Jump to: navigation, search

Difference between revisions of "Trove/cassandra-clustering"

(Impacts)
(Justification/Benefits)
 
(6 intermediate revisions by the same user not shown)
Line 7: Line 7:
  
 
See more info about Cassandra clusters at http://planetcassandra.org/blog/introduction-to-cassandra-clusters/
 
See more info about Cassandra clusters at http://planetcassandra.org/blog/introduction-to-cassandra-clusters/
 +
 +
See more info about cluster deployment at http://pulasthisupun.blogspot.com/2013/12/setting-up-local-3-node-cassandra.html
  
 
== Justification/Benefits ==
 
== Justification/Benefits ==
Cassandra is the easiest to deploy-and-maintain datastore in the NoSQL world.  
+
There are alot overcloud projects that are requesting for NoSQL backend. This is perfect place to propose Trove as deployment tool, but before that we need to cover customer use cases with NoSQL deployment.
 +
From customer and maintainer point of view - Cassandra is the easiest to deploy-and-maintain datastore in the NoSQL eco-system.
  
Since Trove provides framework that allows to implement clustering for certain datastores..
+
Since Trove provides framework that allows to implement clustering for certain datastores,
 +
it would be great value-add for project to allow deployment of CassandraDB clusters through Trove.
  
It would be "nice to have" such ability to deploy C* clusters through Trove.
+
Also, there are several projects, from OpenStack ecosystem, such as Ceilometer and MagnetoDB that may work with Cassandra as well. See
  
See more info at:
+
*https://blueprints.launchpad.net/ceilometer/+spec/support-magnetodb
** http://www.enterprisetech.com/2014/03/20/google-makes-cassandra-nosql-scream-compute-engine/
+
*https://blueprints.launchpad.net/ceilometer/+spec/cassandra-driver
 +
*https://github.com/stackforge/magnetodb/tree/master/magnetodb/storage/driver/cassandra
 +
 
 +
See more info at http://www.enterprisetech.com/2014/03/20/google-makes-cassandra-nosql-scream-compute-engine/
  
 
== Impacts ==
 
== Impacts ==
Line 26: Line 33:
  
  
HOW-TO:
+
We need to add another one cluster task for adding node (existing "add shard" task might work but it would be nice to have consistent states for each datastore), for cassandra we  need to add:
*setup cassandra cluster: http://pulasthisupun.blogspot.com/2013/12/setting-up-local-3-node-cassandra.html
+
 
 +
* "Add node" cluster task
  
 
=== Configuration ===  
 
=== Configuration ===  
 
For Cassandra clustering we need to have next options:
 
For Cassandra clustering we need to have next options:
  
**cluster_support - Boolean
+
*cluster_support - Boolean
  
**api_strategy - fully qualified class name of aAPI strategy implementation
+
*api_strategy - fully qualified class name of aAPI strategy implementation
  
**taskmanager_strategy - fully qualified class name of a Taskmanager implementation
+
*taskmanager_strategy - fully qualified class name of a Taskmanager implementation
  
**guestagent_strategy - fully qualified class name of a Guestagent implementation
+
*guestagent_strategy - fully qualified class name of a Guestagent implementation
  
 
=== Database ===
 
=== Database ===

Latest revision as of 12:34, 15 September 2014

Cassandra clustering

Description

This spec describes Cassandra clustering feature development.

See more info about Cassandra itself at http://cassandra.apache.org/

See more info about Cassandra clusters at http://planetcassandra.org/blog/introduction-to-cassandra-clusters/

See more info about cluster deployment at http://pulasthisupun.blogspot.com/2013/12/setting-up-local-3-node-cassandra.html

Justification/Benefits

There are alot overcloud projects that are requesting for NoSQL backend. This is perfect place to propose Trove as deployment tool, but before that we need to cover customer use cases with NoSQL deployment. From customer and maintainer point of view - Cassandra is the easiest to deploy-and-maintain datastore in the NoSQL eco-system.

Since Trove provides framework that allows to implement clustering for certain datastores, it would be great value-add for project to allow deployment of CassandraDB clusters through Trove.

Also, there are several projects, from OpenStack ecosystem, such as Ceilometer and MagnetoDB that may work with Cassandra as well. See

See more info at http://www.enterprisetech.com/2014/03/20/google-makes-cassandra-nosql-scream-compute-engine/

Impacts

To implement cassandra clustering we need to add 3 new strategies:

  • API strategy implementation
  • Taskmanager implementation
  • Guestagent implementation


We need to add another one cluster task for adding node (existing "add shard" task might work but it would be nice to have consistent states for each datastore), for cassandra we need to add:

  • "Add node" cluster task

Configuration

For Cassandra clustering we need to have next options:

  • cluster_support - Boolean
  • api_strategy - fully qualified class name of aAPI strategy implementation
  • taskmanager_strategy - fully qualified class name of a Taskmanager implementation
  • guestagent_strategy - fully qualified class name of a Guestagent implementation

Database

None

Public API

None

CLI interface

None

ReST Part

None

Internal API

None

RPC API description

None

Guest Agent

According to base guestagent strategy we need to add next calls:

  • Name:
    • add_members
  • Type:
    • call
  • Payload:
    • members
      • a list of instances to join the ring.