fix(gateway/cli): relax local backend self-pairing and harden launchd restarts (#46290)

Signed-off-by: sallyom <somalley@redhat.com>

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

@@ -182,7 +182,7 @@ export async function inspectGatewayRestart(params: {
             return true;
           }
           if (runtimePid == null) {
-            return true;
+            return false;
           }
           return !listenerOwnedByRuntimePid({ listener, runtimePid });
         })

src/cli/daemon-cli/status.gather.test.ts

@@ -1,5 +1,6 @@
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import { captureEnv } from "../../test-utils/env.js";
+import type { GatewayRestartSnapshot } from "./restart-health.js";
 
 const callGatewayStatusProbe = vi.fn(async (_opts?: unknown) => ({ ok: true as const }));
 const loadGatewayTlsRuntime = vi.fn(async (_cfg?: unknown) => ({
@@ -18,6 +19,14 @@ const readLastGatewayErrorLine = vi.fn(async (_env?: NodeJS.ProcessEnv) => null)
 const auditGatewayServiceConfig = vi.fn(async (_opts?: unknown) => undefined);
 const serviceIsLoaded = vi.fn(async (_opts?: unknown) => true);
 const serviceReadRuntime = vi.fn(async (_env?: NodeJS.ProcessEnv) => ({ status: "running" }));
+const inspectGatewayRestart = vi.fn<(opts?: unknown) => Promise<GatewayRestartSnapshot>>(
+  async (_opts?: unknown) => ({
+    runtime: { status: "running", pid: 1234 },
+    portUsage: { port: 19001, status: "busy", listeners: [], hints: [] },
+    healthy: true,
+    staleGatewayPids: [],
+  }),
+);
 const serviceReadCommand = vi.fn<
   (env?: NodeJS.ProcessEnv) => Promise<{
     programArguments: string[];
@@ -117,6 +126,10 @@ vi.mock("./probe.js", () => ({
   probeGatewayStatus: (opts: unknown) => callGatewayStatusProbe(opts),
 }));
 
+vi.mock("./restart-health.js", () => ({
+  inspectGatewayRestart: (opts: unknown) => inspectGatewayRestart(opts),
+}));
+
 const { gatherDaemonStatus } = await import("./status.gather.js");
 
 describe("gatherDaemonStatus", () => {
@@ -139,6 +152,7 @@ describe("gatherDaemonStatus", () => {
     delete process.env.DAEMON_GATEWAY_PASSWORD;
     callGatewayStatusProbe.mockClear();
     loadGatewayTlsRuntime.mockClear();
+    inspectGatewayRestart.mockClear();
     daemonLoadedConfig = {
       gateway: {
         bind: "lan",
@@ -362,4 +376,34 @@ describe("gatherDaemonStatus", () => {
     expect(callGatewayStatusProbe).not.toHaveBeenCalled();
     expect(status.rpc).toBeUndefined();
   });
+
+  it("surfaces stale gateway listener pids from restart health inspection", async () => {
+    inspectGatewayRestart.mockResolvedValueOnce({
+      runtime: { status: "running", pid: 8000 },
+      portUsage: {
+        port: 19001,
+        status: "busy",
+        listeners: [{ pid: 9000, ppid: 8999, commandLine: "openclaw-gateway" }],
+        hints: [],
+      },
+      healthy: false,
+      staleGatewayPids: [9000],
+    });
+
+    const status = await gatherDaemonStatus({
+      rpc: {},
+      probe: true,
+      deep: false,
+    });
+
+    expect(inspectGatewayRestart).toHaveBeenCalledWith(
+      expect.objectContaining({
+        port: 19001,
+      }),
+    );
+    expect(status.health).toEqual({
+      healthy: false,
+      staleGatewayPids: [9000],
+    });
+  });
 });

src/cli/daemon-cli/status.gather.ts

@@ -29,6 +29,7 @@ import {
 import { pickPrimaryTailnetIPv4 } from "../../infra/tailnet.js";
 import { loadGatewayTlsRuntime } from "../../infra/tls/gateway.js";
 import { probeGatewayStatus } from "./probe.js";
+import { inspectGatewayRestart } from "./restart-health.js";
 import { normalizeListenerAddress, parsePortFromArgs, pickProbeHostForBind } from "./shared.js";
 import type { GatewayRpcOpts } from "./types.js";
 
@@ -112,6 +113,10 @@ export type DaemonStatus = {
     error?: string;
     url?: string;
   };
+  health?: {
+    healthy: boolean;
+    staleGatewayPids: number[];
+  };
   extraServices: Array<{ label: string; detail: string; scope: string }>;
 };
 
