CLI: fix gateway restart health ownership for child listener pids (#24696)

Merged via /review-pr -> /prepare-pr -> /merge-pr.

Prepared head SHA: d6d4b43f7e0a59856f40d259053cbf653fac3bc2
Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com>
Co-authored-by: gumadeiras <5599352+gumadeiras@users.noreply.github.com>
Reviewed-by: @gumadeiras

src/cli/daemon-cli/restart-health.test.ts

@@ -0,0 +1,66 @@
+import { beforeEach, describe, expect, it, vi } from "vitest";
+import type { GatewayService } from "../../daemon/service.js";
+import type { PortListenerKind, PortUsage } from "../../infra/ports.js";
+
+const inspectPortUsage = vi.hoisted(() => vi.fn<(port: number) => Promise<PortUsage>>());
+const classifyPortListener = vi.hoisted(() =>
+  vi.fn<(_listener: unknown, _port: number) => PortListenerKind>(() => "gateway"),
+);
+
+vi.mock("../../infra/ports.js", () => ({
+  classifyPortListener: (listener: unknown, port: number) => classifyPortListener(listener, port),
+  formatPortDiagnostics: vi.fn(() => []),
+  inspectPortUsage: (port: number) => inspectPortUsage(port),
+}));
+
+describe("inspectGatewayRestart", () => {
+  beforeEach(() => {
+    inspectPortUsage.mockReset();
+    inspectPortUsage.mockResolvedValue({
+      port: 0,
+      status: "free",
+      listeners: [],
+      hints: [],
+    });
+    classifyPortListener.mockReset();
+    classifyPortListener.mockReturnValue("gateway");
+  });
+
+  it("treats a gateway listener child pid as healthy ownership", async () => {
+    const service = {
+      readRuntime: vi.fn(async () => ({ status: "running", pid: 7000 })),
+    } as unknown as GatewayService;
+
+    inspectPortUsage.mockResolvedValue({
+      port: 18789,
+      status: "busy",
+      listeners: [{ pid: 7001, ppid: 7000, commandLine: "openclaw-gateway" }],
+      hints: [],
+    });
+
+    const { inspectGatewayRestart } = await import("./restart-health.js");
+    const snapshot = await inspectGatewayRestart({ service, port: 18789 });
+
+    expect(snapshot.healthy).toBe(true);
+    expect(snapshot.staleGatewayPids).toEqual([]);
+  });
+
+  it("marks non-owned gateway listener pids as stale while runtime is running", async () => {
+    const service = {
+      readRuntime: vi.fn(async () => ({ status: "running", pid: 8000 })),
+    } as unknown as GatewayService;
+
+    inspectPortUsage.mockResolvedValue({
+      port: 18789,
+      status: "busy",
+      listeners: [{ pid: 9000, ppid: 8999, commandLine: "openclaw-gateway" }],
+      hints: [],
+    });
+
+    const { inspectGatewayRestart } = await import("./restart-health.js");
+    const snapshot = await inspectGatewayRestart({ service, port: 18789 });
+
+    expect(snapshot.healthy).toBe(false);
+    expect(snapshot.staleGatewayPids).toEqual([9000]);
+  });
+});

src/cli/daemon-cli/restart-health.ts

@@ -21,6 +21,13 @@ export type GatewayRestartSnapshot = {
   staleGatewayPids: number[];
 };
 
+function listenerOwnedByRuntimePid(params: {
+  listener: PortUsage["listeners"][number];
+  runtimePid: number;
+}): boolean {
+  return params.listener.pid === params.runtimePid || params.listener.ppid === params.runtimePid;
+}
+
 export async function inspectGatewayRestart(params: {
   service: GatewayService;
   port: number;
@@ -54,18 +61,27 @@ export async function inspectGatewayRestart(params: {
         )
       : [];
   const running = runtime.status === "running";
+  const runtimePid = runtime.pid;
   const ownsPort =
-    runtime.pid != null
-      ? portUsage.listeners.some((listener) => listener.pid === runtime.pid)
+    runtimePid != null
+      ? portUsage.listeners.some((listener) => listenerOwnedByRuntimePid({ listener, runtimePid }))
       : gatewayListeners.length > 0 ||
         (portUsage.status === "busy" && portUsage.listeners.length === 0);
   const healthy = running && ownsPort;
   const staleGatewayPids = Array.from(
     new Set(
       gatewayListeners
-        .map((listener) => listener.pid)
-        .filter((pid): pid is number => Number.isFinite(pid))
-        .filter((pid) => runtime.pid == null || pid !== runtime.pid || !running),
+        .filter((listener) => Number.isFinite(listener.pid))
+        .filter((listener) => {
+          if (!running) {
+            return true;
+          }
+          if (runtimePid == null) {
+            return true;
+          }
+          return !listenerOwnedByRuntimePid({ listener, runtimePid });
+        })
+        .map((listener) => listener.pid as number),
     ),
   );