Forge Home

redis_multinode

Puppet module for building a multi-node Redis failover group with Sentinel

10,600 downloads

9,568 latest version

1.1 quality score

We run a couple of automated
scans to help you access a
module's quality. Each module is
given a score based on how well
the author has formatted their
code and documentation and
modules are also checked for
malware using VirusTotal.

Please note, the information below
is for guidance only and neither of
these methods should be considered
an endorsement by Puppet.

Version information

  • 0.6.0 (latest)
  • 0.5.0
  • 0.2.0
  • 0.1.1
  • 0.1.0
released Aug 2nd 2014
This version is compatible with:
  • Puppet 3.x
  • , , ,

Start using this module

  • r10k or Code Manager
  • Bolt
  • Manual installation
  • Direct download

Add this module to your Puppetfile:

mod 'shanemadden-redis_multinode', '0.6.0'
Learn more about managing modules with a Puppetfile

Add this module to your Bolt project:

bolt module add shanemadden-redis_multinode
Learn more about using this module with an existing project

Manually install this module globally with Puppet module tool:

puppet module install shanemadden-redis_multinode --version 0.6.0

Direct download is not typically how you would use a Puppet module to manage your infrastructure, but you may want to download the module in order to inspect the code.

Download

Documentation

shanemadden/redis_multinode — version 0.6.0 Aug 2nd 2014

redis_multinode

This module is intended to deploy a group of systems running Redis instances, with replication and failover managed by Redis Sentinel.

Since most Redis client stacks don't have any mechanism for finding the master in a group like this, this module also configures HAProxy, with a listening port on each node which will proxy to the writable node. (Tracking the master is done by a Python script running on a cron every minute)

So, for each instance, there will be one listening port which is the Redis listener (for instance, 6379) which will always be available for read-only operations, and may be able to accept write operations if that node's the master, as well as an HAProxy listener (for instance, 6380) which will track the writable node for the instance, on whichever node it resides.

Usage

A note on the instance configuration: you'll be setting a mode for each system; master, or slave (which should then have a master_ip of the node configured as master). Note that this is just for the initial setup - if/when Sentinel fails your instance over to another node due to the master failing, the module makes no attempt to try to enfore which node holds master.

Hiera

I use this with Hiera handling node classifications (if you haven't looked at Hiera, you should!), so it's built to be pleasant if you do the same. For instance, to set up a node group with two separate instances of Redis for two applications, you could do something like this.

common.yaml

redis_multinode::test::listen_reader   : 6379
redis_multinode::test::listen_writer   : 6380
redis_multinode::test::password        : changeme
redis_multinode::test::quorum          : 2

redis_multinode::my_app::listen_reader : 6381
redis_multinode::my_app::listen_writer : 6382
redis_multinode::my_app::password      : 87W98XqulD
redis_multinode::my_app::quorum        : 2

redis1.example.com.yaml

classes:
  - redis_multinode

redis_multinode::instances:
  - test
  - my_app

redis_multinode::version               : 2.8.3
redis_multinode::test::role            : master
redis_multinode::my_app::role          : master

redis2.example.com.yaml (and redis3.example.com.yaml)

classes:
  - redis_multinode

redis_multinode::instances:
  - test
  - my_app

redis_multinode::version               : 2.8.3
redis_multinode::test::role            : slave
redis_multinode::test::master_ip       : 10.0.50.10
redis_multinode::my_app::role          : slave
redis_multinode::my_app::master_ip     : 10.0.50.10

Note that since we have 3 nodes participating, the quorum is set to 2. To avoid a split-brain scenario, the quorum setting must always be configured to a majority of your cluster nodes - so in a 10 node group, quorum should be 6.

Standard Node Definitions

This module also works with standard node definitions.

node 'redis1.example.com' {
  class { "redis_multinode":
    version       => "2.8.3",
  }
  redis_multinode::instance { "test":
    role          => master,
    listen_reader => 6379,
    listen_writer => 6380,
    password      => "changeme",
    quorum        => 2,
  }
  redis_multinode::instance { "my_app":
    role          => master,
    listen_reader => 6381,
    listen_writer => 6382,
    password      => "87W98XqulD",
    quorum        => 2,
  }
}

node 'redis2.example.com', 'redis3.example.com' {
  class { "redis_multinode":
    version       => "2.8.3",
  }
  redis_multinode::instance { "test":
    role          => slave,
    master_ip     => "10.0.50.10"
    listen_reader => 6379,
    listen_writer => 6380,
    password      => "changeme",
    quorum        => 2,
  }
  redis_multinode::instance { "my_app":
    role          => slave,
    master_ip     => "10.0.50.10"
    listen_reader => 6381,
    listen_writer => 6382,
    password      => "87W98XqulD",
    quorum        => 2,
  }
}

Note that since we have 3 nodes participating, the quorum is set to 2. To avoid a split-brain scenario, the quorum setting must always be configured to a majority of your cluster nodes - so in a 10 node group, quorum should be 6.

Important Notes

  • Make sure not to overlap listening ports - keep in mind that each instance gets two listening ports - one for Redis, and one for HAProxy pointing to the master node.

  • The listening ports configuration, as well as the password, should be the same for each node participating in a group for a given instance.. so in the examples above, if the test instance were only configured on one node of the three, you'd still want the ports and password for all nodes with my_app configured to match.

  • Since the master/slave relationship is handled by Sentinel, Puppet can't have complete control over the config files. Because of this, changing the configuration of an instance (the resource title, the listening ports, or the password) is likely to break things - the Redis instance will be reconfigured but the Sentinel instances will end up either configured incorrectly or with duplicate configuration.

  • Tested on RHEL/CentOS 6 (requires EPEL to be enabled) and Ubuntu 13.10 - should work most reasonably recent Debian and EL derivatives. Probably won't work under Puppet versions older than 3.0 without some modifications due to the use of the built-in Hiera functions.