From 560ae95611e8b87d03725346c41d189d5a57d3ef Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=86=99=E4=BA=86=E5=90=97=EF=BC=9F?=
 <30821268+shuang5725@users.noreply.github.com>
Date: Thu, 2 Apr 2026 16:11:03 +0800
Subject: [PATCH] fix(gateway): terminate owned process before retry to prevent
 port conflict on Windows (#724)

---
 electron/gateway/manager.ts              | 27 ++++++++++++++++++--
 electron/gateway/startup-orchestrator.ts | 32 +++++++++++++++++++++++-
 electron/gateway/supervisor.ts           | 14 ++++++++---
 3 files changed, 67 insertions(+), 6 deletions(-)

diff --git a/electron/gateway/manager.ts b/electron/gateway/manager.ts
index 7d23cdd77..415274a6a 100644
--- a/electron/gateway/manager.ts
+++ b/electron/gateway/manager.ts
@@ -259,8 +259,8 @@ export class GatewayManager extends EventEmitter {
 
           this.startHealthCheck();
         },
-        waitForPortFree: async (port) => {
-          await waitForPortFree(port);
+        waitForPortFree: async (port, signal) => {
+          await waitForPortFree(port, 30000, signal);
         },
         startProcess: async () => {
           await this.startProcess();
@@ -282,6 +282,29 @@ export class GatewayManager extends EventEmitter {
         delay: async (ms) => {
           await new Promise((resolve) => setTimeout(resolve, ms));
         },
+        terminateOwnedProcess: async () => {
+          if (this.process && this.ownsProcess) {
+            logger.info('Terminating owned Gateway process before retry...');
+            const proc = this.process;
+            const pid = proc.pid;
+            await terminateOwnedGatewayProcess(proc).catch(() => {});
+            // Only clear the handle if the process has actually exited.
+            // terminateOwnedGatewayProcess may resolve via its timeout path
+            // while the child is still alive; in that case keep the reference
+            // so subsequent retries and stop() can still target it.
+            if (pid != null) {
+              try {
+                process.kill(pid, 0);
+                // Still alive — keep this.process so later cleanup can reach it
+              } catch {
+                // Process is gone — safe to clear the handle
+                this.process = null;
+              }
+            } else {
+              this.process = null;
+            }
+          }
+        },
       });
     } catch (error) {
       if (error instanceof LifecycleSupersededError) {
diff --git a/electron/gateway/startup-orchestrator.ts b/electron/gateway/startup-orchestrator.ts
index ae2c3ee13..102618554 100644
--- a/electron/gateway/startup-orchestrator.ts
+++ b/electron/gateway/startup-orchestrator.ts
@@ -18,13 +18,15 @@ type StartupHooks = {
   findExistingGateway: (port: number) => Promise<ExistingGatewayInfo | null>;
   connect: (port: number, externalToken?: string) => Promise<void>;
   onConnectedToExistingGateway: () => void;
-  waitForPortFree: (port: number) => Promise<void>;
+  waitForPortFree: (port: number, signal?: AbortSignal) => Promise<void>;
   startProcess: () => Promise<void>;
   waitForReady: (port: number) => Promise<void>;
   onConnectedToManagedGateway: () => void;
   runDoctorRepair: () => Promise<boolean>;
   onDoctorRepairSuccess: () => void;
   delay: (ms: number) => Promise<void>;
+  /** Called before a retry to terminate the previously spawned process if still running */
+  terminateOwnedProcess?: () => Promise<void>;
 };
 
 export async function runGatewayStartupSequence(hooks: StartupHooks): Promise<void> {
@@ -97,6 +99,34 @@ export async function runGatewayStartupSequence(hooks: StartupHooks): Promise<vo
       if (recoveryAction === 'retry') {
         logger.warn(`Transient start error: ${String(error)}. Retrying... (${startAttempts}/${maxStartAttempts})`);
         await hooks.delay(1000);
+        // Terminate the previously spawned process before retrying so it doesn't
+        // hold the port and cause another handshake failure.
+        if (hooks.terminateOwnedProcess) {
+          await hooks.terminateOwnedProcess().catch((err) => {
+            logger.warn('Failed to terminate owned process before retry:', err);
+          });
+        }
+        hooks.assertLifecycle('start/retry-pre-port-wait');
+        // Wait for port to become free before retrying (handles lingering processes).
+        // Use a short-polling AbortController so that a superseding stop()/restart()
+        // can cancel the wait promptly instead of blocking for the full 30s timeout.
+        if (hooks.shouldWaitForPortFree) {
+          const abortController = new AbortController();
+          // Poll lifecycle every 500ms and abort the port-wait if superseded
+          const lifecyclePollInterval = setInterval(() => {
+            try {
+              hooks.assertLifecycle('start/retry-port-wait-poll');
+            } catch {
+              abortController.abort();
+            }
+          }, 500);
+          try {
+            await hooks.waitForPortFree(hooks.port, abortController.signal);
+          } finally {
+            clearInterval(lifecyclePollInterval);
+          }
+        }
+        hooks.assertLifecycle('start/retry-post-port-wait');
         continue;
       }
 
diff --git a/electron/gateway/supervisor.ts b/electron/gateway/supervisor.ts
index 36122fbb2..272380348 100644
--- a/electron/gateway/supervisor.ts
+++ b/electron/gateway/supervisor.ts
@@ -34,11 +34,12 @@ export async function terminateOwnedGatewayProcess(child: Electron.UtilityProces
 
     // 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', () => {
+    const exitListener = () => {
       exited = true;
       clearTimeout(timeout);
       resolve();
-    });
+    };
+    child.once('exit', exitListener);
 
     const pid = child.pid;
     logger.info(`Sending kill to Gateway process (pid=${pid ?? 'unknown'})`);
@@ -72,6 +73,8 @@ export async function terminateOwnedGatewayProcess(child: Electron.UtilityProces
           }
         }
       }
+      // Clean up the exit listener on timeout to prevent listener leaks
+      child.off('exit', exitListener);
       resolve();
     }, 5000);
   });
@@ -125,13 +128,18 @@ export async function unloadLaunchctlGatewayService(): Promise<void> {
   }
 }
 
-export async function waitForPortFree(port: number, timeoutMs = 30000): Promise<void> {
+export async function waitForPortFree(port: number, timeoutMs = 30000, signal?: AbortSignal): Promise<void> {
   const net = await import('net');
   const start = Date.now();
   const pollInterval = 500;
   let logged = false;
 
   while (Date.now() - start < timeoutMs) {
+    if (signal?.aborted) {
+      logger.debug(`waitForPortFree: aborted while waiting for port ${port}`);
+      return;
+    }
+
     const available = await new Promise<boolean>((resolve) => {
       const server = net.createServer();
       server.once('error', () => resolve(false));