First half of v2 — mcplocal can now declare a hibernating backend and
respond to a `wake` task by running a configured recipe. v2 Stage 2
will wire mcpd to dispatch the wake task before relaying inference.
Config (LlmProviderFileEntry):
- New \`wake\` block on a published provider:
wake:
type: http # or: command
url: ... # http only
method: POST # http only, default POST
headers: {...} # http only
body: ... # http only
command: ... # command only
args: [...] # command only
maxWaitSeconds: 60 # how long to poll isAvailable() after wake fires
Registrar (mcplocal):
- At publish time, providers with a wake recipe whose isAvailable()
returns false report initialStatus=hibernating to mcpd. Without a
wake recipe (legacy v1) or when already up, status stays active.
- handleWakeTask: runs the recipe (HTTP request OR child-process
spawn), then polls isAvailable() up to maxWaitSeconds, sending a
heartbeat each loop so mcpd's GC sweep doesn't time us out
mid-boot. Reports { ok, ms } on success or { error } on
timeout/recipe failure via the existing _provider-task/:id/result.
- Replaces the v1 stub that rejected wake tasks with "not implemented".
mcpd VirtualLlmService:
- RegisterProviderInput gains optional initialStatus ('active' |
'hibernating'). The register/upsert path uses it for both new and
reconnecting rows. Defaults to 'active' so v1 publishers still
work unchanged.
- Provider-register route's coercer accepts the new field.
Tests: 3 new in registrar.test.ts cover initialStatus selection
(hibernating when wake configured + unavailable, active otherwise,
active when no wake even if unavailable). 8/8 registrar tests, 833/833
mcpd unchanged.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
350 lines
12 KiB
TypeScript
350 lines
12 KiB
TypeScript
import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
|
|
import http from 'node:http';
|
|
import { mkdtempSync, rmSync, readFileSync, writeFileSync } from 'node:fs';
|
|
import { tmpdir } from 'node:os';
|
|
import { join } from 'node:path';
|
|
import {
|
|
VirtualLlmRegistrar,
|
|
type RegistrarPublishedProvider,
|
|
} from '../src/providers/registrar.js';
|
|
import type { LlmProvider, CompletionOptions, CompletionResult } from '../src/providers/types.js';
|
|
|
|
/**
|
|
* The registrar talks HTTP. Spin a tiny in-process server in each test so
|
|
* we can assert what it sends without mocking node:http itself.
|
|
*/
|
|
interface FakeServer {
|
|
url: string;
|
|
close: () => Promise<void>;
|
|
/** Calls observed in arrival order. */
|
|
calls: Array<{ method: string; path: string; body: string; headers: Record<string, string | string[] | undefined> }>;
|
|
/**
|
|
* Optional handler. If set, runs per-request and decides response. If not,
|
|
* defaults to 201 + JSON `{ providerSessionId: 'sess-FAKE' }` for register
|
|
* and 200 + `{}` for everything else.
|
|
*/
|
|
handler?: (req: http.IncomingMessage, res: http.ServerResponse, body: string) => void;
|
|
}
|
|
|
|
async function startFakeServer(): Promise<FakeServer> {
|
|
const calls: FakeServer['calls'] = [];
|
|
let server!: http.Server;
|
|
const ready = new Promise<FakeServer>((resolve, reject) => {
|
|
server = http.createServer((req, res) => {
|
|
const chunks: Buffer[] = [];
|
|
req.on('data', (c: Buffer) => chunks.push(c));
|
|
req.on('end', () => {
|
|
const body = Buffer.concat(chunks).toString('utf-8');
|
|
calls.push({
|
|
method: req.method ?? '',
|
|
path: req.url ?? '',
|
|
body,
|
|
headers: req.headers,
|
|
});
|
|
if (fake.handler !== undefined) {
|
|
fake.handler(req, res, body);
|
|
return;
|
|
}
|
