| | | 1 | | using Hangfire; |
| | | 2 | | using Hangfire.Dashboard; |
| | | 3 | | using Hangfire.MySql; |
| | | 4 | | using Hangfire.Storage.Monitoring; |
| | | 5 | | using Microsoft.Extensions.Configuration; |
| | | 6 | | using Microsoft.EntityFrameworkCore; |
| | | 7 | | using OzonDomains.Models; |
| | | 8 | | using OzonOrdersWeb.WebServices.HangfireAuthorization; |
| | | 9 | | using OzonOrdersWeb.Observability; |
| | | 10 | | using OzonRepositories.Context; |
| | | 11 | | using Serilog; |
| | | 12 | | using Services.HangfireService; |
| | | 13 | | |
| | | 14 | | namespace OzonOrdersWeb.Extensions; |
| | | 15 | | |
| | | 16 | | public static class HangfireExtensions |
| | | 17 | | { |
| | | 18 | | public const string MainQueue = HangfireQueueNames.Main; |
| | | 19 | | public const string OzonWebhookQueue = HangfireQueueNames.OzonWebhook; |
| | | 20 | | public const string OrderStatusSyncQueue = HangfireQueueNames.OrderStatusSync; |
| | | 21 | | public const string OrderSearchQueue = HangfireQueueNames.OrderSearch; |
| | | 22 | | /// <summary> |
| | | 23 | | /// Удаляет задачи Hangfire (очередь, выполнение, отложенные), оставшиеся после предыдущего запуска. |
| | | 24 | | /// Recurring-расписания не трогает — они перерегистрируются при старте отдельно. |
| | | 25 | | /// Вызывать только после инициализации JobStorage (см. HangfireStartupHostedService). |
| | | 26 | | /// </summary> |
| | | 27 | | public static void ClearHangfireStaleJobs(IConfiguration configuration) |
| | | 28 | | { |
| | 0 | 29 | | if (!configuration.GetValue("Hangfire:ClearQueuesOnStartup", true)) |
| | | 30 | | { |
| | 0 | 31 | | Log.Information("Hangfire: очистка очередей при старте отключена (Hangfire:ClearQueuesOnStartup=false)"); |
| | 0 | 32 | | return; |
| | | 33 | | } |
| | | 34 | | |
| | | 35 | | try |
| | | 36 | | { |
| | 0 | 37 | | var monitor = JobStorage.Current.GetMonitoringApi(); |
| | 0 | 38 | | var deleted = 0; |
| | | 39 | | |
| | 0 | 40 | | foreach (var queue in monitor.Queues()) |
| | | 41 | | { |
| | 0 | 42 | | deleted += DeletePagedJobs(queue.Length, from => |
| | 0 | 43 | | monitor.EnqueuedJobs(queue.Name, (int)from, 1000)); |
| | | 44 | | |
| | | 45 | | try |
| | | 46 | | { |
| | 0 | 47 | | deleted += DeletePagedJobs(monitor.FetchedCount(queue.Name), from => |
| | 0 | 48 | | monitor.FetchedJobs(queue.Name, (int)from, 1000)); |
| | 0 | 49 | | } |
| | 0 | 50 | | catch (Exception ex) |
| | | 51 | | { |
| | 0 | 52 | | Log.Warning(ex, "Hangfire: не удалось очистить fetched-задачи очереди {Queue}", queue.Name); |
| | 0 | 53 | | } |
| | | 54 | | |
| | 0 | 55 | | Log.Information("Hangfire: очищена очередь {Queue}", queue.Name); |
| | | 56 | | } |
| | | 57 | | |
| | 0 | 58 | | deleted += DeletePagedJobs(monitor.ProcessingCount(), from => |
| | 0 | 59 | | monitor.ProcessingJobs((int)from, 1000)); |
| | | 60 | | |
| | 0 | 61 | | deleted += DeletePagedJobs(monitor.ScheduledCount(), from => |
| | 0 | 62 | | monitor.ScheduledJobs((int)from, 1000)); |
| | | 63 | | |
| | 0 | 64 | | Log.Information( |
| | 0 | 65 | | "Hangfire: при старте удалено {Count} задач (очередь/fetched/выполнение/отложенные/retry)", deleted); |
| | 0 | 66 | | } |
| | 0 | 67 | | catch (Exception ex) |
