Repository files navigation

Dockerfile forApache Kafka

The image is available directly fromDocker Hub

All versions of the image are built from the same set of scripts with only minor variations (i.e. certain features are not supported on older versions). The version format mirrors the Kafka format,<scala version>-<kafka version>. Initially, all images are built with the recommended version of scala documented onhttp://kafka.apache.org/downloads. To list all available tags:

curl -s https://registry.hub.docker.com/v2/repositories/wurstmeister/kafka/tags\?page_size\=1024 | jq -r '.results[].name' | sort -u | egrep '\d.\d{2}-.*'

Everytime the image is updated, all tags will be pushed with the latest updates. This should allow for greater consistency across tags, as well as any security updates that have been made to the base image.

• 04-Jun-2019 - Update base image to openjdk 212 (Release notes. Please force pull to get these latest updates - including security patches etc.

• install docker-composehttps://docs.docker.com/compose/install/
• modify theKAFKA_ADVERTISED_HOST_NAME indocker-compose.yml to match your docker host IP (Note: Do not use localhost or 127.0.0.1 as the host ip if you want to run multiple brokers.)
• if you want to customize any Kafka parameters, simply add them as environment variables indocker-compose.yml, e.g. in order to increase themessage.max.bytes parameter set the environment toKAFKA_MESSAGE_MAX_BYTES: 2000000. To turn off automatic topic creation setKAFKA_AUTO_CREATE_TOPICS_ENABLE: 'false'
• Kafka's log4j usage can be customized by adding environment variables prefixed withLOG4J_. These will be mapped tolog4j.properties. For example:LOG4J_LOGGER_KAFKA_AUTHORIZER_LOGGER=DEBUG, authorizerAppender

NOTE: There are several 'gotchas' with configuring networking. If you are not sure about what the requirements are, please check out theConnectivity Guide in theWiki

Start a cluster:

• docker-compose up -d

Add more brokers:

• docker-compose scale kafka=3

Destroy a cluster:

• docker-compose stop

The defaultdocker-compose.yml should be seen as a starting point. By default each broker will get a new port number and broker id on restart. Depending on your use case this might not be desirable. If you need to use specific ports and broker ids, modify the docker-compose configuration accordingly, e.g.docker-compose-single-broker.yml:

• docker-compose -f docker-compose-single-broker.yml up

You can configure the broker id in different ways

1. explicitly, usingKAFKA_BROKER_ID
2. via a command, usingBROKER_ID_COMMAND, e.g.BROKER_ID_COMMAND: "hostname | awk -F'-' '{print $$2}'"

If you don't specify a broker id in your docker-compose file, it will automatically be generated (seehttps://issues.apache.org/jira/browse/KAFKA-1070. This allows scaling up and down. In this case it is recommended to use the--no-recreate option of docker-compose to ensure that containers are not re-created and thus keep their names and ids.

If you want to have kafka-docker automatically create topics in Kafka duringcreation, aKAFKA_CREATE_TOPICS environment variable can beadded indocker-compose.yml.

Here is an example snippet fromdocker-compose.yml:

    environment:      KAFKA_CREATE_TOPICS: "Topic1:1:3,Topic2:1:1:compact"

Topic 1 will have 1 partition and 3 replicas,Topic 2 will have 1 partition, 1 replica and acleanup.policy set tocompact. Also, see FAQ:Topic compaction does not work

If you wish to use multi-line YAML or some other delimiter between your topic definitions, override the default, separator by specifying theKAFKA_CREATE_TOPICS_SEPARATOR environment variable.

For example,KAFKA_CREATE_TOPICS_SEPARATOR: "$$'\n'" would use a newline to split the topic definitions. Syntax has to follow docker-compose escaping rules, andANSI-C quoting.

You can configure the advertised hostname in different ways

1. explicitly, usingKAFKA_ADVERTISED_HOST_NAME
2. via a command, usingHOSTNAME_COMMAND, e.g.HOSTNAME_COMMAND: "route -n | awk '/UG[ \t]/{print $$2}'"

When using commands, make sure you review the "Variable Substitution" section inhttps://docs.docker.com/compose/compose-file/

IfKAFKA_ADVERTISED_HOST_NAME is specified, it takes precedence overHOSTNAME_COMMAND

For AWS deployment, you can use the Metadata service to get the container host's IP:

HOSTNAME_COMMAND=wget -t3 -T2 -qO-  http://169.254.169.254/latest/meta-data/local-ipv4

Reference:http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/ec2-instance-metadata.html

If you require the value ofHOSTNAME_COMMAND in any of your otherKAFKA_XXX variables, use the_{HOSTNAME_COMMAND} string in your variable value, i.e.

