| | | 1 | | using Confluent.Kafka; |
| | | 2 | | using Confluent.Kafka.Admin; |
| | | 3 | | using System.Text; |
| | | 4 | | |
| | | 5 | | namespace AsyncResponse.Transports.Kafka; |
| | | 6 | | |
| | | 7 | | /// <summary>One Kafka message header. Values are raw bytes, per the Kafka wire model.</summary> |
| | | 8 | | internal readonly record struct KafkaTransportHeader(string Key, byte[]? Value) |
| | | 9 | | { |
| | | 10 | | public static KafkaTransportHeader Utf8(string key, string value) |
| | | 11 | | => new(key, Encoding.UTF8.GetBytes(value)); |
| | | 12 | | |
| | | 13 | | public string? ValueUtf8 => Value is null ? null : Encoding.UTF8.GetString(Value); |
| | | 14 | | } |
| | | 15 | | |
| | | 16 | | /// <summary>One message pulled from a Kafka topic, decoupled from the vendor client types.</summary> |
| | | 17 | | internal sealed record KafkaIncomingMessage( |
| | | 18 | | string Topic, |
| | | 19 | | int Partition, |
| | | 20 | | long Offset, |
| | | 21 | | byte[]? Payload, |
| | | 22 | | IReadOnlyList<KafkaTransportHeader> Headers); |
| | | 23 | | |
| | | 24 | | /// <summary>Broker coordinates assigned to a produced message.</summary> |
| | | 25 | | internal sealed record KafkaPublishResult(string Topic, int Partition, long Offset); |
| | | 26 | | |
| | | 27 | | /// <summary> |
| | | 28 | | /// Adapter seam over the Kafka producer so the publish and dead-letter paths are unit-testable |
| | | 29 | | /// on fakes. One producer instance is shared process-wide: Kafka producers are thread-safe and |
| | | 30 | | /// expensive to create. |
| | | 31 | | /// </summary> |
| | | 32 | | internal interface IKafkaProducerClient : IDisposable |
| | | 33 | | { |
| | | 34 | | Task<KafkaPublishResult> PublishAsync( |
| | | 35 | | string topic, |
| | | 36 | | string? key, |
| | | 37 | | byte[] payload, |
| | | 38 | | IReadOnlyList<KafkaTransportHeader> headers, |
| | | 39 | | CancellationToken cancellationToken); |
| | | 40 | | } |
| | | 41 | | |
| | | 42 | | /// <summary> |
| | | 43 | | /// Adapter seam over one Kafka consumer. Not thread-safe: each hosted subscriber owns one |
| | | 44 | | /// consumer and touches it only from its own poll loop — detached handlers never call it; their |
| | | 45 | | /// completions are settled by the poll thread (and, after the loop has exited, by the |
| | | 46 | | /// dispatcher's disposal, sequentially). |
| | | 47 | | /// </summary> |
| | | 48 | | internal interface IKafkaConsumerClient : IDisposable |
| | | 49 | | { |
| | | 50 | | /// <summary>Subscribes the consumer group member to the given topic.</summary> |
| | | 51 | | void Subscribe(string topic); |
| | | 52 | | |
| | | 53 | | /// <summary> |
| | | 54 | | /// Polls for one message, waiting at most <paramref name="maxWait"/>. Returns <c>null</c> when |
| | | 55 | | /// no message arrived in time so the caller can re-check cancellation and backpressure state. |
| | | 56 | | /// </summary> |
| | | 57 | | KafkaIncomingMessage? Consume(TimeSpan maxWait); |
| | | 58 | | |
| | | 59 | | /// <summary> |
| | | 60 | | /// Marks the message resolved by storing <c>offset + 1</c> for its partition. The stored |
| | | 61 | | /// offset is committed by the consumer's auto-committer on the configured interval, and on |
| | | 62 | | /// <see cref="Close"/>. |
