badgr-cli 1.1.0 → 1.1.2

package/tests/serve-lifecycle.test.js

@@ -1,931 +0,0 @@
-/**
- * badgr serve — end-to-end lifecycle tests (mocked API)
- *
- * Covers the launch-readiness matrix:
- *   - Known-good model on two GPU types; endpoint health check passes
- *   - Stop with `badgr down` → terminateDeployment called, billing ends
- *   - Deployment fails during startup (OOM/crash) → clear receipt before charges
- *   - Endpoint health-check timeout → receipt shows health_check_timeout
- *   - Invalid model ID / no capacity → error surfaced before meaningful charges
- *   - Private HF repo without credentials → PROVISIONING_FAILED before charges
- *   - Incompatible model/template → PROVISIONING_FAILED before charges
- *   - --no-wait skips health check
- *   - Custom image skips health check automatically
- *   - Tier-1 unavailable → auto-expand to tier-2
- *   - Receipt records providerRoute, tier, gpu, endpointUrl, status
- *
- * Readiness is driven entirely by Badgr's own deployment status
- * (GET /deployments/{id} — endpoint_ready/health_path/readiness_reason/fix_hint).
- * The CLI never fetches the RunPod proxy/pod endpoint directly, so these tests
- * mock api.callApi responses rather than global.fetch.
- */
-import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
-import { serveCommand } from '../src/commands/serve.js';
-
-// ── Module mocks ──────────────────────────────────────────────────────────────
-
-vi.mock('../src/api.js', () => ({
-  callApi:             vi.fn(),
-  terminateDeployment: vi.fn().mockResolvedValue({}),
-  listDeployments:     vi.fn().mockResolvedValue({ deployments: [], count: 0 }),
-}));
-
-vi.mock('../src/store.js', () => ({
-  addDeployment:        vi.fn(),
-  addReceipt:           vi.fn(),
-  updateReceipt:        vi.fn(),
-  generateReceiptId:    vi.fn(() => 'rcpt-serve-001'),
-  generateDeploymentId: vi.fn(() => 'dep-serve-001'),
-  listDeployments:      vi.fn(() => []),
-  listReceipts:         vi.fn(() => []),
-  findDeployment:       vi.fn(() => null),
-  updateDeployment:     vi.fn(),
-  removeDeployment:     vi.fn(),
-}));
-
-import * as api   from '../src/api.js';
-import * as store from '../src/store.js';
-
-// ── Helpers ───────────────────────────────────────────────────────────────────
-
-const config = { apiKey: 'sk-test', baseUrl: 'https://api.test/v1' };
-
-const chalk = {
-  bold:   s => s,
-  dim:    s => s,
-  red:    s => s,
-  yellow: s => s,
-  green:  s => s,
-  cyan:   s => s,
-};
-
-const ENDPOINT_URL = 'https://dep-serve-001.aibadgr.com/v1';
-
-function makeServeDep(overrides = {}) {
-  return {
-    deployment_id: 'dep-serve-001',
-    status:        'running',
-    gpu_type:      'L40S',
-    gpu_count:     1,
-    cost_per_hour: 1.80,
-    provider:      'runpod',
-    receipt_id:    'rcpt-serve-001',
-    tier:          '1',
-    endpoint_url:  ENDPOINT_URL,
-    model:         'meta-llama/Llama-3.1-8B-Instruct',
-    ...overrides,
-  };
-}
-
-// A GET /deployments/{id} response where the backend reports the app-level
-// endpoint is ready (endpoint_ready:true) — this is what the CLI now waits on
-// instead of fetching the pod directly.
