kafka原理和实践（四）spring-kafka消费者源码

系列目录

kafka原理和实践（一）原理：10分钟入门

kafka原理和实践（二）spring-kafka简单实践

kafka原理和实践（三）spring-kafka生产者源码

kafka原理和实践（四）spring-kafka消费者源码

kafka原理和实践（五）spring-kafka配置详解

kafka原理和实践（六）总结升华

==============正文分割线=====================

一、kafkaConsumer消费者模型

如上图所示，spring-kafka消费者模型主要流程：

1.容器启动，轮询执行消费。

2.kafkaConsumer拉取消息流程：

1）Fetcher请求获取器获取请求并存储在unset中

2）ConsumerNetworkClient网络客户端执行poll(),调用NetWlrikClient的send()方法从unset中获取ClientRequest请求转成RequestSend最终塞进Selector的KafkaChannel通道中，Seletcor.send()从kafka集群拉取待消费数据ConsumerRecords

3. 消费者监听器MessageListener.onMessage()执行用户自定义的实际消费业务逻辑。

一、kafkaConsumer构造

 @SuppressWarnings("unchecked")
     private KafkaConsumer(ConsumerConfig config,
                           Deserializer<K> keyDeserializer,
                           Deserializer<V> valueDeserializer) {
         try {
             log.debug("Starting the Kafka consumer");
             this.requestTimeoutMs = config.getInt(ConsumerConfig.REQUEST_TIMEOUT_MS_CONFIG);
             int sessionTimeOutMs = config.getInt(ConsumerConfig.SESSION_TIMEOUT_MS_CONFIG);
             int fetchMaxWaitMs = config.getInt(ConsumerConfig.FETCH_MAX_WAIT_MS_CONFIG);
             if (this.requestTimeoutMs <= sessionTimeOutMs || this.requestTimeoutMs <= fetchMaxWaitMs)
                 throw new ConfigException(ConsumerConfig.REQUEST_TIMEOUT_MS_CONFIG + " should be greater than " + ConsumerConfig.SESSION_TIMEOUT_MS_CONFIG + " and " + ConsumerConfig.FETCH_MAX_WAIT_MS_CONFIG);
             this.time = new SystemTime();
 
             String clientId = config.getString(ConsumerConfig.CLIENT_ID_CONFIG);
             if (clientId.length() <= 0)
                 clientId = "consumer-" + CONSUMER_CLIENT_ID_SEQUENCE.getAndIncrement();
             this.clientId = clientId;
             Map<String, String> metricsTags = new LinkedHashMap<>();
             metricsTags.put("client-id", clientId);
             MetricConfig metricConfig = new MetricConfig().samples(config.getInt(ConsumerConfig.METRICS_NUM_SAMPLES_CONFIG))
                     .timeWindow(config.getLong(ConsumerConfig.METRICS_SAMPLE_WINDOW_MS_CONFIG), TimeUnit.MILLISECONDS)
                     .tags(metricsTags);
             List<MetricsReporter> reporters = config.getConfiguredInstances(ConsumerConfig.METRIC_REPORTER_CLASSES_CONFIG,
                     MetricsReporter.class);
             reporters.add(new JmxReporter(JMX_PREFIX));
             this.metrics = new Metrics(metricConfig, reporters, time);
             this.retryBackoffMs = config.getLong(ConsumerConfig.RETRY_BACKOFF_MS_CONFIG);
 