| | | 63 | | /// </summary> |
| | | 64 | | void StoreOffset(string topic, int partition, long offset); |
| | | 65 | | |
| | | 66 | | /// <summary>Pauses fetching on all currently assigned partitions (backpressure).</summary> |
| | | 67 | | void PauseAssignment(); |
| | | 68 | | |
| | | 69 | | /// <summary>Resumes fetching on all currently assigned partitions.</summary> |
| | | 70 | | void ResumeAssignment(); |
| | | 71 | | |
| | | 72 | | /// <summary> |
| | | 73 | | /// Pauses fetching on one partition while a detached handler runs its message, so the |
| | | 74 | | /// partition's order holds with nothing buffered in-process. Throws when the partition is not |
| | | 75 | | /// currently assigned (revoked by a rebalance); callers treat that as informational. |
| | | 76 | | /// </summary> |
| | | 77 | | void PausePartition(string topic, int partition); |
| | | 78 | | |
| | | 79 | | /// <summary>Resumes fetching on one partition once its detached handler has settled. Throws when it is no longer as |
| | | 80 | | void ResumePartition(string topic, int partition); |
| | | 81 | | |
| | | 82 | | /// <summary>Leaves the group cleanly, committing stored offsets.</summary> |
| | | 83 | | void Close(); |
| | | 84 | | } |
| | | 85 | | |
| | | 86 | | /// <summary>Creates one consumer per hosted subscriber role.</summary> |
| | | 87 | | internal interface IKafkaConsumerClientFactory |
| | | 88 | | { |
| | | 89 | | IKafkaConsumerClient Create(KafkaSubscriberRole role); |
| | | 90 | | } |
| | | 91 | | |
| | | 92 | | /// <summary>Adapter seam over the Kafka admin client for startup topic provisioning.</summary> |
| | | 93 | | internal interface IKafkaAdminClient |
| | | 94 | | { |
| | | 95 | | /// <summary>Creates any missing topics; existing topics are left untouched.</summary> |
| | | 96 | | Task EnsureTopicsAsync( |
| | | 97 | | IReadOnlyList<string> topics, |
| | | 98 | | int numPartitions, |
| | | 99 | | short replicationFactor, |
| | | 100 | | CancellationToken cancellationToken); |
| | | 101 | | } |
| | | 102 | | |
| | | 103 | | internal sealed class KafkaProducerClientAdapter : IKafkaProducerClient |
| | | 104 | | { |
| | | 105 | | private readonly KafkaAsyncResponseTransportOptions _options; |
| | 210 | 106 | | private readonly object _producerGate = new(); |
| | | 107 | | private IProducer<string?, byte[]>? _producer; |
| | | 108 | | private bool _disposed; |
| | | 109 | | |
| | | 110 | | /// <summary>Runs the KafkaProducerClientAdapter operation.</summary> |
| | 210 | 111 | | public KafkaProducerClientAdapter(KafkaAsyncResponseTransportOptions options) |
| | | 112 | | { |
| | 210 | 113 | | _options = options; |
| | 210 | 114 | | } |
| | | 115 | | |
| | | 116 | | // Built lazily so constructing the adapter (e.g. during DI validation) never dials the broker, |
| | | 117 | | // and assigned only on SUCCESS so a faulted build attempt is not cached: this adapter is a |
| | | 118 | | // process-lifetime singleton, and Lazy<T>'s ExecutionAndPublication mode would rethrow one |
| | | 119 | | // transient construction failure on every later publish until restart. Parity with the |
| | | 120 | | // RabbitMQ/Pub-Sub/SQS worker transports, whose comments pin the same rule. |
| | | 121 | | private IProducer<string?, byte[]> Producer |
