.NET Core 下使用 Kafka

安裝

CentOS 安裝 kafka

• Kafka : http://kafka.apache.org/downloads
• ZooLeeper : https://zookeeper.apache.org/releases.html

下載並解壓

# 下載，並解壓
$ wget https://archive.apache.org/dist/kafka/2.1.1/kafka_2.12-2.1.1.tgz
$ tar -zxvf  kafka_2.12-2.1.1.tgz
$ mv kafka_2.12-2.1.1.tgz /data/kafka

# 下載 zookeeper，解壓
$ wget https://mirror.bit.edu.cn/apache/zookeeper/zookeeper-3.5.8/apache-zookeeper-3.5.8-bin.tar.gz
$ tar -zxvf apache-zookeeper-3.5.8-bin.tar.gz
$ mv apache-zookeeper-3.5.8-bin /data/zookeeper

啓動 ZooKeeper

# 複製配置模版
$ cd /data/kafka/conf
$ cp zoo_sample.cfg zoo.cfg

# 看看配置需不需要改
$ vim zoo.cfg

# 命令
$ ./bin/zkServer.sh start    # 啓動
$ ./bin/zkServer.sh status   # 狀態
$ ./bin/zkServer.sh stop     # 停止
$ ./bin/zkServer.sh restart  # 重啓

# 使用客戶端測試
$ ./bin/zkCli.sh -server localhost:2181
$ quit

啓動 Kafka

# 備份配置
$ cd /data/kafka
$ cp config/server.properties config/server.properties_copy

# 修改配置
$ vim /data/kafka/config/server.properties

# 集羣配置下，每個 broker 的 id 是必須不同的
# broker.id=0

# 監聽地址設置（內網）
# listeners=PLAINTEXT://ip:9092

# 對外提供服務的IP、端口
# advertised.listeners=PLAINTEXT://106.75.84.97:9092

# 修改每個topic的默認分區參數num.partitions，默認是1，具體合適的取值需要根據服務器配置進程確定，UCloud.ukafka = 3
# num.partitions=3

# zookeeper 配置
# zookeeper.connect=localhost:2181

# 通過配置啓動 kafka
$  ./bin/kafka-server-start.sh  config/server.properties&

# 狀態查看
$ ps -ef|grep kafka
$ jps

docker下安裝Kafka

docker pull wurstmeister/zookeeper
docker run -d --name zookeeper -p 2181:2181 wurstmeister/zookeeper

docker pull wurstmeister/kafka
docker run -d --name kafka --publish 9092:9092 --link zookeeper --env KAFKA_ZOOKEEPER_CONNECT=zookeeper:2181 --env KAFKA_ADVERTISED_HOST_NAME=192.168.1.111 --env KAFKA_ADVERTISED_PORT=9092 wurstmeister/kafka

介紹

• Broker：消息中間件處理節點，一個Kafka節點就是一個broker，多個broker可以組成一個Kafka集羣。
• Topic：一類消息，例如page view日誌、click日誌等都可以以topic的形式存在，Kafka集羣能夠同時負責多個topic的分發。
• Partition：topic物理上的分組，一個topic可以分爲多個partition，每個partition是一個有序的隊列。
• Segment：partition物理上由多個segment組成，下面2.2和2.3有詳細說明。
• offset：每個partition都由一系列有序的、不可變的消息組成，這些消息被連續的追加到partition中。partition中的每個消息都有一個連續的序列號叫做offset,用於partition唯一標識一條消息。

kafka partition 和 consumer 數目關係

• 如果consumer比partition多是浪費，因爲kafka的設計是在一個partition上是不允許併發的，所以consumer數不要大於partition數 。
• 如果consumer比partition少，一個consumer會對應於多個partitions，這裏主要合理分配consumer數和partition數，否則會導致partition裏面的數據被取的不均勻 。最好partiton數目是consumer數目的整數倍，所以partition數目很重要，比如取24，就很容易設定consumer數目 。
• 如果consumer從多個partition讀到數據，不保證數據間的順序性，kafka只保證在一個partition上數據是有序的，但多個partition，根據你讀的順序會有不同
• 增減consumer，broker，partition會導致rebalance，所以rebalance後consumer對應的partition會發生變化

快速開始

在 .NET Core 項目中安裝組件

Install-Package Confluent.Kafka

開源地址：https://github.com/confluentinc/confluent-kafka-dotnet

添加IKafkaService服務接口

public interface IKafkaService
{
    /// <summary>
    /// 發送消息至指定主題
    /// </summary>
    /// <typeparam name="TMessage"></typeparam>
    /// <param name="topicName"></param>
    /// <param name="message"></param>
    /// <returns></returns>
    Task PublishAsync<TMessage>(string topicName, TMessage message) where TMessage : class;

