lmxopcua/src/ZB.MOM.WW.OtOpcUa.Server/Observability/HealthEndpointsHost.cs

using System.Net;
using System.Text;
using System.Text.Json;
using Microsoft.Extensions.Logging;
using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
using ZB.MOM.WW.OtOpcUa.Core.Hosting;
using ZB.MOM.WW.OtOpcUa.Core.Observability;

namespace ZB.MOM.WW.OtOpcUa.Server.Observability;

/// <summary>
///     Standalone <see cref="HttpListener"/> host for <c>/healthz</c> and <c>/readyz</c>
///     separate from the OPC UA binding. Per <c>docs/v2/implementation/phase-6-1-resilience-
///     and-observability.md</c> §Stream C.1.
/// </summary>
/// <remarks>
///     Binds to <c>http://localhost:4841</c> by default — loopback avoids the Windows URL-ACL
///     elevation requirement that binding to <c>http://+:4841</c> (wildcard) would impose.
///     When a deployment needs remote probing, a reverse proxy or explicit netsh urlacl grant
///     is the expected path; documented in <c>docs/v2/Server-Deployment.md</c> in a follow-up.
/// </remarks>
public sealed class HealthEndpointsHost : IAsyncDisposable
{
    private readonly string _prefix;
    private readonly DriverHost _driverHost;
    private readonly Func<bool> _configDbHealthy;
    private readonly Func<bool> _usingStaleConfig;
    private readonly ILogger<HealthEndpointsHost> _logger;
    private readonly HttpListener _listener = new();
    private readonly DateTime _startedUtc = DateTime.UtcNow;
    private CancellationTokenSource? _cts;
    private Task? _acceptLoop;
    private bool _disposed;

    public HealthEndpointsHost(
        DriverHost driverHost,
        ILogger<HealthEndpointsHost> logger,
        Func<bool>? configDbHealthy = null,
        Func<bool>? usingStaleConfig = null,
        string prefix = "http://localhost:4841/")
    {
        _driverHost = driverHost;
        _logger = logger;
        _configDbHealthy = configDbHealthy ?? (() => true);
        _usingStaleConfig = usingStaleConfig ?? (() => false);
        _prefix = prefix.EndsWith('/') ? prefix : prefix + "/";
        _listener.Prefixes.Add(_prefix);
    }

    public void Start()
    {
        _listener.Start();
        _cts = new CancellationTokenSource();
        _acceptLoop = Task.Run(() => AcceptLoopAsync(_cts.Token));
        _logger.LogInformation("Health endpoints listening on {Prefix}", _prefix);
    }

    private async Task AcceptLoopAsync(CancellationToken ct)
    {
        while (!ct.IsCancellationRequested)
        {
            HttpListenerContext ctx;
            try
            {
                ctx = await _listener.GetContextAsync().ConfigureAwait(false);
            }
            catch (HttpListenerException) when (ct.IsCancellationRequested) { break; }
            catch (ObjectDisposedException) { break; }

            _ = Task.Run(() => HandleAsync(ctx), ct);
        }
    }

    private async Task HandleAsync(HttpListenerContext ctx)
    {
        try
        {
            var path = ctx.Request.Url?.AbsolutePath ?? "/";
            switch (path)
            {
                case "/healthz":
                    await WriteHealthzAsync(ctx).ConfigureAwait(false);
                    break;
                case "/readyz":
                    await WriteReadyzAsync(ctx).ConfigureAwait(false);
                    break;
                default:
                    ctx.Response.StatusCode = 404;
                    break;
            }
        }
        catch (Exception ex)
        {
            _logger.LogWarning(ex, "Health endpoint handler failure");
            try { ctx.Response.StatusCode = 500; } catch { /* ignore */ }
        }
        finally
        {
            try { ctx.Response.Close(); } catch { /* ignore */ }
        }
    }

    private async Task WriteHealthzAsync(HttpListenerContext ctx)
    {
        var configHealthy = _configDbHealthy();
        var staleConfig = _usingStaleConfig();
        // /healthz is 200 when process alive + (config DB reachable OR cache-warm).
        // Stale-config still serves 200 so the process isn't flagged dead when the DB
        // blips; the body surfaces the stale flag for operators.
        var healthy = configHealthy || staleConfig;
        ctx.Response.StatusCode = healthy ? 200 : 503;

        var body = JsonSerializer.Serialize(new
        {
            status = healthy ? "healthy" : "unhealthy",
            uptimeSeconds = (int)(DateTime.UtcNow - _startedUtc).TotalSeconds,
            configDbReachable = configHealthy,
            usingStaleConfig = staleConfig,
        });
        await WriteBodyAsync(ctx, body).ConfigureAwait(false);
    }

    private async Task WriteReadyzAsync(HttpListenerContext ctx)
    {
        var snapshots = BuildSnapshots();
        var verdict = DriverHealthReport.Aggregate(snapshots);
        ctx.Response.StatusCode = DriverHealthReport.HttpStatus(verdict);

        var body = JsonSerializer.Serialize(new
        {
            verdict = verdict.ToString(),
            uptimeSeconds = (int)(DateTime.UtcNow - _startedUtc).TotalSeconds,
            drivers = snapshots.Select(d => new
            {
                id = d.DriverInstanceId,
                state = d.State.ToString(),
                detail = d.DetailMessage,
            }).ToArray(),
            degradedDrivers = snapshots
                .Where(d => d.State == DriverState.Degraded || d.State == DriverState.Reconnecting)
                .Select(d => d.DriverInstanceId)
                .ToArray(),
        });
        await WriteBodyAsync(ctx, body).ConfigureAwait(false);
    }

    private IReadOnlyList<DriverHealthSnapshot> BuildSnapshots()
    {
        var list = new List<DriverHealthSnapshot>();
        foreach (var id in _driverHost.RegisteredDriverIds)
        {
            var driver = _driverHost.GetDriver(id);
            if (driver is null) continue;
            var health = driver.GetHealth();
            list.Add(new DriverHealthSnapshot(driver.DriverInstanceId, health.State, health.LastError));
        }
        return list;
    }

    private static async Task WriteBodyAsync(HttpListenerContext ctx, string body)
    {
        var bytes = Encoding.UTF8.GetBytes(body);
        ctx.Response.ContentType = "application/json; charset=utf-8";
        ctx.Response.ContentLength64 = bytes.LongLength;
        await ctx.Response.OutputStream.WriteAsync(bytes).ConfigureAwait(false);
    }

    public async ValueTask DisposeAsync()
    {
        if (_disposed) return;
        _disposed = true;
        _cts?.Cancel();
        try { _listener.Stop(); } catch { /* ignore */ }
        if (_acceptLoop is not null)
        {
            try { await _acceptLoop.ConfigureAwait(false); } catch { /* ignore */ }
        }
        _listener.Close();
        _cts?.Dispose();
    }
}