|
if (req.url === '/api/v1/llms/_provider-register') {
|
|
res.writeHead(201, { 'content-type': 'application/json' });
|
|
res.end(JSON.stringify({ providerSessionId: 'sess-FAKE', llms: [] }));
|
|
return;
|
|
}
|
|
res.writeHead(200, { 'content-type': 'application/json' });
|
|
res.end('{}');
|
|
});
|
|
});
|
|
server.listen(0, '127.0.0.1', () => {
|
|
const addr = server.address();
|
|
if (addr === null || typeof addr === 'string') {
|
|
reject(new Error('listen failed'));
|
|
return;
|
|
}
|
|
const fakeReady: FakeServer = {
|
|
url: `http://127.0.0.1:${String(addr.port)}`,
|
|
close: () => new Promise<void>((r) => { server.close(() => r()); }),
|
|
calls,
|
|
};
|
|
Object.assign(fake, fakeReady);
|
|
resolve(fake);
|
|
});
|
|
});
|
|
const fake: FakeServer = {} as FakeServer;
|
|
return ready;
|
|
}
|
|
|
|
function makeProvider(name: string, content = 'hi from local'): LlmProvider {
|
|
return {
|
|
name,
|
|
async complete(_opts: CompletionOptions): Promise<CompletionResult> {
|
|
return {
|
|
content,
|
|
toolCalls: [],
|
|
usage: { promptTokens: 1, completionTokens: 1, totalTokens: 2 },
|
|
finishReason: 'stop',
|
|
};
|
|
},
|
|
async listModels() { return []; },
|
|
async isAvailable() { return true; },
|
|
};
|
|
}
|
|
|
|
let tempDir: string;
|
|
|
|
beforeEach(() => {
|
|
tempDir = mkdtempSync(join(tmpdir(), 'mcpctl-registrar-test-'));
|
|
});
|
|
|
|
afterEach(() => {
|
|
rmSync(tempDir, { recursive: true, force: true });
|
|
});
|
|
|
|
function silentLog(): { info: ReturnType<typeof vi.fn>; warn: ReturnType<typeof vi.fn>; error: ReturnType<typeof vi.fn> } {
|
|
return { info: vi.fn(), warn: vi.fn(), error: vi.fn() };
|
|
}
|
|
|
|
describe('VirtualLlmRegistrar', () => {
|
|
it('start() with no published providers is a silent no-op', async () => {
|
|
const log = silentLog();
|
|
const registrar = new VirtualLlmRegistrar({
|
|
mcpdUrl: 'http://unreachable.example',
|
|
token: 'tok',
|
|
publishedProviders: [],
|
|
sessionFilePath: join(tempDir, 'provider-session'),
|
|
log,
|
|
});
|
|
await registrar.start();
|
|
expect(log.info).toHaveBeenCalledWith(expect.stringContaining('nothing to publish'));
|
|
registrar.stop();
|
|
});
|
|
|
|
it('register POSTs to /_provider-register and persists the returned sessionId', async () => {
|
|
const fake = await startFakeServer();
|
|
try {
|
|
const sessionFilePath = join(tempDir, 'provider-session');
|
|
const published: RegistrarPublishedProvider[] = [
|
|
{ provider: makeProvider('vllm-local'), type: 'openai', model: 'qwen', tier: 'fast' },
|
|
];
|
|
const registrar = new VirtualLlmRegistrar({
|
|
mcpdUrl: fake.url,
|
|
token: 'tok-abc',
|
|
publishedProviders: published,
|
|
sessionFilePath,
|
|
log: silentLog(),
|
|
// Make heartbeat huge so it doesn't fire mid-test.
|
|
heartbeatIntervalMs: 60_000,
|
|
});
|
|
await registrar.start();
|
|
// Allow the SSE open to enter flight (we never feed it a response,
|
|
// but the request fires synchronously after register).
|
|
await new Promise((r) => setTimeout(r, 20));
|
|
|
|
const registerCall = fake.calls.find((c) => c.path === '/api/v1/llms/_provider-register');
|
|
expect(registerCall).toBeDefined();
|
|
expect(registerCall!.method).toBe('POST');
|
|
const body = JSON.parse(registerCall!.body) as { providers: Array<Record<string, unknown>> };
|
|
expect(body.providers).toHaveLength(1);
|
|
expect(body.providers[0]).toMatchObject({
|
|
name: 'vllm-local',
|
|
type: 'openai',
|
|
model: 'qwen',
|
|
tier: 'fast',
|
|
// v2 always sends initialStatus; defaults to 'active' when no
|
|
// wake recipe is configured.
