This document summarizes a presentation about using the HBase database with Ruby on Rails applications. HBase is a distributed, column-oriented database that is well-suited for large datasets and integrates with Hadoop. The presentation covers what HBase is, its tradeoffs compared to relational databases, when it may be a good or bad fit, and examples of using HBase with Rails through libraries like hbase-stargate and MassiveRecord. It also includes a demo of running JRuby scripts against HBase and building a Rails frontend.

1. Rails on HBase
Zachary Pinter and Tony Hillerson
RailsConf 2011

2. What we will cover

3. What is it?

4. What are the
tradeoffs that HBase
makes?

5. Why HBase is
probably the wrong
choice for your app

6. Why HBase might be
the right choice for
your app

7. +
How to use HBase
with Rails

8. What we won’t cover

9. This is not a deep
dive

10. This is not a guide to
setting up a
production HBase
cluster
*See http://cloudera.com

11. So...
What is HBase?

12. HBase is based on
Bigtable

13. HBase Grew out of
Hadoop

14. Core Concepts

15. Distributed Column
Oriented Database

16. Tables
column
key core:link
row
llb0ga http://google.com
http://en.oreilly.com/rails2011/public/
llb260
schedule/detail/17886

17. Column Families
core column family stats column family
key core:link stats:hits
llb0ga http://google.com 100504
http://en.oreilly.com/rails2011/
llb260 2
public/schedule/detail/17886

18. Column Families
Must be defined at creation or added to schema
later
Should group data accessed and sized similarly

19. Columns
Can be dynamically added given that the family
exists
Stored as a byte array - do what you need with it

20. Column Families:
Data Locality

21. Regions
master
llb0ga lld92q
... ...
llbee9 lq1rro
Regionserver Regionserver

22. Main Operations
Get
Put
Incr
Delete
Scan

23. JRuby based shell

24. hbase shell
ruby-1.9.2-p180 :002 > status
1 servers, 0 dead, 1.0000 average load
ruby-1.9.2-p180 :003 > (1..10).each{ status }
1 servers, 0 dead, 3.0000 average load
1 servers, 0 dead, 3.0000 average load
1 servers, 0 dead, 3.0000 average load
1 servers, 0 dead, 3.0000 average load
1 servers, 0 dead, 3.0000 average load
1 servers, 0 dead, 3.0000 average load
1 servers, 0 dead, 3.0000 average load
1 servers, 0 dead, 3.0000 average load
1 servers, 0 dead, 3.0000 average load
1 servers, 0 dead, 3.0000 average load
=> 1..10
ruby-1.9.2-p180 :004 >

25. API
REST
Thrift
Apache Avro
Java API

26. What are the tradeoffs that
HBase makes?
NoSQL = Tradeoffs

27. Highly Distributed
Pros Cons
• Built-in sharding • Makes joins difficult/impossible
• No single server holds all the data • Limited real-time queries/sorting
• Different servers operate on different
slices of the data

28. Sorted Row Keys
Bytearray keys
Any kind of data you want
Sorted in byte order
All row access through the key

29. Sorted Row Keys
Pros Cons
• Real-time random lookups by key • Conflates unique identifier with default
• Real-time range queries sort
• Keys are an integral part of schema
design, requiring extra consideration

30. Map / Reduce
Pros Cons
• Scales linearly: more servers = faster • Scans every row per query *
response times • Not real-time (batch operation)
• Operates on huge datasets (billions of
rows)
• Works well with dynamic/non-uniform
schemas

31. HBase vs. X
http://nosql-database.org
http://www.sevenweeks.org/
Seven Databases in Seven Weeks
Eric Redmond
Don’t Miss His Talk!!

32. Why HBase is
probably the wrong
choice for your app

33. HBase Limitations

34. HBase Limitations
Limited real-time queries (key lookup, range lookup)
Limited sorting (one default sort per table)
Limited transactions
Manual indexing
Joins/Normalization
Storage of large binary data *

35. HBase Requirements
• At least 5 Servers (Zookeeper, Region Server, Name Nodes, Data Nodes, etc)
• Memory Intensive (2-4gb bar minimum per server, depending on server role)
• CPU Intensive
• Rough EC2 Cost:
5 c1.xlarge reserved for 1 year
= over $850/month
(not counting bandwidth!)

36. Why HBase might be
the right choice for
your app

37. Large data sets
Ability to aggregate and analyze billions of rows
Starts to shine when relational databases break
down
Is your db already sharded?
Are you using SQL queries that take hours to run?

38. Hadoop Integration
HDFS for storing files
Native Map/Reduce
Supports data locality
Doesn’t require data to be transferred

39. Flexible schema
Supports non-heterogeneous data through column
families with mixed key/value pairs
Ability to refactor the data store via map/reduce
(beats a multi-day ALTER TABLE command)

40. Rails on HBase

41. hbase-stargate
https://github.com/greglu/hbase-stargate
Uses Stargate, HBase REST server

42. Rhino
https://github.com/sqs/rhino/
Thrift API
Seems to be abandoned?

43. Massive Record
https://github.com/CompanyBook/massive_record
It works!
Uses Thrift API

44. Kwisatz: An URL
Shortener

45. Demo:
JRuby Scripts

46. Demo:
Rails Front-End

47. Source
Demo source code
github.com/effectiveui/jruby-hbase-demo
github.com/effectiveui/rails-hbase-massive-record

48. Getting Set Up
(Mac) brew install hbase
Cloudera VM
http://bit.ly/dSfPa9

49. Recap

50. Where Would I Use
HBase?

51. Where is HBase’s
Sweet Spot

52. Where Would I Use
Rails With HBase?

53. Questions?

54. Thank You!
Zachary: @zpinter
Tony: @thillerson
EffectiveUI.com