feat(ops): upload node logs and task traffic [CI]

backend/src/MoShiro.Backend.Application/Agents/MoShiroFleetHooks.cs

@@ -72,9 +72,11 @@ public sealed class MoShiroNodeEventSink : INodeEventSink
     private readonly NodeOwnerService _owners;
     private readonly ProbeQuicRuntime _quicRuntime;
     private readonly ProbeRuntime _probeRuntime;
+    private readonly NodeRuntimeLogService _nodeLogs;
 
-    public MoShiroNodeEventSink(IAgentStore agents, AgentRuntime runtime, IEventStore events, LiveHub live, AgentService service, NodeInfoService nodes, ReleaseService releases, ScheduleService schedule, IProbeStore probes, NodeOwnerService owners, ProbeQuicRuntime quicRuntime, ProbeRuntime probeRuntime)
+    public MoShiroNodeEventSink(IAgentStore agents, AgentRuntime runtime, IEventStore events, LiveHub live, AgentService service, NodeInfoService nodes, ReleaseService releases, ScheduleService schedule, IProbeStore probes, NodeOwnerService owners, ProbeQuicRuntime quicRuntime, ProbeRuntime probeRuntime, NodeRuntimeLogService nodeLogs)
     {
+        _nodeLogs = nodeLogs;
         _owners = owners;
         _quicRuntime = quicRuntime;
         _probeRuntime = probeRuntime;
@@ -125,6 +127,7 @@ public sealed class MoShiroNodeEventSink : INodeEventSink
                     _releases.RecordStatus(id, beat?.Update);
                     if (beat is not null)
                     {
+                        _nodeLogs.Append(id, beat.RuntimeLogs, beat.DroppedRuntimeLogs ?? 0);
                         state.ReplaceProbes(beat.Probes.Select(p => new ConnectedProbeInfo(p.Id.Value, p.Version, p.Remote, p.SinceUtcMs, p.IpVersion)));
                         foreach (ConnectedProbe p in beat.Probes)
                         {
@@ -185,6 +188,7 @@ public sealed class MoShiroNodeEventSink : INodeEventSink
                     _releases.RecordStatus(id, heartbeat?.Update);
                     if (heartbeat is not null)
                     {
+                        _nodeLogs.Append(id, heartbeat.RuntimeLogs, heartbeat.DroppedRuntimeLogs ?? 0);
                         _probeRuntime.RecordHeartbeat(id.Value, nodeEvent.AtMs, heartbeat);
                     }
                     if (heartbeat?.QuicLinks is { } quicLinks)

backend/src/MoShiro.Backend.Application/Agents/NodeRuntimeLogService.cs

@@ -0,0 +1,107 @@
+using System.Collections.Concurrent;
+using Fleet.Core;
+using MoShiro.Protocol;
+
+namespace MoShiro.Backend.Application;
+
+public sealed record NodeRuntimeLog(long Id, long AtUtcMs, string Level, string Message, IReadOnlyDictionary<string, string> Fields);
+
+public sealed record NodeRuntimeLogSnapshot(IReadOnlyList<NodeRuntimeLog> Items, long Dropped);
+
+/// <summary>Bounded in-memory recent logs uploaded by authenticated Agent and Probe heartbeats.</summary>
+public sealed class NodeRuntimeLogService(TimeProvider time)
+{
+    private const int CapacityPerNode = 256;
+    private const int MaxBatch = 32;
+    private readonly ConcurrentDictionary<NodeId, LogBuffer> _nodes = new();
+
+    public void Append(NodeId node, IReadOnlyList<NodeLogRecord>? records, long droppedByNode)
+    {
+        if ((records is null || records.Count == 0) && droppedByNode <= 0)
+        {
+            return;
+        }
+
+        LogBuffer buffer = _nodes.GetOrAdd(node, static _ => new LogBuffer());
+        lock (buffer.Gate)
+        {
+            buffer.DroppedByNode += Math.Max(0, droppedByNode);
+            if (records is null)
+            {
+                return;
+            }
+
+            foreach (NodeLogRecord record in records.Take(MaxBatch))
+            {
+                long now = time.GetUtcNow().ToUnixTimeMilliseconds();
+                long at = record.AtUtcMs < now - 86_400_000 || record.AtUtcMs > now + 86_400_000 ? now : record.AtUtcMs;
+                var entry = new NodeRuntimeLog(
+                    ++buffer.LastId,
+                    at,
+                    record.Level is "info" or "warn" or "error" ? record.Level : "info",
+                    Truncate(record.Message ?? "", 384),
+                    SafeFields(record.Fields));
+
+                if (buffer.Items.Count == CapacityPerNode)
+                {
+                    buffer.Items.Dequeue();
+                    buffer.DroppedByBackend++;
+                }
+
+                buffer.Items.Enqueue(entry);
+            }
+        }
+    }
+
+    public NodeRuntimeLogSnapshot List(NodeId node, int limit)
+    {
+        if (!_nodes.TryGetValue(node, out LogBuffer? buffer))
+        {
+            return new NodeRuntimeLogSnapshot([], 0);
+        }
+
+        lock (buffer.Gate)
+        {
+            return new NodeRuntimeLogSnapshot(
+                [.. buffer.Items.Reverse().Take(Math.Clamp(limit, 1, 200))],
+                buffer.DroppedByNode + buffer.DroppedByBackend);
+        }
+    }
+
+    private static string Truncate(string value, int maxLength) => value.Length <= maxLength ? value : value[..maxLength];
+
+    private static Dictionary<string, string> SafeFields(IReadOnlyDictionary<string, string>? fields)
+    {
+        var safe = new Dictionary<string, string>(StringComparer.Ordinal);
+        if (fields is null)
+        {
+            return safe;
+        }
+
+        foreach ((string key, string value) in fields.Take(8))
+        {
+            string safeKey = Truncate(key, 48);
+            string lower = key.ToLowerInvariant();
+            bool sensitive = lower.Contains("token", StringComparison.Ordinal)
+                || lower.Contains("secret", StringComparison.Ordinal)
+                || lower.Contains("password", StringComparison.Ordinal)
+                || lower.Contains("private", StringComparison.Ordinal)
+                || lower.Contains("signature", StringComparison.Ordinal)
+                || lower.Contains("authorization", StringComparison.Ordinal)
+                || lower.Contains("cookie", StringComparison.Ordinal)
+                || lower == "mac";
+            safe[safeKey] = sensitive ? "[redacted]" : Truncate(value ?? "", 128);
+        }
+
+        return safe;
+    }
+
+    private sealed class LogBuffer
+    {
+        public object Gate { get; } = new();
+        public Queue<NodeRuntimeLog> Items { get; } = [];
+        public long LastId;
+        public long DroppedByNode;
+        public long DroppedByBackend;
+    }
+}

backend/src/MoShiro.Backend.Application/ApplicationServiceCollectionExtensions.cs

@@ -17,6 +17,7 @@ public static class ApplicationServiceCollectionExtensions
 
         services.AddSingleton<AgentRuntime>();
         services.AddSingleton<ProbeRuntime>();
+        services.AddSingleton<NodeRuntimeLogService>();
         services.AddSingleton<ProbeQuicRuntime>();
         services.AddSingleton<LiveHub>();
         services.AddSingleton<HotLayer>();