discord-bot/CompatBot/Watchdog.cs

134 lines
6.3 KiB
C#
Raw Normal View History

using System;
using System.Collections.Concurrent;
using System.Diagnostics;
2020-03-21 10:16:03 +00:00
using System.Linq;
2020-09-09 17:17:54 +00:00
using System.Runtime;
using System.Threading.Tasks;
using CompatBot.Commands;
2020-03-21 10:16:03 +00:00
using CompatBot.Database.Providers;
using CompatBot.EventHandlers;
using DSharpPlus;
2020-07-03 08:57:27 +00:00
using Microsoft.ApplicationInsights;
using NLog;
namespace CompatBot
{
internal static class Watchdog
{
private static readonly TimeSpan CheckInterval = TimeSpan.FromSeconds(10);
public static readonly ConcurrentQueue<DateTime> DisconnectTimestamps = new();
public static readonly Stopwatch TimeSinceLastIncomingMessage = Stopwatch.StartNew();
private static bool IsOk => DisconnectTimestamps.IsEmpty && TimeSinceLastIncomingMessage.Elapsed < Config.IncomingMessageCheckIntervalInMin;
2020-11-11 19:11:36 +00:00
private static DiscordClient? discordClient = null;
public static async Task Watch(DiscordClient client)
{
discordClient = client;
do
{
await Task.Delay(CheckInterval, Config.Cts.Token).ConfigureAwait(false);
2020-03-21 10:16:03 +00:00
foreach (var sudoer in ModProvider.Mods.Values.Where(m => m.Sudoer))
{
var user = await client.GetUserAsync(sudoer.DiscordId).ConfigureAwait(false);
if (user?.Presence?.Activity?.CustomStatus?.Name is string cmd && cmd.StartsWith("restart"))
{
var instance = cmd.Split(' ', StringSplitOptions.RemoveEmptyEntries).LastOrDefault();
if (ulong.TryParse(instance, out var botId) && botId == client.CurrentUser.Id)
{
Config.Log.Warn($"Found request to restart on {user.Username}#{user.Discriminator}'s custom status");
Sudo.Bot.Restart(Program.InvalidChannelId, $"Restarted by request from {user.Username}#{user.Discriminator}'s custom status");
}
}
}
2020-01-03 10:15:19 +00:00
if (IsOk)
continue;
try
{
2020-07-03 08:57:27 +00:00
Config.TelemetryClient?.TrackEvent("socket-deadlock-potential");
2019-04-28 18:08:18 +00:00
Config.Log.Warn("Potential socket deadlock detected, reconnecting...");
2019-04-22 22:49:46 +00:00
await client.ReconnectAsync(true).ConfigureAwait(false);
await Task.Delay(CheckInterval, Config.Cts.Token).ConfigureAwait(false);
2020-01-03 10:15:19 +00:00
if (IsOk)
{
2019-04-28 18:08:18 +00:00
Config.Log.Info("Looks like we're back in business");
2019-04-22 22:49:46 +00:00
continue;
}
2020-01-03 10:15:19 +00:00
2020-07-03 08:57:27 +00:00
Config.TelemetryClient?.TrackEvent("socket-deadlock-for-sure");
2019-04-28 18:08:18 +00:00
Config.Log.Error("Hard reconnect failed, restarting...");
2020-03-21 11:45:04 +00:00
Sudo.Bot.Restart(Program.InvalidChannelId, $@"Restarted to reset potential socket deadlock (last incoming message event: {TimeSinceLastIncomingMessage.Elapsed:h\:mm\:ss} ago)");
}
catch (Exception e)
{
Config.Log.Error(e);
}
} while (!Config.Cts.IsCancellationRequested);
}
2020-07-03 08:57:27 +00:00
public static void OnLogHandler(string level, string message)
{
if (level == nameof(LogLevel.Info))
{
2020-11-11 19:11:36 +00:00
if (message.Contains("Session resumed"))
DisconnectTimestamps.Clear();
}
else if (level == nameof(LogLevel.Warn))
{
2020-11-11 19:11:36 +00:00
if (message.Contains("Dispatch:PRESENCES_REPLACE")
&& discordClient != null)
BotStatusMonitor.RefreshAsync(discordClient).ConfigureAwait(false).GetAwaiter().GetResult();
2020-11-11 19:11:36 +00:00
else if (message.Contains("Pre-emptive ratelimit triggered"))
Config.TelemetryClient?.TrackEvent("preemptive-rate-limit");
}
else if (level == nameof(LogLevel.Fatal))
{
2020-11-11 19:11:36 +00:00
if ((message.Contains("Socket connection terminated"))
|| (message.Contains("heartbeats were skipped. Issuing reconnect.")))
DisconnectTimestamps.Enqueue(DateTime.UtcNow);
}
}
2020-07-03 08:57:27 +00:00
public static async Task SendMetrics(DiscordClient client)
{
do
{
await Task.Delay(Config.MetricsIntervalInSec).ConfigureAwait(false);
2020-11-11 19:11:36 +00:00
var gcMemInfo = GC.GetGCMemoryInfo();
2020-09-09 17:17:54 +00:00
using var process = Process.GetCurrentProcess();
2020-11-14 10:25:20 +00:00
if (Config.TelemetryClient is not TelemetryClient tc)
continue;
tc.TrackMetric("gw-latency", client.Ping);
tc.TrackMetric("time-since-last-incoming-message", TimeSinceLastIncomingMessage.ElapsedMilliseconds);
tc.TrackMetric("memory-gc-total", gcMemInfo.HeapSizeBytes);
tc.TrackMetric("memory-gc-load", gcMemInfo.MemoryLoadBytes);
tc.TrackMetric("memory-gc-commited", gcMemInfo.TotalCommittedBytes);
tc.TrackMetric("memory-process-private", process.PrivateMemorySize64);
tc.TrackMetric("memory-process-ws", process.WorkingSet64);
tc.TrackMetric("github-limit-remaining", GithubClient.Client.RateLimitRemaining);
tc.Flush();
2020-07-03 08:57:27 +00:00
} while (!Config.Cts.IsCancellationRequested);
}
2020-09-09 17:17:54 +00:00
public static async Task CheckGCStats()
{
do
{
2020-11-11 19:11:36 +00:00
var gcMemInfo = GC.GetGCMemoryInfo();
2020-09-09 17:17:54 +00:00
using var process = Process.GetCurrentProcess();
2020-09-10 17:16:10 +00:00
Config.Log.Info($"Process memory stats:\n" +
2020-11-11 19:11:36 +00:00
$"GC Heap: {gcMemInfo.HeapSizeBytes}\n" +
2020-09-10 17:16:10 +00:00
$"Private: {process.PrivateMemorySize64}\n" +
$"Working set: {process.WorkingSet64}\n" +
$"Virtual: {process.VirtualMemorySize64}\n" +
$"Paged: {process.PagedMemorySize64}\n" +
$"Paged sytem: {process.PagedSystemMemorySize64}\n" +
$"Non-pated system: {process.NonpagedSystemMemorySize64}");
await Task.Delay(TimeSpan.FromHours(1)).ConfigureAwait(false);
2020-09-09 17:17:54 +00:00
} while (!Config.Cts.IsCancellationRequested);
}
}
}