How to make Kafka broker failover to work regarding consumers?

Question

It seems very complicated to make a replicated broker work regarding consumers: it seems when stopping certain brokers, some consumers don't work anymore and, when the specific broker is up again, those consumers that didn't work receive all the "missing" messages.

I am using a 2 brokers scenario. Created a replicated topic like this:

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

The excerpt from the server config looks like this ( notice it is the same for both servers except port, broker id and log dir):

# 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

Let's decribe my topic using 2 brokers:

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

Let's see the code for the consumer: Consumer ( 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;
}

Producer and main code:

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);
}

Now let's see the behaviour:

1. All brokers are up:

Output of kafka topic:

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

Output of program:

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)

Since the consumers are in different groups all messages are broadcasted to them, everything is ok.

2 Bring down broker 2:

Describe topic:

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

Output of program:

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)

Now only 1 consumer receives data. Let's bring up broker 2 again: Now the other 2 consumers receive the missing data:

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)

3. Bring down broker 1:

Now only 2 consumers receive data:

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)

If I bring it on the other consumer wil also receive missing data.

My point guys ( sorry for big write but I am trying to capture the context ), is how to make sure that no matter what broker I would stop, the consumers would work correctly? ( receive all messages normally )?

PS: I tried setting the offsets.topic.replication.factor=2 or 3, but it didn't have any effect.

Answer 1

Please make sure you have changed the property called offsets.topic.replication.factor to atleast 3.

This property is used to manage offset and consumer interaction. When a kafka server is started, it auto creates a topic with name __consumer_offsets. So if the replicas are not created in this topic, then a consumer cannot know for sure if something has been pushed to the Topic it was listening to.

Link to detail of this property : https://kafka.apache.org/documentation/#brokerconfigs

Answer 2

Messages to that broker will not be ignored if the no. of alive brokers is lesser than the configured replicas. Whenever a new Kafka broker joins the cluster, the data gets replicated to that node. https://stackoverflow.com/a/38998062/6274525

So when your broker 2 goes down, the messages still get pushed to another alive broker because there is 1 live broker and replication factor is 2. Since your other 2 consumers are subscribed to broker 2 (which is down), they are unable to consume.

When your broker 2 is up again, the data gets duplicated to this new node and hence the consumers attached to this node receive the message (referred by you as "missing" messages).