-function readyStatus(overrides = {}) {
-  return { status: 'running', endpoint_ready: true, health_path: '/v1/models', ...overrides };
-}
-
-function notReadyStatus(overrides = {}) {
-  return { status: 'running', endpoint_ready: false, health_path: '/v1/models', readiness_reason: 'starting', ...overrides };
-}
-
-// ── Setup / teardown ──────────────────────────────────────────────────────────
-
-beforeEach(() => {
-  vi.useFakeTimers();
-  process.exitCode = undefined;
-  vi.spyOn(console, 'log').mockImplementation(() => {});
-  vi.spyOn(console, 'error').mockImplementation(() => {});
-  vi.spyOn(process.stdout, 'write').mockImplementation(() => true);
-  vi.clearAllMocks();
-  store.generateReceiptId.mockReturnValue('rcpt-serve-001');
-  api.terminateDeployment.mockResolvedValue({});
-  // No test should ever hit the network directly — waitForEndpoint polls
-  // Badgr's own API only. Fail loudly if anything calls fetch.
-  global.fetch = vi.fn().mockRejectedValue(new Error('serve.js must not call fetch() directly'));
-});
-
-afterEach(() => {
-  vi.useRealTimers();
-  vi.restoreAllMocks();
-  process.exitCode = undefined;
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 1. Known-good model, endpoint health check passes
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('successful serve', () => {
-  it('Llama-3.1-8B on L40S via RunPod — endpoint ready, receipt records state', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep())                    // POST /serve
-      .mockResolvedValueOnce({ status: 'running' })              // pre-health status check
-      .mockResolvedValueOnce(readyStatus());                     // waitForEndpoint poll
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--gpu', 'L40S', '--max-cost', '5'], chalk);
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    expect(store.addDeployment).toHaveBeenCalledWith(expect.objectContaining({
-      id:          'dep-serve-001',
-      gpu:         'L40S',
-      endpointUrl: ENDPOINT_URL,
-    }));
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({
-      status:      'ready',
-      endpointUrl: ENDPOINT_URL,
-    }));
-    expect(global.fetch).not.toHaveBeenCalled();
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('Mistral-7B on A100 via Vast.ai — endpoint ready', async () => {
-    const dep = makeServeDep({ gpu_type: 'A100', provider: 'vastai', model: 'mistralai/Mistral-7B-v0.1' });
-    api.callApi
-      .mockResolvedValueOnce(dep)
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce(readyStatus());
-
-    const p = serveCommand(config, ['mistralai/Mistral-7B-v0.1', '--gpu', 'A100', '--max-cost', '5'], chalk);
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    expect(store.addReceipt).toHaveBeenCalledWith(expect.objectContaining({
-      gpu:           'A100',
-      providerRoute: 'vastai',
-    }));
-    expect(process.exitCode).toBeFalsy();
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 2. Stop with badgr down — terminateDeployment called (tested via serve's own teardown)
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('billing ends when deployment stops', () => {
-  it('serveCommand calls terminateDeployment on SIGINT (graceful stop)', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep())
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce(readyStatus());
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '5'], chalk);
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    // Endpoint started and receipt updated — billing tracking is in the receipt
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({
-      status: 'ready',
-    }));
-    // badgr down is a separate command; here we verify receipt has endpointUrl and deploymentId for it
-    const addCall = store.addReceipt.mock.calls[0];
-    expect(addCall[0]).toMatchObject({
-      deploymentId: 'dep-serve-001',
-    });
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 3. Deployment fails during startup (OOM / crash)
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('deployment fails during startup', () => {
-  it('reports failure and records failed receipt before health check runs', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep()) // POST /serve
-      .mockResolvedValueOnce({ status: 'failed', error: 'OOM: not enough VRAM' }); // pre-health check
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '5'], chalk);
-    await vi.advanceTimersByTimeAsync(2000);
-    await p;
-
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({
-      status: 'failed',
-    }));
-    expect(process.exitCode).toBe(1);
-  });
-
-  it('detects dep failure mid-health-check via waitForEndpoint dep status poll', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep()) // POST /serve
-      .mockResolvedValueOnce({ status: 'running' }) // pre-health check ok
-      .mockResolvedValueOnce({ status: 'failed', error: 'CUDA error' }); // poll inside waitForEndpoint
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '5'], chalk);
-    // need to advance past the 8000ms sleep in waitForEndpoint loop
-    await vi.advanceTimersByTimeAsync(10000);
-    await p;
-
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({
-      status: 'failed',
-    }));
-    expect(process.exitCode).toBe(1);
-  });
-