@@ -331,6 +336,14 @@ export async function gatherDaemonStatus(
         configPath: daemonConfigSummary.path,
       })
     : undefined;
+  const health =
+    opts.probe && loaded
+      ? await inspectGatewayRestart({
+          service,
+          port: daemonPort,
+          env: serviceEnv,
+        }).catch(() => undefined)
+      : undefined;
 
   let lastError: string | undefined;
   if (loaded && runtime?.status === "running" && portStatus && portStatus.status !== "busy") {
@@ -357,6 +370,14 @@ export async function gatherDaemonStatus(
     ...(portCliStatus ? { portCli: portCliStatus } : {}),
     lastError,
     ...(rpc ? { rpc: { ...rpc, url: gateway.probeUrl } } : {}),
+    ...(health
+      ? {
+          health: {
+            healthy: health.healthy,
+            staleGatewayPids: health.staleGatewayPids,
+          },
+        }
+      : {}),
     extraServices,
   };
 }

src/cli/daemon-cli/status.print.test.ts

@@ -0,0 +1,116 @@
+import { beforeEach, describe, expect, it, vi } from "vitest";
+
+const runtime = vi.hoisted(() => ({
+  log: vi.fn<(line: string) => void>(),
+  error: vi.fn<(line: string) => void>(),
+}));
+
+vi.mock("../../runtime.js", () => ({
+  defaultRuntime: runtime,
+}));
+
+vi.mock("../../terminal/theme.js", () => ({
+  colorize: (_rich: boolean, _theme: unknown, text: string) => text,
+}));
+
+vi.mock("../../commands/onboard-helpers.js", () => ({
+  resolveControlUiLinks: () => ({ httpUrl: "http://127.0.0.1:18789" }),
+}));
+
+vi.mock("../../daemon/inspect.js", () => ({
+  renderGatewayServiceCleanupHints: () => [],
+}));
+
+vi.mock("../../daemon/launchd.js", () => ({
+  resolveGatewayLogPaths: () => ({
+    stdoutPath: "/tmp/gateway.out.log",
+    stderrPath: "/tmp/gateway.err.log",
+  }),
+}));
+
+vi.mock("../../daemon/systemd-hints.js", () => ({
+  isSystemdUnavailableDetail: () => false,
+  renderSystemdUnavailableHints: () => [],
+}));
+
+vi.mock("../../infra/wsl.js", () => ({
+  isWSLEnv: () => false,
+}));
+
+vi.mock("../../logging.js", () => ({
+  getResolvedLoggerSettings: () => ({ file: "/tmp/openclaw.log" }),
+}));
+
+vi.mock("./shared.js", () => ({
+  createCliStatusTextStyles: () => ({
+    rich: false,
+    label: (text: string) => text,
+    accent: (text: string) => text,
+    infoText: (text: string) => text,
+    okText: (text: string) => text,
+    warnText: (text: string) => text,
+    errorText: (text: string) => text,
+  }),
+  filterDaemonEnv: () => ({}),
+  formatRuntimeStatus: () => "running (pid 8000)",
+  resolveRuntimeStatusColor: () => "",
+  renderRuntimeHints: () => [],
+  safeDaemonEnv: () => [],
+}));
+
+vi.mock("./status.gather.js", () => ({
+  renderPortDiagnosticsForCli: () => [],
+  resolvePortListeningAddresses: () => ["127.0.0.1:18789"],
+}));
+
+const { printDaemonStatus } = await import("./status.print.js");
+
+describe("printDaemonStatus", () => {
+  beforeEach(() => {
+    runtime.log.mockReset();
+    runtime.error.mockReset();
+  });
+
+  it("prints stale gateway pid guidance when runtime does not own the listener", () => {
+    printDaemonStatus(
+      {
+        service: {
+          label: "LaunchAgent",
+          loaded: true,
+          loadedText: "loaded",
+          notLoadedText: "not loaded",
+          runtime: { status: "running", pid: 8000 },
+        },
+        gateway: {
+          bindMode: "loopback",
+          bindHost: "127.0.0.1",
+          port: 18789,
+          portSource: "env/config",
+          probeUrl: "ws://127.0.0.1:18789",
+        },
+        port: {
+          port: 18789,
+          status: "busy",
+          listeners: [{ pid: 9000, ppid: 8999, address: "127.0.0.1:18789" }],
+          hints: [],
+        },
+        rpc: {
+          ok: false,
+          error: "gateway closed (1006 abnormal closure (no close frame))",
+          url: "ws://127.0.0.1:18789",
+        },
+        health: {
+          healthy: false,
+          staleGatewayPids: [9000],
+        },
+        extraServices: [],
+      },
+      { json: false },
+    );
+
+    expect(runtime.error).toHaveBeenCalledWith(
+      expect.stringContaining("Gateway runtime PID does not own the listening port"),
+    );
+    expect(runtime.error).toHaveBeenCalledWith(expect.stringContaining("openclaw gateway restart"));
+  });
+});

src/cli/daemon-cli/status.print.ts

@@ -194,6 +194,25 @@ export function printDaemonStatus(status: DaemonStatus, opts: { json: boolean })
     spacer();
   }
 
+  if (
+    status.health &&
+    status.health.staleGatewayPids.length > 0 &&
+    service.runtime?.status === "running" &&
+    typeof service.runtime.pid === "number"
+  ) {
+    defaultRuntime.error(
+      errorText(
+        `Gateway runtime PID does not own the listening port. Other gateway process(es) are listening: ${status.health.staleGatewayPids.join(", ")}`,
+      ),
+    );
+    defaultRuntime.error(
+      errorText(
+        `Fix: run ${formatCliCommand("openclaw gateway restart")} and re-check with ${formatCliCommand("openclaw gateway status --deep")}.`,
+      ),
+    );
+    spacer();
+  }
+
   const systemdUnavailable =
     process.platform === "linux" && isSystemdUnavailableDetail(service.runtime?.detail);
   if (systemdUnavailable) {