| | | 122 | | { |
| | | 123 | | get |
| | | 124 | | { |
| | 426 | 125 | | if (Volatile.Read(ref _producer) is { } existing) |
| | 219 | 126 | | return existing; |
| | | 127 | | |
| | 207 | 128 | | lock (_producerGate) |
| | | 129 | | { |
| | | 130 | | // Checked under the same gate Dispose latches under: a build racing Dispose must |
| | | 131 | | // either be flushed and disposed by Dispose (build won the gate) or never happen |
| | | 132 | | // (Dispose won) — a producer built after the latch would leak its librdkafka |
| | | 133 | | // threads and silently drop its buffered jobs at shutdown. |
| | 207 | 134 | | ObjectDisposedException.ThrowIf(_disposed, this); |
| | 203 | 135 | | return _producer ??= CreateProducer(); |
| | | 136 | | } |
| | 199 | 137 | | } |
| | | 138 | | } |
| | | 139 | | |
| | | 140 | | /// <summary>Publishes the supplied message.</summary> |
| | | 141 | | public async Task<KafkaPublishResult> PublishAsync( |
| | | 142 | | string topic, |
| | | 143 | | string? key, |
| | | 144 | | byte[] payload, |
| | | 145 | | IReadOnlyList<KafkaTransportHeader> headers, |
| | | 146 | | CancellationToken cancellationToken) |
| | | 147 | | { |
| | 416 | 148 | | var message = new Message<string?, byte[]> |
| | 416 | 149 | | { |
| | 416 | 150 | | Key = key, |
| | 416 | 151 | | Value = payload |
| | 416 | 152 | | }; |
| | | 153 | | |
| | 416 | 154 | | if (headers.Count > 0) |
| | | 155 | | { |
| | 97 | 156 | | message.Headers = []; |
| | 408 | 157 | | foreach (var header in headers) |
| | 107 | 158 | | message.Headers.Add(header.Key, header.Value); |
| | | 159 | | } |
| | | 160 | | |
| | 416 | 161 | | var result = await Producer |
| | 416 | 162 | | .ProduceAsync(topic, message, cancellationToken) |
| | 416 | 163 | | .ConfigureAwait(false); |
| | | 164 | | |
| | 412 | 165 | | return new KafkaPublishResult(result.Topic, result.Partition.Value, result.Offset.Value); |
| | 412 | 166 | | } |
| | | 167 | | |
| | | 168 | | /// <summary>Releases resources held by this instance.</summary> |
| | | 169 | | public void Dispose() |
| | | 170 | | { |
| | | 171 | | // The whole read/flush/dispose runs under the build gate: reading the field outside it |
| | | 172 | | // raced a concurrent build — whichever the read missed was never flushed (dropping its |
| | | 173 | | // buffered jobs and leaking librdkafka threads), and the reverse order left later |
| | | 174 | | // publishes producing onto a disposed handle. |
| | 208 | 175 | | lock (_producerGate) |
| | | 176 | | { |
| | 208 | 177 | | if (_disposed) |
| | 2 | 178 | | return; |
| | | 179 | | |
| | 206 | 180 | | _disposed = true; |
| | 206 | 181 | | var producer = _producer; |
| | 206 | 182 | | _producer = null; |
| | 206 | 183 | | if (producer is null) |
| | 5 | 184 | | return; |
| | | 185 | | |
| | | 186 | | try |
| | | 187 | | { |
| | 201 | 188 | | producer.Flush(_options.OperationTimeout); |
| | 199 | 189 | | } |
| | 2 | 190 | | catch (KafkaException) |
| | | 191 | | { |
| | | 192 | | // Best-effort drain of in-flight messages on shutdown; Dispose below always runs. |
| | 2 | 193 | | } |
| | | 194 | | |
| | 201 | 195 | | producer.Dispose(); |
| | 201 | 196 | | } |
| | 208 | 197 | | } |
| | | 198 | | |
| | | 199 | | private IProducer<string?, byte[]> CreateProducer() |