-  it('surfaces backend fix_hint as the failure reason (pod exited / readiness timeout)', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep())
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce({
-        status: 'failed',
-        error: 'pod_exited',
-        fix_hint: 'The pod stopped running before the endpoint ever responded — likely an OOM.',
-      });
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '5'], chalk);
-    await vi.advanceTimersByTimeAsync(10000);
-    await p;
-
-    expect(process.exitCode).toBe(1);
-    const errLines = console.error.mock.calls.flat().join('\n');
-    expect(errLines).toContain('likely an OOM');
-  });
-
-  it('prints failure_class/next_action from the backend using the shared vocabulary', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep())
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce({
-        status: 'failed',
-        error: 'pod_exited',
-        fix_hint: 'The pod stopped running before the endpoint ever responded.',
-        failure_class: 'container_start_failed',
-        next_action: 'Check the image and command, then retry.',
-      });
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '5'], chalk);
-    await vi.advanceTimersByTimeAsync(10000);
-    await p;
-
-    const errLines = console.error.mock.calls.flat().join('\n');
-    expect(errLines).toContain('Class: container_start_failed');
-    expect(errLines).toContain('Next: Check the image and command, then retry.');
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 4. Endpoint health-check timeout
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('health check timeout', () => {
-  it('records health_check_timeout if endpoint_ready never becomes true within 15 minutes', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep()) // POST /serve
-      .mockResolvedValueOnce({ status: 'running' }) // pre-health check
-      .mockResolvedValue(notReadyStatus()); // all subsequent dep polls: still not ready
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '5'], chalk);
-    // waitForEndpoint timeout is VLLM_SERVE_WAIT_MS = 15 * 60 * 1000
-    await vi.advanceTimersByTimeAsync(910000);
-    await p;
-
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({
-      status: 'health_check_timeout',
-    }));
-    // Should NOT set exitCode = 1 (still starting, user can check badgr logs)
-    // The command returns normally (exitCode is set only when failing to start)
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 5. Invalid model ID / no capacity
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('invalid model ID or no capacity', () => {
-  it('surfaces CapacityError for unknown model / no matching GPU', async () => {
-    const capacityErr = Object.assign(new Error('no capacity'), {
-      errorData: { code: 'NO_CAPACITY_MATCH' },
-      httpStatus: 503,
-    });
-    // Both tier-1 and tier-2 fail
-    api.callApi.mockRejectedValue(capacityErr);
-
-    const errLines = [];
-    console.error.mockImplementation(msg => errLines.push(msg));
-
-    await serveCommand(config, ['invalid/nonexistent-model-xyz', '--max-cost', '5'], chalk);
-
-    expect(process.exitCode).toBe(1);
-    const combined = errLines.join('\n');
-    expect(combined).toMatch(/No suitable GPU capacity|capacity/i);
-    // No receipt should reflect billing — addDeployment was never called
-    expect(store.addDeployment).not.toHaveBeenCalled();
-  });
-
-  it('surfaces PROVISIONING_FAILED for private HF repo without credentials', async () => {
-    const provErr = Object.assign(new Error('provisioning failed'), {
-      errorData: { code: 'PROVISIONING_FAILED', debug_error: 'HF_TOKEN missing, repo is private' },
-      httpStatus: 503,
-    });
-    // tier-1 provisioning fails; tier-2 is also tried and fails
-    api.callApi.mockRejectedValue(provErr);
-
-    const errLines = [];
-    console.error.mockImplementation(msg => errLines.push(msg));
-
-    await serveCommand(config, ['my-org/private-model', '--max-cost', '5'], chalk);
-
-    expect(process.exitCode).toBe(1);
-    // Debug info hidden by default (BADGR_DEBUG not set)
-    expect(errLines.join('\n')).not.toContain('HF_TOKEN missing');
-    // Clear failure message provided
-    expect(errLines.join('\n')).toMatch(/failed to start|try again/i);
-    expect(store.addDeployment).not.toHaveBeenCalled();
-  });
-
-  it('surfaces PROVISIONING_FAILED for incompatible model/template combination', async () => {
-    const provErr = Object.assign(new Error('provisioning failed'), {
-      errorData: { code: 'PROVISIONING_FAILED', debug_error: 'chat template mismatch' },
-      httpStatus: 503,
-    });
-    api.callApi.mockRejectedValue(provErr);
-
-    await serveCommand(config, ['some/model', '--task', 'unknown_task', '--max-cost', '5'], chalk);
-
-    expect(process.exitCode).toBe(1);
-    expect(store.addDeployment).not.toHaveBeenCalled();
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 6. --no-wait skips health check
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('--no-wait flag', () => {
-  it('returns immediately without polling deployment status for readiness', async () => {
-    api.callApi.mockResolvedValueOnce(makeServeDep());
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--no-wait', '--max-cost', '5'], chalk);
-    await vi.advanceTimersByTimeAsync(100);
-    await p;
-
-    // Only the POST /serve call — no readiness polling at all.