|
|
initialStatus: 'active',
|
|
});
|
|
expect(registerCall!.headers['authorization']).toBe('Bearer tok-abc');
|
|
|
|
// Sticky session id persisted.
|
|
expect(readFileSync(sessionFilePath, 'utf-8').trim()).toBe('sess-FAKE');
|
|
expect(registrar.getSessionId()).toBe('sess-FAKE');
|
|
|
|
registrar.stop();
|
|
} finally {
|
|
await fake.close();
|
|
}
|
|
});
|
|
|
|
it('reuses an existing sticky session id from disk on next start', async () => {
|
|
const fake = await startFakeServer();
|
|
try {
|
|
const sessionFilePath = join(tempDir, 'provider-session');
|
|
writeFileSync(sessionFilePath, 'sess-existing\n', 'utf-8');
|
|
|
|
const registrar = new VirtualLlmRegistrar({
|
|
mcpdUrl: fake.url,
|
|
token: 't',
|
|
publishedProviders: [
|
|
{ provider: makeProvider('vllm-local'), type: 'openai', model: 'm' },
|
|
],
|
|
sessionFilePath,
|
|
log: silentLog(),
|
|
heartbeatIntervalMs: 60_000,
|
|
});
|
|
await registrar.start();
|
|
await new Promise((r) => setTimeout(r, 20));
|
|
|
|
const registerCall = fake.calls.find((c) => c.path === '/api/v1/llms/_provider-register');
|
|
const body = JSON.parse(registerCall!.body) as { providerSessionId?: string };
|
|
expect(body.providerSessionId).toBe('sess-existing');
|
|
|
|
registrar.stop();
|
|
} finally {
|
|
await fake.close();
|
|
}
|
|
});
|
|
|
|
it('heartbeat ticker POSTs the session id at the configured interval', async () => {
|
|
const fake = await startFakeServer();
|
|
try {
|
|
const registrar = new VirtualLlmRegistrar({
|
|
mcpdUrl: fake.url,
|
|
token: 't',
|
|
publishedProviders: [
|
|
{ provider: makeProvider('vllm-local'), type: 'openai', model: 'm' },
|
|
],
|
|
sessionFilePath: join(tempDir, 'provider-session'),
|
|
log: silentLog(),
|
|
heartbeatIntervalMs: 30, // tight so the test doesn't drag
|
|
});
|
|
await registrar.start();
|
|
// Wait long enough for at least 2 heartbeats to fire.
|
|
await new Promise((r) => setTimeout(r, 100));
|
|
registrar.stop();
|
|
|
|
const heartbeats = fake.calls.filter((c) => c.path === '/api/v1/llms/_provider-heartbeat');
|
|
expect(heartbeats.length).toBeGreaterThanOrEqual(2);
|
|
for (const h of heartbeats) {
|
|
const body = JSON.parse(h.body) as { providerSessionId: string };
|
|
expect(body.providerSessionId).toBe('sess-FAKE');
|
|
}
|
|
} finally {
|
|
await fake.close();
|
|
}
|
|
});
|
|
|
|
// ── v2: hibernating + wake recipe ──
|
|
|
|
it('publishes initialStatus=hibernating when provider is unavailable AND wake is configured', async () => {
|
|
const fake = await startFakeServer();
|
|
try {
|
|
const sleeping: LlmProvider = {
|
|
name: 'vllm-local',
|
|
async complete() { throw new Error('not running'); },
|
|
async listModels() { return []; },
|
|
async isAvailable() { return false; },
|
|
};
|
|
const registrar = new VirtualLlmRegistrar({
|
|
mcpdUrl: fake.url,
|
|
token: 't',
|
|
publishedProviders: [{
|
|
provider: sleeping,
|
|
type: 'openai',
|
|
model: 'm',
|
|
wake: { type: 'http', url: 'http://localhost:9999/wake', maxWaitSeconds: 1 },
|
|
}],
|
|
sessionFilePath: join(tempDir, 'provider-session'),
|
|
log: silentLog(),
|
|
heartbeatIntervalMs: 60_000,
|
|
});
|
|
await registrar.start();
|
|
await new Promise((r) => setTimeout(r, 20));
|
|
|
|
const registerCall = fake.calls.find((c) => c.path === '/api/v1/llms/_provider-register');
|
|
const body = JSON.parse(registerCall!.body) as { providers: Array<{ initialStatus?: string }> };