| | | 68 | | { |
| | 0 | 69 | | Log.Warning(ex, "Hangfire: не удалось очистить очереди при старте"); |
| | 0 | 70 | | } |
| | 0 | 71 | | } |
| | | 72 | | |
| | | 73 | | private static int DeletePagedJobs<T>(long total, Func<long, JobList<T>> fetchPage) |
| | | 74 | | { |
| | 0 | 75 | | if (total <= 0) |
| | | 76 | | { |
| | 0 | 77 | | return 0; |
| | | 78 | | } |
| | | 79 | | |
| | 0 | 80 | | var deleted = 0; |
| | 0 | 81 | | for (long from = 0; from < total; from += 1000) |
| | | 82 | | { |
| | 0 | 83 | | var jobs = fetchPage(from); |
| | 0 | 84 | | foreach (var job in jobs) |
| | | 85 | | { |
| | 0 | 86 | | if (BackgroundJob.Delete(job.Key)) |
| | | 87 | | { |
| | 0 | 88 | | deleted++; |
| | | 89 | | } |
| | | 90 | | } |
| | | 91 | | |
| | 0 | 92 | | if (jobs.Count < 1000) |
| | | 93 | | { |
| | | 94 | | break; |
| | | 95 | | } |
| | | 96 | | } |
| | | 97 | | |
| | 0 | 98 | | return deleted; |
| | | 99 | | } |
| | | 100 | | |
| | | 101 | | public static void AddHangfireServices(this IServiceCollection services, IConfiguration configuration, bool isProd) |
| | | 102 | | { |
| | | 103 | | // Размер пула задаётся в connection string. Для трёх серверов Hangfire 12 соединений |
| | | 104 | | // недостаточно: служебные процессы и воркеры одновременно используют storage. |
| | 0 | 105 | | string hangfireConnection = (isProd |
| | 0 | 106 | | ? configuration.GetConnectionString("HangfireConnectionProd") |
| | 0 | 107 | | : configuration.GetConnectionString("HangfireConnectionLocal")) |
| | 0 | 108 | | ?? throw new InvalidOperationException("Hangfire connection string is not configured."); |
| | | 109 | | |
| | 0 | 110 | | hangfireConnection = EnsureHangfireCommandTimeout(hangfireConnection, 120); |
| | 0 | 111 | | services.AddSingleton(new HangfireStorageConnection(hangfireConnection)); |
| | 0 | 112 | | services.AddSingleton<HangfireSchedulerWatchdogStorage>(); |
| | 0 | 113 | | var slowJobMilliseconds = configuration.GetValue<int?>( |
| | 0 | 114 | | $"{ObservabilityOptions.SectionName}:SlowLongOperationMilliseconds") ?? 30_000; |
| | | 115 | | |
| | 0 | 116 | | services.AddHangfire((serviceProvider, configuration) => |
| | 0 | 117 | | { |
| | 0 | 118 | | #pragma warning disable CS0618 // Помечено Obsolete в расчёте на новый heartbeat-протокол Hangfire.SqlServer, |
| | 0 | 119 | | // которого в этом MySQL-порте (arnoldasgudas/Hangfire.MySqlStorage 2.0.3) нет — FetchedAt/ |
| | 0 | 120 | | // InvisibilityTimeout по-прежнему единственный механизм переоткрытия «зависших» джоб (см. схему |
| | 0 | 121 | | // HangfireJobQueue). Default 30 мин реально отменяет CancellationToken у живого воркера — |
| | 0 | 122 | | // задевает BitrixStock этап 5 (полный скан каталога Ozon для крупных пресетов превышает 30 мин). |
| | 0 | 123 | | var mySqlStorageOptions = new MySqlStorageOptions |
| | 0 | 124 | | { |
| | 0 | 125 | | TablesPrefix = "Hangfire", |
| | 0 | 126 | | InvisibilityTimeout = TimeSpan.FromHours(6) |
| | 0 | 127 | | }; |
| | 0 | 128 | | #pragma warning restore CS0618 |
| | 0 | 129 | | |
| | 0 | 130 | | configuration.UseStorage(new MySqlStorage(hangfireConnection, mySqlStorageOptions)); |
| | 0 | 131 | | |