-    expect(api.callApi).toHaveBeenCalledTimes(1);
-    // Receipt records 'starting' (not 'ready') when --no-wait is used
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({
-      status: 'starting',
-    }));
-    expect(process.exitCode).toBeFalsy();
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 7. Custom image health check behaviour
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('custom image health check', () => {
-  it('skips health check for non-comfyui --image (no known readiness path)', async () => {
-    api.callApi.mockResolvedValueOnce(makeServeDep({ model: null }));
-
-    const p = serveCommand(
-      config,
-      ['--image', 'ghcr.io/my-org/diffusers-api:latest', '--gpu', 'L40S', '--max-cost', '5'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(100);
-    await p;
-
-    expect(api.callApi).toHaveBeenCalledTimes(1);
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('auto-detects comfyui image and waits on backend-reported endpoint_ready', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ model: null }))  // POST /serve
-      .mockResolvedValueOnce({ status: 'running' })           // pre-health dep status check
-      .mockResolvedValueOnce({ status: 'running', endpoint_ready: true, health_path: '/system_stats' });
-
-    const p = serveCommand(
-      config,
-      ['--image', 'ghcr.io/my-org/comfyui-custom:latest', '--gpu', 'A100', '--max-cost', '5'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({ status: 'ready' }));
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('sends --health-path override to the backend when explicitly provided', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ model: null }))  // POST /serve
-      .mockResolvedValueOnce({ status: 'running' })           // pre-health dep status check
-      .mockResolvedValueOnce({ status: 'running', endpoint_ready: true, health_path: '/health' });
-
-    const p = serveCommand(
-      config,
-      ['--image', 'ghcr.io/my-org/stable-diffusion:latest', '--health-path', '/health', '--max-cost', '5'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    const body = api.callApi.mock.calls[0][1].body;
-    expect(body.health_path).toBe('/health');
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('sends --idle-timeout to the backend as idle_timeout_minutes', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep())
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce(readyStatus());
-
-    const p = serveCommand(
-      config,
-      ['meta-llama/Llama-3.1-8B-Instruct', '--idle-timeout', '30', '--max-cost', '5'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    const body = api.callApi.mock.calls[0][1].body;
-    expect(body.idle_timeout_minutes).toBe(30);
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('omits idle_timeout_minutes when --idle-timeout is not passed', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep())
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce(readyStatus());
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '5'], chalk);
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    const body = api.callApi.mock.calls[0][1].body;
-    expect(body.idle_timeout_minutes).toBeUndefined();
-  });
-
-  it('prints the per-endpoint API key exactly once when the backend returns one', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ endpoint_api_key: 'bge_test_key_xyz' }))
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce(readyStatus());
-
-    const lines = [];
-    console.log.mockImplementation((...args) => lines.push(args.join(' ')));
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '5'], chalk);
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    const keyLine = lines.find(l => l.includes('API key:'));
-    expect(keyLine).toContain('bge_test_key_xyz');
-    const sdkLine = lines.find(l => l.includes('api_key='));
-    expect(sdkLine).toContain('bge_test_key_xyz');
-  });
-
-  it('prints a heartbeat reminder when --idle-timeout is set', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ endpoint_api_key: 'bge_test_key_xyz' }))
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce(readyStatus());
-
-    const lines = [];
-    console.log.mockImplementation((...args) => lines.push(args.join(' ')));
-
-    const p = serveCommand(
-      config,
-      ['meta-llama/Llama-3.1-8B-Instruct', '--idle-timeout', '15', '--max-cost', '5'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    expect(lines.some(l => l.includes('badgr heartbeat dep-serve-001'))).toBe(true);
-  });
-
