From 28508bc64312a5e0acab86e83eef490b30f4ab48 Mon Sep 17 00:00:00 2001
From: paisley <8197966+su8su@users.noreply.github.com>
Date: Thu, 2 Apr 2026 16:45:23 +0800
Subject: [PATCH] fix(gateway): refine process termination and wait logic on
 retry (#750)

---
 electron/gateway/manager.ts              |  4 +++-
 electron/gateway/startup-orchestrator.ts | 13 ++++++++++---
 electron/gateway/supervisor.ts           |  2 +-
 3 files changed, 14 insertions(+), 5 deletions(-)

diff --git a/electron/gateway/manager.ts b/electron/gateway/manager.ts
index 415274a6a..e61172087 100644
--- a/electron/gateway/manager.ts
+++ b/electron/gateway/manager.ts
@@ -297,11 +297,13 @@ export class GatewayManager extends EventEmitter {
                 process.kill(pid, 0);
                 // Still alive — keep this.process so later cleanup can reach it
               } catch {
-                // Process is gone — safe to clear the handle
+                // Process is gone — safe to clear the handle and ownership flag
                 this.process = null;
+                this.ownsProcess = false;
               }
             } else {
               this.process = null;
+              this.ownsProcess = false;
             }
           }
         },
diff --git a/electron/gateway/startup-orchestrator.ts b/electron/gateway/startup-orchestrator.ts
index 102618554..4036a6efe 100644
--- a/electron/gateway/startup-orchestrator.ts
+++ b/electron/gateway/startup-orchestrator.ts
@@ -98,14 +98,15 @@ 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.
+        // Terminate the previously spawned process before the backoff delay so
+        // it releases the port as early as possible; the subsequent delay gives
+        // the OS time to recycle the port (especially TCP TIME_WAIT on Windows).
         if (hooks.terminateOwnedProcess) {
           await hooks.terminateOwnedProcess().catch((err) => {
             logger.warn('Failed to terminate owned process before retry:', err);
           });
         }
+        await hooks.delay(1000);
         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()
@@ -122,6 +123,12 @@ export async function runGatewayStartupSequence(hooks: StartupHooks): Promise<vo
           }, 500);
           try {
             await hooks.waitForPortFree(hooks.port, abortController.signal);
+          } catch (portWaitError) {
+            // If the wait was aborted due to lifecycle supersede, convert to
+            // LifecycleSupersededError so the outer handler propagates correctly.
+            hooks.assertLifecycle('start/retry-port-wait-aborted');
+            // If assertLifecycle didn't throw, it's a genuine port-wait error — rethrow.
+            throw portWaitError;
           } finally {
             clearInterval(lifecyclePollInterval);
           }
diff --git a/electron/gateway/supervisor.ts b/electron/gateway/supervisor.ts
index 272380348..9186986fd 100644
--- a/electron/gateway/supervisor.ts
+++ b/electron/gateway/supervisor.ts
@@ -137,7 +137,7 @@ export async function waitForPortFree(port: number, timeoutMs = 30000, signal?:
   while (Date.now() - start < timeoutMs) {
     if (signal?.aborted) {
       logger.debug(`waitForPortFree: aborted while waiting for port ${port}`);
-      return;
+      throw new Error(`waitForPortFree: aborted while waiting for port ${port}`);
     }
 
     const available = await new Promise<boolean>((resolve) => {