Initial commit: JMon cross-platform monitoring system
- gRPC client-server metrics collection - PeriodicTimer-based 5-second sampling - Delta encoding with configurable thresholds - Queue/retry mechanism for resilience - ProcessId tracking for duplicate detection - Server-side state reconstruction - Native AOT compilation support - Docker/Podman containerization - PostgreSQL + TimescaleDB persistence layer
This commit is contained in:
205
JMonServer/Program.cs
Normal file
205
JMonServer/Program.cs
Normal file
@@ -0,0 +1,205 @@
|
||||
using JMonAgent;
|
||||
using JMon.Protos;
|
||||
|
||||
var builder = WebApplication.CreateBuilder(args);
|
||||
|
||||
builder.Services.AddGrpc();
|
||||
|
||||
// Track latest full metrics per machine (keyed by HWID:PID)
|
||||
var latestMetrics = new Dictionary<string, SystemMetrics>();
|
||||
|
||||
// Track pending metrics per machine for replay in timestamp order
|
||||
var pendingMetrics = new Dictionary<string, List<JMon.Protos.DeltaMetrics>>();
|
||||
|
||||
// Track which process is active per HWID (for duplicate detection)
|
||||
var activeProcesses = new Dictionary<string, (int ProcessId, DateTime LastSeen)>();
|
||||
|
||||
builder.Services.AddSingleton(latestMetrics);
|
||||
builder.Services.AddSingleton(pendingMetrics);
|
||||
builder.Services.AddSingleton(activeProcesses);
|
||||
|
||||
var app = builder.Build();
|
||||
app.MapGrpcService<MetricsServiceImpl>();
|
||||
// Use environment variables for URL, fallback to HTTP on 0.0.0.0:5001
|
||||
var urls = Environment.GetEnvironmentVariable("ASPNETCORE_URLS") ?? "http://0.0.0.0:5001";
|
||||
app.Run(urls);
|
||||
|
||||
public class MetricsServiceImpl : MetricsService.MetricsServiceBase
|
||||
{
|
||||
private readonly Dictionary<string, SystemMetrics> _latestMetrics;
|
||||
private readonly Dictionary<string, List<JMon.Protos.DeltaMetrics>> _pendingMetrics;
|
||||
private readonly Dictionary<string, (int ProcessId, DateTime LastSeen)> _activeProcesses;
|
||||
|
||||
public MetricsServiceImpl(Dictionary<string, SystemMetrics> latestMetrics, Dictionary<string, List<JMon.Protos.DeltaMetrics>> pendingMetrics, Dictionary<string, (int ProcessId, DateTime LastSeen)> activeProcesses)
|
||||
{
|
||||
_latestMetrics = latestMetrics;
|
||||
_pendingMetrics = pendingMetrics;
|
||||
_activeProcesses = activeProcesses;
|
||||
}
|
||||
|
||||
public override async Task<JMon.Protos.MetricsResponse> ReportMetrics(JMon.Protos.DeltaMetrics request, Grpc.Core.ServerCallContext context)
|
||||
{
|
||||
var hwid = request.Hwid;
|
||||
|
||||
// Check for duplicate processes on same HWID
|
||||
var command = MetricsResponse.Types.CommandType.Ack;
|
||||
var message = "Metrics received";
|
||||
|
||||
if (_activeProcesses.ContainsKey(hwid))
|
||||
{
|
||||
var (activePid, lastSeen) = _activeProcesses[hwid];
|
||||
|
||||
if (activePid != request.ProcessId)
|
||||
{
|
||||
// Multiple processes detected - SIGTERM the older one, SYNC the newer
|
||||
if (request.ProcessId > activePid)
|
||||
{
|
||||
// New process is newer - terminate the old one
|
||||