-  it('--health-path also works for model-based serve (overrides /models default)', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep())                 // POST /serve
-      .mockResolvedValueOnce({ status: 'running' })           // pre-health dep status check
-      .mockResolvedValueOnce({ status: 'running', endpoint_ready: true, health_path: '/v1/health' });
-
-    const p = serveCommand(
-      config,
-      ['meta-llama/Llama-3.1-8B-Instruct', '--health-path', '/v1/health', '--max-cost', '5'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    const body = api.callApi.mock.calls[0][1].body;
-    expect(body.health_path).toBe('/v1/health');
-    expect(process.exitCode).toBeFalsy();
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 8. Routing: tier-1 unavailable → auto-expand to tier-2
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('tier-2 fallback for serve', () => {
-  it('expands to tier-2 when tier-1 returns NO_CAPACITY_MATCH', async () => {
-    const capacityErr = Object.assign(new Error('no capacity'), {
-      errorData: { code: 'NO_CAPACITY_MATCH' },
-      httpStatus: 503,
-    });
-    api.callApi
-      .mockRejectedValueOnce(capacityErr)                                               // tier-1
-      .mockResolvedValueOnce(makeServeDep({ provider: 'vast', tier: '2' }))             // tier-2
-      .mockResolvedValueOnce({ status: 'running' })                                    // pre-health
-      .mockResolvedValueOnce(readyStatus());
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '5'], chalk);
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    const postCalls = api.callApi.mock.calls.filter(c => c[1]?.method === 'POST');
-    expect(postCalls.length).toBeGreaterThanOrEqual(2);
-    expect(store.addReceipt).toHaveBeenCalledWith(expect.objectContaining({ tier: '2' }));
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('--no-fallback prevents tier-2 expansion for serve', async () => {
-    const capacityErr = Object.assign(new Error('no capacity'), {
-      errorData: { code: 'NO_CAPACITY_MATCH' },
-      httpStatus: 503,
-    });
-    api.callApi.mockRejectedValueOnce(capacityErr);
-
-    await serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--no-fallback', '--max-cost', '5'], chalk);
-
-    const postCalls = api.callApi.mock.calls.filter(c => c[1]?.method === 'POST');
-    expect(postCalls.length).toBe(1);
-    expect(process.exitCode).toBe(1);
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 9. Missing endpoint URL guard
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('missing endpoint URL guard', () => {
-  it('exits with error when backend returns no endpoint_url', async () => {
-    api.callApi.mockResolvedValueOnce(makeServeDep({ endpoint_url: null, openai_base_url: null }));
-
-    await serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '5'], chalk);
-
-    expect(process.exitCode).toBe(1);
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({
-      status: 'no_endpoint_url',
-    }));
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 10. Input validation
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('input validation for serve', () => {
-  it('rejects missing model and --image', async () => {
-    await serveCommand(config, [], chalk);
-    expect(api.callApi).not.toHaveBeenCalled();
-  });
-
-  it('rejects --count 0', async () => {
-    await serveCommand(config, ['my/model', '--count', '0'], chalk);
-    expect(api.callApi).not.toHaveBeenCalled();
-    expect(process.exitCode).toBe(1);
-  });
-
-  it('rejects invalid --region', async () => {
-    await serveCommand(config, ['my/model', '--region', 'MOON'], chalk);
-    expect(api.callApi).not.toHaveBeenCalled();
-    expect(process.exitCode).toBe(1);
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// --list-aliases — discoverability for blessed vLLM aliases
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('--list-aliases', () => {
-  it('prints tested model routes without requiring a model or API call', async () => {
-    await serveCommand({ apiKey: null, baseUrl: 'https://api.test/v1' }, ['--list-aliases'], chalk);
-    expect(api.callApi).not.toHaveBeenCalled();
-    expect(process.exitCode).toBeFalsy();
-    const logged = console.log.mock.calls.flat().join('\n');
-    expect(logged).toContain('Tested model routes:');
-    expect(logged).toContain('qwen-7b');
-    expect(logged).toContain('You can also try a Hugging Face model ID:');
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// Model support-level messaging — blessed route vs best-effort HF model vs
-// custom container, plus gated-model HF_TOKEN guidance shown only on failure.
