Fix desktop daemon stale PID startup (#913)

packages/desktop/src/daemon/daemon-manager.test.ts

@@ -1,3 +1,4 @@
+import { EventEmitter } from "node:events";
 import { beforeEach, describe, expect, it, vi } from "vitest";
 
 import { DEFAULT_DESKTOP_SETTINGS } from "../settings/desktop-settings";
@@ -70,6 +71,34 @@ function desktopSettingsWithManagement(enabled: boolean) {
   };
 }
 
+type MockChildProcess = EventEmitter & {
+  stdout: EventEmitter;
+  stderr: EventEmitter;
+  pid: number;
+  spawnfile: string;
+  spawnargs: string[];
+  unref: ReturnType<typeof vi.fn>;
+};
+
+function createMockChildProcess(): MockChildProcess {
+  const child = new EventEmitter() as MockChildProcess;
+  child.stdout = new EventEmitter();
+  child.stderr = new EventEmitter();
+  child.pid = 1234;
+  child.spawnfile = "node";
+  child.spawnargs = ["node", "daemon.js"];
+  child.unref = vi.fn();
+  return child;
+}
+
+function scheduleFailedStartupOutput(child: MockChildProcess): void {
+  setImmediate(() => {
+    child.stdout.emit("data", Buffer.from(`${"x".repeat(80_000)}stdout-tail`));
+    child.stderr.emit("data", Buffer.from(`${"y".repeat(80_000)}stderr-tail`));
+    child.emit("exit", 1, null);
+  });
+}
+
 describe("daemon-manager commands", () => {
   beforeEach(() => {
     mocks.settings = DEFAULT_DESKTOP_SETTINGS;
@@ -165,4 +194,61 @@ describe("daemon-manager commands", () => {
       "--json",
     ]);
   });
+
+  it("uses a reachable daemon when the PID file is stale", async () => {
+    mocks.runExternalCliJsonCommand.mockResolvedValue({
+      localDaemon: "stale_pid",
+      connectedDaemon: "reachable",
+      serverId: "server-1",
+      pid: 7675,
+      listen: "127.0.0.1:6767",
+      hostname: "dev-host",
+      daemonVersion: "1.2.2",
+      desktopManaged: true,
+    });
+    const handlers = createDaemonCommandHandlers();
+
+    await expect(handlers.start_desktop_daemon()).resolves.toEqual({
+      serverId: "server-1",
+      status: "running",
+      listen: "127.0.0.1:6767",
+      hostname: "dev-host",
+      pid: null,
+      home: "/tmp/paseo-home",
+      version: "1.2.2",
+      desktopManaged: false,
+      error: null,
+    });
+
+    expect(mocks.spawnProcess).not.toHaveBeenCalled();
+  });
+
+  it("bounds captured daemon startup output", async () => {
+    mocks.runExternalCliJsonCommand.mockResolvedValue({
+      localDaemon: "stopped",
+      connectedDaemon: "unreachable",
+      serverId: "",
+    });
+    mocks.spawnProcess.mockImplementation(() => {
+      const child = createMockChildProcess();
+      scheduleFailedStartupOutput(child);
+      return child;
+    });
+    const handlers = createDaemonCommandHandlers();
+
+    let thrown: Error | null = null;
+    try {
+      await handlers.start_desktop_daemon();
+    } catch (error) {
+      thrown = error instanceof Error ? error : new Error(String(error));
+    }
+
+    expect(thrown).toBeInstanceOf(Error);
+    const message = thrown?.message ?? "";
+    expect(message).toContain("Daemon failed to start: exit code 1");
+    expect(message).toContain("output truncated to the last 65536 chars");
+    expect(message).toContain("stdout-tail");
+    expect(message).toContain("stderr-tail");
+    expect(message.length).toBeLessThan(150_000);
+  });
 });

packages/desktop/src/daemon/daemon-manager.ts

@@ -43,6 +43,7 @@ const DAEMON_LOG_FILENAME = "daemon.log";
 const STARTUP_POLL_INTERVAL_MS = 200;
 const STARTUP_POLL_MAX_ATTEMPTS = 150;
 const DETACHED_STARTUP_GRACE_MS = 1200;
+const STARTUP_OUTPUT_CAPTURE_LIMIT_CHARS = 64 * 1024;
 
 type DesktopDaemonState = "starting" | "running" | "stopped" | "errored";
 
@@ -69,6 +70,11 @@ interface DesktopPairingOffer {
   qr: string | null;
 }
 
+interface StartupOutputCapture {
+  text: string;
+  truncated: boolean;
+}
+
 function parseReleaseChannel(
   args: Record<string, unknown> | undefined,
 ): AppReleaseChannel | undefined {
@@ -145,6 +151,30 @@ function tailFile(filePath: string, lines = 50): string {
   }
 }
 