| | 0 | 132 | | // Без этого BackgroundJob.Enqueue попадает в default, которую воркеры не обрабатывают. |
| | 0 | 133 | | configuration.UseFilter(new QueueAttribute(MainQueue)); |
| | 0 | 134 | | configuration.UseFilter(new HangfireDistributedLockContentionFilter()); |
| | 0 | 135 | | configuration.UseFilter(new HangfireObservabilityFilter(slowJobMilliseconds)); |
| | 0 | 136 | | configuration.UseFilter(new HangfireFailureTelegramFilter(serviceProvider)); |
| | 0 | 137 | | configuration.UseFilter(new HangfireSafeDistributedLockFilter()); |
| | 0 | 138 | | }); |
| | | 139 | | |
| | | 140 | | |
| | 0 | 141 | | if (!configuration.GetValue("Hangfire:ServerEnabled", true)) |
| | | 142 | | { |
| | 0 | 143 | | return; |
| | | 144 | | } |
| | | 145 | | |
| | | 146 | | // Hangfire: WorkerCount воркеров + ~6 служебных потоков с соединениями к storage. |
| | | 147 | | // Пул Hangfire — отдельный (Application Name=st2-hangfire в connection string). |
| | 0 | 148 | | var configuredWorkers = configuration.GetValue<int?>("Hangfire:WorkerCount"); |
| | 0 | 149 | | var workerCount = configuredWorkers is > 0 |
| | 0 | 150 | | ? configuredWorkers.Value |
| | 0 | 151 | | : Math.Min(2, Math.Max(1, Environment.ProcessorCount)); |
| | | 152 | | |
| | 0 | 153 | | var pollIntervalSeconds = configuration.GetValue<int?>("Hangfire:PollIntervalSeconds"); |
| | 0 | 154 | | services.AddHangfireServer(options => |
| | 0 | 155 | | { |
| | 0 | 156 | | // ozon-webhook — отдельная очередь, чтобы всплеск вебхуков не блокировал main. |
| | 0 | 157 | | options.Queues = [MainQueue, OzonWebhookQueue]; |
| | 0 | 158 | | options.WorkerCount = workerCount; |
| | 0 | 159 | | if (pollIntervalSeconds is > 0) |
| | 0 | 160 | | options.SchedulePollingInterval = TimeSpan.FromSeconds(pollIntervalSeconds.Value); |
| | 0 | 161 | | }); |
| | | 162 | | |
| | | 163 | | // Опрос Ozon не должен конкурировать с импортами, 1С и вебхуками за воркеры и пул ST2. |
| | | 164 | | // Отдельный сервер с одним воркером сохраняет порядок и жёстко ограничивает параллелизм. |
| | 0 | 165 | | services.AddHangfireServer(options => |
| | 0 | 166 | | { |
| | 0 | 167 | | options.ServerName = $"{Environment.MachineName}:order-status-sync"; |
| | 0 | 168 | | options.Queues = [OrderStatusSyncQueue]; |
| | 0 | 169 | | options.WorkerCount = 1; |
| | 0 | 170 | | if (pollIntervalSeconds is > 0) |
| | 0 | 171 | | options.SchedulePollingInterval = TimeSpan.FromSeconds(pollIntervalSeconds.Value); |
| | 0 | 172 | | }); |
| | | 173 | | |
| | | 174 | | // Инкрементальная и полная переиндексация могут долго дренировать накопившиеся записи. |
| | | 175 | | // Она выполняется последовательно и не занимает воркеры прикладной очереди. |
| | 0 | 176 | | services.AddHangfireServer(options => |
| | 0 | 177 | | { |
| | 0 | 178 | | options.ServerName = $"{Environment.MachineName}:order-search"; |
| | 0 | 179 | | options.Queues = [OrderSearchQueue]; |
| | 0 | 180 | | options.WorkerCount = 1; |
| | 0 | 181 | | if (pollIntervalSeconds is > 0) |
| | 0 | 182 | | options.SchedulePollingInterval = TimeSpan.FromSeconds(pollIntervalSeconds.Value); |
| | 0 | 183 | | }); |
| | 0 | 184 | | } |
| | | 185 | | |
| | | 186 | | /// <summary> |