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('model support-level messaging', () => {
-  it('labels a full Hugging Face model ID as best-effort, not tested', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ model: 'Qwen/Qwen2.5-7B-Instruct' }))
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce(readyStatus());
-
-    const p = serveCommand(config, ['Qwen/Qwen2.5-7B-Instruct', '--max-cost', '10'], chalk);
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    const logged = console.log.mock.calls.flat().join('\n');
-    expect(logged).toContain('Route: best-effort Hugging Face model');
-    expect(logged).not.toContain('Route: tested');
-  });
-
-  it('labels a blessed alias as a tested route without extra caveats', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ model: 'Qwen/Qwen2.5-7B-Instruct' }))
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce(readyStatus());
-
-    const p = serveCommand(config, ['qwen-7b', '--max-cost', '10'], chalk);
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    const logged = console.log.mock.calls.flat().join('\n');
-    expect(logged).toContain('Route: tested');
-    expect(logged).not.toContain('best-effort');
-  });
-
-  it('does not show HF_TOKEN guidance up front for a gated model that launches fine', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ model: 'meta-llama/Llama-3.1-8B-Instruct' }))
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce(readyStatus());
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '10'], chalk);
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    const logged = console.log.mock.calls.flat().join('\n');
-    expect(logged).not.toContain('may require Hugging Face access');
-  });
-
-  it('shows HF_TOKEN guidance only once a gated model actually fails to start', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ model: 'meta-llama/Llama-3.1-8B-Instruct' }))
-      .mockResolvedValueOnce({ status: 'failed', error: 'gated repo — 401' });
-
-    const p = serveCommand(config, ['meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '10'], chalk);
-    await p;
-
-    const logged = console.error.mock.calls.flat().join('\n');
-    expect(logged).toContain('may require Hugging Face access');
-    expect(logged).toContain('--env HF_TOKEN=$HF_TOKEN');
-    expect(process.exitCode).toBe(1);
-  });
-
-  it('omits HF_TOKEN guidance on failure when HF_TOKEN is already provided', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ model: 'meta-llama/Llama-3.1-8B-Instruct' }))
-      .mockResolvedValueOnce({ status: 'failed', error: 'crashed' });
-
-    const p = serveCommand(config, [
-      'meta-llama/Llama-3.1-8B-Instruct', '--max-cost', '10', '--env', 'HF_TOKEN=hf_abc123',
-    ], chalk);
-    await p;
-
-    const logged = console.error.mock.calls.flat().join('\n');
-    expect(logged).not.toContain('may require Hugging Face access');
-  });
-
-  it('labels a custom container as "custom server", not a Hugging Face model', async () => {
-    api.callApi.mockResolvedValueOnce(makeServeDep({ model: undefined, image: 'my/custom-server:latest' }));
-
-    const p = serveCommand(config, [
-      '--image', 'my/custom-server:latest', '--max-cost', '10', '--no-wait',
-    ], chalk);
-    await p;
-
-    const logged = console.log.mock.calls.flat().join('\n');
-    expect(logged).toContain('custom server');
-    expect(logged).toContain('Your container owns the app behavior.');
