Impala

Table of Contents

1. Module Description - What the module does and why it is useful
2. Setup - The basics of getting started with impala
   • What impala affects
   • Setup requirements
3. Usage - Configuration options and additional functionality
   • Beginning with Impala
   • Cluster with more HDFS Name nodes
4. Reference - An under-the-hood peek at what the module is doing and how
   • Module Parameters (impala class)
5. Limitations - OS compatibility, etc.
6. Development - Guide for contributing to the module

Module Description

Management of Cloudera Impala cluster. All services can be separated across all nodes or collocated as needed.

Puppet >= 3.x is required.

Supported are:

• Debian 7/wheezy: tested with CDH 5.4.7/5.5.0, Impala 2.2.0/2.3.0
• Ubuntu 14/trusty
• RHEL 6 and clones

Security is not working (IMPALA-2645).

Setup

What impala affects

• Packages: installs Impala packages as needed
• Files modified:
  • /etc/impala/conf/*
  • /etc/default/impala
  • /usr/local/sbin/impmanager (when requested in features)
• Alternatives:
  • alternatives are used for /etc/impala/conf
  • this module switches to the new alternative by default, so the original configuration can be kept intact
• Helper files:
  • /var/lib/impala/.puppet-*
• Users:
  • impala system user and group are created (from packages or by puppet)
• Secret Files:
  • permissions of keytab are modified (default location: /etc/security/keytabs/impala.service.keytab)
• Services: setup and start Impala services as needed (catalog, server, statestore)

Setup Requirements

Hadoop cluster

Hadoop cluster is required. This puppet module should be applied after the Hadoop cluster is orchestrated. For example when using cesnet-hadoop module:

Class['::hadoop::common::hdfs::config'] -> Class['::impala::common::config']
Class['::hbase::common::config'] -> Class['::impala::common::config']

Some settings of Hadoop cluster are required:

• dfs.datanode.hdfs-blocks-metadata.enabled=true
• dfs.client.read.shortcircuit=true
• dfs.domain.socket.path set to location writable by datanodes

Native Hadoop library should be installed.

It is not recommended to install impala nodes on HDFS Name Node: installing on Name Nodes provides no additional data locality, and executing queries with such a configuration might cause memory contention and negatively impact the HDFS Name Node.

Hive

Hive metastore service is required. You can use cesnet-hive puppet module for it (hive and hive::metastore classes), recommended with MySQL or PostgreSQL database backends (puppetlabs-postgresql or puppetlabs-mysql puppet modules for example).

Hive group (group ownership of HDFS directory /user/hive/warehouse) must be assigned to the impala user.

This is handled automatically. Just make sure the group parameter has the proper value (the same as used in hive), defaults are OK.

1. In deployment with Sentry: we're is using 'hive' group by default and impala is assigned to that group:

   usermod -G hive -a impala

2. In deployment without Sentry: we're using 'users' group by default and impala is assigned to that group:

   usermod -G users -a impala

Usage

Beginning with Impala

Basic example:

class { 'impala':
}

include ::impala::frontend
include ::impala::statestore
include ::impala::catalog
include ::impala::server

Cluster example:

class { 'impala':
  catalog_hostname    => 'master.example.com',
  statestore_hostname => 'master.example.com',
  servers             => ['node1.example.com', 'node2.example.com', 'node3.example.com'],
}

node example.com {
  include ::impala::frontend
}

node master.example.com {
  include ::impala::statestore
  include ::impala::catalog
}

node /node\d+\.example\.com/ {
  include ::impala::server
}

Cluster with more HDFS Name nodes

If there are used more HDFS Name Nodes in the Hadoop cluster (high availability, namespaces, ...), it is needed to have 'impala' system user on all of them to authorization work properly. You could install Impala daemons (using impala::server), but just creating the user is enough (using impala::user).

Note, the impala::hdfs class is available too. It is not needed to use this class (it is here only for similarity with other addons), but it includes the impala::user, so impala::hdfs can be called instead.

Example:

node <HDFS_NAMENODE> {
  include ::impala::hdfs
}

node <HDFS_OTHER_NAMENODE> {
  include ::impala::user
}

Reference

• impala
• impala::catalog: Impala Catalog (one in the cluster)
• impala::catalog::config
• impala::catalog::install
• impala::catalog::service
• impala::common::config
• impala::common::daemon
• impala::common::install
• impala::common::postinstall
• impala::frontend: Impala Frontend (on client)
• impala::frontend::config
• impala::frontend::install
• impala::hdfs: HDFS initialization
• impala::params
• impala::server: Impala Server (mostly on each Hadoop Data Node)
• impala::server::config
• impala::server::install
• impala::server::service
• impala::statestore: Impala Statestore (one in the cluster)
• impala::statestore::config
• impala::statestore::install
• impala::statestore::service
• impala::user: Create impala system user, if needed

impala class

Configure Impala Puppet Module.

alternatives

Switches the alternatives used for the configuration. Default: 'cluster' (Debian) or undef.

It can be used only when supported (for example with Cloudera distribution).

catalog_hostname

Default: $::fqdn.

debug_enable

Install also debug package and enable core dumps. Default: false.

group

Hive group on HDFS. Default: 'users' (without sentry), 'hive' (with sentry).

impala user is added to this group.

features

Enable additional features. Default: {}.

Available features:

• manager: script in /usr/local to start/stop all daemons relevant for given node

keytab

Impala keytab file for catalog and statestore services. Default: '/etc/security/keytab/impala.service.keytab'.

It must contain impala principal:

• impala/<HOSTNAME>@<REALM>

parameters

Daemon parameters to set. Default: undef.

Value is a hash with catalog, server, and statestore keys.

realm

Kerberos realm. Default: undef.

Non-empty value enables Kerberos security.

servers

Array of Impala server hostnames. Default: [$::fqdn].

statestore_hostname

Statestore service hostname. Default: $::fqdn.

udf_enable

Installs also headers on the frontend for development of user defined queries. Default: true.

Limitations

Hadoop cluster and Hive metastore is required. See Setup Requirements.

This module setup Impala cluster and tries to not limit generic usage by doing other stuff. You can have your own repository with Hadoop SW, you can select which Kerberos implementation, or Java version to use. All of that you will probably already have for the Hadoop cluster.

Security is not working yet and usage with sentry has not been tested.

Development

• Repository: https://github.com/MetaCenterCloudPuppet/cesnet-impala
• Issues: https://github.com/MetaCenterCloudPuppet/cesnet-impala/issues
• Tests: see .travis.yml
• Email: František Dvořák <valtri@civ.zcu.cz>