             // load interceptors and make sure they get clientId
             Map<String, Object> userProvidedConfigs = config.originals();
             userProvidedConfigs.put(ConsumerConfig.CLIENT_ID_CONFIG, clientId);
             List<ConsumerInterceptor<K, V>> interceptorList = (List) (new ConsumerConfig(userProvidedConfigs)).getConfiguredInstances(ConsumerConfig.INTERCEPTOR_CLASSES_CONFIG,
                     ConsumerInterceptor.class);
             this.interceptors = interceptorList.isEmpty() ? null : new ConsumerInterceptors<>(interceptorList);
             if (keyDeserializer == null) {
                 this.keyDeserializer = config.getConfiguredInstance(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG,
                         Deserializer.class);
                 this.keyDeserializer.configure(config.originals(), true);
             } else {
                 config.ignore(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG);
                 this.keyDeserializer = keyDeserializer;
             }
             if (valueDeserializer == null) {
                 this.valueDeserializer = config.getConfiguredInstance(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG,
                         Deserializer.class);
                 this.valueDeserializer.configure(config.originals(), false);
             } else {
                 config.ignore(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG);
                 this.valueDeserializer = valueDeserializer;
             }
             ClusterResourceListeners clusterResourceListeners = configureClusterResourceListeners(keyDeserializer, valueDeserializer, reporters, interceptorList);
             this.metadata = new Metadata(retryBackoffMs, config.getLong(ConsumerConfig.METADATA_MAX_AGE_CONFIG), false, clusterResourceListeners);
             List<InetSocketAddress> addresses = ClientUtils.parseAndValidateAddresses(config.getList(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG));
             this.metadata.update(Cluster.bootstrap(addresses), 0);
             String metricGrpPrefix = "consumer";
             ChannelBuilder channelBuilder = ClientUtils.createChannelBuilder(config.values());
             NetworkClient netClient = new NetworkClient(
                     new Selector(config.getLong(ConsumerConfig.CONNECTIONS_MAX_IDLE_MS_CONFIG), metrics, time, metricGrpPrefix, channelBuilder),
                     this.metadata,
                     clientId,
                     100, // a fixed large enough value will suffice
                     config.getLong(ConsumerConfig.RECONNECT_BACKOFF_MS_CONFIG),
                     config.getInt(ConsumerConfig.SEND_BUFFER_CONFIG),
                     config.getInt(ConsumerConfig.RECEIVE_BUFFER_CONFIG),
                     config.getInt(ConsumerConfig.REQUEST_TIMEOUT_MS_CONFIG), time);
             this.client = new ConsumerNetworkClient(netClient, metadata, time, retryBackoffMs,
                     config.getInt(ConsumerConfig.REQUEST_TIMEOUT_MS_CONFIG));
             OffsetResetStrategy offsetResetStrategy = OffsetResetStrategy.valueOf(config.getString(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG).toUpperCase(Locale.ROOT));
             this.subscriptions = new SubscriptionState(offsetResetStrategy);
             List<PartitionAssignor> assignors = config.getConfiguredInstances(
                     ConsumerConfig.PARTITION_ASSIGNMENT_STRATEGY_CONFIG,
                     PartitionAssignor.class);
             this.coordinator = new ConsumerCoordinator(this.client,
                     config.getString(ConsumerConfig.GROUP_ID_CONFIG),
                     config.getInt(ConsumerConfig.MAX_POLL_INTERVAL_MS_CONFIG),
                     config.getInt(ConsumerConfig.SESSION_TIMEOUT_MS_CONFIG),
                     config.getInt(ConsumerConfig.HEARTBEAT_INTERVAL_MS_CONFIG),
                     assignors,
                     this.metadata,
                     this.subscriptions,
                     metrics,
                     metricGrpPrefix,
                     this.time,
                     retryBackoffMs,
                     new ConsumerCoordinator.DefaultOffsetCommitCallback(),
                     config.getBoolean(ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG),
                     config.getInt(ConsumerConfig.AUTO_COMMIT_INTERVAL_MS_CONFIG),
                     this.interceptors,
                     config.getBoolean(ConsumerConfig.EXCLUDE_INTERNAL_TOPICS_CONFIG));
             this.fetcher = new Fetcher<>(this.client,
                     config.getInt(ConsumerConfig.FETCH_MIN_BYTES_CONFIG),
                     config.getInt(ConsumerConfig.FETCH_MAX_BYTES_CONFIG),
                     config.getInt(ConsumerConfig.FETCH_MAX_WAIT_MS_CONFIG),
                     config.getInt(ConsumerConfig.MAX_PARTITION_FETCH_BYTES_CONFIG),
                     config.getInt(ConsumerConfig.MAX_POLL_RECORDS_CONFIG),
                     config.getBoolean(ConsumerConfig.CHECK_CRCS_CONFIG),
                     this.keyDeserializer,
                     this.valueDeserializer,
                     this.metadata,
                     this.subscriptions,
                     metrics,
                     metricGrpPrefix,
                     this.time,
                     this.retryBackoffMs);
 
             config.logUnused();
             AppInfoParser.registerAppInfo(JMX_PREFIX, clientId);
 
             log.debug("Kafka consumer created");
         } catch (Throwable t) {
             // call close methods if internal objects are already constructed
             // this is to prevent resource leak. see KAFKA-2121
             close(true);
             // now propagate the exception
             throw new KafkaException("Failed to construct kafka consumer", t);
         }
     }

从KafkaConsumer构造函数来看，核心组件有：

1.Metadata：封装了元数据的一些逻辑的类。元数据仅保留一个主题的子集，随着时间的推移可以添加。当我们请求一个主题的元数据时，我们没有任何元数据会触发元数据更新。如果对元数据启用了主题过期，那么在更新之后，在过期时间间隔内未使用的任何主题都将从元数据刷新集中删除。

2.ConsumerNetworkClient：高等级消费者访问网络层，为请求Future任务提供基本支持。这个类是线程安全的，但是不提供响应回调的同步。这保证在调用它们时不会持有锁。