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 11. llama.cpp runtime (--runtime llama.cpp --gguf)
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('--runtime llama.cpp', () => {
-  const HF_REPO = 'HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive';
-  const HF_FILE = 'Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf';
-
-  it('errors when --hf-repo or --hf-file are missing', async () => {
-    await serveCommand(config, ['--runtime', 'llama.cpp', '--max-cost', '10'], chalk);
-    expect(process.exitCode).toBe(1);
-    expect(api.callApi).not.toHaveBeenCalled();
-  });
-
-  it('errors when only --hf-repo is given (missing --hf-file)', async () => {
-    await serveCommand(config, ['--runtime', 'llama.cpp', '--hf-repo', HF_REPO, '--max-cost', '10'], chalk);
-    expect(process.exitCode).toBe(1);
-    expect(api.callApi).not.toHaveBeenCalled();
-  });
-
-  it('uses llama.cpp image with LLAMA_ARG_HF_REPO and LLAMA_ARG_HF_FILE env vars', async () => {
-    // --no-wait: only POST /serve called (1 mock needed)
-    api.callApi.mockResolvedValueOnce(makeServeDep());
-
-    const p = serveCommand(
-      config,
-      ['--runtime', 'llama.cpp', '--hf-repo', HF_REPO, '--hf-file', HF_FILE, '--max-cost', '10', '--no-wait'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(100);
-    await p;
-
-    const body = api.callApi.mock.calls[0][1].body;
-    expect(body.image).toBe('michaelmanleyx/llama-cpp:server-cuda');
-    expect(body.env.LLAMA_ARG_HF_REPO).toBe(HF_REPO);
-    expect(body.env.LLAMA_ARG_HF_FILE).toBe(HF_FILE);
-    expect(body.model).toBeUndefined();
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('records deployment and receipt as ready on success', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep())
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce({ status: 'running', endpoint_ready: true, health_path: '/health' });
-
-    const p = serveCommand(
-      config,
-      ['--runtime', 'llama.cpp', '--hf-repo', HF_REPO, '--hf-file', HF_FILE, '--max-cost', '10'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    expect(store.addDeployment).toHaveBeenCalledWith(expect.objectContaining({
-      id: 'dep-serve-001',
-    }));
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({
-      status: 'ready',
-    }));
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('merges user --env with HF env vars', async () => {
-    // --no-wait: only POST /serve called (1 mock needed)
-    api.callApi.mockResolvedValueOnce(makeServeDep());
-
-    const p = serveCommand(
-      config,
-      ['--runtime', 'llama.cpp', '--hf-repo', HF_REPO, '--hf-file', HF_FILE,
-       '--env', 'CTX_SIZE=8192', '--max-cost', '10', '--no-wait'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(100);
-    await p;
-
-    const body = api.callApi.mock.calls[0][1].body;
-    expect(body.env.LLAMA_ARG_HF_REPO).toBe(HF_REPO);
-    expect(body.env.LLAMA_ARG_HF_FILE).toBe(HF_FILE);
-    expect(body.env.CTX_SIZE).toBe('8192');
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 12. E2E smoke test — tiny HF GGUF (ggml-org/tiny-llamas / stories260K.gguf)
-//
-// Uses a real, publicly-available tiny GGUF (~1 MB) to verify the full
-// request lifecycle: arg parsing → correct serve body → readiness polling via
-// Badgr's own deployment status → receipt recorded as ready → OpenAI-compatible
-// base URL returned.
