本地应用看门狗测试通过

NewLifeX · Nov 2, 2023 · 6885f0e · 6885f0e
1 parent 05e9d24
commit 6885f0e
Show file tree

Hide file tree

Showing 5 changed files with 52 additions and 15 deletions.
diff --git a/StarAgent/StarAgent.csproj b/StarAgent/StarAgent.csproj
@@ -40,7 +40,7 @@
   </ItemGroup>
 
   <ItemGroup>
-    <PackageReference Include="NewLife.Agent" Version="10.6.2023.1016-beta1607" />
+    <PackageReference Include="NewLife.Agent" Version="10.6.2023.1102" />
   </ItemGroup>
 
   <ItemGroup>

diff --git a/StarAgent/StarService.cs b/StarAgent/StarService.cs
@@ -1,5 +1,7 @@
 using System.Collections.Concurrent;
 using System.Diagnostics;
+using System.Security.Cryptography;
+using System.Threading;
 using NewLife;
 using NewLife.Agent;
 using NewLife.Log;
@@ -94,6 +96,10 @@ public AgentInfo Info(AgentInfo info)
         return ai;
     }
 
+    /// <summary>本地心跳</summary>
+    /// <param name="info"></param>
+    /// <returns></returns>
+    [Api(nameof(Ping))]
     public PingResponse Ping(LocalPingInfo info)
     {
         if (info != null && info.ProcessId > 0)
@@ -261,14 +267,16 @@ private void CheckLocal()
     #region 看门狗
     static ConcurrentDictionary<Int32, DateTime> _dogs = new();
     static TimerX _dogTimer;
-    static void FeedDog(Int32 pid, Int32 timeout)
+    void FeedDog(Int32 pid, Int32 timeout)
     {
         if (pid <= 0 || timeout <= 0) return;
 
+        using var span = DefaultTracer.Instance?.NewSpan(nameof(FeedDog), new { pid, timeout });
+
         // 更新过期时间，超过该时间未收到心跳，将会重启本应用进程
         _dogs[pid] = DateTime.Now.AddSeconds(timeout);
 
-        _dogTimer ??= new TimerX(CheckDog, null, 1000, 15000) { Async = true };
+        _dogTimer ??= new TimerX(CheckDog, Manager, 1000, 15000) { Async = true };
     }
 
     static void CheckDog(Object state)
@@ -287,20 +295,33 @@ static void CheckDog(Object state)
             }
         }
 
+        if (ks.Count == 0 && ds.Count == 0) return;
+
+        using var span = DefaultTracer.Instance?.NewSpan(nameof(CheckDog));
+
         // 重启超时的进程
         foreach (var item in ks)
         {
-            var p = Process.GetProcessById(item);
-            if (p == null || p.HasExited)
-                _dogs.Remove(item);
-            else
+            try
             {
-                XTrace.WriteLine("进程[{0}/{1}]超过一定时间没有心跳，可能已经假死，准备重启。", p.ProcessName, p.Id);
+                var p = Process.GetProcessById(item);
+                if (p == null || p.HasExited)
+                    _dogs.Remove(item);
+                else
+                {
+                    XTrace.WriteLine("进程[{0}/{1}]超过一定时间没有心跳，可能已经假死，准备重启。", p.ProcessName, p.Id);
 
-                p.SafetyKill();
+                    span?.AppendTag($"SafetyKill {p.ProcessName}/{p.Id}");
+                    p.SafetyKill();
 
-                //todo 启动应用。暂时不需要，因为StarAgent会自动启动
-                if (state is ServiceManager manager) manager.CheckNow();
+                    //todo 启动应用。暂时不需要，因为StarAgent会自动启动
+                    if (state is ServiceManager manager) manager.CheckNow();
+                }
+            }
+            catch (Exception ex)
+            {
+                _dogs.Remove(item);
+                XTrace.WriteException(ex);
             }
         }
 
@@ -313,6 +334,9 @@ static void CheckDog(Object state)
     #endregion
 
     #region 日志
+    /// <summary>链路追踪</summary>
+    public ITracer Tracer { get; set; }
+
     /// <summary>日志</summary>
     public ILog Log { get; set; }
 

diff --git a/Stardust/Managers/ServiceController.cs b/Stardust/Managers/ServiceController.cs
@@ -556,7 +556,7 @@ public Boolean Check()
                     // 遍历所有进程，从命令行参数中找到启动文件名一致的进程
                     foreach (var item in Process.GetProcesses())
                     {
-                        if (item.Id == mypid) continue;
+                        if (item.Id == mypid || item.HasExited) continue;
                         if (!item.ProcessName.EqualIgnoreCase(ProcessName)) continue;
 
                         var name = AppInfo.GetProcessName(item);
@@ -575,7 +575,7 @@ public Boolean Check()
             {
                 span?.AppendTag($"GetProcessesByName({ProcessName})");
 
-                var ps = Process.GetProcessesByName(ProcessName).Where(e => e.Id != mypid).ToArray();
+                var ps = Process.GetProcessesByName(ProcessName).Where(e => e.Id != mypid && !e.HasExited).ToArray();
                 if (ps.Length > 0) return TakeOver(ps[0], $"按[Name={ProcessName}]查找");
             }
         }
@@ -587,7 +587,7 @@ public Boolean Check()
         p = Process;
         if (p != null && EventProvider is StarClient client)
         {
-            if (_appInfo == null)
+            if (_appInfo == null || _appInfo.Id != p.Id)
                 _appInfo = new AppInfo(p) { AppName = inf.Name };
             else
                 _appInfo.Refresh();

diff --git a/Stardust/Stardust.csproj b/Stardust/Stardust.csproj
@@ -88,7 +88,7 @@
 
   <ItemGroup Condition="'$(TargetFramework)'!='net40'">
     <PackageReference Include="NewLife.Core" Version="10.6.2023.1101" />
-    <PackageReference Include="NewLife.Remoting" Version="2.5.2023.1101" />
+    <PackageReference Include="NewLife.Remoting" Version="2.5.2023.1102" />
   </ItemGroup>
   <ItemGroup Condition="'$(TargetFramework)'=='net40'">
     <PackageReference Include="NewLife.Core" Version="10.6.2023.1101-net40" />

diff --git a/Test/Program.cs b/Test/Program.cs
@@ -20,6 +20,7 @@
 using Stardust;
 using Stardust.Data;
 using Stardust.Data.Nodes;
+using Stardust.Models;
 using Stardust.Monitors;
 
 namespace Test;
@@ -130,6 +131,18 @@ static void Test3()
         //client.ProbeAndInstall(null, "1.6");
         var info = client.GetInfo();
 
+        var appInfo = new AppInfo(Process.GetCurrentProcess());
+
+        for (var i = 0; i < 5; i++)
+        {
+            _ = client.PingAsync(appInfo, 5);
+
+            Thread.Sleep(2000);
+        }
+
+        Console.WriteLine("等待");
+        Console.ReadLine();
+
         //var p = Process.GetCurrentProcess();
         //var name = p.MainModule.FileName;
         //var str = name + Environment.NewLine + name.ToJson();