3.SubscriptionState：订阅的TopicPartition的offset状态维护

4.ConsumerCoordinator：消费者的协调者，负责partitiion的分配，reblance

5.Fetcher：从brokers上按照配置获取消息。

二、消费者容器启动流程

kafka消费者有两种常见的实现方式：

1.xml配置文件

2.基于注解实现

其实，不管哪种方式，本质只是生成Spring Bean的方式不同而已。我们就以xml的实现方式来追踪源码。

基于xml的总体配置如下：

 1 <!-- 1.定义consumer的参数 -->
 2     <bean id="consumerProperties" class="java.util.HashMap">
 3         <constructor-arg>
 4             <map>
 5                 <entry key="bootstrap.servers" value="${bootstrap.servers}" />
 6                 <entry key="group.id" value="${group.id}" />
 7                 <entry key="enable.auto.commit" value="${enable.auto.commit}" />
 8                 <entry key="session.timeout.ms" value="${session.timeout.ms}" />
 9                 <entry key="key.deserializer"
10                     value="org.apache.kafka.common.serialization.StringDeserializer" />
11                 <entry key="value.deserializer"
12                     value="org.apache.kafka.common.serialization.StringDeserializer" />
13             </map>
14         </constructor-arg>
15     </bean>
16
17     <!-- 2.创建consumerFactory bean -->
18     <bean id="consumerFactory"
19         class="org.springframework.kafka.core.DefaultKafkaConsumerFactory" >
20         <constructor-arg>
21             <ref bean="consumerProperties" />
22         </constructor-arg>
23     </bean>
24
25     <!-- 3.定义消费实现类 -->
26     <bean id="kafkaConsumerService" class="xxx.service.impl.KafkaConsumerSerivceImpl" />
27
28     <!-- 4.消费者容器配置信息 -->
29     <bean id="containerProperties" class="org.springframework.kafka.listener.config.ContainerProperties">
30         <!-- topic -->
31         <constructor-arg name="topics">
32             <list>
33                 <value>${kafka.consumer.topic.credit.for.lease}</value>
34                 <value>${loan.application.feedback.topic}</value>
35                 <value>${templar.agreement.feedback.topic}</value>
36                 <value>${templar.aggrement.active.feedback.topic}</value>
37                 <value>${templar.aggrement.agreementRepaid.topic}</value>
38                 <value>${templar.aggrement.agreementWithhold.topic}</value>
39                 <value>${templar.aggrement.agreementRepayRemind.topic}</value>
40             </list>
41         </constructor-arg>
42         <property name="messageListener" ref="kafkaConsumerService" />
43     </bean>
44     <!-- 5.消费者并发消息监听容器，执行doStart()方法 -->
45     <bean id="messageListenerContainer" class="org.springframework.kafka.listener.ConcurrentMessageListenerContainer" init-method="doStart" >
46         <constructor-arg ref="consumerFactory" />
47         <constructor-arg ref="containerProperties" />
48         <property name="concurrency" value="${concurrency}" />
49     </bean>

分为5个步骤：

2.1.定义消费参数bean

consumerProperties ,就是个map<key,value>

2.2.创建consumerFactory bean

DefaultKafkaConsumerFactory 实现了ConsumerFactory接口，提供创建消费者和判断是否自动提交2个方法。通过consumerProperties作为参数构造。

 public interface ConsumerFactory<K, V> {
 
     Consumer<K, V> createConsumer();
 
     boolean isAutoCommit();
 
 }

2.3.定义消费实现类

自定义一个类实现MessageListener接口，接口设计如下：

实现onMessage方法，去消费接收到的消息。两种方案：

1）MessageListener 消费完消息后自动提交offset(enable.auto.commit=true时),可提高效率，存在消费失败但移动了偏移量的风险。

2）AcknowledgingMessageListener 消费完消息后手动提交offset(enable.auto.commit=false时)效率降低，无消费失败但移动偏移量的风险。

2.4.监听容器配置信息

ContainerProperties：包含了一个监听容器的运行时配置信息，主要定义了监听的主题、分区、初始化偏移量，还有消息监听器。

 public class ContainerProperties {
 
     private static final int DEFAULT_SHUTDOWN_TIMEOUT = 10000;
 
     private static final int DEFAULT_QUEUE_DEPTH = 1;
 
     private static final int DEFAULT_PAUSE_AFTER = 10000;
 
     /**
      * Topic names.监听的主题字符串数组
      */
     private final String[] topics;
 
     /**
      * Topic pattern.监听的主题模板
      */
     private final Pattern topicPattern;
 
     /**
      * Topics/partitions/initial offsets.
      */
     private final TopicPartitionInitialOffset[] topicPartitions;
 
     /**
      * 确认模式（自动确认属性为false时使用）
      * <ul>
      * <li>1.RECORD逐条确认: 每条消息被发送给监听者后确认</li>
      * <li>2.BATCH批量确认: 当批量消息记录被消费者接收到并传送给监听器时确认</li>
      * <li>3.TIME超时确认:当超过设置的超时时间毫秒数时确认(should be greater than
      * {@code #setPollTimeout(long) pollTimeout}.</li>
      * <li>4.COUNT计数确认: 当接收到指定数量之后确认</li>
      * <li>5.MANUAL手动确认：由监听器负责确认（AcknowledgingMessageListener）</ul>
      */
     private AbstractMessageListenerContainer.AckMode ackMode = AckMode.BATCH;
 
     /**
      * The number of outstanding record count after which offsets should be
      * committed when {@link AckMode#COUNT} or {@link AckMode#COUNT_TIME} is being
      * used.
      */
     private int ackCount;
 
     /**
      * The time (ms) after which outstanding offsets should be committed when
      * {@link AckMode#TIME} or {@link AckMode#COUNT_TIME} is being used. Should be
      * larger than
      */
     private long ackTime;
 
     /**
      * 消息监听器，必须是 MessageListener或者AcknowledgingMessageListener两者中的一个      *      */
     private Object messageListener;
 
     /**
      * The max time to block in the consumer waiting for records.
      */
     private volatile long pollTimeout = 1000;
 
