refactor: split node agent and Go control plane

2026-09-11 16:57:31 +08:00
parent 078c22c73b
commit c86ba9c4d7
129 changed files with 654 additions and 77 deletions
@@ -0,0 +1,106 @@
+using System.Text.Json;
+
+namespace WxAgent.Core;
+
+public sealed record WechatStabilitySample(
+    DateTimeOffset Timestamp,
+    int ProcessCount,
+    long WorkingSetBytes,
+    int HandleCount,
+    int ThreadCount,
+    bool WindowFound,
+    IReadOnlyList<WxAgentErrorCode> Errors,
+    long MessageEvents,
+    long ReconnectEvents,
+    long AgentWorkingSetBytes = 0,
+    int AgentHandleCount = 0,
+    int AgentThreadCount = 0,
+    DateTimeOffset? LastSnapshotAt = null);
+
+public sealed record WechatStabilityReport(
+    DateTimeOffset StartedAt,
+    DateTimeOffset FinishedAt,
+    IReadOnlyList<WechatStabilitySample> Samples,
+    long MessageEvents,
+    long ReconnectEvents,
+    long WorkingSetGrowthBytes,
+    int HandleGrowth,
+    int ThreadGrowth,
+    bool Healthy,
+    IReadOnlyList<string> Findings,
+    double RequestedSeconds = 0,
+    double ObservedSeconds = 0,
+    bool CompletedDuration = false);
+
+public static class WechatStabilityAnalyzer
+{
+    public static WechatStabilityReport Analyze(
+        DateTimeOffset startedAt,
+        DateTimeOffset finishedAt,
+        IReadOnlyList<WechatStabilitySample> samples,
+        long messageEvents,
+        long reconnectEvents)
+    {
+        ArgumentNullException.ThrowIfNull(samples);
+        if (samples.Count == 0)
+            throw new WxAgentException(WxAgentErrorCode.InvalidArgument, "At least one stability sample is required.");
+
+        var first = samples[0];
+        var last = samples[^1];
+        var memoryGrowth = last.WorkingSetBytes - first.WorkingSetBytes;
+        var handleGrowth = last.HandleCount - first.HandleCount;
+        var threadGrowth = last.ThreadCount - first.ThreadCount;
+        var findings = new List<string>();
+        var recoveredWindowLoss = last.WindowFound && reconnectEvents > 0;
+        if (samples.Any(sample => !sample.WindowFound) && !recoveredWindowLoss) findings.Add("wechat-window-unavailable");
+        foreach (var error in last.Errors) findings.Add("final-error:" + error);
+        if (last.AgentWorkingSetBytes - first.AgentWorkingSetBytes > 128L * 1024 * 1024) findings.Add("agent-working-set-growth-over-128mb");
+        if (last.AgentHandleCount - first.AgentHandleCount > 100) findings.Add("agent-handle-growth-over-100");
+        if (last.AgentThreadCount - first.AgentThreadCount > 20) findings.Add("agent-thread-growth-over-20");
+        if (memoryGrowth > 256L * 1024 * 1024) findings.Add("working-set-growth-over-256mb");
+        if (handleGrowth > 500) findings.Add("handle-growth-over-500");
+        if (threadGrowth > 100) findings.Add("thread-growth-over-100");
+        if (samples.Any(sample => sample.Timestamp - (sample.LastSnapshotAt ?? startedAt) > TimeSpan.FromMinutes(5)))
+            findings.Add("listener-snapshot-gap-over-5min");
+
+        return new WechatStabilityReport(
+            startedAt,
+            finishedAt,
+            samples,
+            messageEvents,
+            reconnectEvents,
+            memoryGrowth,
+            handleGrowth,
+            threadGrowth,
+            findings.Count == 0,
+            findings);
+    }
+
+    public static WechatStabilityReport CheckRun(
+        WechatStabilityReport report, TimeSpan requested, TimeSpan observed, bool cancelled,
+        string? listenerFailure = null, string? samplingFailure = null)
+    {
+        var findings = report.Findings.ToList();
+        var completed = observed + TimeSpan.FromSeconds(1) >= requested && !cancelled;
+        if (!completed) findings.Add("duration-incomplete");
+        if (listenerFailure is not null) findings.Add("listener-failed:" + listenerFailure);
+        if (samplingFailure is not null) findings.Add("sampling-failed:" + samplingFailure);
+        if (report.Samples[^1].LastSnapshotAt is not { } snapshot
+            || report.FinishedAt - snapshot > TimeSpan.FromSeconds(10)) findings.Add("listener-snapshot-stale");
+        return report with { Healthy = findings.Count == 0, Findings = findings, RequestedSeconds = requested.TotalSeconds,
+            ObservedSeconds = observed.TotalSeconds, CompletedDuration = completed };
+    }
+
+    public static async Task WriteJsonAsync(WechatStabilityReport report, string path, CancellationToken cancellationToken)
+    {
+        var fullPath = Path.GetFullPath(path);
+        Directory.CreateDirectory(Path.GetDirectoryName(fullPath)!);
+        var temporary = fullPath + ".tmp";
+        try
+        {
+            await File.WriteAllTextAsync(temporary, JsonSerializer.Serialize(report, new JsonSerializerOptions { WriteIndented = true }), cancellationToken);
+            File.Move(temporary, fullPath, overwrite: true);
+        }
+        finally { if (File.Exists(temporary)) File.Delete(temporary); }
+    }
+}