Managing the Basho Data Platform with the Cloudsoft UX, including Riak blueprints in Apache Brooklyn and building up to tiered dynamic IoT analytics management
1. Managing the Basho Data Platform with the Cloudsoft UX
RICON Nov 2015
Alex Heneveld @ahtweetin
Mike Zaccardo @ItsMeMikeZ
Cuyler Jones (Basho)
@cuylerjones
2. @ahtweetin
Recap from 2014
Riak Blueprints in Apache Brooklyn
The easiest way to deploy and manage Riak anywhere!
Blueprints for Riak nodes and clusters
available off-the-shelf in Apache Brooklyn
Node and cluster metrics & auto-scaling policies
Supporting AWS, Azure, Openstack, GCE, Softlayer,
Docker, VMware, Virtustream, Bluebox, and more
3. @ahtweetin
What is a Brooklyn Blueprint?
location: aws-ec2
services:
- type: riak-cluster
initialSize: 5
4. @ahtweetin
What is a Brooklyn Blueprint?
location:
byon:
hosts: [ 10.11.12.{101-105} ]
services:
- type: riak-cluster
initialSize: 5
12. @ahtweetin
New for 2015 (and 2016)
Bring your own data tools blueprints
Blueprints for the Basho Data Platform
Consistent across many infrastructures,
now including Docker and Mesos
Riak Enterprise support in Cloudsoft AMP
13. @ahtweetin
The Basho Data Platform
BDP: Combines Riak with other analytics tools to give a simpler solution to a
much larger class of analysis challenges, including Riak, Cache-Proxy, and Spark
Coming soon: Riak TS and range scan queries
14. @ahtweetin
The Basho Data Platform
Redis and Cache-Proxy: Enables natural use of Redis as a read cache for Riak KV
Coming soon: Write-through support for SET, DEL, and PEXPIRE to invalidate cache
BDP: Combines Riak with other analytics tools to give a simpler solution to a
much larger class of analysis challenges, including Riak, Cache-Proxy, and Spark
Coming soon: Riak TS and range scan queries
15. @ahtweetin
The Basho Data Platform
Redis and Cache-Proxy: Enables natural use of Redis as a read cache for Riak KV
Coming soon: Write-through support for SET, DEL, and PEXPIRE to invalidate cache
Spark: Streaming optimization for Riak KV, with 1000 key-object pairs per request.
BDP Leader Election Service allowing HA without ZK, simplifying cluster operation.
BDP: Combines Riak with other analytics tools to give a simpler solution to a
much larger class of analysis challenges, including Riak, Cache-Proxy, and Spark
Coming soon: Riak TS and range scan queries
17. @ahtweetin
The Basho Data Platform
BDP typically runs with a secondary Riak cluster, alongside a primary Riak cluster.
After installing BDP on nodes, start services by running:
data-platform-admin start-service »node« »group« »service«
18. @ahtweetin
The Basho Data Platform
BDP typically runs with a secondary Riak cluster, alongside a primary Riak cluster.
After installing BDP on nodes, start services by running:
data-platform-admin start-service »node« »group« »service«
Considerations for Spark
Spark workers should be run on different nodes from Riak to avoid interference.
Start with a one-to-one correspondence between Riak nodes and Spark nodes.
Follow general Spark provisioning guidelines for Disk, Memory, CPU, and Network.
Considerations for Redis and Cache-Proxy
Redis should be run on different nodes from Riak and Spark to avoid interference.
Cache Proxy and Redis should be run together to create a mesh of cache servers.
Set CACHE_TTL to meet the Eventual Consistency SLA of the application.
Monitor usage/allocation via Redis INFO (used_memory and maxmemory).
35. @ahtweetin
The BDP and Cloudsoft AMP
One-click deployment of BDP, following best practices
Scale each sub-system, manually or automatically
Consistency across bare metal, many clouds, containers, and more
Handle replication, migration and DR
Blueprints extensible for the systems you need today
and the flexibility and agility you will need in the future