     /**
      * 线程执行器：轮询消费者
      */
     private AsyncListenableTaskExecutor consumerTaskExecutor;
 
     /**
      * 线程执行器：调用监听器
      */
     private AsyncListenableTaskExecutor listenerTaskExecutor;
 
     /**
      * 错误回调，当监听器抛出异常时
      */
     private GenericErrorHandler<?> errorHandler;
 
     /**
      * When using Kafka group management and {@link #setPauseEnabled(boolean)} is
      * true, the delay after which the consumer should be paused. Default 10000.
      */
     private long pauseAfter = DEFAULT_PAUSE_AFTER;
 
     /**
      * When true, avoids rebalancing when this consumer is slow or throws a
      * qualifying exception - pauses the consumer. Default: true.
      * @see #pauseAfter
      */
     private boolean pauseEnabled = true;
 
     /**
      * Set the queue depth for handoffs from the consumer thread to the listener
      * thread. Default 1 (up to 2 in process).
      */
     private int queueDepth = DEFAULT_QUEUE_DEPTH;
 
     /**
      * 停止容器超时时间    */
     private long shutdownTimeout = DEFAULT_SHUTDOWN_TIMEOUT;
 
     /**
      * 用户定义的消费者再平衡监听器实现类     */
     private ConsumerRebalanceListener consumerRebalanceListener;
 
     /**
      * 提交回调，默认记录日志。      */
     private OffsetCommitCallback commitCallback;
 
     /**
      * Whether or not to call consumer.commitSync() or commitAsync() when the
      * container is responsible for commits. Default true. See
      * https://github.com/spring-projects/spring-kafka/issues/62 At the time of
      * writing, async commits are not entirely reliable.
      */
     private boolean syncCommits = true;
 
     private boolean ackOnError = true;
 
     private Long idleEventInterval;
 
     public ContainerProperties(String... topics) {
         Assert.notEmpty(topics, "An array of topicPartitions must be provided");
         this.topics = Arrays.asList(topics).toArray(new String[topics.length]);
         this.topicPattern = null;
         this.topicPartitions = null;
     }
 
     public ContainerProperties(Pattern topicPattern) {
         this.topics = null;
         this.topicPattern = topicPattern;
         this.topicPartitions = null;
     }
 
     public ContainerProperties(TopicPartitionInitialOffset... topicPartitions) {
         this.topics = null;
         this.topicPattern = null;
         Assert.notEmpty(topicPartitions, "An array of topicPartitions must be provided");
         this.topicPartitions = new LinkedHashSet<>(Arrays.asList(topicPartitions))
                 .toArray(new TopicPartitionInitialOffset[topicPartitions.length]);
     }
 ...省略各种set、get
 
 }

2.5.启动并发消息监听容器

核心类ConcurrentMessageListenerContainer,继承自抽象类AbstractMessageListenerContainer，类图如下：

看上图可知AbstractMessageListenerContainer有2个实现类分别对应单线程和多线程，建议采用多线程消费。下面分析一下主要ConcurrentMessageListenerContainer类，注意2个方法：

1.构造函数，入参：消费者工厂ConsumerFactory+容器配置ContainerProperties

2.doStart()：核心方法KafkaMessageListenerContainer的start()方法。源码如下：

 public class ConcurrentMessageListenerContainer<K, V> extends AbstractMessageListenerContainer<K, V> {
 
     private final ConsumerFactory<K, V> consumerFactory;
 
     private final List<KafkaMessageListenerContainer<K, V>> containers = new ArrayList<>();
 
     private int concurrency = 1;
 
     /**
      * Construct an instance with the supplied configuration properties.
      * The topic partitions are distributed evenly across the delegate
      * {@link KafkaMessageListenerContainer}s.
      * @param consumerFactory the consumer factory.
      * @param containerProperties the container properties.
      */
     public ConcurrentMessageListenerContainer(ConsumerFactory<K, V> consumerFactory,
             ContainerProperties containerProperties) {
         super(containerProperties);
         Assert.notNull(consumerFactory, "A ConsumerFactory must be provided");
         this.consumerFactory = consumerFactory;
     }
 
     public int getConcurrency() {
         return this.concurrency;
     }
 
     /**
      * The maximum number of concurrent {@link KafkaMessageListenerContainer}s running.
      * Messages from within the same partition will be processed sequentially.
      * @param concurrency the concurrency.
      */
     public void setConcurrency(int concurrency) {
         Assert.isTrue(concurrency > 0, "concurrency must be greater than 0");
         this.concurrency = concurrency;
     }
 
     /**
      * Return the list of {@link KafkaMessageListenerContainer}s created by
      * this container.
      * @return the list of {@link KafkaMessageListenerContainer}s created by
      * this container.
      */
     public List<KafkaMessageListenerContainer<K, V>> getContainers() {
         return Collections.unmodifiableList(this.containers);
     }
 