command = MetricsResponse.Types.CommandType.Sigterm;
|
||||
message = $"Duplicate process detected. Terminating PID {activePid}";
|
||||
_activeProcesses[hwid] = (request.ProcessId, DateTime.UtcNow);
|
||||
}
|
||||
else
|
||||
{
|
||||
// Current request is older - ask for full sync
|
||||
command = MetricsResponse.Types.CommandType.Sync;
|
||||
message = $"Duplicate process detected. Newer process (PID {activePid}) is active";
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
// Same process, just update last seen
|
||||
_activeProcesses[hwid] = (request.ProcessId, DateTime.UtcNow);
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
// First process for this HWID
|
||||
_activeProcesses[hwid] = (request.ProcessId, DateTime.UtcNow);
|
||||
}
|
||||
|
||||
// If this is a retried metric, queue it for ordered replay
|
||||
if (request.Retried)
|
||||
{
|
||||
if (!_pendingMetrics.ContainsKey(hwid))
|
||||
_pendingMetrics[hwid] = new List<JMon.Protos.DeltaMetrics>();
|
||||
_pendingMetrics[hwid].Add(request);
|
||||
}
|
||||
|
||||
// Process all pending metrics for this machine in timestamp order
|
||||
if (_pendingMetrics.ContainsKey(hwid) && _pendingMetrics[hwid].Count > 0)
|
||||
{
|
||||
var pending = _pendingMetrics[hwid];
|
||||
// Sort by timestamp (oldest first)
|
||||
var sorted = pending.OrderBy(m => m.TimestampUnixMs).ToList();
|
||||
|
||||
foreach (var metric in sorted)
|
||||
{
|
||||
ApplyMetric(hwid, metric);
|
||||
}
|
||||
|
||||
// Clear processed pending metrics
|
||||
_pendingMetrics[hwid].Clear();
|
||||
}
|
||||
|
||||
// Apply the current metric
|
||||
if (!request.Retried)
|
||||
{
|
||||
ApplyMetric(hwid, request);
|
||||
}
|
||||
|
||||
var serverTime = DateTime.UtcNow;
|
||||
var agentTime = DateTime.UnixEpoch.AddMilliseconds(request.TimestampUnixMs);
|
||||
var timeDeltaMs = Math.Abs((serverTime - agentTime).TotalMilliseconds);
|
||||
|
||||
var key = $"{hwid}:{request.ProcessId}";
|
||||
if (_latestMetrics.TryGetValue(key, out var latest))
|
||||
{
|
||||
DisplayMetrics("[gRPC]", latest, timeDeltaMs, request.Retried);
|
||||
}
|
||||
|
||||
return await Task.FromResult(new JMon.Protos.MetricsResponse
|
||||
{
|
||||
Success = true,
|
||||
Command = command,
|
||||
Message = message
|
||||
});
|
||||
}
|
||||
|
||||
private void ApplyMetric(string hwid, JMon.Protos.DeltaMetrics delta)
|
||||
{
|
||||
var agentTime = DateTime.UnixEpoch.AddMilliseconds(delta.TimestampUnixMs);
|
||||
|
||||
// Use HWID + PID as key
|
||||
var key = $"{hwid}:{delta.ProcessId}";
|
||||
|
||||
// Reconstruct full metrics by merging delta with previous state
|
||||
SystemMetrics full;
|
||||
if (_latestMetrics.TryGetValue(key, out var previous))
|
||||
{
|
||||
full = new SystemMetrics(
|
||||
delta.MachineName,
|
||||
delta.HasOsDescription ? delta.OsDescription : previous.OSDescription,
|
||||
delta.HasCpuUsagePercentage ? delta.CpuUsagePercentage : previous.CpuUsagePercentage,
|
||||
delta.HasMemoryUsedBytes ? delta.MemoryUsedBytes : previous.MemoryUsedBytes,
|
||||
delta.HasMemoryTotalBytes ? delta.MemoryTotalBytes : previous.MemoryTotalBytes,
