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<Node> 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<Integer, List<ProducerBatch>> 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<ProducerBatch> batches) {
if (batches.isEmpty())
return;
Map<TopicPartition, MemoryRecords> produceRecordsByPartition = new HashMap<>(batches.size());
final Map<TopicPartition, ProducerBatch> 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<ClientResponse> 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<ClientResponse> 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