apache_spark

This cookbook installs and configures Apache Spark.

• GitHub: https://github.com/clearstorydata-cookbooks/apache_spark
• Chef Supermarket: https://supermarket.chef.io/cookbooks/apache_spark
• Travis CI: https://travis-ci.org/clearstorydata-cookbooks/apache_spark
• Documentation: http://clearstorydata-cookbooks.github.io/apache_spark/chef/apache_spark.html

Overview

This cookbook installs and configures Apache Spark. Currently, only the standalone deployment mode is supported. Future work:

• YARN and Mesos deployment modes
• Support installing from Cloudera and HDP Spark packages.

Compatibility

The following platforms are currently tested:

• Ubuntu 12.04
• CentOS 6.5

The following platforms are not tested but will probably work (tests coming soon):

• Fedora 21
• Ubuntu 14.04

Configuration

• node['apache_spark']['install_mode']: tarball to install from a downloaded tarball, or package to install from an OS-specific package.
• node['apache_spark']['download_url']: the URL to download Apache Spark binary distribution tarball in the tarball installation mode.
• node['apache_spark']['checksum']: SHA256 checksum for the Apache Spark binary distribution tarball.
• node['apache_spark']['pkg_name']: package name to install in the package installation mode.
• node['apache_spark']['pkg_version']: package version to install in the package installation mode.
• node['apache_spark']['install_dir']: target directory to install Spark to in the tarball installation mode. In the package mode, this must be set to the directory that the package installs Spark into.
• node['apache_spark']['install_base_dir']: in the tarball installation mode, this is where the tarball is actually extracted, and a symlink pointing to the subdirectory containing a specific Spark version is created at node['apache_spark']['install_dir'].
• node['apache_spark']['user']: UNIX user to create for running Spark.
• node['apache_spark']['group']: UNIX group to create for running Spark.
• node['apache_spark']['standalone']['master_host']: Spark standalone-mode workers will connect to this host.
• node['apache_spark']['standalone']['master_bind_ip']: the IP the master should bind to. This should be set in such a way that workers will be able to connect to the master.
• node['apache_spark']['standalone']['master_port']: the port for the Spark standalone master to listen on.
• node['apache_spark']['standalone']['master_webui_port']: Spark standalone master web UI port.
• node['apache_spark']['standalone']['worker_bind_ip']: the IP address workers bind to. They bind to all network interfaces by default.
• node['apache_spark']['standalone']['worker_webui_port']: the port for the Spark worker web UI to listen on.
• node['apache_spark']['standalone']['job_dir_days_retained']: app-... subdirectories of node['apache_spark']['standalone']['worker_work_dir'] older than this number of days will be deleted periodically on worker nodes to prevent unbounded accumulation. These directories contain Spark executor stdout/stderr logs. The directories will still be retained to honor node['apache_spark']['standalone']['job_dir_num_retained'].
• node['apache_spark']['standalone']['job_dir_num_retained']: the minimum number of Spark executor log directories (app-...) to retain, regardless of creation time.
• node['apache_spark']['standalone']['worker_dir_cleanup_log']: log file path for the Spark executor log directories cleanup script.
• node['apache_spark']['standalone']['worker_cores']: the number of "cores" (threads) to allocate on each worker node.
• node['apache_spark']['standalone']['worker_work_dir']: the directory to store Spark executor logs and Spark job jars.
• node['apache_spark']['standalone']['worker_memory_mb']: the amount of memory in MB to allocate to each worker (i.e. the maximum total memory used by different applications' executors running on a worker node).
• node['apache_spark']['standalone']['default_executor_mem_mb']: the default amount of memory to be allocated to a Spark application's executor on each node.
• node['apache_spark']['standalone']['log_dir']: the log directory for Spark masters and workers.
• node['apache_spark']['standalone']['daemon_root_logger']: the spark.root.logger property is set to this.
• node['apache_spark']['standalone']['max_num_open_files']: the maximum number of open files to set using ulimit before launching a worker.
• node['apache_spark']['standalone']['java_debug_enabled']: whether Java debugging options are to be enabled for Spark processes. Note: currently, this option is not working as intended.
• node['apache_spark']['standalone']['default_debug_port']: default Java debug port to use. A free port is chosen if this port is unavailable.
• node['apache_spark']['standalone']['master_debug_port']: default Java debug port to use for Spark masters. A free port is chosen if this port is unavailable.
• node['apache_spark']['standalone']['worker_debug_port']: default Java debug port to use for Spark workers. A free port is chosen if this port is unavailable.
• node['apache_spark']['standalone']['executor_debug_port']: default Java debug port to use for Spark standalone executors. A free port is chosen if this port is unavailable.
• node['apache_spark']['standalone']['common_extra_classpath_items']: common classpath items to add to Spark application driver and executors (but not Spark master and worker processes).
• node['apache_spark']['standalone']['worker_dir']: Set to a non-nil value to tell the spark worker to use an alternate directory for spark scratch space
• node['apache_spark']['standalone']['worker_opts']: Set to a non-nil value to pass along any additional settings to the spark worker. E.G.: -Dspark.worker.cleanup.enabled=true -Dspark.worker.cleanup.appDataTtl=86400. Ideal for worker options only that you do not want in the default configuration file.
• node['apache_spark']['conf']['...']: Spark configuration options that go into the default Spark configuration file. See https://spark.apache.org/docs/latest/configuration.html for details.
• node['apache_spark']['standalone']['local_dirs']: a list of local directories to use on workers. This is where map output files are stored, so these directories should have enough space available.
• node['apache_spark']['standalone']['ha_recovery_mode']: This setting can be used to run spark master in HA mode. More details in: http://spark.apache.org/docs/latest/spark-standalone.html#standby-masters-with-zookeeper
• node['apache_spark']['standalone']['ha_zookeeper_url']: The ZooKeeper cluster url (e.g., 192.168.1.100:2181,192.168.1.101:2181).
• node['apache_spark']['standalone']['ha_zookeeper_dir']: The directory in ZooKeeper to store recovery state
• node['apache_spark']['standalone']['master_url']: For HA via zookeeper the spark url needs to point to more than one master hosts If this is not defined it is derived from master_host and master_port
• node['apache_spark']['standalone']['worker_jmx_enabled']: true|false, to enable or disable JMX for worker nodes
• node['apache_spark']['standalone']['worker_jmx_port']: JMX port number for worker
• node['apache_spark']['standalone']['worker_jmx_authenticate']: Whether JMX requires authentication or not for worker
• node['apache_spark']['standalone']['worker_jmx_ssl']: Whether JMX requires SSL or not for worker
• node['apache_spark']['standalone']['master_jmx_enabled']: true|false, to enable or disable JMX for master nodes
• node['apache_spark']['standalone']['master_jmx_port']: JMX port number for master
• node['apache_spark']['standalone']['master_jmx_authenticate']: Whether JMX requires authentication or not for master
• node['apache_spark']['standalone']['master_jmx_ssl']: Whether JMX requires SSL or not for master

Testing

ChefSpec

bundle install
bundle exec rspec

Test Kitchen

bundle install
bundle exec kitchen test

Contributing

If you would like to contribute this cookbook's development, please follow the steps below:

• Fork this repository on GitHub
• Make your changes
• Run tests
• Submit a pull request

License

Apache License 2.0

https://www.apache.org/licenses/LICENSE-2.0