| | | 200 | | { |
| | 203 | 201 | | var config = new ProducerConfig |
| | 203 | 202 | | { |
| | 203 | 203 | | BootstrapServers = KafkaTransportOptionsValidator.Required( |
| | 203 | 204 | | _options.BootstrapServers, |
| | 203 | 205 | | nameof(_options.BootstrapServers)), |
| | 203 | 206 | | ClientId = KafkaTransportClientDefaults.ResolveClientId(_options), |
| | 203 | 207 | | // Worker jobs must not be silently reordered or duplicated by producer-side retries. |
| | 203 | 208 | | Acks = Acks.All, |
| | 203 | 209 | | EnableIdempotence = true |
| | 203 | 210 | | }; |
| | | 211 | | |
| | 203 | 212 | | _options.ConfigureProducer?.Invoke(config); |
| | 199 | 213 | | return new ProducerBuilder<string?, byte[]>(config).Build(); |
| | | 214 | | } |
| | | 215 | | } |
| | | 216 | | |
| | | 217 | | internal sealed class KafkaConsumerClientAdapter(IConsumer<string?, byte[]> _consumer) : IKafkaConsumerClient |
| | | 218 | | { |
| | | 219 | | /// <summary>Subscribes the consumer group member to the given topic.</summary> |
| | | 220 | | public void Subscribe(string topic) |
| | | 221 | | => _consumer.Subscribe(topic); |
| | | 222 | | |
| | | 223 | | /// <summary>Polls for one message.</summary> |
| | | 224 | | public KafkaIncomingMessage? Consume(TimeSpan maxWait) |
| | | 225 | | { |
| | | 226 | | var result = _consumer.Consume(maxWait); |
| | | 227 | | if (result is null || result.IsPartitionEOF) |
| | | 228 | | return null; |
| | | 229 | | |
| | | 230 | | return new KafkaIncomingMessage( |
| | | 231 | | result.Topic, |
| | | 232 | | result.Partition.Value, |
| | | 233 | | result.Offset.Value, |
| | | 234 | | result.Message?.Value, |
| | | 235 | | ReadHeaders(result.Message?.Headers)); |
| | | 236 | | } |
| | | 237 | | |
| | | 238 | | /// <summary>Marks the message resolved by storing the next offset for its partition.</summary> |
| | | 239 | | public void StoreOffset(string topic, int partition, long offset) |
| | | 240 | | => _consumer.StoreOffset(new TopicPartitionOffset(topic, new Partition(partition), new Offset(offset + 1))); |
| | | 241 | | |
| | | 242 | | /// <summary>Pauses fetching on all currently assigned partitions.</summary> |
| | | 243 | | public void PauseAssignment() |
| | | 244 | | => _consumer.Pause(_consumer.Assignment); |
| | | 245 | | |
| | | 246 | | /// <summary>Resumes fetching on all currently assigned partitions.</summary> |
| | | 247 | | public void ResumeAssignment() |
| | | 248 | | => _consumer.Resume(_consumer.Assignment); |
| | | 249 | | |
| | | 250 | | /// <summary>Pauses fetching on one partition.</summary> |
| | | 251 | | public void PausePartition(string topic, int partition) |
| | | 252 | | => _consumer.Pause([new TopicPartition(topic, new Partition(partition))]); |
| | | 253 | | |
| | | 254 | | /// <summary>Resumes fetching on one partition.</summary> |
| | | 255 | | public void ResumePartition(string topic, int partition) |
| | | 256 | | => _consumer.Resume([new TopicPartition(topic, new Partition(partition))]); |
| | | 257 | | |
| | | 258 | | /// <summary>Leaves the group cleanly, committing stored offsets.</summary> |
| | | 259 | | public void Close() |
| | | 260 | | => _consumer.Close(); |
| | | 261 | | |
| | | 262 | | /// <summary>Releases resources held by this instance.</summary> |