| | | 187 | | /// Помечает «зависшие» прикладные выполнения со статусом Running как прерванные. |
| | | 188 | | /// Такие записи остаются после падения/перезапуска приложения, когда задача Hangfire |
| | | 189 | | /// не успела проставить финальный статус. Вызывать при старте приложения. |
| | | 190 | | /// </summary> |
| | | 191 | | public static async Task AbortStaleRunningExecutionsAsync(IServiceProvider services) |
| | | 192 | | { |
| | | 193 | | try |
| | | 194 | | { |
| | 0 | 195 | | using var scope = services.CreateScope(); |
| | 0 | 196 | | var db = scope.ServiceProvider.GetRequiredService<ApplicationDbContext>(); |
| | 0 | 197 | | var now = DateTime.UtcNow; |
| | | 198 | | const string reason = "Прервано при перезапуске приложения"; |
| | | 199 | | |
| | 0 | 200 | | var background = await db.BackgroundJobExecutions |
| | 0 | 201 | | .Where(e => e.Status == "Running") |
| | 0 | 202 | | .ExecuteUpdateAsync(s => s |
| | 0 | 203 | | .SetProperty(e => e.Status, "Failed") |
| | 0 | 204 | | .SetProperty(e => e.CompletedAt, now) |
| | 0 | 205 | | .SetProperty(e => e.ErrorMessage, reason)); |
| | | 206 | | |
| | 0 | 207 | | var clear = await db.BitrixStockClearExecutions |
| | 0 | 208 | | .Where(e => e.Status == BitrixStockClearExecutionStatus.Running) |
| | 0 | 209 | | .ExecuteUpdateAsync(s => s |
| | 0 | 210 | | .SetProperty(e => e.Status, BitrixStockClearExecutionStatus.Failed) |
| | 0 | 211 | | .SetProperty(e => e.CompletedAt, now) |
| | 0 | 212 | | .SetProperty(e => e.ErrorMessage, reason)); |
| | | 213 | | |
| | 0 | 214 | | var push = await db.BitrixStockPushExecutions |
| | 0 | 215 | | .Where(e => e.Status == BitrixStockPushExecutionStatus.Running) |
| | 0 | 216 | | .ExecuteUpdateAsync(s => s |
| | 0 | 217 | | .SetProperty(e => e.Status, BitrixStockPushExecutionStatus.Failed) |
| | 0 | 218 | | .SetProperty(e => e.CompletedAt, now) |
| | 0 | 219 | | .SetProperty(e => e.ErrorMessage, reason)); |
| | | 220 | | |
| | 0 | 221 | | var stockSync = await db.WarehouseStockSyncRuns |
| | 0 | 222 | | .Where(e => e.Status == WarehouseStockSyncRunStatus.Running) |
| | 0 | 223 | | .ExecuteUpdateAsync(s => s |
| | 0 | 224 | | .SetProperty(e => e.Status, WarehouseStockSyncRunStatus.Failed) |
| | 0 | 225 | | .SetProperty(e => e.CompletedAtUtc, now) |
| | 0 | 226 | | .SetProperty(e => e.ErrorMessage, reason)); |
| | | 227 | | |
| | 0 | 228 | | var emailRouting = await db.EmailRoutingJobExecutions |
| | 0 | 229 | | .Where(e => e.Status == EmailRoutingJobExecutionStatus.Running) |
| | 0 | 230 | | .ExecuteUpdateAsync(s => s |
| | 0 | 231 | | .SetProperty(e => e.Status, EmailRoutingJobExecutionStatus.Failed) |
| | 0 | 232 | | .SetProperty(e => e.CompletedAtUtc, now) |
| | 0 | 233 | | .SetProperty(e => e.ErrorMessage, reason)); |
| | | 234 | | |
| | 0 | 235 | | var archive = await db.BitrixStockArchiveExecutions |
| | 0 | 236 | | .Where(e => e.Status == BitrixStockArchiveExecutionStatus.Running) |
| | 0 | 237 | | .ExecuteUpdateAsync(s => s |
| | 0 | 238 | | .SetProperty(e => e.Status, BitrixStockArchiveExecutionStatus.Failed) |
| | 0 | 239 | | .SetProperty(e => e.CompletedAt, now) |
| | 0 | 240 | | .SetProperty(e => e.ErrorMessage, reason)); |
| | | 241 | | |
