kafka使用者获取API不返回正确的偏移量值

Question

我已经建立了一个实验性的Kafka环境，有3个代理和一个有3个分区的主题，我有一个生产者和一个消费者。我想为特定使用者修改分区的偏移量。我在kafka文档中读到，kafka中的使用者提交/获取API可以提交特定的偏移量或获取使用者读取的最新偏移量。以下是API的链接：

但是，我已经生成了一些消息，我的使用者已经使用了这些消息并输出每个已读消息的偏移量。

如果有人能帮上忙，我将不胜感激。我想知道我的代码的哪一部分是错的。或者可能API有问题。请不要犹豫，提出任何有用的意见。我的代码与我提供的链接中的代码完全相同。但是，如果你需要看我的代码，请告诉我把它放在这里。

Kafka版本是0.10.2.0

主题：“TestPIC3”

………………………………。

使用者配置：

props.put("group.id", "test");

props.put("client.id", "MyConsumer");

public class KafkaOffsetManage {

public static void main(String[] args) {


    BlockingChannel channel = new BlockingChannel("localhost", 9095,
            BlockingChannel.UseDefaultBufferSize(),
            BlockingChannel.UseDefaultBufferSize(),
            5000 /* read timeout in millis */);
    channel.connect();
    final String MY_GROUP = "test";
    final String MY_CLIENTID = "MyConsumer";
    int correlationId = 0;
    final TopicAndPartition testPartition0 = new TopicAndPartition("testpic3",0);
    final TopicAndPartition testPartition1 = new TopicAndPartition("testpic3",1);
    final TopicAndPartition testPartition2 = new TopicAndPartition("testpic3",2);
    channel.send(new ConsumerMetadataRequest(MY_GROUP, ConsumerMetadataRequest.CurrentVersion(), correlationId++, MY_CLIENTID));
    ConsumerMetadataResponse metadataResponse = ConsumerMetadataResponse.readFrom(channel.receive().buffer());
    System.out.println("+++++++++++++++++++++++++++");

    System.out.println(metadataResponse.errorCode());

    if (metadataResponse.errorCode() == ErrorMapping.NoError()) {
        Broker offsetManager = metadataResponse.coordinator();
        // if the coordinator is different, from the above channel's host then reconnect
        channel.disconnect();
        channel = new BlockingChannel(offsetManager.host(), offsetManager.port(),
                BlockingChannel.UseDefaultBufferSize(),
                BlockingChannel.UseDefaultBufferSize(),
                5000 /* read timeout in millis */);
        channel.connect();
        System.out.println("Connected to Offset Manager");
        System.out.println(offsetManager.host() + ",  Port:"+ offsetManager.port());

    } else {
        // retry (after backoff)
    }



    // How to fetch offsets


    List<TopicAndPartition> partitions = new ArrayList<TopicAndPartition>();
    partitions.add(testPartition0);
    //partitions.add(testPartition1);
    OffsetFetchRequest fetchRequest = new OffsetFetchRequest(
            MY_GROUP,
            partitions,
            (short) 2 /* version */, // version 1 and above fetch from Kafka, version 0 fetches from ZooKeeper
            correlationId,
            MY_CLIENTID);
    try {
        channel.send(fetchRequest.underlying());
        OffsetFetchResponse fetchResponse = OffsetFetchResponse.readFrom(channel.receive().buffer());
        OffsetMetadataAndError result = fetchResponse.offsets().get(testPartition0);

        short offsetFetchErrorCode = result.error();
        if (offsetFetchErrorCode == ErrorMapping.NotCoordinatorForConsumerCode()) {
            channel.disconnect();
            // Go to step 1 and retry the offset fetch
        } else if (offsetFetchErrorCode  == ErrorMapping.OffsetsLoadInProgressCode()) {
            // retry the offset fetch (after backoff)
        } else {
            long retrievedOffset = result.offset();
            String retrievedMetadata = result.metadata();
            System.out.println("The retrieved offset is:"+ Long.toString(retrievedOffset));
            System.out.println(retrievedMetadata);
            System.out.println(result.toString());
        }
    }
    catch (Exception e) {
        channel.disconnect();
        // Go to step 1 and then retry offset fetch after backoff
    }
 }
}

+++++++++++++++++++++++++++
0

Connected to Offset Manager

user-virtual-machine,  Port:9093
------------------------
The retrieved offset is:-1

OffsetMetadataAndError[-1,,3]

Process finished with exit code 0

<artifactId>kafka_2.10</artifactId> 
<version>0.10.2.0</version> 

<artifactId>kafka_2.10</artifactId> 
<version>0.8.2.0</version>

谢谢，

Answer 1

这种情况发生是因为偏移量过期。Kafka有两个参数控制着这种行为。首先是“__consumer_offsets”主题的“retention.ms”设置。它应该等于-1来禁用该主题内记录的过期。我假设使用Kafka版本1.1.x。使用命令检查主题配置：

$ ./kafka-configs.sh --entity-type topics \
                     --entity-name __consumer_offsets \
                     --zookeeper localhost:2181 \
                     --describe
Configs for topic '__consumer_offsets' are compression.type=producer,cleanup.policy=compact,min.insync.replicas=2,segment.bytes=104857600,retention.ms=-1,unclean.leader.election.enable=false

如果不满足配置设置，请使用命令进行更改：

$ ./kafka-configs.sh --entity-type topics \
                     --entity-name __consumer_offsets \
                     --zookeeper localhost:2181 \
                     --alter \
                     --add-config retention.ms=-1

假设设置了保留策略，接下来需要检查主题中是否有任何提交的消息。默认情况下，Kafka不允许阅读内部主题。要更改此行为，请创建一个具有使用者设置的文件：

$ echo exclude.internal.topics=false > consumer.properties

$ ./kafka-console-consumer.sh --consumer.config consumer.properties \
                              --from-beginning \
                              --topic __consumer_offsets \
                              --zookeeper localhost:2181 \
                              --formatter "kafka.coordinator.group.GroupMetadataManager\$OffsetsMessageFormatter"

[test_client,Test.Purposes,2]::[OffsetMetadata[13,NO_METADATA],CommitTime 1534165245681,ExpirationTime 1534251645681]
[test_client,Test.Purposes,0]::[OffsetMetadata[14,NO_METADATA],CommitTime 1534165245776,ExpirationTime 1534251645776]
[test_client,Test.Purposes,1]::[OffsetMetadata[8,NO_METADATA],CommitTime 1534165690946,ExpirationTime 1534252090946]

ExpirationTime = CommitTime + offsets.retention.minutes

$ date -d @1534165245
Mon Aug 13 16:00:45 UTC 2018

$ date -d @1534251645
Tue Aug 14 16:00:45 UTC 2018

正好是24小时。

因此，解决不正确偏移量问题的方法是增加“offsets.retention.minutes”设置，记住这会影响代理内存的使用，当系统中有许多死亡的使用者组时，还会定期提交未更改的偏移量以增加过期时间。