KAFKA_ADVERTISED_LISTENERS=SSL://_{HOSTNAME_COMMAND}:9093,PLAINTEXT://9092

If the required advertised port is not static, it may be necessary to determine this programatically. This can be done with thePORT_COMMAND environment variable.

PORT_COMMAND: "docker port $$(hostname) 9092/tcp | cut -d: -f2"

This can be then interpolated in any otherKAFKA_XXX config using the_{PORT_COMMAND} string, i.e.

KAFKA_ADVERTISED_LISTENERS: PLAINTEXT://1.2.3.4:_{PORT_COMMAND}

It may be useful to have theKafka Documentation open, to understand the various broker listener configuration options.

Since 0.9.0, Kafka has supportedmultiple listener configurations for brokers to help support different protocols and discriminate between internal and external traffic. Later versions of Kafka have deprecatedadvertised.host.name andadvertised.port.

NOTE:advertised.host.name andadvertised.port still work as expected, but should not be used if configuring the listeners.

The example environment below:

HOSTNAME_COMMAND: curl http://169.254.169.254/latest/meta-data/public-hostnameKAFKA_ADVERTISED_LISTENERS: INSIDE://:9092,OUTSIDE://_{HOSTNAME_COMMAND}:9094KAFKA_LISTENERS: INSIDE://:9092,OUTSIDE://:9094KAFKA_LISTENER_SECURITY_PROTOCOL_MAP: INSIDE:PLAINTEXT,OUTSIDE:PLAINTEXTKAFKA_INTER_BROKER_LISTENER_NAME: INSIDE

Will result in the following broker config:

advertised.listeners = OUTSIDE://ec2-xx-xx-xxx-xx.us-west-2.compute.amazonaws.com:9094,INSIDE://:9092listeners = OUTSIDE://:9094,INSIDE://:9092inter.broker.listener.name = INSIDE

• No listeners may share a port number.
• An advertised.listener must be present by protocol name and port number in the list of listeners.

You can configure the broker rack affinity in different ways

1. explicitly, usingKAFKA_BROKER_RACK
2. via a command, usingRACK_COMMAND, e.g.RACK_COMMAND: "curl http://169.254.169.254/latest/meta-data/placement/availability-zone"

In the above example the AWS metadata service is used to put the instance's availability zone in thebroker.rack property.

For monitoring purposes you may wish to configure JMX. Additional to the standard JMX parameters, problems could arise from the underlying RMI protocol used to connect

• java.rmi.server.hostname - interface to bind listening port
• com.sun.management.jmxremote.rmi.port - The port to service RMI requests

For example, to connect to a kafka running locally (assumes exposing port 1099)

  KAFKA_JMX_OPTS: "-Dcom.sun.management.jmxremote -Dcom.sun.management.jmxremote.authenticate=false -Dcom.sun.management.jmxremote.ssl=false -Djava.rmi.server.hostname=127.0.0.1 -Dcom.sun.management.jmxremote.rmi.port=1099"  JMX_PORT: 1099

Jconsole can now connect atjconsole 192.168.99.100:1099

The listener configuration above is necessary when deploying Kafka in a Docker Swarm using an overlay network. By separating OUTSIDE and INSIDE listeners, a host can communicate with clients outside the overlay network while still benefiting from it from within the swarm.

In addition to the multiple-listener configuration, additional best practices for operating Kafka in a Docker Swarm include:

• Use "deploy: global" in a compose file to launch one and only one Kafka broker per swarm node.
• Use compose file version '3.2' (minimum Docker version 16.04) and the "long" port definition with the port in "host" mode instead of the default "ingress" load-balanced port binding. This ensures that outside requests are always routed to the correct broker. For example:

ports:   - target: 9094     published: 9094     protocol: tcp     mode: host

Older compose files using the short-version of port mapping may encounter Kafka client issues if their connection to individual brokers cannot be guaranteed.

See the included sample compose filedocker-compose-swarm.yml

See thewiki for information on adding or updating versions to release to Dockerhub.

http://wurstmeister.github.io/kafka-docker/