fix(processes): fix multiple clawx processes running concurently (#589)

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Haze <hazeone@users.noreply.github.com>
Co-authored-by: paisley <8197966+su8su@users.noreply.github.com>
Co-authored-by: Felix <24791380+vcfgv@users.noreply.github.com>

electron/gateway/manager.ts

@@ -242,6 +242,7 @@ export class GatewayManager extends EventEmitter {
           await this.connect(port, externalToken);
         },
         onConnectedToExistingGateway: () => {
+
           // If the existing gateway is actually our own spawned UtilityProcess
           // (e.g. after a self-restart code=1012), keep ownership so that
           // stop() can still terminate the process during a restart() cycle.
@@ -250,6 +251,7 @@ export class GatewayManager extends EventEmitter {
             this.ownsProcess = false;
             this.setStatus({ pid: undefined });
           }
+
           this.startHealthCheck();
         },
         waitForPortFree: async (port) => {
@@ -356,6 +358,25 @@ export class GatewayManager extends EventEmitter {
     this.setStatus({ state: 'stopped', error: undefined, pid: undefined, connectedAt: undefined, uptime: undefined });
   }
 
+  /**
+   * Best-effort emergency cleanup for app-quit timeout paths.
+   * Only terminates a process this manager still owns.
+   */
+  async forceTerminateOwnedProcessForQuit(): Promise<boolean> {
+    if (!this.process || !this.ownsProcess) {
+      return false;
+    }
+
+    const child = this.process;
+    await terminateOwnedGatewayProcess(child);
+    if (this.process === child) {
+      this.process = null;
+    }
+    this.ownsProcess = false;
+    this.setStatus({ pid: undefined });
+    return true;
+  }
+
   /**
    * Restart Gateway process
    */
@@ -724,6 +745,7 @@ export class GatewayManager extends EventEmitter {
 
     this.process = child;
     this.ownsProcess = true;
+    logger.debug(`Gateway manager now owns process pid=${child.pid ?? 'unknown'}`);
     this.lastSpawnSummary = lastSpawnSummary;
   }
 

electron/gateway/supervisor.ts

@@ -22,39 +22,58 @@ export function warmupManagedPythonReadiness(): void {
 }
 
 export async function terminateOwnedGatewayProcess(child: Electron.UtilityProcess): Promise<void> {
-  let exited = false;
+  const terminateWindowsProcessTree = async (pid: number): Promise<void> => {
+    const cp = await import('child_process');
+    await new Promise<void>((resolve) => {
+      cp.exec(`taskkill /F /PID ${pid} /T`, { timeout: 5000, windowsHide: true }, () => resolve());
+    });
+  };
 
   await new Promise<void>((resolve) => {
+    let exited = false;
+
+    // Register a single exit listener before any kill attempt to avoid
+    // the race where exit fires between two separate `once('exit')` calls.
     child.once('exit', () => {
       exited = true;
+      clearTimeout(timeout);
       resolve();
     });
 
     const pid = child.pid;
     logger.info(`Sending kill to Gateway process (pid=${pid ?? 'unknown'})`);
-    try {
-      child.kill();
-    } catch {
-      // ignore if already exited
+
+    if (process.platform === 'win32' && pid) {
+      void terminateWindowsProcessTree(pid).catch((err) => {
+        logger.warn(`Windows process-tree kill failed for Gateway pid=${pid}:`, err);
+      });
+    } else {
+      try {
+        child.kill();
+      } catch {
+        // ignore if already exited
+      }
     }
 
     const timeout = setTimeout(() => {
       if (!exited) {
         logger.warn(`Gateway did not exit in time, force-killing (pid=${pid ?? 'unknown'})`);
         if (pid) {
-          try {
-            process.kill(pid, 'SIGKILL');
-          } catch {
-            // ignore
+          if (process.platform === 'win32') {
+            void terminateWindowsProcessTree(pid).catch((err) => {
+              logger.warn(`Forced Windows process-tree kill failed for Gateway pid=${pid}:`, err);
+            });
+          } else {
+            try {
+              process.kill(pid, 'SIGKILL');
+            } catch {
+              // ignore
+            }
           }
         }
       }
       resolve();
     }, 5000);
-
-    child.once('exit', () => {
-      clearTimeout(timeout);
-    });
   });
 }
 
@@ -226,6 +245,9 @@ export async function findExistingGatewayProcess(options: {
       const pids = await getListeningProcessIds(port);
       if (pids.length > 0 && (!ownedPid || !pids.includes(String(ownedPid)))) {
         await terminateOrphanedProcessIds(port, pids);
+        if (process.platform === 'win32') {
+          await waitForPortFree(port, 10000);
+        }
         return null;
       }
     } catch (err) {