Skip to content

Garry's Opinionated OpenClaw/Hermes Agent Brain

TypeScript

test(serve): answered-tool-call boot test runs a 200 ms window over 40 RPCs of 20 ms, so runner stalls can't open a real no-progress gap (#8446)

garrytan7de252f7a018
diff --git a/test/serve-stdio-lifecycle.test.ts b/test/serve-stdio-lifecycle.test.ts
index cc661f0ddf..99b0832963 100644
--- a/test/serve-stdio-lifecycle.test.ts
+++ b/test/serve-stdio-lifecycle.test.ts
@@ -716,16 +716,19 @@ describe('boot-readiness deadline (#3273)', () => {
});
// The 52k-document serve kept answering tool calls while its boot waited on
- // the engine behind them; completed requests are progress too.
+ // the engine behind them; completed requests are progress too. The window is
+ // 10x one request so scheduling stalls on a loaded runner never open a real
+ // no-progress gap as long as the window (#8446), and the boot still runs
+ // about four windows, which only per-request progress keeps alive.
test('a boot phase waiting behind answered tool calls is not killed', async () => {
const { trackStdioRpc } = await import('../src/mcp/server.ts');
const h = makeHarness();
h.opts.startMcpServer = async (_engine, startOpts) => {
startOpts?.onBootPhase?.('mcp_connect');
startOpts?.onBootPhase?.('persistence_consumer');
- for (let i = 0; i < 12; i++) await trackStdioRpc(() => new Promise(r => setTimeout(r, 20)));
+ for (let i = 0; i < 40; i++) await trackStdioRpc(() => new Promise(r => setTimeout(r, 20)));
};
- h.opts.bootTimeoutMs = 60;
+ h.opts.bootTimeoutMs = 200;
await runServe(h.engine as unknown as BrainEngine, [], h.opts);
await new Promise(r => setTimeout(r, 100));
expect(h.engine.disconnectCalls).toBe(0);