如何使 Kafka 代理故障转移对消费者起作用？答案

【问题标题】：How to make Kafka broker failover to work regarding consumers?如何使 Kafka 代理故障转移对消费者起作用？
【发布时间】：2019-09-04 22:53:21
【问题描述】：

让复制的代理对消费者起作用似乎非常复杂：似乎在停止某些代理时，一些消费者不再工作，而当特定代理再次启动时，那些没有工作的消费者会收到所有的“丢失”消息。

我正在使用 2 个代理方案。像这样创建了一个复制主题：

  $KAFKA_HOME/bin/kafka-topics.sh --create \
  --zookeeper localhost:2181 \
  --replication-factor 2 \
  --partitions 3 \
  --topic replicated_topic

服务器配置的摘录如下所示（请注意，除了端口、代理 ID 和日志目录外，两台服务器都是相同的）：

# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements.  See the NOTICE file distributed with
# this work for additional information regarding copyright ownership.
# The ASF licenses this file to You under the Apache License, Version 2.0
# (the "License"); you may not use this file except in compliance with
# the License.  You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

# see kafka.server.KafkaConfig for additional details and defaults

############################# Server Basics #############################

# The id of the broker. This must be set to a unique integer for each broker.
broker.id=0
############################# Socket Server Settings #############################

# The address the socket server listens on. It will get the value returned from 
# java.net.InetAddress.getCanonicalHostName() if not configured.
#   FORMAT:
#     listeners = listener_name://host_name:port
#   EXAMPLE:
#     listeners = PLAINTEXT://your.host.name:9092
listeners=PLAINTEXT://:9092

# Hostname and port the broker will advertise to producers and consumers. If not set, 
# it uses the value for "listeners" if configured.  Otherwise, it will use the value
# returned from java.net.InetAddress.getCanonicalHostName().
#advertised.listeners=PLAINTEXT://your.host.name:9092

# Maps listener names to security protocols, the default is for them to be the same. See the config documentation for more details
#listener.security.protocol.map=PLAINTEXT:PLAINTEXT,SSL:SSL,SASL_PLAINTEXT:SASL_PLAINTEXT,SASL_SSL:SASL_SSL

# The number of threads that the server uses for receiving requests from the network and sending responses to the network
num.network.threads=3

# The number of threads that the server uses for processing requests, which may include disk I/O
num.io.threads=8

# The send buffer (SO_SNDBUF) used by the socket server
socket.send.buffer.bytes=102400

# The receive buffer (SO_RCVBUF) used by the socket server
socket.receive.buffer.bytes=102400

# The maximum size of a request that the socket server will accept (protection against OOM)
socket.request.max.bytes=104857600


############################# Log Basics #############################

# A comma seperated list of directories under which to store log files
log.dirs=/tmp/kafka-logs0

# The default number of log partitions per topic. More partitions allow greater
# parallelism for consumption, but this will also result in more files across
# the brokers.
num.partitions=1

# The number of threads per data directory to be used for log recovery at startup and flushing at shutdown.
# This value is recommended to be increased for installations with data dirs located in RAID array.
num.recovery.threads.per.data.dir=1

############################# Internal Topic Settings  #############################
# The replication factor for the group metadata internal topics "__consumer_offsets" and "__transaction_state"
# For anything other than development testing, a value greater than 1 is recommended for to ensure availability such as 3.
offsets.topic.replication.factor=2
transaction.state.log.replication.factor=1
transaction.state.log.min.isr=1

############################# Log Flush Policy #############################

# Messages are immediately written to the filesystem but by default we only fsync() to sync
# the OS cache lazily. The following configurations control the flush of data to disk.
# There are a few important trade-offs here:
#    1. Durability: Unflushed data may be lost if you are not using replication.
#    2. Latency: Very large flush intervals may lead to latency spikes when the flush does occur as there will be a lot of data to flush.
#    3. Throughput: The flush is generally the most expensive operation, and a small flush interval may lead to exceessive seeks.
# The settings below allow one to configure the flush policy to flush data after a period of time or
# every N messages (or both). This can be done globally and overridden on a per-topic basis.

