nexus/backend/Services/StaleTaskRecoveryService.cs

using Nexus.Api.Data;
using Nexus.Api.Models;
using Nexus.Api.Repositories;

namespace Nexus.Api.Services;

public sealed class StaleTaskRecoveryService(
    ITaskRepository taskRepository,
    IActivityRepository activityRepository,
    ILiveUpdateService liveUpdateService,
    INotificationService notificationService) : IStaleTaskRecoveryService
{
    private const string StalledActivityType = "stalled";

    /// <summary>
    /// NICHT-destruktiver Watchdog: markiert „In progress"-Tasks ohne Aktivität seit
    /// <paramref name="stalledThreshold"/> als hängend (Activity-Event + Notification an Iris),
    /// OHNE die Spalte zu ändern oder Arbeit zu verwerfen. Iris eskaliert dann (nachfragen,
    /// neu delegieren, ggf. auf Blocked setzen). Dedup: bereits gemeldete Hänger werden nicht
    /// erneut gemeldet, solange kein neuer Fortschritt (andere Activity) dazwischen liegt.
    /// </summary>
    public async Task<int> FlagStalledInProgressTasksAsync(TimeSpan stalledThreshold, CancellationToken ct = default)
    {
        var now = DateTimeOffset.UtcNow;
        var threshold = now - stalledThreshold;
        var allTasks = await taskRepository.GetAllAsync(ct);

        var inProgress = allTasks
            .Where(t => string.Equals(t.State, TaskStateHelper.ToStateString(TaskState.InProgress), StringComparison.OrdinalIgnoreCase))
            .ToList();
        if (inProgress.Count == 0)
            return 0;

        var activities = await activityRepository.GetRecentForTasksAsync(inProgress.Select(t => t.Id), ct);
        var activityByTask = activities
            .Where(a => a.TaskId.HasValue)
            .GroupBy(a => a.TaskId!.Value)
            .ToDictionary(g => g.Key, g => g.OrderByDescending(a => a.CreatedAt).ToList());

        var flaggedCount = 0;

        foreach (var task in inProgress)
        {
            activityByTask.TryGetValue(task.Id, out var taskActivity);
            var latest = taskActivity?.FirstOrDefault();
            var lastProgressAt = latest?.CreatedAt ?? task.UpdatedAt;

            if (lastProgressAt >= threshold)
                continue;

            // Dedup: schon als hängend gemeldet und seither kein neuer Fortschritt.
            if (latest is not null && string.Equals(latest.Type, StalledActivityType, StringComparison.OrdinalIgnoreCase))
                continue;

            var silentFor = now - lastProgressAt;
            await activityRepository.AddAsync(new ActivityEvent
            {
                Type = StalledActivityType,
                Message = $"Watchdog: keine Aktivität seit {FormatDuration(silentFor)} (Schwelle {FormatDuration(stalledThreshold)}). Task bleibt In progress, Iris zur Eskalation benachrichtigt.",
                TaskId = task.Id
            }, ct);

            await notificationService.CreateAsync(
                "task_stalled",
                $"Task hängt: {task.Title}",
                $"Seit {FormatDuration(silentFor)} keine Aktivität. Bitte nachfassen, neu delegieren oder blockieren.",
                "iris",
                task.Id,
                ct);

            flaggedCount++;
        }

        if (flaggedCount > 0)
            liveUpdateService.Publish("tasks.board.snapshot", await BuildBoardSnapshotAsync(ct), "board");

        return flaggedCount;
    }

    /// <summary>
    /// Destruktiver Fallback (nur manuell via Endpoint / expliziter Cron): setzt hängende
    /// „In progress"-Tasks hart auf Backlog zurück. Verwirft laufenden Kontext — daher NICHT
    /// mehr der Standard-Watchdog, sondern nur noch auf Anforderung.
    /// </summary>
    public async Task<int> ResetStaleInProgressTasksAsync(TimeSpan staleThreshold, CancellationToken ct = default)
    {
        var threshold = DateTimeOffset.UtcNow - staleThreshold;
        var staleTasks = await GetStaleTasksAsync(threshold, ct);
        if (staleTasks.Count == 0)
            return 0;

