31 lines
1.0 KiB
JavaScript
31 lines
1.0 KiB
JavaScript
import test from 'node:test';
|
|
import assert from 'node:assert';
|
|
import { safeImport } from '../_helpers/path.js';
|
|
import { FakeLLMClient } from '../_helpers/fakeLLMClient.js';
|
|
|
|
function skip(msg){ console.warn('[SKIP]', msg); }
|
|
|
|
test('LLMManager: enforces concurrency limit', async () => {
|
|
const res = await safeImport('LLMManager');
|
|
if (!res.ok) { skip(res.reason); return; }
|
|
const L = res.mod;
|
|
|
|
if (typeof L.__setClient !== 'function' || typeof L.__setConcurrency !== 'function') {
|
|
skip('Missing __setClient/__setConcurrency hooks');
|
|
return;
|
|
}
|
|
|
|
const fake = new FakeLLMClient({ delayMs: 50 });
|
|
L.__setClient('mock', fake);
|
|
L.__setConcurrency(2); // on demande 2 max
|
|
|
|
const N = 5;
|
|
const calls = Array.from({length:N}, (_,i)=> L.callModel({ provider:'mock', model:'fake-1', input:`#${i}` }));
|
|
const t0 = Date.now();
|
|
await Promise.all(calls);
|
|
const t1 = Date.now();
|
|
|
|
// Avec delay 50ms et concu=2, 5 requêtes => ~3 vagues => durée >= 150ms (avec marge)
|
|
assert.ok((t1 - t0) >= 120, `expected batching effect, got ${t1-t0}ms`);
|
|
});
|