| | | 263 | | public void Dispose() |
| | | 264 | | => _consumer.Dispose(); |
| | | 265 | | |
| | | 266 | | private static KafkaTransportHeader[] ReadHeaders(Headers? headers) |
| | | 267 | | { |
| | | 268 | | if (headers is null || headers.Count == 0) |
| | | 269 | | return []; |
| | | 270 | | |
| | | 271 | | var result = new KafkaTransportHeader[headers.Count]; |
| | | 272 | | for (var i = 0; i < headers.Count; i++) |
| | | 273 | | result[i] = new KafkaTransportHeader(headers[i].Key, headers[i].GetValueBytes()); |
| | | 274 | | |
| | | 275 | | return result; |
| | | 276 | | } |
| | | 277 | | } |
| | | 278 | | |
| | | 279 | | internal sealed class KafkaConsumerClientFactory(KafkaAsyncResponseTransportOptions _options) : IKafkaConsumerClientFact |
| | | 280 | | { |
| | | 281 | | /// <summary>Creates one consumer for the given subscriber role.</summary> |
| | | 282 | | public IKafkaConsumerClient Create(KafkaSubscriberRole role) |
| | | 283 | | { |
| | | 284 | | var subscriberOptions = role is KafkaSubscriberRole.Worker |
| | | 285 | | ? _options.WorkerSubscriber |
| | | 286 | | : _options.ResponseSubscriber; |
| | | 287 | | var config = new ConsumerConfig |
| | | 288 | | { |
| | | 289 | | BootstrapServers = KafkaTransportOptionsValidator.Required( |
| | | 290 | | _options.BootstrapServers, |
| | | 291 | | nameof(_options.BootstrapServers)), |
| | | 292 | | GroupId = role is KafkaSubscriberRole.Worker |
| | | 293 | | ? _options.WorkerConsumerGroup |
| | | 294 | | : _options.ResponseConsumerGroup, |
| | | 295 | | ClientId = $"{KafkaTransportClientDefaults.ResolveClientId(_options)}-{role.ToString().ToLowerInvariant()}", |
| | | 296 | | // Manual offset management: offsets are stored per resolved message (StoreOffset) and |
| | | 297 | | // flushed by the auto-committer, so a crash redelivers at-least-once instead of losing work. |
| | | 298 | | EnableAutoCommit = true, |
| | | 299 | | EnableAutoOffsetStore = false, |
| | | 300 | | AutoCommitIntervalMs = (int)Math.Max(1, _options.OffsetCommitInterval.TotalMilliseconds), |
| | | 301 | | // The poll thread's longest gap (one inline handler wait of DetachHandlerAfter plus |
| | | 302 | | // one poll) is validated against this deadline at startup, so it is set explicitly |
| | | 303 | | // instead of trusting the librdkafka default to line up. |
| | | 304 | | MaxPollIntervalMs = (int)Math.Max(1, subscriberOptions.MaxPollInterval.TotalMilliseconds), |
| | | 305 | | // Start new consumer groups at the beginning of the topic so messages published before |
| | | 306 | | // the first subscriber starts are not skipped (mirrors the other transports). |
| | | 307 | | AutoOffsetReset = AutoOffsetReset.Earliest, |
| | | 308 | | EnablePartitionEof = false |
| | | 309 | | }; |
| | | 310 | | |
| | | 311 | | _options.ConfigureConsumer?.Invoke(config); |
| | | 312 | | return new KafkaConsumerClientAdapter(new ConsumerBuilder<string?, byte[]>(config).Build()); |
| | | 313 | | } |
| | | 314 | | } |
| | | 315 | | |
| | | 316 | | [System.Diagnostics.CodeAnalysis.ExcludeFromCodeCoverage] |
| | | 317 | | internal sealed class KafkaAdminClientAdapter(KafkaAsyncResponseTransportOptions _options) : IKafkaAdminClient |
| | | 318 | | { |
| | | 319 | | /// <summary>Creates any missing topics; existing topics are left untouched.</summary> |