        var latestActivityByTaskId = await GetLatestActivityByTaskIdAsync(staleTasks.Select(task => task.Id), ct);
        var now = DateTimeOffset.UtcNow;
        var resetCount = 0;

        foreach (var task in staleTasks)
        {
            var currentTask = await taskRepository.GetByIdAsync(task.Id, ct);
            if (currentTask is null || !IsStaleInProgress(currentTask, threshold))
                continue;

            latestActivityByTaskId.TryGetValue(currentTask.Id, out var lastActivityAt);
            var message = BuildActivityMessage(currentTask, staleThreshold, now, lastActivityAt);
            var updated = await taskRepository.TryResetStaleInProgressToBacklogAsync(
                currentTask.Id,
                threshold,
                now,
                ct);
            if (!updated)
                continue;

            await activityRepository.AddAsync(new ActivityEvent
            {
                Type = "task",
                Message = message,
                TaskId = task.Id
            }, ct);

            resetCount++;
        }

        if (resetCount > 0)
            liveUpdateService.Publish("tasks.board.snapshot", await BuildBoardSnapshotAsync(ct), "board");

        return resetCount;
    }

    private async Task<List<WorkTask>> GetStaleTasksAsync(DateTimeOffset threshold, CancellationToken ct)
    {
        var allTasks = await taskRepository.GetAllAsync(ct);

        return allTasks
            .Where(task => IsStaleInProgress(task, threshold))
            .ToList();
    }

    private static bool IsStaleInProgress(WorkTask task, DateTimeOffset threshold)
        => string.Equals(task.State, TaskStateHelper.ToStateString(TaskState.InProgress), StringComparison.OrdinalIgnoreCase)
           && task.UpdatedAt < threshold;

    private async Task<Dictionary<Guid, DateTimeOffset>> GetLatestActivityByTaskIdAsync(
        IEnumerable<Guid> taskIds,
        CancellationToken ct)
    {
        var activities = await activityRepository.GetRecentForTasksAsync(taskIds, ct);

        return activities
            .Where(activity => activity.TaskId.HasValue)
            .GroupBy(activity => activity.TaskId!.Value)
            .ToDictionary(group => group.Key, group => group.Max(activity => activity.CreatedAt));
    }

    private async Task<BoardResponse> BuildBoardSnapshotAsync(CancellationToken ct)
    {
        var allTasks = await taskRepository.GetAllAsync(ct);
        var activity = await activityRepository.GetRecentForTasksAsync(allTasks.Select(task => task.Id), ct);
        return TaskService.BuildMasterBoard(allTasks, activity);
    }

    private static string BuildActivityMessage(
        WorkTask task,
        TimeSpan staleThreshold,
        DateTimeOffset now,
        DateTimeOffset? lastActivityAt)
    {
        var staleAge = now - task.UpdatedAt;
        var details = new List<string>
        {
            "reason=stale-recovery",
            "previous status In progress",
            $"stale reference {now:O}",
            $"stale age {FormatDuration(staleAge)}",
            $"threshold {FormatDuration(staleThreshold)}"
        };

        if (lastActivityAt.HasValue)
            details.Add($"last activity {lastActivityAt.Value:O}");

        details.Add($"last update {task.UpdatedAt:O}");
        details.Add("new status Backlog");

        return $"Task \"{task.Title}\" reset from In progress to Backlog by stale recovery ({string.Join("; ", details)})";
    }

    private static string FormatDuration(TimeSpan duration)
    {
        if (duration.TotalHours >= 1)
            return $"{(int)duration.TotalHours}h {duration.Minutes}min";
        return $"{Math.Max(0, (int)duration.TotalMinutes)}min";
    }
}