消费者而在消费了消息之后会把消费的offset提交到 __consumer_offsets-
的内置Topic中;每个消费者组都有维护一个当前消费者组的offset。那么问题来了: 消费组什么时候把offset更新到broker中的分区中呢?
Kafka消费者的配置信息
Name | 描述 | default |
---|---|---|
enable.auto.commit | 如果为true,消费者的offset将在后台周期性的提交 | true |
auto.commit.interval.ms | 如果enable.auto.commit设置为true,则消费者偏移量自动提交给Kafka的频率(以毫秒为单位) | 5000 |
消费者端开启了自动提交之后,每隔auto.commit.interval.ms
自动提交一次;
public static void main(String[] args) {
//创建kafka消费者配置对象以及配置信息
Properties props = new Properties();
props.put("bootstrap.servers", "hadoop102:9092,hadoop103:9092,hadoop104:9092");
props.put("group.id", "hy-local-consumer");
props.put("enable.auto.commit", "true");
props.put("auto.commit.interval.ms", "5000");
props.put("key.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");
props.put("value.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");
//创建kafka消费者对象
KafkaConsumer<String, String> kafkaConsumer = new KafkaConsumer<>(props);
//消费消息
kafkaConsumer.subscribe(Arrays.asList("hy1-test-topic"));
while (true) {
ConsumerRecords<String, String> records = kafkaConsumer.poll(Duration.ofSeconds(5));
for (ConsumerRecord<String, String> record : records) {
System.out.printf("------offset-- = %d, key = %s, value = %s%n", record.offset(), record.key(), record.value());
}
}
}
/*
output:
------offset-- = 5, key = null, value = NBA
------offset-- = 4, key = null, value = CBA
------offset-- = 5, key = null, value = CUBA
------offset-- = 6, key = null, value = NCAA
------offset-- = 6, key = null, value = ABA
------offset-- = 5, key = null, value = NBL
*/
假如Consumer在获取了消息消费成功但是在提交offset之前服务挂掉了,会出现什么情况?
答:重复消费
虽然自动提交 offset 十分简介便利,但由于其是基于时间提交的,开发人员难以把握 offset 提交的时机。因此 Kafka 还提供了手动提交 offset 的 API。
手动提交 offset 的方法有两种: commitSync(同步提交)
和 commitAsync(异步 提交)
commitSync
阻塞当前线程,一直到提交成功,并且会自动失败重试(由不可控因素导致, 也会出现提交失败);而commitAsync
则没有失败重试机制,故有可能提交失败。public static void main(String[] args) {
//创建kafka消费者配置对象以及配置信息
Properties props = new Properties();
props.put("bootstrap.servers", "hadoop102:9092,hadoop103:9092,hadoop104:9092");
props.put("group.id", "hy-local-consumer");
props.put("enable.auto.commit", "true");
props.put("auto.commit.interval.ms", "5000");
props.put("key.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");
props.put("value.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");
//创建kafka消费者对象
KafkaConsumer<String, String> kafkaConsumer = new KafkaConsumer<>(props);
//订阅消息
kafkaConsumer.subscribe(Arrays.asList("hy2-test-topic"));
while (true) {
ConsumerRecords<String, String> records = kafkaConsumer.poll(Duration.ofSeconds(2));
for (ConsumerRecord<String, String> record : records) {
System.out.printf("------offset-- = %d, key = %s, value = %s%n", record.offset(), record.key(), record.value())
}
//同步提交,当前线程会阻塞直到 offset 提交成功
kafkaConsumer.commitSync();
}
}
虽然同步提交 offset 更可靠一些,但是由于其会阻塞当前线程,直到提交成功。因此吞 吐量会收到很大的影响。因此更多的情况下,会选用异步提交 offset 的方式。
public static void main(String[] args) {
//创建kafka消费者配置对象以及配置信息
Properties props = new Properties();
props.put("bootstrap.servers", "hadoop102:9092,hadoop103:9092,hadoop104:9092");
props.put("group.id", "hy-local-consumer");
props.put("enable.auto.commit", "true");
props.put("auto.commit.interval.ms", "5000");
props.put("key.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");
props.put("value.deserializer", "org.apache.kafka.common.serialization.StringDeserializer");
//创建kafka消费者对象
KafkaConsumer<String, String> kafkaConsumer = new KafkaConsumer<>(props);
//订阅消息
kafkaConsumer.subscribe(Arrays.asList("hy2-test-topic"));
while (true) {
ConsumerRecords<String, String> records = kafkaConsumer.poll(Duration.ofSeconds(2));
for (ConsumerRecord<String, String> record : records) {
System.out.printf("------offset-- = %d, key = %s, value = %s%n", record.offset(), record.key(), record.value());
}
//异步提交
kafkaConsumer.commitAsync(new OffsetCommitCallback() {
@Override
public void onComplete(Map<TopicPartition, OffsetAndMetadata> offsets, Exception exception) {
if (exception != null) {
System.err.println("异常.....");
}
}
});
}
}
无论是同步提交还是异步提交 offset,都有可能会造成数据的漏消费或者重复消费。先提交 offset 后消费,有可能造成数据的漏消费;而先消费后提交 offset,有可能会造成数据的重复消费