# The number of messages to accept before forcing a flush of data to disk
#log.flush.interval.messages=10000

# The maximum amount of time a message can sit in a log before we force a flush
#log.flush.interval.ms=1000

############################# Log Retention Policy #############################

# The following configurations control the disposal of log segments. The policy can
# be set to delete segments after a period of time, or after a given size has accumulated.
# A segment will be deleted whenever *either* of these criteria are met. Deletion always happens
# from the end of the log.

# The minimum age of a log file to be eligible for deletion due to age
log.retention.hours=168

# A size-based retention policy for logs. Segments are pruned from the log unless the remaining
# segments drop below log.retention.bytes. Functions independently of log.retention.hours.
#log.retention.bytes=1073741824

# The maximum size of a log segment file. When this size is reached a new log segment will be created.
log.segment.bytes=1073741824

# The interval at which log segments are checked to see if they can be deleted according
# to the retention policies
log.retention.check.interval.ms=300000

############################# Zookeeper #############################

# Zookeeper connection string (see zookeeper docs for details).
# This is a comma separated host:port pairs, each corresponding to a zk
# server. e.g. "127.0.0.1:3000,127.0.0.1:3001,127.0.0.1:3002".
# You can also append an optional chroot string to the urls to specify the
# root directory for all kafka znodes.
zookeeper.connect=localhost:2181

# Timeout in ms for connecting to zookeeper
zookeeper.connection.timeout.ms=6000


############################# Group Coordinator Settings #############################

# The following configuration specifies the time, in milliseconds, that the GroupCoordinator will delay the initial consumer rebalance.
# The rebalance will be further delayed by the value of group.initial.rebalance.delay.ms as new members join the group, up to a maximum of max.poll.interval.ms.
# The default value for this is 3 seconds.
# We override this to 0 here as it makes for a better out-of-the-box experience for development and testing.
# However, in production environments the default value of 3 seconds is more suitable as this will help to avoid unnecessary, and potentially expensive, rebalances during application startup.
group.initial.rebalance.delay.ms=0

让我们用 2 个经纪人来描述我的主题：

Topic:replicated_topic  PartitionCount:3    ReplicationFactor:2 Configs:
    Topic: replicated_topic Partition: 0    Leader: 1   Replicas: 1,0   Isr: 1,0
    Topic: replicated_topic Partition: 1    Leader: 0   Replicas: 0,1   Isr: 1,0
    Topic: replicated_topic Partition: 2    Leader: 1   Replicas: 1,0   Isr: 1,0

让我们看看消费者的代码：消费者（impl Callable）

@Override
public Void call() throws Exception {
    final Properties props = new Properties();
    props.put(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG,
            bootstrapServers);
    props.put(ConsumerConfig.GROUP_ID_CONFIG,
            groupId);
    props.put(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG,
            IntegerDeserializer.class.getName());
    props.put(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG,
            StringDeserializer.class.getName());

    final Consumer<Integer, String> consumer = new KafkaConsumer<>(props);

    consumer.subscribe(Collections.singletonList(topicName));

    ConsumerRecords<Integer, String> records = null;
    while (!Thread.currentThread().isInterrupted()) {
        records = consumer.poll(1000);
        if (records.isEmpty()) {
            continue;
        }
        records.forEach(rec -> LOGGER.debug("{}@{} consumed from topic {}, partition {} pair ({},{})",
                ConsumerCallable.class.getSimpleName(), hashCode(), rec.topic(), rec.partition(), rec.key(), rec.value()));
        consumer.commitAsync();
    }

    consumer.close();
    return null;
}

生产者和主要代码：

private static final String TOPIC_NAME = "replicated_topic";
private static final String BOOTSTRAP_SERVERS = "localhost:9092, localhost:9093";
private static final Logger LOGGER = LoggerFactory.getLogger(Main.class);