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('llama.cpp E2E smoke — tiny GGUF (ggml-org/tiny-llamas)', () => {
-  const TINY_REPO = 'ggml-org/tiny-llamas';
-  const TINY_FILE = 'stories260K.gguf';
-
-  it('full lifecycle: arg parse → serve body → readiness poll → ready receipt', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ model: undefined }))  // POST /serve
-      .mockResolvedValueOnce({ status: 'running' })                 // pre-health dep check
-      .mockResolvedValueOnce({ status: 'running', endpoint_ready: true, health_path: '/health' });
-
-    const p = serveCommand(
-      config,
-      ['--runtime', 'llama.cpp',
-       '--hf-repo', TINY_REPO,
-       '--hf-file', TINY_FILE,
-       '--max-cost', '1'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    // Correct image and env vars in the serve request
-    const body = api.callApi.mock.calls[0][1].body;
-    expect(body.image).toBe('michaelmanleyx/llama-cpp:server-cuda');
-    expect(body.env.LLAMA_ARG_HF_REPO).toBe(TINY_REPO);
-    expect(body.env.LLAMA_ARG_HF_FILE).toBe(TINY_FILE);
-    expect(body.model).toBeUndefined();
-
-    // Receipt finalised as ready with endpoint URL
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({
-      status: 'ready',
-      endpointUrl: ENDPOINT_URL,
-    }));
-
-    expect(process.exitCode).toBeFalsy();
-  });
-});
-
-// ─────────────────────────────────────────────────────────────────────────────
-// 13. --task routing: transcribe and image use /health; embed uses /models
-// ─────────────────────────────────────────────────────────────────────────────
-
-describe('--task routing for managed runtimes', () => {
-  it('--task transcribe becomes ready via backend readiness (health_path=/health)', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ model: 'large-v3' }))  // POST /serve
-      .mockResolvedValueOnce({ status: 'running' })                  // pre-health dep check
-      .mockResolvedValueOnce({ status: 'running', endpoint_ready: true, health_path: '/health' });
-
-    const p = serveCommand(
-      config,
-      ['large-v3', '--task', 'transcribe', '--max-cost', '5'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({ status: 'ready' }));
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('--task transcribe sends task field to backend', async () => {
-    api.callApi.mockResolvedValueOnce(makeServeDep({ model: 'large-v3' }));
-
-    const p = serveCommand(
-      config,
-      ['large-v3', '--task', 'transcribe', '--max-cost', '5', '--no-wait'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(100);
-    await p;
-
-    const body = api.callApi.mock.calls[0][1].body;
-    expect(body.task).toBe('transcribe');
-    expect(body.model).toBe('large-v3');
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('--task image becomes ready via backend readiness (health_path=/health)', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ model: 'black-forest-labs/FLUX.1-schnell' }))
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce({ status: 'running', endpoint_ready: true, health_path: '/health' });
-
-    const p = serveCommand(
-      config,
-      ['black-forest-labs/FLUX.1-schnell', '--task', 'image', '--max-cost', '10'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({ status: 'ready' }));
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('--task image sends task field to backend', async () => {
-    api.callApi.mockResolvedValueOnce(makeServeDep({ model: 'black-forest-labs/FLUX.1-schnell' }));
-
-    const p = serveCommand(
-      config,
-      ['black-forest-labs/FLUX.1-schnell', '--task', 'image', '--max-cost', '10', '--no-wait'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(100);
-    await p;
-
-    const body = api.callApi.mock.calls[0][1].body;
-    expect(body.task).toBe('image');
-    expect(body.model).toBe('black-forest-labs/FLUX.1-schnell');
-    expect(process.exitCode).toBeFalsy();
-  });
-
-  it('--task embed becomes ready via backend readiness (health_path=/v1/models)', async () => {
-    api.callApi
-      .mockResolvedValueOnce(makeServeDep({ model: 'BAAI/bge-large-en-v1.5' }))
-      .mockResolvedValueOnce({ status: 'running' })
-      .mockResolvedValueOnce(readyStatus());
-
-    const p = serveCommand(
-      config,
-      ['BAAI/bge-large-en-v1.5', '--task', 'embed', '--max-cost', '5'],
-      chalk,
-    );
-    await vi.advanceTimersByTimeAsync(5000);
-    await p;
-
-    expect(store.updateReceipt).toHaveBeenCalledWith('rcpt-serve-001', expect.objectContaining({ status: 'ready' }));
-    expect(process.exitCode).toBeFalsy();
-  });
-});