     /*
      * Under lifecycle lock.
      */
     @Override
     protected void doStart() {
         if (!isRunning()) {
             ContainerProperties containerProperties = getContainerProperties();
             TopicPartitionInitialOffset[] topicPartitions = containerProperties.getTopicPartitions();
             if (topicPartitions != null//校验并发数>分区数，报错。
                     && this.concurrency > topicPartitions.length) {
                 this.logger.warn("When specific partitions are provided, the concurrency must be less than or "
                         + "equal to the number of partitions; reduced from " + this.concurrency + " to "
                         + topicPartitions.length);
                 this.concurrency = topicPartitions.length;//并发数最大只能=分区数
             }
             setRunning(true);
             //遍历创建监听器容器
             for (int i = 0; i < this.concurrency; i++) {
                 KafkaMessageListenerContainer<K, V> container;
                 if (topicPartitions == null) {
                     container = new KafkaMessageListenerContainer<>(this.consumerFactory, containerProperties);
                 }
                 else {
                     container = new KafkaMessageListenerContainer<>(this.consumerFactory, containerProperties,
                             partitionSubset(containerProperties, i));
                 }
                 if (getBeanName() != null) {
                     container.setBeanName(getBeanName() + "-" + i);
                 }
                 if (getApplicationEventPublisher() != null) {
                     container.setApplicationEventPublisher(getApplicationEventPublisher());
                 }
                 container.setClientIdSuffix("-" + i);
                 container.start();//核心方法，启动容器
                 this.containers.add(container);
             }
         }
     } ...省略
 }

继续追踪，调用AbstractMessageListenerContainer的doStart(),值得注意的是start()和stop方法加了同一把锁，用于锁住生命周期。

 private final Object lifecycleMonitor = new Object();
 
 @Override
     public final void start() {
         synchronized (this.lifecycleMonitor) {
             Assert.isTrue(
                     this.containerProperties.getMessageListener() instanceof KafkaDataListener,
                     "A " + KafkaDataListener.class.getName() + " implementation must be provided");
             doStart();
         }
     }
 
     protected abstract void doStart();

最终调用的是KafkaMessageListenerContainer的doStart()

 @Override
     protected void doStart() {
         if (isRunning()) {
             return;
         }
         ContainerProperties containerProperties = getContainerProperties();
 
         if (!this.consumerFactory.isAutoCommit()) {
             AckMode ackMode = containerProperties.getAckMode();
             if (ackMode.equals(AckMode.COUNT) || ackMode.equals(AckMode.COUNT_TIME)) {
                 Assert.state(containerProperties.getAckCount() > 0, "'ackCount' must be > 0");
             }
             if ((ackMode.equals(AckMode.TIME) || ackMode.equals(AckMode.COUNT_TIME))
                     && containerProperties.getAckTime() == 0) {
                 containerProperties.setAckTime(5000);
             }
         }
 
         Object messageListener = containerProperties.getMessageListener();
         Assert.state(messageListener != null, "A MessageListener is required");
         if (messageListener instanceof GenericAcknowledgingMessageListener) {
             this.acknowledgingMessageListener = (GenericAcknowledgingMessageListener<?>) messageListener;
         }
         else if (messageListener instanceof GenericMessageListener) {
             this.listener = (GenericMessageListener<?>) messageListener;
         }
         else {
             throw new IllegalStateException("messageListener must be 'MessageListener' "
                     + "or 'AcknowledgingMessageListener', not " + messageListener.getClass().getName());
         }
         if (containerProperties.getConsumerTaskExecutor() == null) {
             SimpleAsyncTaskExecutor consumerExecutor = new SimpleAsyncTaskExecutor(
                     (getBeanName() == null ? "" : getBeanName()) + "-C-");
             containerProperties.setConsumerTaskExecutor(consumerExecutor);
         }
         if (containerProperties.getListenerTaskExecutor() == null) {
             SimpleAsyncTaskExecutor listenerExecutor = new SimpleAsyncTaskExecutor(
                     (getBeanName() == null ? "" : getBeanName()) + "-L-");
             containerProperties.setListenerTaskExecutor(listenerExecutor);
         }//1.构建 监听消费者
         this.listenerConsumer = new ListenerConsumer(this.listener, this.acknowledgingMessageListener);
         setRunning(true);
　　　　　　//2.异步提交 监听消费者任务，返回Future并赋值。
         this.listenerConsumerFuture = containerProperties
                 .getConsumerTaskExecutor()
                 .submitListenable(this.listenerConsumer);
     }

doStart主要包含2个操作：构建内部类ListenerConsumer和提交 监听消费者任务，返回Future并赋值。

1.构建内部类ListenerConsumer

ListenerConsumer类图如下：

ListenerConsumer构造函数源码如下：

 @SuppressWarnings("unchecked")
         ListenerConsumer(GenericMessageListener<?> listener, GenericAcknowledgingMessageListener<?> ackListener) {
             Assert.state(!this.isAnyManualAck || !this.autoCommit,
                     "Consumer cannot be configured for auto commit for ackMode " + this.containerProperties.getAckMode());
             @SuppressWarnings("deprecation")
             final Consumer<K, V> consumer =
                     KafkaMessageListenerContainer.this.consumerFactory instanceof
                                     org.springframework.kafka.core.ClientIdSuffixAware
                             ? ((org.springframework.kafka.core.ClientIdSuffixAware<K, V>) KafkaMessageListenerContainer
                                     .this.consumerFactory)
                                         .createConsumer(KafkaMessageListenerContainer.this.clientIdSuffix)
                             : KafkaMessageListenerContainer.this.consumerFactory.createConsumer();
 