| | | 320 | | public async Task EnsureTopicsAsync( |
| | | 321 | | IReadOnlyList<string> topics, |
| | | 322 | | int numPartitions, |
| | | 323 | | short replicationFactor, |
| | | 324 | | CancellationToken cancellationToken) |
| | | 325 | | { |
| | | 326 | | var config = new AdminClientConfig |
| | | 327 | | { |
| | | 328 | | BootstrapServers = KafkaTransportOptionsValidator.Required( |
| | | 329 | | _options.BootstrapServers, |
| | | 330 | | nameof(_options.BootstrapServers)), |
| | | 331 | | ClientId = $"{KafkaTransportClientDefaults.ResolveClientId(_options)}-admin" |
| | | 332 | | }; |
| | | 333 | | |
| | | 334 | | _options.ConfigureAdminClient?.Invoke(config); |
| | | 335 | | using var adminClient = new AdminClientBuilder(config).Build(); |
| | | 336 | | |
| | | 337 | | var specifications = topics |
| | | 338 | | .Select(topic => new TopicSpecification |
| | | 339 | | { |
| | | 340 | | Name = topic, |
| | | 341 | | NumPartitions = numPartitions, |
| | | 342 | | ReplicationFactor = replicationFactor |
| | | 343 | | }) |
| | | 344 | | .ToArray(); |
| | | 345 | | |
| | | 346 | | try |
| | | 347 | | { |
| | | 348 | | await adminClient.CreateTopicsAsync( |
| | | 349 | | specifications, |
| | | 350 | | new CreateTopicsOptions { RequestTimeout = _options.OperationTimeout }) |
| | | 351 | | .WaitAsync(cancellationToken) |
| | | 352 | | .ConfigureAwait(false); |
| | | 353 | | } |
| | | 354 | | catch (CreateTopicsException ex) when (ex.Results.All(result => |
| | | 355 | | result.Error.Code is ErrorCode.NoError or ErrorCode.TopicAlreadyExists)) |
| | | 356 | | { |
| | | 357 | | // The topics already exist; this is the expected path after the first app instance. |
| | | 358 | | } |
| | | 359 | | } |
| | | 360 | | } |
| | | 361 | | |
| | | 362 | | internal static class KafkaTransportClientDefaults |
| | | 363 | | { |
| | | 364 | | private static readonly string GeneratedClientId = $"asyncresponse-{Environment.MachineName}-{Environment.ProcessId} |
| | | 365 | | |
| | | 366 | | /// <summary>Runs the ResolveClientId operation.</summary> |
| | | 367 | | public static string ResolveClientId(KafkaAsyncResponseTransportOptions options) |
| | | 368 | | => !string.IsNullOrWhiteSpace(options.ClientId) |
| | | 369 | | ? options.ClientId! |
| | | 370 | | : GeneratedClientId; |
| | | 371 | | } |
| | | 372 | | |
| | | 373 | | internal static class KafkaTransportRetry |
| | | 374 | | { |
| | | 375 | | /// <summary>Runs this background operation until cancellation is requested.</summary> |
| | | 376 | | public static Task<T> ExecuteAsync<T>( |
| | | 377 | | Func<CancellationToken, Task<T>> action, |
| | | 378 | | int maxAttempts, |
| | | 379 | | TimeSpan baseDelay, |
| | | 380 | | TimeSpan maxDelay, |
| | | 381 | | CancellationToken cancellationToken) |
| | | 382 | | => AsyncResponseRetry.ExecuteAsync(action, IsTransient, maxAttempts, baseDelay, maxDelay, cancellationToken); |
| | | 383 | | |
| | | 384 | | /// <summary>Runs the IsTransient operation.</summary> |
| | | 385 | | public static bool IsTransient(Exception exception) |
| | | 386 | | => exception is TimeoutException |
| | | 387 | | || (exception is KafkaException kafkaException && !kafkaException.Error.IsFatal); |
| | | 388 | | } |