kafka发送者线程一

kafka生产者线程负责生产消息,而将消息发送给broker是有一个专门的发送者线程来处理的,也称之为IO Thread,实现了消息的生产与发送解耦,提高吞吐量。
IO Thread是随着生产者的产生而启动的,在启动发送者线程之前会先初始化一个消息累加器

this.accumulator = new RecordAccumulator(logContext, config.getInt(ProducerConfig.BATCH_SIZE_CONFIG), this.compressionType, lingerMs(config), retryBackoffMs, deliveryTimeoutMs, metrics, PRODUCER_METRIC_GROUP_NAME, time, apiVersions, transactionManager, new BufferPool(this.totalMemorySize, config.getInt(ProducerConfig.BATCH_SIZE_CONFIG), metrics, time, PRODUCER_METRIC_GROUP_NAME)); 这里有几个参数需要注意下: kafka会将发往同一个分区的消息累积到同一个批次中,就会涉及到两个维度,空间和时间,不然会陷入无限等待批次消息的累积 batchSize,该参数就是指的空间,也就是每个批次累积多少消息,或者每个批次的分配的内存大小,太大会造成内存浪费, 太小的话,当消息一多时,会有多个批次需要发送,降低了吞吐量 lingerMs,该参数就是指的时间,就是每个批次延迟时间的上限,但是假如批次信息达到batchSize的值后,lingerMs就会失效 当批次的大小还未达到batchSize时,发送消息会延迟,即使当前kafka没有负载压力的情况下 deliveryTimeoutMs,调send方法到return的时间限制,包含消息的延迟发送时间、收到broker响应的时间、发送失败重试的时间三者总和的时间限制。 该值应该大于或等于request.timeout.ms+linger.ms的时间

当对生产者的相关属性值和累加器都初始化后,kafka会自动启动一个io thread
this.sender = newSender(logContext, kafkaClient, this.metadata); String ioThreadName = NETWORK_THREAD_PREFIX + " | " + clientId; this.ioThread = new KafkaThread(ioThreadName, this.sender, true); this.ioThread.start();

其中,KafkaClient是实际负责发送消息的客户端,底层基于nio实现的,与broker进行网络通信;KafkaThread继承了Thread;设置了发送者线程的名称、运行的任务runnable、以及线程模式为daemon,这里为true,因此发送者线程的主要逻辑在于runnable的run方法,也就是Sender的run
public void run() { log.debug("Starting Kafka producer I/O thread."); // main loop, runs until close is called while (running) { try { runOnce(); } catch (Exception e) { log.error("Uncaught error in kafka producer I/O thread: ", e); } }log.debug("Beginning shutdown of Kafka producer I/O thread, sending remaining records."); 省略.... }

从run方法可看出,就是不断执行runOnce方法,因此来看下runOnce方法的逻辑
void runOnce() { if (transactionManager != null) { //忽略.... //处理事务相关的 }long currentTimeMs = time.milliseconds(); //建立与broker的连接,准备待发送消息的数据 long pollTimeout = sendProducerData(currentTimeMs); //处理连接上发生的各种IO事件,包含获取来自broker的数据,发送实际的消息对象 client.poll(pollTimeout, currentTimeMs); }