             this.theListener = listener == null ? ackListener : listener;
             ConsumerRebalanceListener rebalanceListener = createRebalanceListener(consumer);
 
             if (KafkaMessageListenerContainer.this.topicPartitions == null) {
                 if (this.containerProperties.getTopicPattern() != null) {
                     consumer.subscribe(this.containerProperties.getTopicPattern(), rebalanceListener);
                 }
                 else {
                     consumer.subscribe(Arrays.asList(this.containerProperties.getTopics()), rebalanceListener);
                 }
             }
             else {
                 List<TopicPartitionInitialOffset> topicPartitions =
                         Arrays.asList(KafkaMessageListenerContainer.this.topicPartitions);
                 this.definedPartitions = new HashMap<>(topicPartitions.size());
                 for (TopicPartitionInitialOffset topicPartition : topicPartitions) {
                     this.definedPartitions.put(topicPartition.topicPartition(),
                             new OffsetMetadata(topicPartition.initialOffset(), topicPartition.isRelativeToCurrent()));
                 }
                 consumer.assign(new ArrayList<>(this.definedPartitions.keySet()));
             }
             this.consumer = consumer;
             GenericErrorHandler<?> errHandler = this.containerProperties.getGenericErrorHandler();
             this.genericListener = listener; 
               //1.
               if (this.theListener instanceof BatchAcknowledgingMessageListener) {
                 this.listener = null;
                 this.batchListener = null;
                 this.acknowledgingMessageListener = null;
                 this.batchAcknowledgingMessageListener = (BatchAcknowledgingMessageListener<K, V>) this.theListener;
                 this.isBatchListener = true;
             }//2.
             else if (this.theListener instanceof AcknowledgingMessageListener) {
                 this.listener = null;
                 this.acknowledgingMessageListener = (AcknowledgingMessageListener<K, V>) this.theListener;
                 this.batchListener = null;
                 this.batchAcknowledgingMessageListener = null;
                 this.isBatchListener = false;
             }//3.
             else if (this.theListener instanceof BatchMessageListener) {
                 this.listener = null;
                 this.batchListener = (BatchMessageListener<K, V>) this.theListener;
                 this.acknowledgingMessageListener = null;
                 this.batchAcknowledgingMessageListener = null;
                 this.isBatchListener = true;
             }//4.
             else if (this.theListener instanceof MessageListener) {
                 this.listener = (MessageListener<K, V>) this.theListener;
                 this.batchListener = null;
                 this.acknowledgingMessageListener = null;
                 this.batchAcknowledgingMessageListener = null;
                 this.isBatchListener = false;
             }
             else {
                 throw new IllegalArgumentException("Listener must be one of 'MessageListener', "
                         + "'BatchMessageListener', 'AcknowledgingMessageListener', "
                         + "'BatchAcknowledgingMessageListener', not " + this.theListener.getClass().getName());
             }
             if (this.isBatchListener) {
                 validateErrorHandler(true);
                 this.errorHandler = new LoggingErrorHandler();
                 this.batchErrorHandler = errHandler == null ? new BatchLoggingErrorHandler()
                         : (BatchErrorHandler) errHandler;
             }
             else {
                 validateErrorHandler(false);
                 this.errorHandler = errHandler == null ? new LoggingErrorHandler() : (ErrorHandler) errHandler;
                 this.batchErrorHandler = new BatchLoggingErrorHandler();
             }
             Assert.state(!this.isBatchListener || !this.isRecordAck, "Cannot use AckMode.RECORD with a batch listener");
         }