    /// <summary>
    /// 從指定主題訂閱消息
    /// </summary>
    /// <typeparam name="TMessage"></typeparam>
    /// <param name="topics"></param>
    /// <param name="messageFunc"></param>
    /// <param name="cancellationToken"></param>
    /// <returns></returns>
    Task SubscribeAsync<TMessage>(IEnumerable<string> topics, Action<TMessage> messageFunc, CancellationToken cancellationToken) where TMessage : class;
}

實現IKafkaService

public class KafkaService : IKafkaService
{
    public async Task PublishAsync<TMessage>(string topicName, TMessage message) where TMessage : class
    {
        var config = new ProducerConfig
        {
            BootstrapServers = "127.0.0.1:9092"
        };
        using var producer = new ProducerBuilder<string, string>(config).Build();
        await producer.ProduceAsync(topicName, new Message<string, string>
        {
            Key = Guid.NewGuid().ToString(),
            Value = message.SerializeToJson()
        });
    }

    public async Task SubscribeAsync<TMessage>(IEnumerable<string> topics, Action<TMessage> messageFunc, CancellationToken cancellationToken) where TMessage : class
    {
        var config = new ConsumerConfig
        {
            BootstrapServers = "127.0.0.1:9092",
            GroupId = "crow-consumer",
            EnableAutoCommit = false,
            StatisticsIntervalMs = 5000,
            SessionTimeoutMs = 6000,
            AutoOffsetReset = AutoOffsetReset.Earliest,
            EnablePartitionEof = true
        };
        //const int commitPeriod = 5;
        using var consumer = new ConsumerBuilder<Ignore, string>(config)
                             .SetErrorHandler((_, e) =>
                             {
                                 Console.WriteLine($"Error: {e.Reason}");
                             })
                             .SetStatisticsHandler((_, json) =>
                             {
                                 Console.WriteLine($" - {DateTime.Now:yyyy-MM-dd HH:mm:ss} > 消息監聽中..");
                             })
                             .SetPartitionsAssignedHandler((c, partitions) =>
                             {
                                 string partitionsStr = string.Join(", ", partitions);
                                 Console.WriteLine($" - 分配的 kafka 分區: {partitionsStr}");
                             })
                             .SetPartitionsRevokedHandler((c, partitions) =>
                             {
                                 string partitionsStr = string.Join(", ", partitions);
                                 Console.WriteLine($" - 回收了 kafka 的分區: {partitionsStr}");
                             })
                             .Build();
        consumer.Subscribe(topics);
        try
        {
            while (true)
            {
                try
                {
                    var consumeResult = consumer.Consume(cancellationToken);
                    Console.WriteLine($"Consumed message '{consumeResult.Message?.Value}' at: '{consumeResult?.TopicPartitionOffset}'.");
                    if (consumeResult.IsPartitionEOF)
                    {
                        Console.WriteLine($" - {DateTime.Now:yyyy-MM-dd HH:mm:ss} 已經到底了：{consumeResult.Topic}, partition {consumeResult.Partition}, offset {consumeResult.Offset}.");
                        continue;
                    }
                    TMessage messageResult = null;
                    try
                    {
                        messageResult = JsonConvert.DeserializeObject<TMessage>(consumeResult.Message.Value);
                    }
                    catch (Exception ex)
                    {
                        var errorMessage = $" - {DateTime.Now:yyyy-MM-dd HH:mm:ss}【Exception 消息反序列化失敗，Value：{consumeResult.Message.Value}】 ：{ex.StackTrace?.ToString()}";
                        Console.WriteLine(errorMessage);
                        messageResult = null;
                    }
                    if (messageResult != null/* && consumeResult.Offset % commitPeriod == 0*/)
                    {
                        messageFunc(messageResult);
                        try
                        {
                            consumer.Commit(consumeResult);
                        }
                        catch (KafkaException e)
                        {
                            Console.WriteLine(e.Message);
                        }
                    }
                }
                catch (ConsumeException e)
                {
                    Console.WriteLine($"Consume error: {e.Error.Reason}");
                }
            }
        }
        catch (OperationCanceledException)
        {
            Console.WriteLine("Closing consumer.");
            consumer.Close();
        }
        await Task.CompletedTask;
    }
}

注入IKafkaService，在需要使用的地方直接調用即可。

public class MessageService : IMessageService, ITransientDependency
{
    private readonly IKafkaService _kafkaService;
    public MessageService(IKafkaService kafkaService)
    {
        _kafkaService = kafkaService;
    }

    public async Task RequestTraceAdded(XxxEventData eventData)
    {
        await _kafkaService.PublishAsync(eventData.TopicName, eventData);
    }
}

以上相當於一個生產者，當我們消息隊列發出後，還需一個消費者進行消費，所以可以使用一個控制檯項目接收消息來處理業務。

var cts = new CancellationTokenSource();
Console.CancelKeyPress += (_, e) =>
{
    e.Cancel = true;
    cts.Cancel();
};

await kafkaService.SubscribeAsync<XxxEventData>(topics, async (eventData) =>
{
    // Your logic

    Console.WriteLine($" - {eventData.EventTime:yyyy-MM-dd HH:mm:ss} 【{eventData.TopicName}】- > 已處理");
}, cts.Token);