在sendProducerData的方法中,看下以下几个关键的逻辑
Cluster cluster = metadata.fetch(); // 通过累加器获取到准备发送数据的分区 RecordAccumulator.ReadyCheckResult result = this.accumulator.ready(cluster, now); //和这些分区的节点建立连接,底层是通过nio的方式建立连接的 Iterator iter = result.readyNodes.iterator(); long notReadyTimeout = Long.MAX_VALUE; while (iter.hasNext()) { Node node = iter.next(); if (!this.client.ready(node, now)) { iter.remove(); notReadyTimeout = Math.min(notReadyTimeout, this.client.pollDelayMs(node, now)); } } //获取待发送的批次信息 Map batches = this.accumulator.drain(cluster, result.readyNodes, this.maxRequestSize, now); //省略...//将批次信息转化为具体的网络请求信息,看如下重载的方法 sendProduceRequests(batches, now); return pollTimeout; //简单的说,这个方法就是将与节点连接的channel的监听事件设置为OP_WRITE表示可写的, //然后构建相应的请求消息体,放置到ByteBuffer中 private void sendProduceRequest(long now, int destination, short acks, int timeout, List【kafka发送者线程一】 batches) { if (batches.isEmpty()) return; Map produceRecordsByPartition = new HashMap<>(batches.size()); final Map recordsByPartition = new HashMap<>(batches.size()); // ProduceRequest.Builder requestBuilder = ProduceRequest.Builder.forMagic(minUsedMagic, acks, timeout, produceRecordsByPartition, transactionalId); RequestCompletionHandler callback = new RequestCompletionHandler() { public void onComplete(ClientResponse response) { handleProduceResponse(response, recordsByPartition, time.milliseconds()); } }; String nodeId = Integer.toString(destination); ClientRequest clientRequest = client.newClientRequest(nodeId, requestBuilder, now, acks != 0, requestTimeoutMs, callback); client.send(clientRequest, now); }

sendProducerData方法就是建立与broker的连接,注册到selector,设置监听事件为op_write,准备好相应的请求数据(待发送的消息)。需要注意的是,此时,消息还没有真正的发送出去。真正发送的消息在runOnce方法内部调用client.poll方法
public List poll(long timeout, long now) { ensureActive(); //省略..... long metadataTimeout = metadataUpdater.maybeUpdate(now); try { //这里底层就是调用selector的select方法,获取到可以处理IO事件的channel,根据selectionKey的类型进行相对应的处理, //分别为isConnectable、isReadable、isWritable this.selector.poll(Utils.min(timeout, metadataTimeout, defaultRequestTimeoutMs)); } catch (IOException e) { log.error("Unexpected error during I/O", e); } // process completed actions long updatedNow = this.time.milliseconds(); List responses = new ArrayList<>(); handleCompletedSends(responses, updatedNow); handleCompletedReceives(responses, updatedNow); handleDisconnections(responses, updatedNow); handleConnections(); handleInitiateApiVersionRequests(updatedNow); handleTimedOutRequests(responses, updatedNow); completeResponses(responses); return responses; }

下面的两个方法为与broker节点建立连接的底层nio实现
private void initiateConnect(Node node, long now) { String nodeConnectionId = node.idString(); try { connectionStates.connecting(nodeConnectionId, now, node.host(), clientDnsLookup); InetAddress address = connectionStates.currentAddress(nodeConnectionId); log.debug("Initiating connection to node {} using address {}", node, address); selector.connect(nodeConnectionId, new InetSocketAddress(address, node.port()), this.socketSendBuffer, this.socketReceiveBuffer); } catch (IOException e) { //省略 }public void connect(String id, InetSocketAddress address, int sendBufferSize, int receiveBufferSize) throws IOException { ensureNotRegistered(id); SocketChannel socketChannel = SocketChannel.open(); SelectionKey key = null; try { //配置channel为非阻塞模式,并设置channel的发送和接收缓冲区大小 configureSocketChannel(socketChannel, sendBufferSize, receiveBufferSize); //建立连接 boolean connected = doConnect(socketChannel, address); //将channel注册到selector上 key = registerChannel(id, socketChannel, SelectionKey.OP_CONNECT); if (connected) { immediatelyConnectedKeys.add(key); //若连接成功了,改变channel的监听事件 key.interestOps(0); } } catch (IOException | RuntimeException e) { if (key != null) immediatelyConnectedKeys.remove(key); channels.remove(id); socketChannel.close(); throw e; } }

总结,kafka的发送者线程底层使用nio来与broker建立连接与数据通信,因此涉及到如何构造发送消息的ByteBuffer对象,处理来自broker的响应数据等,但是本文只是介绍个大体的方向,并没有对细节进行详情的说明

    推荐阅读