1.定义消费者订阅topic或者指定分区

2.设置监听器，支持4种：

　　1）BatchAcknowledgingMessageListener批量需确认消息监听器

　　2）AcknowledgingMessageListener需确认消息监听器

　　3）BatchMessageListener批量消息监听器

　　4）MessageListener消息监听器（用的最多，一次消费一条）

2.提交 监听消费者任务（ListenerConsumer），返回Future并赋值。

这里我们看一下任务Runnable接口的run方法，分两种情况

1.如果自定义了分区,没必要再平衡分配分区了，直接回调

2.未指定分区，进入自旋消费

 @Override
         public void run() {
             if (this.genericListener instanceof ConsumerSeekAware) {
                 ((ConsumerSeekAware) this.genericListener).registerSeekCallback(this);
             }
             this.count = 0;
             this.last = System.currentTimeMillis();
             if (isRunning() && this.definedPartitions != null) {// 1.如果运行中且自定义了分区,没必要再平衡分配分区了，直接回调
                 initPartitionsIfNeeded();// 有需要就初始化分区
                 // 回调
                 if (!this.autoCommit) {
                     startInvoker();
                 }
             }
             long lastReceive = System.currentTimeMillis();
             long lastAlertAt = lastReceive;
             while (isRunning()) {//2.未指定分区，进入自旋消费
                 try {
                     if (!this.autoCommit) {
                         processCommits();// 如果手动提交，处理提交
                     }
                     processSeeks();// 重新定位偏移量，下一次消费时使用
                     if (this.logger.isTraceEnabled()) {
                         this.logger.trace("Polling (paused=" + this.paused + ")...");
                     }// 1）拉取消费记录
                     ConsumerRecords<K, V> records = this.consumer.poll(this.containerProperties.getPollTimeout());
                     if (records != null && this.logger.isDebugEnabled()) {
                         this.logger.debug("Received: " + records.count() + " records");
                     }
                     if (records != null && records.count() > 0) {
                         if (this.containerProperties.getIdleEventInterval() != null) {
                             lastReceive = System.currentTimeMillis();
                         }// 2）如果设置了自动提交，直接在当前线程执行
                         if (this.autoCommit) {
                             invokeListener(records);
                         }
                         else {// 3）否则发送消息进缓存队列
                             if (sendToListener(records)) {
                                 if (this.assignedPartitions != null) {
                                     // avoid group management rebalance due to a slow
                                     // consumer
                                     this.consumer.pause(this.assignedPartitions);
                                     this.paused = true;
                                     this.unsent = records;
                                 }
                             }
                         }
                     }
                     else {
                         if (this.containerProperties.getIdleEventInterval() != null) {
                             long now = System.currentTimeMillis();
                             if (now > lastReceive + this.containerProperties.getIdleEventInterval()
                                     && now > lastAlertAt + this.containerProperties.getIdleEventInterval()) {
                                 publishIdleContainerEvent(now - lastReceive);
                                 lastAlertAt = now;
                                 if (this.genericListener instanceof ConsumerSeekAware) {
                                     seekPartitions(getAssignedPartitions(), true);
                                 }
                             }
                         }
                     }
                     this.unsent = checkPause(this.unsent);
                 }
                 catch (WakeupException e) {
                     this.unsent = checkPause(this.unsent);
                 }
                 catch (Exception e) {
                     if (this.containerProperties.getGenericErrorHandler() != null) {
                         this.containerProperties.getGenericErrorHandler().handle(e, null);
                     }
                     else {
                         this.logger.error("Container exception", e);
                     }
                 }
             }
             if (this.listenerInvokerFuture != null) {
                 stopInvoker();
                 commitManualAcks();
             }
             try {
                 this.consumer.unsubscribe();
             }
             catch (WakeupException e) {
                 // No-op. Continue process
             }
             this.consumer.close();
             if (this.logger.isInfoEnabled()) {
                 this.logger.info("Consumer stopped");
             }
         }

1.如果用户自定义了分区且非自动提交，那么开启异步线程执行ListenerInvoker任务，源码如下：

 private void startInvoker() {
             ListenerConsumer.this.invoker = new ListenerInvoker();
             ListenerConsumer.this.listenerInvokerFuture = this.containerProperties.getListenerTaskExecutor()
                     .submit(ListenerConsumer.this.invoker);
         }

执行ListenerInvoker的run方法，实际上就执行一遍，因为CountDownLatch初始化为1

 private final class ListenerInvoker implements SchedulingAwareRunnable {
 
             private final CountDownLatch exitLatch = new CountDownLatch(1);
 
             private volatile boolean active = true;
 
             private volatile Thread executingThread;
 
             ListenerInvoker() {
                 super();
             }
 
             @Override
             public void run() {
                 Assert.isTrue(this.active, "This instance is not active anymore");
                 if (ListenerConsumer.this.theListener instanceof ConsumerSeekAware) {
                     ((ConsumerSeekAware) ListenerConsumer.this.theListener).registerSeekCallback(ListenerConsumer.this);
                 }
                 try {
                     this.executingThread = Thread.currentThread();
                     while (this.active) {
                         try {// 从阻塞队列LinkedBlockingQueue recordsToProcess中拉取 待消费记录
                             ConsumerRecords<K, V> records = ListenerConsumer.this.recordsToProcess.poll(1,
                                     TimeUnit.SECONDS);
                             if (this.active) {
                                 if (records != null) {
                                     invokeListener(records);// 消费
                                 }
                                 else {
                                     if (ListenerConsumer.this.logger.isTraceEnabled()) {
                                         ListenerConsumer.this.logger.trace("No records to process");
                                     }
                                 }
                             }
                         }
                         catch (InterruptedException e) {
                             if (!this.active) {
                                 Thread.currentThread().interrupt();
                             }
                             else {
                                 ListenerConsumer.this.logger.debug("Interrupt ignored");
                             }
                         }
                     }
                 }
                 finally {
                     this.active = false;
                     this.exitLatch.countDown();
                 }
             }
 
             @Override
             public boolean isLongLived() {
                 return true;
             }
         }

 private void invokeListener(final ConsumerRecords<K, V> records) {
             if (this.isBatchListener) {
                 invokeBatchListener(records);
             }
             else {
                 invokeRecordListener(records);
             }
         }

