+        code,
+        message: `Peer ${peerId} returned ${status}: ${serverMessage}`,
+        peerId,
+      });
+    }
+
+    try {
+      return JSON.parse(rawBody) as unknown;
+    } catch (err) {
+      throw new FederationClientError({
+        code: 'INVALID_RESPONSE',
+        message: `Peer ${peerId} returned non-JSON body`,
+        peerId,
+        cause: err,
+      });
+    }
+  }
+
+  /**
+   * Parse and validate a response body against a Zod schema.
+   *
+   * For list/get, callers pass the result of `FederationListResponseSchema(z.unknown())`
+   * so that the envelope structure is validated without requiring a concrete item schema
+   * at the client level.  The generic `T` provides compile-time typing.
+   *
+   * Throws `FederationClientError({ code: 'INVALID_RESPONSE' })` on parse failure.
+   */
+  private parseWith<T>(peerId: string, body: unknown, schema: z.ZodTypeAny): T {
+    const result = schema.safeParse(body);
+    if (!result.success) {
+      const issues = result.error.issues
+        .map((e: z.ZodIssue) => `[${e.path.join('.') || 'root'}] ${e.message}`)
+        .join('; ');
+      throw new FederationClientError({
+        code: 'INVALID_RESPONSE',
+        message: `Peer ${peerId} returned invalid response shape: ${issues}`,
+        peerId,
+      });
+    }
+    return result.data as T;
+  }
+}

apps/gateway/src/federation/client/index.ts

@@ -0,0 +1,13 @@
+/**
+ * Federation client barrel — re-exports for FederationModule consumers.
+ *
+ * M3-09 (QuerySourceService) and future milestones should import from here,
+ * not directly from the implementation file.
+ */
+
+export {
+  FederationClientService,
+  FederationClientError,
+  type FederationClientErrorCode,
+  type FederationClientErrorOptions,
+} from './federation-client.service.js';

apps/gateway/src/federation/federation.module.ts

@@ -5,10 +5,11 @@ import { EnrollmentController } from './enrollment.controller.js';
 import { EnrollmentService } from './enrollment.service.js';
 import { FederationController } from './federation.controller.js';
 import { GrantsService } from './grants.service.js';
+import { FederationClientService } from './client/index.js';
 
 @Module({
   controllers: [EnrollmentController, FederationController],
-  providers: [AdminGuard, CaService, EnrollmentService, GrantsService],
-  exports: [CaService, EnrollmentService, GrantsService],
+  providers: [AdminGuard, CaService, EnrollmentService, GrantsService, FederationClientService],
+  exports: [CaService, EnrollmentService, GrantsService, FederationClientService],
 })
 export class FederationModule {}

pnpm-lock.yaml

@@ -179,6 +179,9 @@ importers:
       socket.io:
         specifier: ^4.8.0
         version: 4.8.3
+      undici:
+        specifier: ^7.24.6
+        version: 7.24.6
       uuid:
         specifier: ^11.0.0
         version: 11.1.0
@@ -713,10 +716,10 @@ importers:
     dependencies:
       '@mariozechner/pi-agent-core':
         specifier: ^0.63.1
-        version: 0.63.2(@modelcontextprotocol/sdk@1.28.0(zod@4.3.6))(ws@8.20.0)(zod@3.25.76)
+        version: 0.63.2(@modelcontextprotocol/sdk@1.28.0(zod@4.3.6))(ws@8.20.0)(zod@4.3.6)
       '@mariozechner/pi-ai':
         specifier: ^0.63.1
-        version: 0.63.2(@modelcontextprotocol/sdk@1.28.0(zod@4.3.6))(ws@8.20.0)(zod@3.25.76)
+        version: 0.63.2(@modelcontextprotocol/sdk@1.28.0(zod@4.3.6))(ws@8.20.0)(zod@4.3.6)
       '@sinclair/typebox':
         specifier: ^0.34.41
         version: 0.34.48
@@ -6993,10 +6996,6 @@ packages:
     resolution: {integrity: sha512-gBLkYIlEnSp8pFbT64yFgGE6UIB9tAkhukC23PmMDCe5Nd+cRqKxSjw5y54MK2AZMgZfJWMaNE4nYUHgi1XEOw==}
     engines: {node: '>=18.17'}
 
-  undici@7.24.3:
-    resolution: {integrity: sha512-eJdUmK/Wrx2d+mnWWmwwLRyA7OQCkLap60sk3dOK4ViZR7DKwwptwuIvFBg2HaiP9ESaEdhtpSymQPvytpmkCA==}
-    engines: {node: '>=20.18.1'}
-
   undici@7.24.6:
     resolution: {integrity: sha512-Xi4agocCbRzt0yYMZGMA6ApD7gvtUFaxm4ZmeacWI4cZxaF6C+8I8QfofC20NAePiB/IcvZmzkJ7XPa471AEtA==}
     engines: {node: '>=20.18.1'}
@@ -7329,12 +7328,6 @@ snapshots:
       '@jridgewell/gen-mapping': 0.3.13
       '@jridgewell/trace-mapping': 0.3.31
 