+function createStartupOutputCapture(): StartupOutputCapture {
+  return { text: "", truncated: false };
+}
+
+function appendStartupOutput(capture: StartupOutputCapture, chunk: Buffer): StartupOutputCapture {
+  const nextText = capture.text + chunk.toString();
+  if (nextText.length <= STARTUP_OUTPUT_CAPTURE_LIMIT_CHARS) {
+    return { text: nextText, truncated: capture.truncated };
+  }
+
+  return {
+    text: nextText.slice(-STARTUP_OUTPUT_CAPTURE_LIMIT_CHARS),
+    truncated: true,
+  };
+}
+
+function formatStartupOutput(capture: StartupOutputCapture): string {
+  if (!capture.truncated) {
+    return capture.text;
+  }
+
+  return `[output truncated to the last ${STARTUP_OUTPUT_CAPTURE_LIMIT_CHARS} chars]\n${capture.text}`;
+}
+
 function logDesktopDaemonLifecycle(message: string, details?: Record<string, unknown>): void {
   log.info("[desktop daemon]", message, {
     pid: process.pid,
@@ -197,17 +227,28 @@ export async function resolveDesktopDaemonStatus(): Promise<DesktopDaemonStatus>
       unknown
     >;
     const localDaemon = typeof payload.localDaemon === "string" ? payload.localDaemon : "stopped";
-    const running = localDaemon === "running";
+    const connectedDaemon =
+      typeof payload.connectedDaemon === "string" ? payload.connectedDaemon : "not_probed";
+    const hasRunningLocalProcess = localDaemon === "running";
+    const hasLocalProcess = hasRunningLocalProcess || localDaemon === "unresponsive";
+    const apiReachable = connectedDaemon === "reachable";
+    let status: DesktopDaemonState = "stopped";
+    if (apiReachable || hasRunningLocalProcess) {
+      status = "running";
+    } else if (localDaemon === "unresponsive") {
+      status = "errored";
+    }
 
     return {
       serverId: typeof payload.serverId === "string" ? payload.serverId : "",
-      status: running ? "running" : "stopped",
+      status,
       listen: typeof payload.listen === "string" ? payload.listen : null,
-      hostname: running && typeof payload.hostname === "string" ? payload.hostname : null,
-      pid: running && typeof payload.pid === "number" ? payload.pid : null,
+      hostname:
+        status === "running" && typeof payload.hostname === "string" ? payload.hostname : null,
+      pid: hasLocalProcess && typeof payload.pid === "number" ? payload.pid : null,
       home,
       version: typeof payload.daemonVersion === "string" ? payload.daemonVersion : null,
-      desktopManaged: payload.desktopManaged === true,
+      desktopManaged: hasRunningLocalProcess && payload.desktopManaged === true,
       error: null,
     };
   } catch (error) {
@@ -248,15 +289,17 @@ function assertBuiltInDaemonManagementEnabled(settings: DesktopSettings): void {
 
 function buildStartupFailureError(
   result: { code: number | null; signal: string | null; error?: Error },
-  stdout: string,
-  stderr: string,
+  stdout: StartupOutputCapture,
+  stderr: StartupOutputCapture,
 ): Error {
   const reason = result.error
     ? result.error.message
     : `exit code ${result.code ?? "unknown"}${result.signal ? ` (${result.signal})` : ""}`;
   const parts = [`Daemon failed to start: ${reason}`];
-  if (stderr.trim()) parts.push(`stderr:\n${stderr.trim()}`);
-  if (stdout.trim()) parts.push(`stdout:\n${stdout.trim()}`);
+  const formattedStderr = formatStartupOutput(stderr).trim();
+  const formattedStdout = formatStartupOutput(stdout).trim();
+  if (formattedStderr) parts.push(`stderr:\n${formattedStderr}`);
+  if (formattedStdout) parts.push(`stdout:\n${formattedStdout}`);
   const logs = tailFile(logFilePath(), 15);
   if (logs) parts.push(`Recent logs (${logFilePath()}):\n${logs}`);
   return new Error(parts.join("\n\n"));
@@ -337,13 +380,13 @@ async function startDaemon(): Promise<DesktopDaemonStatus> {
     stdio: ["ignore", "pipe", "pipe"],
   });
 
-  let stdout = "";
-  let stderr = "";
+  let stdout = createStartupOutputCapture();
+  let stderr = createStartupOutputCapture();
   child.stdout!.on("data", (data: Buffer) => {
-    stdout += data.toString();
+    stdout = appendStartupOutput(stdout, data);
   });
   child.stderr!.on("data", (data: Buffer) => {
-    stderr += data.toString();
+    stderr = appendStartupOutput(stderr, data);
   });
 
   logDesktopDaemonLifecycle("detached spawn returned", {
@@ -381,8 +424,8 @@ async function startDaemon(): Promise<DesktopDaemonStatus> {
   logDesktopDaemonLifecycle("detached startup grace period completed", {
     childPid: child.pid ?? null,
     exitedEarly: result.exitedEarly,
-    stdout: stdout.slice(0, 2000),
-    stderr: stderr.slice(0, 2000),
+    stdout: formatStartupOutput(stdout).slice(0, 2000),
+    stderr: formatStartupOutput(stderr).slice(0, 2000),
     ...(result.exitedEarly
       ? {
           exitCode: result.code,