test(daemon): stabilize restart diagnostics

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

@@ -3,14 +3,16 @@ import { invokeDesktopCommand } from "@/desktop/electron/invoke";
 import type { AgentSkillSelection } from "@getpaseo/protocol/messages";
 
 export type DesktopDaemonState = "starting" | "running" | "stopped" | "errored";
-export type DesktopDaemonStopReason =
-  | "manual_ipc"
-  | "settings"
-  | "host_remove"
-  | "quit"
-  | "app_update"
-  | "version_mismatch"
-  | "restart";
+export const DESKTOP_DAEMON_STOP_REASONS = [
+  "manual_ipc",
+  "settings",
+  "host_remove",
+  "quit",
+  "app_update",
+  "version_mismatch",
+  "restart",
+] as const;
+export type DesktopDaemonStopReason = (typeof DESKTOP_DAEMON_STOP_REASONS)[number];
 
 export interface DesktopDaemonStatus {
   serverId: string;

packages/cli/tests/25-daemon-restart-supervisor.test.ts

@@ -45,7 +45,7 @@ function readWorkerPid(supervisorPid: number): number | null {
     return null;
   }
 
-  const result = spawnSync("ps", ["ax", "-o", "pid=,ppid="], { encoding: "utf8" });
+  const result = spawnSync("ps", ["ax", "-o", "pid=,ppid=,stat="], { encoding: "utf8" });
   if (result.status !== 0 || result.error) {
     return null;
   }
@@ -55,12 +55,15 @@ function readWorkerPid(supervisorPid: number): number | null {
     if (!trimmed) {
       continue;
     }
-    const [pidToken, ppidToken] = trimmed.split(/\s+/);
+    const [pidToken, ppidToken, statToken] = trimmed.split(/\s+/);
     const pid = Number.parseInt(pidToken ?? "", 10);
     const ppid = Number.parseInt(ppidToken ?? "", 10);
     if (!Number.isInteger(pid) || !Number.isInteger(ppid)) {
       continue;
     }
+    if (statToken?.includes("Z")) {
+      continue;
+    }
     if (ppid === supervisorPid && pid > 0) {
       return pid;
     }
@@ -197,16 +200,27 @@ try {
     await client?.close().catch(() => undefined);
   }
 
-  await waitFor(
-    () => {
-      const workerPid = readWorkerPid(supervisorPid);
-      return (
-        workerPid !== null && workerPid !== workerPidBeforeRestart && isProcessRunning(workerPid)
-      );
-    },
-    20000,
-    "worker pid did not change after restart request",
-  );
+  try {
+    await waitFor(
+      () => {
+        const workerPid = readWorkerPid(supervisorPid);
+        return (
+          workerPid !== null && workerPid !== workerPidBeforeRestart && isProcessRunning(workerPid)
+        );
+      },
+      30000,
+      "worker pid did not change after restart request",
+    );
+  } catch (error) {
+    const capturedSupervisorLogs = await readCapturedSupervisorLogs(
+      paseoHome,
+      recentSupervisorLogs,
+    );
+    throw new Error(
+      `worker pid did not change after restart request, logs:\n${capturedSupervisorLogs}`,
+      { cause: error },
+    );
+  }
 
   const workerPidAfterRestart = readWorkerPid(supervisorPid);
   assert(workerPidAfterRestart !== null, "worker process should exist after restart");
@@ -216,6 +230,28 @@ try {
     "worker pid should change after restart",
   );
 
+  try {
+    await waitFor(
+      async () => {
+        const status = await readDaemonStatus(paseoHome);
+        return status.localDaemon === "running" && status.pid === supervisorPid;
+      },
+      30000,
+      "daemon did not return to running after restart",
+    );
+  } catch (error) {
+    const capturedSupervisorLogs = await readCapturedSupervisorLogs(
+      paseoHome,
+      recentSupervisorLogs,
+    );
+    throw new Error(
+      `daemon did not return to running after restart, logs:\n${capturedSupervisorLogs}`,
+      {
+        cause: error,
+      },
+    );
+  }
+
   const statusAfterRestart = await readDaemonStatus(paseoHome);
   assert.strictEqual(
     statusAfterRestart.localDaemon,

packages/server/src/server/lifecycle-reasons.ts

@@ -1,5 +1,6 @@
 export const CLIENT_SHUTDOWN_RPC_REASON = "client_shutdown_rpc";
 export const DEFAULT_CLIENT_RESTART_RPC_REASON = "client_restart_rpc";
+export const DAEMON_UPDATE_RPC_REASON = "daemon_update";
 
 export function normalizeClientRestartRpcReason(reason: string | undefined): string {
   return reason?.trim() || DEFAULT_CLIENT_RESTART_RPC_REASON;

packages/server/src/server/session/daemon/daemon-self-update-session-controller.ts

@@ -7,6 +7,7 @@ import {
   type DaemonSelfUpdater,
 } from "./daemon-self-updater.js";
 import { getErrorMessage } from "@getpaseo/protocol/error-utils";
+import { DAEMON_UPDATE_RPC_REASON } from "../../lifecycle-reasons.js";
 
 type DaemonUpdateRequest = Extract<SessionInboundMessage, { type: "daemon.update.request" }>;
 
@@ -87,7 +88,7 @@ export class DaemonSelfUpdateSessionController {
         type: "restart",
         clientId: this.clientId,
         requestId: msg.requestId,
-        reason: "daemon_update",
+        reason: DAEMON_UPDATE_RPC_REASON,
       });
     } catch (error) {
       if (error instanceof DaemonSelfUpdateInProgressError) {

packages/server/src/server/websocket-server.ts

@@ -87,6 +87,7 @@ import { ProviderUsageService } from "../services/quota-fetcher/service.js";
 import { getProcessMemoryDiagnostics, getProcessUptimeSeconds } from "./process-diagnostics.js";
 import {
   CLIENT_SHUTDOWN_RPC_REASON,
+  DAEMON_UPDATE_RPC_REASON,
   normalizeClientRestartRpcReason,
 } from "./lifecycle-reasons.js";
 import { CLIENT_CAPS } from "@getpaseo/protocol/client-capabilities";
@@ -2919,7 +2920,7 @@ function getControlRpcLogInfo(
     return {
       requestType: message.type,
       requestId: message.requestId,
-      reason: "daemon_update",
+      reason: DAEMON_UPDATE_RPC_REASON,
     };
   }
   return null;