-  '@anthropic-ai/sdk@0.73.0(zod@3.25.76)':
-    dependencies:
-      json-schema-to-ts: 3.1.1
-    optionalDependencies:
-      zod: 3.25.76
-
   '@anthropic-ai/sdk@0.73.0(zod@4.3.6)':
     dependencies:
       json-schema-to-ts: 3.1.1
@@ -8676,18 +8669,6 @@ snapshots:
       - ws
       - zod
 
-  '@mariozechner/pi-agent-core@0.63.2(@modelcontextprotocol/sdk@1.28.0(zod@4.3.6))(ws@8.20.0)(zod@3.25.76)':
-    dependencies:
-      '@mariozechner/pi-ai': 0.63.2(@modelcontextprotocol/sdk@1.28.0(zod@4.3.6))(ws@8.20.0)(zod@3.25.76)
-    transitivePeerDependencies:
-      - '@modelcontextprotocol/sdk'
-      - aws-crt
-      - bufferutil
-      - supports-color
-      - utf-8-validate
-      - ws
-      - zod
-
   '@mariozechner/pi-agent-core@0.63.2(@modelcontextprotocol/sdk@1.28.0(zod@4.3.6))(ws@8.20.0)(zod@4.3.6)':
     dependencies:
       '@mariozechner/pi-ai': 0.63.2(@modelcontextprotocol/sdk@1.28.0(zod@4.3.6))(ws@8.20.0)(zod@4.3.6)
@@ -8736,30 +8717,6 @@ snapshots:
       - ws
       - zod
 
-  '@mariozechner/pi-ai@0.63.2(@modelcontextprotocol/sdk@1.28.0(zod@4.3.6))(ws@8.20.0)(zod@3.25.76)':
-    dependencies:
-      '@anthropic-ai/sdk': 0.73.0(zod@3.25.76)
-      '@aws-sdk/client-bedrock-runtime': 3.1008.0
-      '@google/genai': 1.45.0(@modelcontextprotocol/sdk@1.28.0(zod@4.3.6))
-      '@mistralai/mistralai': 1.14.1
-      '@sinclair/typebox': 0.34.48
-      ajv: 8.18.0
-      ajv-formats: 3.0.1(ajv@8.18.0)
-      chalk: 5.6.2
-      openai: 6.26.0(ws@8.20.0)(zod@3.25.76)
-      partial-json: 0.1.7
-      proxy-agent: 6.5.0
-      undici: 7.24.3
-      zod-to-json-schema: 3.25.1(zod@3.25.76)
-    transitivePeerDependencies:
-      - '@modelcontextprotocol/sdk'
-      - aws-crt
-      - bufferutil
-      - supports-color
-      - utf-8-validate
-      - ws
-      - zod
-
   '@mariozechner/pi-ai@0.63.2(@modelcontextprotocol/sdk@1.28.0(zod@4.3.6))(ws@8.20.0)(zod@4.3.6)':
     dependencies:
       '@anthropic-ai/sdk': 0.73.0(zod@4.3.6)
@@ -8773,7 +8730,7 @@ snapshots:
       openai: 6.26.0(ws@8.20.0)(zod@4.3.6)
       partial-json: 0.1.7
       proxy-agent: 6.5.0
-      undici: 7.24.3
+      undici: 7.24.6
       zod-to-json-schema: 3.25.1(zod@4.3.6)
     transitivePeerDependencies:
       - '@modelcontextprotocol/sdk'
@@ -12632,7 +12589,7 @@ snapshots:
       saxes: 6.0.0
       symbol-tree: 3.2.4
       tough-cookie: 6.0.1
-      undici: 7.24.3
+      undici: 7.24.6
       w3c-xmlserializer: 5.0.0
       webidl-conversions: 8.0.1
       whatwg-mimetype: 5.0.0
@@ -13352,11 +13309,6 @@ snapshots:
     dependencies:
       mimic-function: 5.0.1
 
-  openai@6.26.0(ws@8.20.0)(zod@3.25.76):
-    optionalDependencies:
-      ws: 8.20.0
-      zod: 3.25.76
-
   openai@6.26.0(ws@8.20.0)(zod@4.3.6):
     optionalDependencies:
       ws: 8.20.0
@@ -14488,8 +14440,6 @@ snapshots:
 
   undici@6.21.3: {}
 
-  undici@7.24.3: {}
-
   undici@7.24.6: {}
 
   unhomoglyph@1.0.6: {}