kafka cookbook

Installs and configures Kafka >= v0.8.1.

Initially based on the Kafka cookbook released by Webtrends (thanks!), but with a few notable differences:

does not depend on runit cookbook.
does not depend on zookeeper cookbook, thus it will not search for nodes with a specific role or such, that is left up to you to decide.
intended to be used by wrapper cookbooks.

Maintainers

This cookbook is maintained by the Sous Chefs. The Sous Chefs are a community of Chef cookbook maintainers working together to maintain important cookbooks. If you’d like to know more please visit sous-chefs.org or come chat with us on the Chef Community Slack in #sous-chefs.

Requirements

This cookbook does not depend on any specific cookbooks, but it requires that java is installed on the system, thus the java cookbook is recommended.

Furthermore, Kafka requires ZooKeeper for coordination, and this cookbook does not install or manage ZooKeeper to any extent. A general recommendation is to not run Kafka and ZooKeeper on the same hardware.

Chef 13+

Platform

Amazon Linux
CentOS 6 and 7
Debian 7 and 8
Fedora 21
Ubuntu 14.04

The platforms / versions listed above are the ones that are included in .kitchen.yml and/or tested in the wild, so it might work on other platforms as well, YMMV.

Attributes

In order to keep the README in some kind of manageable state (and thus in sync with attributes), attributes are documented inline (in the attribute files that is).

Attributes concerning configuration of a Kafka broker are to be set under the broker namespace, as such:

node.default['kafka']['broker']['log.dirs'] = %w[/tmp/kafka-logs]
node.default['kafka']['broker']['num.io.threads'] = 4
node.default['kafka']['broker']['zookeeper.connect'] = %w[localhost:2181]

The attribute names match the configuration names that Kafka uses to make it easier to support new and old versions of Kafka simultaneously, by avoiding "hardcoded" attribute names for configuration options, so please refer to the official documentation for the release at your hand.

Refer to the official documentation for the version of Kafka that you're installing. Documentation for the latest release can be found over here.

Recipes

This section describes the different recipes that are available.

default

Downloads and installs Kafka from the official binary releases. Defaults to installing v1.1.1 of Kafka.

Controlling restart of Kafka brokers in a cluster

Any changes made to the broker configuration could result in a restart of the Kafka broker, if the kafka.automatic_restart attribute is set to true. If Chef runs as a daemon on all of the nodes this could result in all of the Kafka brokers being brought down at the same time, resulting in unavailability of service.

If unavailability is an issue, this cookbook provides an option to implement custom logic to control the restart of Kafka brokers so that not all of the brokers in a cluster are stopped at the same time. For example the custom logic can be something along the lines of acquiring a lock in ZooKeeper and when held the broker is allowed to restart. Be aware that a restart might take quite some time if you're using controlled shutdown and have a lot of partitions, and Chef usually have some timeout for each resource.

By default the resources in the _coordinate recipe performs the start/restart of the kafka service. If custom logic needs to be implemented, this recipe can be replaced with another recipe, but don't forget to update the kafka.start_coordination.recipe attribute.

The only requirement is that the new recipe has a ruby_block resource with 'coordinate-kafka-start' as ID. The following is a sample recipe that shows roughly what one can do with this feature.

ruby_block 'coordinate-kafka-start' do
  block do
    Chef::Log.info 'Custom recipe to coordinate Kafka start/restart'
  end
  action :nothing
  notifies :create, 'ruby_block[restart-coordination]', :delayed
end

ruby_block 'restart-coordination' do
  block do
    Chef::Log.info 'Implement the process to coordinate the restart, like using ZK'
  end
  action :nothing
  # `kafka_service_resource` is a helper method that will return the correct
  # resource name depending on whether you're using `runit` or not.
  notifies :restart, kafka_service_resource, :delayed
  notifies :create, 'ruby_block[restart-coordination-cleanup]', :delayed
end

ruby_block 'restart-coordination-cleanup' do
  block do
    Chef::Log.info 'Implement any cleanup logic required after restart like releasing locks'
  end
  action :nothing
end

Please refer to issue #58 for background of this feature.

FAQ

Kafka dies for no apparent reason (ulimit)

Depending on your system / infrastructure setup you might run into issues with Kafka just sporadically dying for no obvious reason. One thing you might want to look into is the file handle limit as Kafka tend to keep a lot file handles open due to socket connections (depends on the number of brokers, producers and consumers) and the actual data log files (depends on the number of partitions and log segment and/or log roll settings).

It's possible to set a specific ulimit for Kafka using the node['kafka']['ulimit_file'] attribute. If this value is not set, Kafka will use whatever the system default is, which as stated previously might not be enough, so it might be wise to set a higher limit.

How do I get started locally? (minimal required setup)

If you want to hit the ground running and just setup a single broker (or a cluster for that matter) locally, these are the necessary broker attributes that needs to be set (assumes that you have ZooKeeper running on port 2181 locally):

node.default['kafka']['broker']['zookeeper.connect'] = 'localhost:2181'
# This shouldn't normally be necessary, but might need to be set explicitly
# if you're having trouble connecting to the brokers.
node.default['kafka']['broker']['hostname'] = '127.0.0.1' # or perhaps 'localhost'

If you plan on running a cluster locally you will want to set separate values for the following configuration options:

node.default['kafka']['broker']['broker.id'] = <id>
node.default['kafka']['broker']['port'] = <port>
node.default['kafka']['broker']['log.dirs'] = <dir path>

Other than that things should work as expected, though depending on what platform you're running on, you might want to change the install and config directories as well. See attributes/default.rb and recipes/_defaults.rb for the default path regarding directories that Kafka will use.

Kafka killed prematurely (kill timeout)

When using controlled shutdown and either systemd or upstart as init system you might run into issues with Kafka being killed before it has managed to shutdown completely, resulting in long recovery times.

Not sure if it's possible to configure either systemd or upstart to not automatically kill processes, but a workaround is to set kafka.kill_timeout to a sufficiently high value.