SystemDManager.cs

Go to the documentation of this file.

using System;
using System.Globalization;
using System.Threading;
using System.Threading.Tasks;
 
using Microsoft.Extensions.Hosting;
using Microsoft.Extensions.Logging;
 
using Mono.Unix;
 
using Tgstation.Server.Host.Components;
using Tgstation.Server.Host.Core;
 
namespace Tgstation.Server.Host.System
{
    sealed class SystemDManager : BackgroundService, IRestartHandler, IDisposable
    {
        const string SDNotifyWatchdog = "WATCHDOG=1";
 
        readonly IHostApplicationLifetime applicationLifetime;
 
        readonly IInstanceManager instanceManager;
 
        readonly IRestartRegistration restartRegistration;
 
        readonly ILogger<SystemDManager> logger;
 
        bool restartInProgress;
 
        static long GetMonotonicUsec() => global::System.Diagnostics.Stopwatch.GetTimestamp(); // HACK: https://github.com/dotnet/runtime/blob/v8.0.0-preview.6.23329.7/src/native/libs/System.Native/pal_time.c#L84 clock_gettime_nsec_np is an OSX only thing apparently...
 
        public SystemDManager(
            IHostApplicationLifetime applicationLifetime,
            IInstanceManager instanceManager,
            IServerControl serverControl,
            ILogger<SystemDManager> logger)
        {
            this.applicationLifetime = applicationLifetime ?? throw new ArgumentNullException(nameof(applicationLifetime));
            this.instanceManager = instanceManager ?? throw new ArgumentNullException(nameof(instanceManager));
 
            ArgumentNullException.ThrowIfNull(serverControl);
 
            this.logger = logger ?? throw new ArgumentNullException(nameof(logger));
 
            restartRegistration = serverControl.RegisterForRestart(this);
        }
 
        public override void Dispose()
        {
            base.Dispose();
            restartRegistration.Dispose();
        }
 
        public ValueTask HandleRestart(Version? updateVersion, bool handlerMayDelayShutdownWithExtremelyLongRunningTasks, CancellationToken cancellationToken)
        {
            // If this is set, we know a gracefule SHUTDOWN was requested
            restartInProgress = !handlerMayDelayShutdownWithExtremelyLongRunningTasks;
            return ValueTask.CompletedTask;
        }
 
        protected override async Task ExecuteAsync(CancellationToken cancellationToken)
        {
            if (!SendSDNotify(SDNotifyWatchdog))
            {
                logger.LogDebug("SystemD not detected");
                return;
            }
 
            logger.LogDebug("SystemD detected");
 
            if (applicationLifetime.ApplicationStarted.IsCancellationRequested)
                throw new InvalidOperationException("RunAsync called after application started!");
 
            logger.LogTrace("Installing lifetime handlers...");
 
            var readyCounts = 0;
            void CheckReady()
            {
                if (Interlocked.Increment(ref readyCounts) < 2)
                    return;
 
                SendSDNotify("READY=1");
            }
 
            applicationLifetime.ApplicationStarted.Register(() => CheckReady());
            applicationLifetime.ApplicationStopping.Register(
                () => SendSDNotify(
                    restartInProgress
                        ? $"RELOADING=1\nMONOTONIC_USEC={GetMonotonicUsec()}"
                        : "STOPPING=1"));
 
            try
            {
                await instanceManager.Ready.WaitAsync(cancellationToken);
                CheckReady();
 
                var watchdogUsec = Environment.GetEnvironmentVariable("WATCHDOG_USEC");
                if (String.IsNullOrWhiteSpace(watchdogUsec))
                {
                    logger.LogDebug("WATCHDOG_USEC not present, not starting watchdog loop");
                    return;
                }
 
                var microseconds = UInt64.Parse(watchdogUsec, CultureInfo.InvariantCulture);
                var timeoutIntervalMillis = (int)(microseconds / 1000);
 
                logger.LogDebug("Starting watchdog loop with interval of {timeoutInterval}ms", timeoutIntervalMillis);
 
                var timeoutInterval = TimeSpan.FromMilliseconds(timeoutIntervalMillis);
                var nextExpectedTimeout = DateTimeOffset.UtcNow + timeoutInterval;
                var timeToNextExpectedTimeout = nextExpectedTimeout - DateTimeOffset.UtcNow;
                while (!cancellationToken.IsCancellationRequested)
                {
                    var delayInterval = timeToNextExpectedTimeout / 2;
                    await Task.Delay(delayInterval, cancellationToken);
 
                    var notifySuccess = SendSDNotify(SDNotifyWatchdog);
 
                    var now = DateTimeOffset.UtcNow;
                    if (notifySuccess)
                        nextExpectedTimeout = now + timeoutInterval;
 
                    timeToNextExpectedTimeout = nextExpectedTimeout - now;
 
                    if (!notifySuccess)
                        logger.LogWarning("Missed systemd heartbeat! Expected timeout in {timeoutMs}ms...", timeToNextExpectedTimeout.TotalMilliseconds);
                }
            }
            catch (OperationCanceledException ex)
            {
                logger.LogTrace(ex, "Watchdog loop cancelled!");
            }
            catch (Exception ex)
            {
                logger.LogError(ex, "Watchdog loop crashed!");
            }
 
            logger.LogDebug("Exited watchdog loop");
        }
 
        bool SendSDNotify(string command)
        {
            logger.LogTrace("Sending sd_notify {message}...", command);
            int result;
            try
            {
                result = NativeMethods.sd_notify(0, command);
            }
            catch (Exception ex)
            {
                logger.LogInformation(ex, "Exception attempting to invoke sd_notify!");
                return false;
            }
 
            if (result > 0)
                return true;
 
            if (result < 0)
                logger.LogError(new UnixIOException(result), "sd_notify {message} failed!", command);
            else
                logger.LogTrace("Could not send sd_notify {message}. Socket closed!", command);
 
            return false;
        }
    }
}