如上图，从阻塞队列中取得待消费记录，用迭代器iterator消费，根据自定义消费类型，用不同listener来执行onMessage方法（用户自定义MessageListener接口的onMessage方法，实现用户自己的消费业务逻辑）

 private void invokeRecordListener(final ConsumerRecords<K, V> records) {
             Iterator<ConsumerRecord<K, V>> iterator = records.iterator();
             while (iterator.hasNext() && (this.autoCommit || (this.invoker != null && this.invoker.active))) {
                 final ConsumerRecord<K, V> record = iterator.next();
                 if (this.logger.isTraceEnabled()) {
                     this.logger.trace("Processing " + record);
                 }
                 try {
                     if (this.acknowledgingMessageListener != null) {
                         this.acknowledgingMessageListener.onMessage(record,// 终极核心方法，用户自定义的MessageListener接口的onMessage方法
                                 this.isAnyManualAck
                                         ? new ConsumerAcknowledgment(record, this.isManualImmediateAck)
                                         : null);
                     }
                     else {
                         this.listener.onMessage(record);// 终极核心方法，用户自定义的MessageListener接口的onMessage方法
                     }
                     if (!this.isAnyManualAck && !this.autoCommit) {
                         this.acks.add(record);
                     }
                 }
                 catch (Exception e) {
                     if (this.containerProperties.isAckOnError() && !this.autoCommit) {
                         this.acks.add(record);
                     }
                     try {
                         this.errorHandler.handle(e, record);
                     }
                     catch (Exception ee) {
                         this.logger.error("Error handler threw an exception", ee);
                     }
                     catch (Error er) { //NOSONAR
                         this.logger.error("Error handler threw an error", er);
                         throw er;
                     }
                 }
             }
         }

2.未指定分区，进入自旋

// 1）拉取消费记录
ConsumerRecords<K, V> records = this.consumer.poll(this.containerProperties.getPollTimeout());

 2）如果设置了自动提交，直接在当前线程执行
invokeListener(records);

// 3）否则发送消息进缓存队列
sendToListener(records)

1）在每个轮询中，消费者将尝试使用最后一个被使用的偏移量作为起始偏移量，并按顺序提取。最后一个被消费的偏移量可以通过 seek(TopicPartition,long)或自动设置为最后一个被订阅的分区列表的偏移量获得。

 @Override
     public ConsumerRecords<K, V> poll(long timeout) {
         acquire();
         try {
             if (timeout < 0)
                 throw new IllegalArgumentException("Timeout must not be negative");
 
             if (this.subscriptions.hasNoSubscriptionOrUserAssignment())
                 throw new IllegalStateException("Consumer is not subscribed to any topics or assigned any partitions");
 
             // poll for new data until the timeout expires
             long start = time.milliseconds();
             long remaining = timeout;
             do {
                 Map<TopicPartition, List<ConsumerRecord<K, V>>> records = pollOnce(remaining);
                 if (!records.isEmpty()) {
                     fetcher.sendFetches();// 在返回所获取的记录之前，我们可以发送下一轮的fetches并避免阻塞等待它们的响应，以便在用户处理获取的记录时进行流水线操作。
                     client.pollNoWakeup();//由于已经更新了所使用的位置，所以我们不允许在返回所获取的记录之前触发wakeups或任何其他错误。
 
                     if (this.interceptors == null)
                         return new ConsumerRecords<>(records);
                     else// 如果存在消费者拦截器执行拦截
                         return this.interceptors.onConsume(new ConsumerRecords<>(records));
                 }
 
                 long elapsed = time.milliseconds() - start;
                 remaining = timeout - elapsed;
             } while (remaining > 0);
 
             return ConsumerRecords.empty();
         } finally {
             release();
         }
     }

pollOnce：

 private Map<TopicPartition, List<ConsumerRecord<K, V>>> pollOnce(long timeout) {
         coordinator.poll(time.milliseconds());
 
         // 遍历所有的TopicPartition，如果有未知偏移量（分区的），那么更新。涉及coordinator刷新已提交分区偏移量+fetcher更新获取位置
         if (!subscriptions.hasAllFetchPositions())
             updateFetchPositions(this.subscriptions.missingFetchPositions());
 
         // 返回已获取到的记录
         Map<TopicPartition, List<ConsumerRecord<K, V>>> records = fetcher.fetchedRecords();
         if (!records.isEmpty())
             return records;
 
         // 发送fetch请求
         fetcher.sendFetches();
 
         long now = time.milliseconds();
         long pollTimeout = Math.min(coordinator.timeToNextPoll(now), timeout);
         // 执行IO，拉取数据
         client.poll(pollTimeout, now, new PollCondition() {
             @Override
             public boolean shouldBlock() {
                 // since a fetch might be completed by the background thread, we need this poll condition
                 // to ensure that we do not block unnecessarily in poll()
                 return !fetcher.hasCompletedFetches();
             }
         });
         if (coordinator.needRejoin())
             return Collections.emptyMap();
 
         return fetcher.fetchedRecords();
     }

好吧，再往下涉及到通信IO层了，这里不再多说。将来补全了kafka通信协议相关文章后再加上飞机票。

2）invokeListener和分支1一样，最终调用的是用户自定义的MessageListener接口的onMessage方法，不再重复。

3) sendToListener,这里塞进缓存队列LinkedBlockingQueue<ConsumerRecords<K, V>> recordsToProcess,塞进队列后，何时再消费？ListenerInvoker的run方法执行了recordsToProcess.poll进行了消费，