| | 0 | 242 | | var unarchive = await db.BitrixStockUnarchiveExecutions |
| | 0 | 243 | | .Where(e => e.Status == BitrixStockUnarchiveExecutionStatus.Running) |
| | 0 | 244 | | .ExecuteUpdateAsync(s => s |
| | 0 | 245 | | .SetProperty(e => e.Status, BitrixStockUnarchiveExecutionStatus.Failed) |
| | 0 | 246 | | .SetProperty(e => e.CompletedAt, now) |
| | 0 | 247 | | .SetProperty(e => e.ErrorMessage, reason)); |
| | | 248 | | |
| | 0 | 249 | | var returnRuns = await db.OrderReturnSyncRuns |
| | 0 | 250 | | .Where(e => e.Status == OrderReturnSyncRunStatus.Running) |
| | 0 | 251 | | .ExecuteUpdateAsync(s => s |
| | 0 | 252 | | .SetProperty(e => e.Status, OrderReturnSyncRunStatus.Failed) |
| | 0 | 253 | | .SetProperty(e => e.CompletedAtUtc, now) |
| | 0 | 254 | | .SetProperty(e => e.ErrorMessage, reason)); |
| | | 255 | | |
| | 0 | 256 | | var syncReports = await db.Database.ExecuteSqlRawAsync( |
| | 0 | 257 | | "UPDATE BitrixStockSyncReports SET Status = 'Failed', CompletedAtUtc = UTC_TIMESTAMP(6) WHERE Status = ' |
| | | 258 | | |
| | 0 | 259 | | Log.Information( |
| | 0 | 260 | | "Startup cleanup: прерваны зависшие Running-выполнения (background={Bg}, clear={Clear}, push={Push}, sto |
| | 0 | 261 | | background, clear, push, stockSync, archive, unarchive, returnRuns, syncReports); |
| | 0 | 262 | | } |
| | 0 | 263 | | catch (Exception ex) |
| | | 264 | | { |
| | 0 | 265 | | Log.Warning(ex, "Startup cleanup: не удалось прервать зависшие Running-выполнения"); |
| | 0 | 266 | | } |
| | 0 | 267 | | } |
| | | 268 | | |
| | | 269 | | /// <summary> |
| | | 270 | | /// Переносит enqueued/recurring из default в main (и webhook-задачи в ozon-webhook). |
| | | 271 | | /// Вызывать при старте приложения. |
| | | 272 | | /// </summary> |
| | | 273 | | public static async Task RepairHangfireDefaultQueuesAsync(IServiceProvider services) |
| | | 274 | | { |
| | | 275 | | try |
| | | 276 | | { |
| | 0 | 277 | | using var scope = services.CreateScope(); |
| | 0 | 278 | | var hangfireStorage = scope.ServiceProvider.GetRequiredService<HangfireStorageConnection>(); |
| | 0 | 279 | | await HangfireQueueRepair.RepairDefaultQueuesAsync(hangfireStorage); |
| | 0 | 280 | | Log.Information("Hangfire: migrated jobs from default queue to main/ozon-webhook"); |
| | 0 | 281 | | } |
| | 0 | 282 | | catch (Exception ex) when (HangfireQueueRepair.IsLockWaitTimeout(ex)) |
| | | 283 | | { |
| | 0 | 284 | | Log.Information(ex, |
| | 0 | 285 | | "Hangfire: queue repair skipped because of transient storage lock contention; it will be retried at next |
| | 0 | 286 | | } |
| | 0 | 287 | | catch (Exception ex) |
| | | 288 | | { |
| | 0 | 289 | | Log.Warning(ex, "Hangfire: failed to repair default queue jobs"); |
| | 0 | 290 | | } |
| | 0 | 291 | | } |
| | | 292 | | |
| | | 293 | | private static string EnsureHangfireCommandTimeout(string connectionString, int seconds) |
| | | 294 | | { |
| | 0 | 295 | | if (connectionString.Contains("Default Command Timeout", StringComparison.OrdinalIgnoreCase)) |
| | | 296 | | { |
| | 0 | 297 | | return connectionString; |
| | | 298 | | } |
| | | 299 | | |
| | 0 | 300 | | return connectionString.TrimEnd(';') + $";Default Command Timeout={seconds}"; |
| | | 301 | | } |
| | | 302 | | } |