Riak at Engine Yard Cloud
0 likes2,952 views
The document discusses using Riak, an open source NoSQL database, in the cloud to provide highly available and fault tolerant storage for applications. It describes how the authors' application struggled with availability when hosted on AWS, and their research led them to choose Riak as a solution. Riak provides features like consistent hashing, hinted handoff, and active anti-entropy that allow it to scale linearly and remain available even during failures or added/reduced nodes. The document provides guidance on deploying a Riak cluster in the cloud, including choosing instance types, sizing the cluster, disabling swap, using the right filesystem and scheduler, and monitoring and scaling the cluster over time.
![Mount & SCHEDULER
^ Mount with noatime flag
Use deadline scheduler
on EBS
cat /sys/block/xvdj1/queue/scheduler
noop [deadline] cfq
echo deadline > /sys/block/xvdj1/queue/
scheduler](https://image.slidesharecdn.com/riakinthecloud-130404124546-phpapp02/85/Riak-at-Engine-Yard-Cloud-35-320.jpg)
Riak at Engine Yard Cloud
- 1. Getting Started with Riak in the Cloud Big Nerd Ranch
- 3. Our Agenda About Us Our Customers The Problem The Solution
- 4. Engine Yard
- 5. About US Clouds Regions 100% Cloud Based Stacks Support Critical services DR MUST be up HA 24/7/365
- 7. An APPLICATION App App DB Master us-east-1a DB Replica App us-east-1b
- 8. Then AWS Crashed http://www.flickr.com/photos/brandongrasley/8227882239/
- 9. Our App AFTER App App XDB Master us-east-1a DB Replica App us-east-1b
- 10. This sucked, A LOT ! Time to strengthen our HA/DR Story $ Customers and us need to be up all the time!
- 11. Our Research Data Models Consistency Partitions Relational Are ACID Consistency properties Document Availability really Key/Value needed? Column
- 12. Impedance Mismatch Relational Model
- 14. We needed a fault tolerant and highly available (clustered) datastore
- 15. OSS, Dynamo-based, distributed K/V store Hello RIAK! Linearly scalable Fault-tolerant Finally provide HA/DR solution
- 16. Riak’s Use Cases Application Type Key Value Session User/Session id Data Advertising Campaign id Data Logs Date Log file text, json, Content title, integer xml, media..
- 17. When is Riak a good fit When availability is more important than consistency When data can be modeled as keys/values When the problem fits
- 18. Riak is OPS FRIENDLY X node 1.2.1 Growth node node 1.2.1 1.2.1 Reduction node 1.2.1 node 1.2.1 Upgrades node 1.3.0 Failures
- 19. World’s most famous slide
- 20. Riak is SO MUCH MORE Consistent Hashing Hinted Handoff Active Anti Entropy Tuneable consistency N|W|R
- 21. OUR APP AFTER App node 1.3.0 App node node 1.3.0 1.3.0 node node App 1.3.0 1.3.0
- 22. We need a Riak Cluster but we have no hardware
- 23. RIAK In the Wild 18th and Harrison, San Francisco, CA
- 24. Let’s build A PRODUCT http://www.flickr.com/photos/worldbank/8261699153/
- 25. Your CHOICES You are trying to create a Riak cluster. Do you need any help? 1 Number of Nodes 2 Instance Type 3 Data Location 4 Backend Type
- 26. What we GIVE YOU
- 27. Our SUPPORT X2
- 28. HOWTO What you OPS to know
- 29. Prepare Yourself More instances to compensate for the performance variability of virtualized resources
- 30. Choosing INSTANCES ^ m1.large & m1.xlarge EBS-optimized or PIOPs High I/O Quadruple XL (hi1.4xlarge) - SSDs!
- 31. Cluster Sizes 5 nodes or more Best performance & growth Scales linearly Don’t be stingy
- 32. Ring SIZE Change Default CANNOT BE RESET Ring Size at 256 (8-64 vnodes per node) DONT GO OVER 512
- 33. Our PROVISIONING Riak Cluster Riak Cluster App MyRiakCluster: node-0 haproxy HTTP or MyRiakCluster: node-1 Protobuff App haproxy MyRiakCluster: node-2 App MyRiakCluster: node-3 haproxy MyRiakCluster: node-4
- 34. Disable SWAP ^ Riak process pages should not hit swap Swap usage can result in unresponsive server Let the kernel kill it
- 35. Mount & SCHEDULER ^ Mount with noatime flag Use deadline scheduler on EBS cat /sys/block/xvdj1/queue/scheduler noop [deadline] cfq echo deadline > /sys/block/xvdj1/queue/ scheduler
- 36. Linux FILESYSTEMS ^ xfs, ext3 ext4 barrier = 0 data = writeback ZFS not recommended
- 37. Backup STRATEGY ^ Backup both the ring and data directories scp or rsync data dir EBS snapshots
- 38. Monitoring STRATEGY Stats Interface riak-admin status Collectd Nagios
- 39. Scaling RIAK ^ Vertical riak-admin cluster replace <old> <new> ^ Horizontal riak-admin cluster add <new>
- 40. How we SCALE RIAK
- 41. Hybrid Solutions FTW RIAK + PostgreSQL RIAK + FTS Engine* RIAK + Redis
- 42. Try OUR RIAK
- 43. How difficult IS IT?
- 45. Polyglot PERSISTENCE Pick the right tool for the job Apps & Services can have multiple databases
- 46. On SOA My super nice e-commerce site Shopping Cart & Completed Inventory and Recommendations Session Data Orders Item Pricing Engine Session Order Inventory & Nodes and Storage Persistence Price service relations service service service RDBMS K/V Store Document Graph Store Store NoSQL Distilled: Fowler & Sadalage
- 47. We RECOMMEND Don’t optimize too quickly Take time to understand your problem Complexity increases with more databases
- 48. Do KEEP IN MIND Different paradigms are different Document Relational != Store
- 50. Great RESOURCES Little Riak Book (https://github.com/coderoshi/little_riak_book/) docs.basho.com blog.engineyard.com nosql.mypopescu.com
- 51. Questions?