From f9be98e8015c925143ea17d45d6232f04aa2cb15 Mon Sep 17 00:00:00 2001
From: JustMe <mr.toing@outlook.com>
Date: Sun, 18 Oct 2026 10:22:34 +0800
Subject: [PATCH] fix(hosting): exit on our own when the orderly shutdown hangs
 [CI]

---
 backend/fleet/src/Fleet.Hosting/Hosting.cs | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/backend/fleet/src/Fleet.Hosting/Hosting.cs b/backend/fleet/src/Fleet.Hosting/Hosting.cs
index 1a5fe44..5a3a03e 100644
--- a/backend/fleet/src/Fleet.Hosting/Hosting.cs
+++ b/backend/fleet/src/Fleet.Hosting/Hosting.cs
@@ -156,6 +156,9 @@ public static class ServiceShell
             log.Info("stopping", ("result", "SIGHUP"));
             cts.Cancel();
         });
+        // 有序退出卡住（比如某个已经死掉的会话收不到对端的关闭回复）时不能一直挂着：supervisor 的 stop 超时之前自己退出，由它重新拉起
+        using var exited = new CancellationTokenSource();
+        using CancellationTokenRegistration deadline = cts.Token.Register(() => _ = ExitIfStuckAsync(time ?? TimeProvider.System, log, exited.Token));
         Task trimmer = TrimHeapAsync(cts.Token);
         Task scheduled = restart is null ? Task.CompletedTask : WaitForRestartAsync(restart, time ?? TimeProvider.System, log, cts);
         try
@@ -171,9 +174,28 @@ public static class ServiceShell
             await cts.CancelAsync();
             await trimmer;
             await scheduled;
+            await exited.CancelAsync();
         }
     }
 
+    /// <summary>退出请求发出之后给有序退出的时间（要短于容器的 StopTimeout，否则 supervisor 会先强杀）。</summary>
+    public static readonly TimeSpan ShutdownGrace = TimeSpan.FromSeconds(10);
+
+    private static async Task ExitIfStuckAsync(TimeProvider time, INodeLog log, CancellationToken exited)
+    {
+        try
+        {
+            await Task.Delay(ShutdownGrace, time, exited);
+        }
+        catch (OperationCanceledException)
+        {
+            return; // 正常退出了
+        }
+
+        log.Warn("orderly shutdown did not finish in time, exiting", ("grace_s", (long)ShutdownGrace.TotalSeconds));
+        Environment.Exit(0);
+    }
+
     private static async Task WaitForRestartAsync(RestartController restart, TimeProvider time, INodeLog log, CancellationTokenSource cts)
     {
         while (!cts.IsCancellationRequested)