Kafka Administration Using Command Line Tools

The following tools can be found as part of the Kafka distribution, but their use is generally discouraged for various reasons as documented here.

Here are some additional points to be aware of regarding Kafka administration:

• Use Cloudera Manager to start and stop Kafka and Zookeeper services. Do not use the kafka-server-start, kafka-server-stop, zookeeper-server-start, or zookeeper-server-stop commands.
• For a parcel installation, all Kafka command line tools are located in /opt/cloudera/parcels/KAFKA/lib/kafka/bin/. For a package installation, all such tools can be found in /usr/bin/.
• Ensure that the JAVA_HOME environment variable is set to your JDK installation directory before using the command-line tools. For example:

  export JAVA_HOME=/usr/java/jdk1.8.0_144-cloudera
  

• Using any Zookeeper command manually can be very difficult to get right when it comes to interaction with Kafka. Cloudera recommends that you avoid doing any write operations or ACL modifications in Zookeeper.

Use the kafka-topics tool to generate a snapshot of topics in the Kafka cluster.

kafka-topics --zookeeper zkhost --describe

Topic: topic-a1      PartitionCount:3      ReplicationFactor:3      Configs:
         Topic: topic-a1      Partition: 0      Leader: 64      Replicas: 64,62,63      Isr: 64,62,63
         Topic: topic-a1      Partition: 1      Leader: 62      Replicas: 62,63,64      Isr: 62,63,64
         Topic: topic-a1      Partition: 2      Leader: 63      Replicas: 63,64,62      Isr: 63,64,62
Topic: topic-a2      PartitionCount:1      ReplicationFactor:3      Configs:
         Topic: topic-a2      Partition: 0      Leader: 64      Replicas: 64,62,63      Isr: 64,62,63

The output lists each topic and basic partition information. Note the following about the output:

• Partition count: The more partitions, the higher the possible parallelism among consumers and producers.
• Replication factor: Shows 1 for no redundancy and higher for more redundancy.
• Replicas and in-sync replicas (ISR): Shows which broker ID’s have the partitions and which replicas are current.

There are situations where this tool shows an invalid value for the leader broker ID or the number of ISRs is fewer than the number of replicas. In those cases, there may be something wrong with those specific topics.

It is possible to change topic configuration properties using this tool. Increasing the partition count, the replication factor or both is not recommended.

The kafka-console-consumer tool can be useful in a couple of ways:

• Acting as an independent consumer of particular topics. This can be useful to compare results against a consumer program that you’ve written.
• To test general topic consumption without the need to write any consumer code.

Examples of usage:

$ kafka-console-consumer --bootstrap-server <broker1>,<broker2>... --topic <topic> --from-beginning
<record-earliest-offset>
<record-earliest-offset+1>

Note the following about the tool:

• This tool prints all records and keeps outputting as more records are written to the topic.
• If the kafka-console-consumer tool is given no flags, it displays the full help message.
• In older versions of Kafka, it may have been necessary to use the --new-consumer flag. As of Apache Kafka version 0.10.2, this is no longer necessary.

This tool is used to write messages to a topic. It is typically not as useful as the console consumer, but it can be useful when the messages are in a text based format. In general, the usage will be something like:

cat file | kafka-console-producer args

The basic usage of the kafka-consumer-groups tool is:

kafka-consumer-groups --bootstrap-server broker1,broker2... --describe --group GROUP_ID

This tool is primarily useful for debugging consumer offset issues. The output from the tool shows the log and consumer offsets for each partition connected to the consumer group corresponding to GROUP_ID. You can see at a glance which consumers are current with their partition and which ones are behind. From there, you can determine which partitions (and likely the corresponding brokers) are slow.

Beyond this debugging usage, there are other more advanced options to this tool:

• --execute --reset-offsets SCENARIO_OPTION: Resets the offsets for a consumer group to a particular value based on the SCENARIO_OPTION flag given.

  Valid flags for SCENARIO_OPTION are:

  • --to-datetime
  • --by-period
  • --to-earliest
  • --to-latest
  • --shift-by
  • --from-file
  • --to-current

  You will likely want to set the --topic flag to restrict this change to a specific topic or a specific set of partitions within that topic.

This tool can be used to reset all offsets on all topics. This is something you probably won’t ever want to do. It is highly recommended that you use this command carefully.

This tool provides substantial control over partitions in a Kafka cluster. It is mainly used to balance storage loads across brokers through the following reassignment actions:

• Change the ordering of the partition assignment list. Used to control leader imbalances between brokers.
• Reassign partitions from one broker to another. Used to expand existing clusters.
• Reassign partitions between log directories on the same broker. Used to resolve storage load imbalance among available disks in the broker.
• Reassign partitions between log directories across multiple brokers. Used to resolve storage load imbalance across multiple brokers.

The kafka-log-dirs tool allows user to query a list of replicas per log directory on a broker. The tool provides information that is required for optimizing replica assignment across brokers.

{
    "brokers": [
        {
            "broker": 86,
            "logDirs": [
                {
                    "error": null,
                    "logDir": "/var/local/kafka/data",
                    "partitions": [
                        {
                            "isFuture": false,
                            "offsetLag": 0,
                            "partition": "mytopic1-2",
                            "size": 0
                        }
                    ]
                }
            ]
        },
        ...
    ],
    "version": 1
}

The Contents of the kafka-log-dirs Output table gives an overview of the information provided by the kafka-log-dirs tool.

The kafka-*-perf-test tool can be used in several ways. In general, it is expected that these tools should be used on a test or development cluster.

• Measuring read and/or write throughput.
• Stress testing the cluster based on specific parameters (such as message size).
• Load testing for the purpose of evaluating specific metrics or determining the impact of cluster configuration changes.

The kafka-producer-perf-test script can either create a randomly generated byte record:

kafka-producer-perf-test --topic TOPIC --record-size SIZE_IN_BYTES

or randomly read from a set of provided records:

kafka-producer-perf-test --topic TOPIC --payload-delimiter DELIMITER --payload-file INPUT_FILE

where the INPUT_FILE is a concatenated set of pre-generated messages separated by DELIMITER. This script keeps producing messages or limited based on the --num-records flag.

The kafka-consumer-perf-test is:

kafka-consumer-perf-test --broker-list host1:port1,host2:port2,... --zookeeper zk1:port1,zk2:port2,... --topic TOPIC

The flags of most interest for this command are:

• --group gid: If you run more than one instance of this test, you will want to set different ids for each instance.
• --num-fetch-threads: Defaults to 1. Increase if higher throughput testing is needed.
• --from-latest: To start consuming from the latest offset. May be needed for certain types of testing.

Almost all the provided Kafka tools eventually call the kafka-run-class script. This script is generally not called directly. However, if you are proficient with bash and want to understand certain features available in all Kafka scripts as well as some potential debugging scenarios, familiarity with the kafka-run-class script can prove highly beneficial.

For example, there are some useful environment variables that affect all the command line scripts:

• KAFKA_DEBUG allows a Java debugger to attach to the JVM launched by the particular script. Setting KAFKA_DEBUG also allows some further debugging customization:
  • JAVA_DEBUG_PORT sets the JVM debugging port.
  • JAVA_DEBUG_OPTS can be used to override the default debugging arguments being passed to the JVM.
• KAFKA_HEAP_OPTS can be used to pass memory setting arguments to the JVM.
• KAFKA_JVM_PERFORMANCE_OPTS can be used to pass garbage collection flags to the JVM.