public static void main(String[] args) {

    ExecutorService executor = Executors.newCachedThreadPool();
    executor.submit(new ConsumerCallable(TOPIC_NAME, BOOTSTRAP_SERVERS, "group1"));
    executor.submit(new ConsumerCallable(TOPIC_NAME, BOOTSTRAP_SERVERS, "group2"));
    executor.submit(new ConsumerCallable(TOPIC_NAME, BOOTSTRAP_SERVERS, "group3"));

    try (Producer<Integer, String> producer = createProducer()) {
        Scanner scanner = new Scanner(System.in);
        String line = null;
        LOGGER.debug("Please enter 'k v' on the command line to send to Kafka or 'quit' to exit");
        while (scanner.hasNextLine()) {
            line = scanner.nextLine();
            if (line.trim().toLowerCase().equals("quit")) {
                break;
            }
            String[] elements = line.split(" ");
            Integer key = Integer.parseInt(elements[0]);
            String value = elements[1];
            producer.send(new ProducerRecord<>(TOPIC_NAME, key, value));
            producer.flush();
        }
    }
    executor.shutdownNow();
}

private static Producer<Integer, String> createProducer() {
    Properties props = new Properties();
    props.put(ProducerConfig.BOOTSTRAP_SERVERS_CONFIG,
            BOOTSTRAP_SERVERS);
    props.put(ProducerConfig.CLIENT_ID_CONFIG, "KafkaExampleProducer");
    props.put(ProducerConfig.KEY_SERIALIZER_CLASS_CONFIG,
            IntegerSerializer.class.getName());
    props.put(ProducerConfig.VALUE_SERIALIZER_CLASS_CONFIG,
            StringSerializer.class.getName());
    return new KafkaProducer<>(props);
}

现在让我们看看行为：

所有代理都已上线：

kafka 主题的输出：

Topic:replicated_topic  PartitionCount:3    ReplicationFactor:2 Configs:
    Topic: replicated_topic Partition: 0    Leader: 1   Replicas: 1,0   Isr: 1,0
    Topic: replicated_topic Partition: 1    Leader: 0   Replicas: 0,1   Isr: 1,0
    Topic: replicated_topic Partition: 2    Leader: 1   Replicas: 1,0   Isr: 1,0

程序输出：

12:52:30.460 DEBUG Main - Please enter 'k v' on the command line to send to Kafka or 'quit' to exit
1 u
12:52:35.555 DEBUG ConsumerCallable - ConsumerCallable@1241910294 consumed from topic replicated_topic, partition 0 pair (1,u)
12:52:35.559 DEBUG ConsumerCallable - ConsumerCallable@1361430455 consumed from topic replicated_topic, partition 0 pair (1,u)
12:52:35.559 DEBUG ConsumerCallable - ConsumerCallable@186743616 consumed from topic replicated_topic, partition 0 pair (1,u)
2 d
12:52:38.096 DEBUG ConsumerCallable - ConsumerCallable@186743616 consumed from topic replicated_topic, partition 2 pair (2,d)
12:52:38.098 DEBUG ConsumerCallable - ConsumerCallable@1361430455 consumed from topic replicated_topic, partition 2 pair (2,d)
12:52:38.100 DEBUG ConsumerCallable - ConsumerCallable@1241910294 consumed from topic replicated_topic, partition 2 pair (2,d)

由于消费者在不同的组中，所有消息都被广播给他们，一切都很好。

2 关闭代理 2：

描述主题：

Topic:replicated_topic  PartitionCount:3    ReplicationFactor:2 Configs:
    Topic: replicated_topic Partition: 0    Leader: 0   Replicas: 1,0   Isr: 0
    Topic: replicated_topic Partition: 1    Leader: 0   Replicas: 0,1   Isr: 0
    Topic: replicated_topic Partition: 2    Leader: 0   Replicas: 1,0   Isr: 0