|
||||
delta.HasSwapUsedBytes ? delta.SwapUsedBytes : previous.SwapUsedBytes,
|
||||
delta.HasSwapTotalBytes ? delta.SwapTotalBytes : previous.SwapTotalBytes,
|
||||
delta.HasDiskUsedBytes ? delta.DiskUsedBytes : previous.DiskUsedBytes,
|
||||
delta.HasDiskTotalBytes ? delta.DiskTotalBytes : previous.DiskTotalBytes,
|
||||
delta.HasDiskMountPoint ? delta.DiskMountPoint : previous.DiskMountPoint,
|
||||
agentTime,
|
||||
delta.HasIpAddress ? delta.IpAddress : previous.IpAddress,
|
||||
hwid,
|
||||
delta.ProcessId
|
||||
);
|
||||
}
|
||||
else
|
||||
{
|
||||
// First time - initialize with sent values or defaults
|
||||
full = new SystemMetrics(
|
||||
delta.MachineName,
|
||||
delta.HasOsDescription ? delta.OsDescription : "Unknown",
|
||||
delta.HasCpuUsagePercentage ? delta.CpuUsagePercentage : 0,
|
||||
delta.HasMemoryUsedBytes ? delta.MemoryUsedBytes : 0,
|
||||
delta.HasMemoryTotalBytes ? delta.MemoryTotalBytes : 0,
|
||||
delta.HasSwapUsedBytes ? delta.SwapUsedBytes : 0,
|
||||
delta.HasSwapTotalBytes ? delta.SwapTotalBytes : 0,
|
||||
delta.HasDiskUsedBytes ? delta.DiskUsedBytes : 0,
|
||||
delta.HasDiskTotalBytes ? delta.DiskTotalBytes : 0,
|
||||
delta.HasDiskMountPoint ? delta.DiskMountPoint : "/",
|
||||
agentTime,
|
||||
delta.HasIpAddress ? delta.IpAddress : "",
|
||||
hwid,
|
||||
delta.ProcessId
|
||||
);
|
||||
}
|
||||
|
||||
_latestMetrics[key] = full;
|
||||
}
|
||||
|
||||
private void DisplayMetrics(string source, SystemMetrics full, double timeDeltaMs, bool retried)
|
||||
{
|
||||
var tag = retried ? "[RETRIED]" : "";
|
||||
Console.WriteLine($"{source} {tag} {full.MachineName} ({full.HwId}) [PID: {full.ProcessId}]:");
|
||||
Console.WriteLine($" Time Skew: {timeDeltaMs:F1}ms");
|
||||
|
||||
if (timeDeltaMs > 60000)
|
||||
{
|
||||
Console.ForegroundColor = ConsoleColor.Yellow;
|
||||
Console.WriteLine($" ⚠️ ALERT: Clock skew detected! {timeDeltaMs:F0}ms difference");
|
||||
Console.ResetColor();
|
||||
}
|
||||
|
||||
if (!string.IsNullOrEmpty(full.IpAddress))
|
||||
Console.WriteLine($" IP: {full.IpAddress}");
|
||||
Console.WriteLine($" CPU: {full.CpuUsagePercentage:F2}%");
|
||||
var memPercent = (full.MemoryTotalBytes > 0) ? (full.MemoryUsedBytes * 100.0 / full.MemoryTotalBytes).ToString("F1") : "0.0";
|
||||
Console.WriteLine($" Memory: {full.MemoryUsedBytes / (1024*1024)} MB / {full.MemoryTotalBytes / (1024*1024)} MB ({memPercent}%)");
|
||||
var swapPercent = (full.SwapTotalBytes > 0) ? (full.SwapUsedBytes * 100.0 / full.SwapTotalBytes).ToString("F1") : "0.0";
|
||||
Console.WriteLine($" Swap: {full.SwapUsedBytes / (1024*1024)} MB / {full.SwapTotalBytes / (1024*1024)} MB ({swapPercent}%)");
|
||||
var diskPercent = (full.DiskTotalBytes > 0) ? (full.DiskUsedBytes * 100.0 / full.DiskTotalBytes).ToString("F1") : "0.0";
|
||||
Console.WriteLine($" Disk ({full.DiskMountPoint}): {full.DiskUsedBytes / (1024*1024)} MB / {full.DiskTotalBytes / (1024*1024)} MB ({diskPercent}%)");
|
||||
Console.WriteLine($" OS: {full.OSDescription}\n");
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user