|
|
expect(body.providers[0]!.initialStatus).toBe('hibernating');
|
|
registrar.stop();
|
|
} finally {
|
|
await fake.close();
|
|
}
|
|
});
|
|
|
|
it('publishes initialStatus=active when provider is available even with a wake recipe', async () => {
|
|
const fake = await startFakeServer();
|
|
try {
|
|
const awake: LlmProvider = {
|
|
name: 'vllm-local',
|
|
async complete() { throw new Error('not used'); },
|
|
async listModels() { return []; },
|
|
async isAvailable() { return true; },
|
|
};
|
|
const registrar = new VirtualLlmRegistrar({
|
|
mcpdUrl: fake.url,
|
|
token: 't',
|
|
publishedProviders: [{
|
|
provider: awake,
|
|
type: 'openai',
|
|
model: 'm',
|
|
wake: { type: 'http', url: 'http://localhost:9999/wake' },
|
|
}],
|
|
sessionFilePath: join(tempDir, 'provider-session'),
|
|
log: silentLog(),
|
|
heartbeatIntervalMs: 60_000,
|
|
});
|
|
await registrar.start();
|
|
await new Promise((r) => setTimeout(r, 20));
|
|
|
|
const registerCall = fake.calls.find((c) => c.path === '/api/v1/llms/_provider-register');
|
|
const body = JSON.parse(registerCall!.body) as { providers: Array<{ initialStatus?: string }> };
|
|
expect(body.providers[0]!.initialStatus).toBe('active');
|
|
registrar.stop();
|
|
} finally {
|
|
await fake.close();
|
|
}
|
|
});
|
|
|
|
it('publishes initialStatus=active when no wake recipe is configured (legacy path)', async () => {
|
|
const fake = await startFakeServer();
|
|
try {
|
|
// Provider intentionally returns false but has no wake recipe →
|
|
// legacy v1 publishers don't get hibernation behavior.
|
|
const sleeping: LlmProvider = {
|
|
name: 'vllm-local',
|
|
async complete() { return { content: '', toolCalls: [], usage: { promptTokens: 0, completionTokens: 0, totalTokens: 0 }, finishReason: 'stop' }; },
|
|
async listModels() { return []; },
|
|
async isAvailable() { return false; },
|
|
};
|
|
const registrar = new VirtualLlmRegistrar({
|
|
mcpdUrl: fake.url,
|
|
token: 't',
|
|
publishedProviders: [{ provider: sleeping, type: 'openai', model: 'm' }],
|
|
sessionFilePath: join(tempDir, 'provider-session'),
|
|
log: silentLog(),
|
|
heartbeatIntervalMs: 60_000,
|
|
});
|
|
await registrar.start();
|
|
await new Promise((r) => setTimeout(r, 20));
|
|
|
|
const registerCall = fake.calls.find((c) => c.path === '/api/v1/llms/_provider-register');
|
|
const body = JSON.parse(registerCall!.body) as { providers: Array<{ initialStatus?: string }> };
|
|
expect(body.providers[0]!.initialStatus).toBe('active');
|
|
registrar.stop();
|
|
} finally {
|
|
await fake.close();
|
|
}
|
|
});
|
|
|
|
it('throws when mcpd returns non-201 from /_provider-register', async () => {
|
|
const fake = await startFakeServer();
|
|
fake.handler = (_req, res, _body) => {
|
|
res.writeHead(409, { 'content-type': 'application/json' });
|
|
res.end(JSON.stringify({ error: 'Cannot publish over public LLM: vllm-local' }));
|
|
};
|
|
try {
|
|
const registrar = new VirtualLlmRegistrar({
|
|
mcpdUrl: fake.url,
|
|
token: 't',
|
|
publishedProviders: [
|
|
{ provider: makeProvider('vllm-local'), type: 'openai', model: 'm' },
|
|
],
|
|
sessionFilePath: join(tempDir, 'provider-session'),
|
|
log: silentLog(),
|
|
heartbeatIntervalMs: 60_000,
|
|
});
|
|
await expect(registrar.start()).rejects.toThrow(/HTTP 409/);
|
|
} finally {
|
|
await fake.close();
|
|
}
|
|
});
|
|
});
|