fix(agents): detect signaled local service exits

src/agents/provider-local-service.test.ts

@@ -8,6 +8,7 @@ import {
   attachModelProviderLocalService,
   ensureModelProviderLocalService,
   getModelProviderLocalService,
+  hasLocalServiceProcessExited,
   stopManagedProviderLocalServicesForTest,
 } from "./provider-local-service.js";
 
@@ -65,6 +66,12 @@ describe("provider local service", () => {
     });
   });
 
+  it("treats signaled local service children as exited", () => {
+    expect(hasLocalServiceProcessExited({ exitCode: null, signalCode: "SIGTERM" })).toBe(true);
+    expect(hasLocalServiceProcessExited({ exitCode: 0, signalCode: null })).toBe(true);
+    expect(hasLocalServiceProcessExited({ exitCode: null, signalCode: null })).toBe(false);
+  });
+
   it("starts an on-demand local service and stops it after idle", async () => {
     const port = await freePort();
     const healthUrl = `http://127.0.0.1:${port}/v1/models`;
@@ -323,6 +330,29 @@ describe("provider local service", () => {
     expect(Date.now() - startedAt).toBeLessThan(5_000);
   });
 
+  it("reports a local service startup signal exit without waiting for readiness timeout", async () => {
+    const port = await freePort();
+    const model = attachModelProviderLocalService(
+      {
+        id: "demo",
+        provider: "local-signal-exit",
+        api: "openai-completions",
+        baseUrl: `http://127.0.0.1:${port}/v1`,
+      } as unknown as Model<"openai-completions">,
+      {
+        command: process.execPath,
+        args: ["-e", "process.kill(process.pid, 'SIGTERM')"],
+        readyTimeoutMs: 60_000,
+      },
+    );
+
+    const startedAt = Date.now();
+    await expect(ensureModelProviderLocalService(model)).rejects.toThrow(
+      "local-signal-exit local service exited before readiness with signal SIGTERM",
+    );
+    expect(Date.now() - startedAt).toBeLessThan(5_000);
+  });
+
   it("honors request aborts while waiting for local service readiness", async () => {
     const port = await freePort();
     const healthUrl = `http://127.0.0.1:${port}/v1/models`;

src/agents/provider-local-service.ts

@@ -87,7 +87,8 @@ export async function ensureModelProviderLocalService(
 
   try {
     if (
-      managed.process?.exitCode === null &&
+      managed.process &&
+      !hasLocalServiceProcessExited(managed.process) &&
       (await probeHealth(healthUrl, healthHeaders, signal))
     ) {
       return { release };
@@ -110,7 +111,7 @@ export async function ensureModelProviderLocalService(
       });
     }
     await waitForAbort(managed.starting, signal);
-    if (!managed.process || managed.process.exitCode !== null) {
+    if (!managed.process || hasLocalServiceProcessExited(managed.process)) {
       release();
       return undefined;
     }
@@ -232,7 +233,7 @@ async function startAndWaitForLocalService(params: {
   if (await probeHealth(healthUrl, healthHeaders, signal)) {
     return;
   }
-  if (managed.process?.exitCode === null) {
+  if (managed.process && !hasLocalServiceProcessExited(managed.process)) {
     log.info(`restarting unhealthy ${provider} local service`);
     await stopManagedProcessForRestart(managed, signal);
   }
@@ -321,7 +322,7 @@ function stopManagedService(key: string, managed: ManagedLocalService, reason: s
   managed.process = undefined;
   managed.lastExit = undefined;
   services.delete(key);
-  if (child && child.exitCode === null) {
+  if (child && !hasLocalServiceProcessExited(child)) {
     log.info(`stopping local model service: reason=${reason}`);
     child.kill("SIGTERM");
   }
@@ -334,12 +335,12 @@ async function stopManagedProcessForRestart(
   const child = managed.process;
   managed.process = undefined;
   managed.lastExit = undefined;
-  if (!child || child.exitCode !== null) {
+  if (!child || hasLocalServiceProcessExited(child)) {
     return;
   }
   child.kill("SIGTERM");
   await waitForChildExit(child, signal, DEFAULT_PROBE_TIMEOUT_MS);
-  if (child.exitCode === null) {
+  if (!hasLocalServiceProcessExited(child)) {
     child.kill("SIGKILL");
     await waitForChildExit(child, signal, DEFAULT_PROBE_TIMEOUT_MS);
   }
@@ -464,7 +465,7 @@ function waitForChildExit(
   signal: AbortSignal,
   timeoutMs: number,
 ): Promise<void> {
-  if (child.exitCode !== null) {
+  if (hasLocalServiceProcessExited(child)) {
     return Promise.resolve();
   }
   throwIfAborted(signal);
@@ -490,3 +491,9 @@ function waitForChildExit(
     signal.addEventListener("abort", onAbort, { once: true });
   });
 }
+
+export function hasLocalServiceProcessExited(
+  child: Pick<ChildProcess, "exitCode" | "signalCode">,
+): boolean {
+  return child.exitCode !== null || child.signalCode !== null;
+}