TwitchDownloader/dotnet/src/TwitchArchive.Core/Recovery/RecoveryPolicy.cs

using System;

namespace TwitchArchive.Core.Recovery
{
    public enum RecoveryState
    {
        Monitoring,
        Recording,
        FastReconnect,
        SlowReconnect,
        NetworkFault
    }

    public enum RecoveryAction
    {
        None,
        StartRecording,
        StartProcessing,
        SleepOnly
    }

    public sealed class RecoveryDecision
    {
        public RecoveryAction Action { get; init; }
        public TimeSpan Sleep { get; init; }

        public static RecoveryDecision SleepFor(TimeSpan t) => new RecoveryDecision { Action = RecoveryAction.SleepOnly, Sleep = t };
        public static RecoveryDecision StartRecordingNow() => new RecoveryDecision { Action = RecoveryAction.StartRecording, Sleep = TimeSpan.Zero };
        public static RecoveryDecision StartProcessingAndSleep(TimeSpan sleep) => new RecoveryDecision { Action = RecoveryAction.StartProcessing, Sleep = sleep };
        public static RecoveryDecision None() => new RecoveryDecision { Action = RecoveryAction.None, Sleep = TimeSpan.Zero };
    }

    /// <summary>
    /// Pure, testable recovery policy state machine.
    /// It decides whether to start recording immediately, wait a short period (fast reconnect),
    /// wait longer (slow reconnect), or enter a network fault backoff.
    ///</summary>
    public class RecoveryPolicy
    {
        private RecoveryState _state = RecoveryState.Monitoring;
        private readonly TimeSpan _refreshInterval;
        private DateTime? _fastReconnectStartUtc;
        private int _networkFaultAttempts;

        // constants matching requested behavior
        private static readonly TimeSpan FastReconnectInterval = TimeSpan.FromSeconds(10);
        private static readonly TimeSpan FastReconnectWindow = TimeSpan.FromMinutes(2);
        private static readonly TimeSpan SlowReconnectInterval = TimeSpan.FromSeconds(60);
        private static readonly TimeSpan NetworkFaultBase = TimeSpan.FromSeconds(30);
        private static readonly TimeSpan NetworkFaultMax = TimeSpan.FromMinutes(10);

        public RecoveryPolicy(TimeSpan? refreshInterval = null)
        {
            _refreshInterval = refreshInterval ?? TimeSpan.FromSeconds(60);
        }

        /// <summary>
        /// Evaluate the policy given current time, whether the streamer is live (null = unknown),
        /// and whether there was a network error reaching Twitch APIs.
        /// Returns a RecoveryDecision describing the next action and how long to wait if sleeping.
        /// </summary>
        public RecoveryDecision Tick(DateTime nowUtc, bool? isLive, bool networkError)
        {
            // Network error handling: enter NetworkFault state with exponential backoff
            if (networkError)
            {
                _networkFaultAttempts++;
                _state = RecoveryState.NetworkFault;
                var backoff = NetworkFaultBase * Math.Pow(2, Math.Max(0, _networkFaultAttempts - 1));
                if (backoff > NetworkFaultMax) backoff = NetworkFaultMax;
                return RecoveryDecision.SleepFor(backoff);
            }

            // If we recover from network fault, reset attempts
            if (_state == RecoveryState.NetworkFault && !networkError)
            {
                _networkFaultAttempts = 0;
                _state = RecoveryState.Monitoring;
            }

            // If live -> start recording immediately
            if (isLive == true)
            {
                _state = RecoveryState.Recording;
                _fastReconnectStartUtc = null;
                return RecoveryDecision.StartRecordingNow();
            }

            // If unknown live status, treat conservatively: sleep a short amount (refresh)
            if (isLive == null)
            {
                return RecoveryDecision.SleepFor(_refreshInterval);
            }

            // isLive == false here
            switch (_state)
            {
                case RecoveryState.Recording:
                    // Just transitioned from recording to not-live: start fast reconnect window
                    _state = RecoveryState.FastReconnect;
                    _fastReconnectStartUtc = nowUtc;
                    return RecoveryDecision.SleepFor(FastReconnectInterval);

                case RecoveryState.FastReconnect:
                    if (_fastReconnectStartUtc.HasValue && (nowUtc - _fastReconnectStartUtc.Value) < FastReconnectWindow)
                    {
                        // stay in fast reconnect, poll frequently
                        return RecoveryDecision.SleepFor(FastReconnectInterval);
                    }
                    else
                    {
                        // fast reconnect window expired -> move to slow reconnect and trigger processing
                        _state = RecoveryState.SlowReconnect;
                        _fastReconnectStartUtc = null;
                        return RecoveryDecision.StartProcessingAndSleep(SlowReconnectInterval);
                    }

                case RecoveryState.SlowReconnect:
                    // In slow reconnect, poll less frequently
                    return RecoveryDecision.SleepFor(SlowReconnectInterval);

                case RecoveryState.Monitoring:
                default:
                    // Normal monitoring cadence
                    return RecoveryDecision.SleepFor(_refreshInterval);
            }
        }

        public RecoveryState CurrentState => _state;
    }
}