在IKafkaService中已經寫了訂閱消息的接口，這裏也是注入後直接使用即可。

生產者消費者示例

生產者

static async Task Main(string[] args)
{
    if (args.Length != 2)
    {
        Console.WriteLine("Usage: .. brokerList topicName");
        // 127.0.0.1:9092 helloTopic
        return;
    }

    var brokerList = args.First();
    var topicName = args.Last();

    var config = new ProducerConfig { BootstrapServers = brokerList };

    using var producer = new ProducerBuilder<string, string>(config).Build();

    Console.WriteLine("\n-----------------------------------------------------------------------");
    Console.WriteLine($"Producer {producer.Name} producing on topic {topicName}.");
    Console.WriteLine("-----------------------------------------------------------------------");
    Console.WriteLine("To create a kafka message with UTF-8 encoded key and value:");
    Console.WriteLine("> key value<Enter>");
    Console.WriteLine("To create a kafka message with a null key and UTF-8 encoded value:");
    Console.WriteLine("> value<enter>");
    Console.WriteLine("Ctrl-C to quit.\n");

    var cancelled = false;

    Console.CancelKeyPress += (_, e) =>
    {
        e.Cancel = true;
        cancelled = true;
    };

    while (!cancelled)
    {
        Console.Write("> ");

        var text = string.Empty;

        try
        {
            text = Console.ReadLine();
        }
        catch (IOException)
        {
            break;
        }

        if (string.IsNullOrWhiteSpace(text))
        {
            break;
        }

        var key = string.Empty;
        var val = text;

        var index = text.IndexOf(" ");
        if (index != -1)
        {
            key = text.Substring(0, index);
            val = text.Substring(index + 1);
        }

        try
        {
            var deliveryResult = await producer.ProduceAsync(topicName, new Message<string, string>
            {
                Key = key,
                Value = val
            });

            Console.WriteLine($"delivered to: {deliveryResult.TopicPartitionOffset}");
        }
        catch (ProduceException<string, string> e)
        {
            Console.WriteLine($"failed to deliver message: {e.Message} [{e.Error.Code}]");
        }
    }
}

消費者

static void Main(string[] args)
{
    if (args.Length != 2)
    {
        Console.WriteLine("Usage: .. brokerList topicName");
        // 127.0.0.1:9092 helloTopic
        return;
    }

    var brokerList = args.First();
    var topicName = args.Last();

    Console.WriteLine($"Started consumer, Ctrl-C to stop consuming");

    var cts = new CancellationTokenSource();
    Console.CancelKeyPress += (_, e) =>
    {
        e.Cancel = true;
        cts.Cancel();
    };

    var config = new ConsumerConfig
    {
        BootstrapServers = brokerList,
        GroupId = "consumer",
        EnableAutoCommit = false,
        StatisticsIntervalMs = 5000,
        SessionTimeoutMs = 6000,
        AutoOffsetReset = AutoOffsetReset.Earliest,
        EnablePartitionEof = true
    };

    const int commitPeriod = 5;

    using var consumer = new ConsumerBuilder<Ignore, string>(config)
                         .SetErrorHandler((_, e) =>
                         {
                             Console.WriteLine($"Error: {e.Reason}");
                         })
                         .SetStatisticsHandler((_, json) =>
                         {
                             Console.WriteLine($" - {DateTime.Now:yyyy-MM-dd HH:mm:ss} > monitoring..");
                             //Console.WriteLine($"Statistics: {json}");
                         })
                         .SetPartitionsAssignedHandler((c, partitions) =>
                         {
                             Console.WriteLine($"Assigned partitions: [{string.Join(", ", partitions)}]");
                         })
                         .SetPartitionsRevokedHandler((c, partitions) =>
                         {
                             Console.WriteLine($"Revoking assignment: [{string.Join(", ", partitions)}]");
                         })
                         .Build();
    consumer.Subscribe(topicName);

    try
    {
        while (true)
        {
            try
            {
                var consumeResult = consumer.Consume(cts.Token);

                if (consumeResult.IsPartitionEOF)
                {
                    Console.WriteLine($"Reached end of topic {consumeResult.Topic}, partition {consumeResult.Partition}, offset {consumeResult.Offset}.");

                    continue;
                }

                Console.WriteLine($"Received message at {consumeResult.TopicPartitionOffset}: {consumeResult.Message.Value}");

                if (consumeResult.Offset % commitPeriod == 0)
                {
                    try
                    {
                        consumer.Commit(consumeResult);
                    }
                    catch (KafkaException e)
                    {
                        Console.WriteLine($"Commit error: {e.Error.Reason}");
                    }
                }
            }
            catch (ConsumeException e)
            {
                Console.WriteLine($"Consume error: {e.Error.Reason}");
            }
        }
    }
    catch (OperationCanceledException)
    {
        Console.WriteLine("Closing consumer.");
        consumer.Close();
    }
}