vfxreview/RenderWorker/VFXReviewWorker/WorkerService.cs

using Microsoft.Extensions.Hosting;
using Microsoft.Extensions.Logging;

namespace VFXReviewWorker;

/// <summary>
/// Main loop (§7.2): register on startup, heartbeat on its own timer (the
/// cancel-signal channel), claim-poll only when idle, one job at a time.
/// Workers poll dumbly — all scheduling policy (render windows, Render Now,
/// urgent priority) lives server-side in the claim endpoint (§7.10).
/// </summary>
public sealed class WorkerService : BackgroundService
{
    private readonly WorkerOptions _options;
    private readonly ApiClient _api;
    private readonly JobExecutor _executor;
    private readonly ReportSpooler _spooler;
    private readonly ILogger<WorkerService> _log;

    private string? _machineId;
    private ServerConfig _config = new();
    private volatile string? _currentJobId;
    private CancellationTokenSource? _cancelCurrentJob;
    private bool? _lastAeOpen;

    public WorkerService(WorkerOptions options, ApiClient api, JobExecutor executor, ReportSpooler spooler, ILogger<WorkerService> log)
    {
        _options = options;
        _api = api;
        _executor = executor;
        _spooler = spooler;
        _log = log;
    }

    protected override async Task ExecuteAsync(CancellationToken stoppingToken)
    {
        _log.LogInformation("VFXReview RenderWorker {Version} starting as \"{Name}\" → {Server}",
            _options.WorkerVersion, _options.MachineName, _options.ServerUrl);

        _ = Task.Run(() => _spooler.RunAsync(stoppingToken), stoppingToken);

        await RegisterWithRetryAsync(stoppingToken);
        await _executor.ReconcileAsync(_machineId!, stoppingToken);

        _ = Task.Run(() => HeartbeatLoopAsync(stoppingToken), stoppingToken);

        // Claim loop — only while idle
        while (!stoppingToken.IsCancellationRequested)
        {
            try
            {
                // Preview builds drive a full AE instance, which would hijack an
                // artist's open session (§9.1). Simply don't ask for those jobs
                // while AE is running — they stay queued for a quiet moment
                // instead of being claimed and failed.
                var aeOpen = AeSession.InteractiveRunning();
                if (aeOpen != _lastAeOpen)
                {
                    _log.LogInformation(aeOpen
                        ? "Interactive After Effects detected — not claiming preview jobs until it closes"
                        : "No interactive After Effects — preview jobs enabled");
                    _lastAeOpen = aeOpen;
                }
                var types = aeOpen
                    ? new[] { "AE_RENDER" }
                    : new[] { "AE_RENDER", "PREVIEW_ONLY" };

                var job = await _api.ClaimAsync(_machineId!, types, stoppingToken);
                if (job is not null)
                {
                    _log.LogInformation("Claimed job {JobId} (attempt {Attempt}/{Max}, export {ExportId})",
                        job.Id, job.Attempt, job.MaxAttempts, job.ExportId);
                    _cancelCurrentJob = CancellationTokenSource.CreateLinkedTokenSource(CancellationToken.None);
                    _currentJobId = job.Id;
                    try
                    {
                        await _executor.RunAsync(job, _machineId!, _config, _cancelCurrentJob.Token, stoppingToken);
                    }
                    finally
                    {
                        _currentJobId = null;
                        _cancelCurrentJob.Dispose();
                        _cancelCurrentJob = null;
                    }
                    continue; // immediately look for the next job
                }
            }
            catch (OperationCanceledException) when (stoppingToken.IsCancellationRequested)
            {
                break;
            }
            catch (Exception ex)
            {
                _log.LogWarning(ex, "Claim poll failed (server unreachable?) — retrying on normal cadence");
            }
            try { await Task.Delay(TimeSpan.FromSeconds(_config.PollSeconds), stoppingToken); }
            catch (OperationCanceledException) { break; }
        }
        _log.LogInformation("RenderWorker stopping");
    }

    private async Task RegisterWithRetryAsync(CancellationToken ct)
    {
        var backoff = TimeSpan.FromSeconds(1);
        while (!ct.IsCancellationRequested)
        {
            try
            {
                var reg = await _api.RegisterAsync(_options, ct);
                _machineId = reg.Machine.Id;
                _config = reg.Config;
                _log.LogInformation(
                    "Registered as machine {Id} ({Name}); poll={Poll}s heartbeat={Hb}s lease={Lease}s stall={Stall}s enabled={Enabled}",
                    reg.Machine.Id, reg.Machine.Name, _config.PollSeconds, _config.HeartbeatSeconds,
                    _config.LeaseSeconds, _config.StallTimeoutSeconds, reg.Machine.Enabled);
                return;
            }
            catch (OperationCanceledException) when (ct.IsCancellationRequested)
            {
                return;
            }
            catch (Exception ex)
            {
                _log.LogWarning("Registration failed ({Error}); retrying in {Backoff}s", ex.Message, backoff.TotalSeconds);
                try { await Task.Delay(backoff, ct); } catch (OperationCanceledException) { return; }
                backoff = TimeSpan.FromSeconds(Math.Min(backoff.TotalSeconds * 2, 60));
            }
        }
    }

    private async Task HeartbeatLoopAsync(CancellationToken ct)
    {
        while (!ct.IsCancellationRequested)
        {
            try
            {
                var resp = await _api.HeartbeatAsync(_machineId!, _currentJobId, ct);
                foreach (var cmd in resp.Commands)
                {
                    if (cmd.Type == "CANCEL_JOB" && cmd.JobId is not null && cmd.JobId == _currentJobId)
                    {
                        _log.LogWarning("Heartbeat carried CANCEL_JOB for current job {JobId}", cmd.JobId);
                        _cancelCurrentJob?.Cancel();
                    }
                }
            }
            catch (OperationCanceledException) when (ct.IsCancellationRequested)
            {
                return;
            }
            catch (Exception ex)
            {
                _log.LogDebug(ex, "Heartbeat failed (non-fatal)");
            }
            try { await Task.Delay(TimeSpan.FromSeconds(_config.HeartbeatSeconds), ct); }
            catch (OperationCanceledException) { return; }
        }
    }
}