程序输出：

3 t
12:57:03.898 DEBUG ConsumerCallable - ConsumerCallable@186743616 consumed from topic replicated_topic, partition 1 pair (3,t)
4 p
12:57:06.058 DEBUG ConsumerCallable - ConsumerCallable@186743616 consumed from topic replicated_topic, partition 1 pair (4,p)

现在只有 1 个消费者接收数据。让我们再次提出经纪人 2：现在其他 2 个消费者收到了缺失的数据：

12:57:50.863 DEBUG ConsumerCallable - ConsumerCallable@1241910294 consumed from topic replicated_topic, partition 1 pair (3,t)
12:57:50.863 DEBUG ConsumerCallable - ConsumerCallable@1241910294 consumed from topic replicated_topic, partition 1 pair (4,p)
12:57:50.870 DEBUG ConsumerCallable - ConsumerCallable@1361430455 consumed from topic replicated_topic, partition 1 pair (3,t)
12:57:50.870 DEBUG ConsumerCallable - ConsumerCallable@1361430455 consumed from topic replicated_topic, partition 1 pair (4,p)

关闭代理 1：

现在只有 2 个消费者接收数据：

5 c
12:59:13.718 DEBUG ConsumerCallable - ConsumerCallable@1361430455 consumed from topic replicated_topic, partition 2 pair (5,c)
12:59:13.737 DEBUG ConsumerCallable - ConsumerCallable@1241910294 consumed from topic replicated_topic, partition 2 pair (5,c)
6 s
12:59:16.437 DEBUG ConsumerCallable - ConsumerCallable@1361430455 consumed from topic replicated_topic, partition 2 pair (6,s)
12:59:16.438 DEBUG ConsumerCallable - ConsumerCallable@1241910294 consumed from topic replicated_topic, partition 2 pair (6,s)

如果我把它带到其他消费者身上，也会收到丢失的数据。

我的意思是伙计们（很抱歉，我很抱歉，但我正在尝试捕捉上下文），是如何确保无论我停止哪个经纪人，消费者都能正常工作？（正常接收所有消息）？

PS：我尝试设置 offsets.topic.replication.factor=2 或 3，但没有任何效果。

【问题讨论】：

标签： java apache-kafka

【解决方案1】：

如果否，则不会忽略发给该代理的消息。活动代理的数量少于配置的副本。每当有新的 Kafka 代理加入集群时，数据就会被复制到该节点。 https://stackoverflow.com/a/38998062/6274525

因此，当您的代理 2 出现故障时，消息仍会被推送到另一个活跃的代理，因为有 1 个实时代理且复制因子为 2。由于您的其他 2 个消费者订阅了代理 2（已关闭），因此它们是无法消费。

当您的代理 2 再次启动时，数据会复制到此新节点，因此附加到此节点的消费者会收到消息（您将其称为“丢失”消息）。

【讨论】：

但基本上我怎样才能让订阅死节点的消费者再次订阅新节点？我必须在代码中做些什么还是驱动程序应该做？
尝试增加 2 个代理方案的副本数。因为根据您的数据，当代理 2 出现故障时，将会有更多副本从领导副本同步，因此您的消费者可以从中消费。可能是！！！
好吧，kafka 可以容忍 RF-1 故障，所以在我的情况下，系统应该可以在 2-1=1 节点上正常工作。现在因为我有 2 个节点，这就是我将 RF 设置为 2 的原因。增加副本数是什么意思？我只有 2 个节点，最多 2 个。还是我错过了什么？
但是，当一个代理被不干净地关闭时（例如，kill -9），观察到的不可用性可能与分区数量成正比......阅读此confluent.io/blog/…
@adragomir 你能找到解决方案吗？

【解决方案2】：

请确保您已将名为 offsets.topic.replication.factor 的属性更改为至少 3。

此属性用于管理偏移量和消费者交互。启动 kafka 服务器时，它会自动创建一个名为 __consumer_offsets 的主题。因此，如果未在该主题中创建副本，则消费者无法确定是否已将某些内容推送到它正在收听的主题。

此属性详细信息的链接：https://kafka.apache.org/documentation/#brokerconfigs

【讨论】：