HBaseCon 2014-Just the Basics

1

HBase: Just the BasicsJesse Anderson – Curriculum Developer and Instructor

v2

2 ©2014 Cloudera, Inc. All rights reserved.2

What Is HBase?

• NoSQL datastore built on top of HDFS (Hadoop)• An Apache Top Level Project• Handles the various manifestations of Big Data• Based on Google’s BigTable paper

3 ©2014 Cloudera, Inc. All rights reserved.3

Why Use HBase?

• Storing large amounts of data (TB/PB)• High throughput for a large number of requests• Storing unstructured or variable column data• Big Data with random read and writes

4 ©2014 Cloudera, Inc. All rights reserved.4

When to Consider Not Using HBase?

• Only use with Big Data problems• Read straight through files• Write all at once or append new files

• Not random reads or writes• Access patterns of the data are ill-defined

5

HBase ArchitectureHow it works

6 ©2014 Cloudera, Inc. All rights reserved.6

Meet the Daemons

• HBase Master• RegionServer• ZooKeeper• HDFS

• NameNode/Standby NameNode• DataNode

7 ©2014 Cloudera, Inc. All rights reserved.7

Daemon Locations

Master Nodes

Slave Nodes

8 ©2014 Cloudera, Inc. All rights reserved.8

Tables and Column Families

Column Family “contactinfo” Column Family “profilephoto”

Tables are broken into groupings called Column Families.

Group data frequently accessed together and compress it Group photos with different settings

9 ©2014 Cloudera, Inc. All rights reserved.9

Rows and Columns

Row key Column Family “contactinfo” Column Family “profilephoto”adupont fname: Andre lname: Dupontjsmith fname: John lname: Smith image: <smith.jpg>mrossi fname: Mario lname: Rossi image: <mario.jpg>

Row keys identify a row

No storage penalty for unused columns

Each Column Family can have many columns

10 ©2014 Cloudera, Inc. All rights reserved.10

Regions

Row key Column Family “contactinfo”adupont fname: Andre lname: Dupontjsmith fname: John lname: Smith

A table is broken into regions

Row key Column Family “contactinfo”

mrossi fname: Mario lname: Rossi

zstevens fname: Zack lname: Stevens

Regions are served by RegionServers

11 ©2014 Cloudera, Inc. All rights reserved.11

Write Path

1. Which RegionServer is serving the Region?

2. Write to RegionServer

12 ©2014 Cloudera, Inc. All rights reserved.12

Read Path

1. Which RegionServer is serving the Region?

2. Read from RegionServer

13

HBase APIHow to access the data

14 ©2014 Cloudera, Inc. All rights reserved.14

No SQL Means No SQL

• Data is not accessed over SQL• You must:

• Create your own connections• Keep track of the type of data in a column• Give each row a key• Access a row by its key

15 ©2014 Cloudera, Inc. All rights reserved.15

Types of Access

• Gets• Gets a row’s data based on the row key

• Puts• Upserts a row with data based on the row key

• Scans• Finds all matching rows based on the row key• Scan logic can be increased by using filters

16 ©2014 Cloudera, Inc. All rights reserved.16

Gets

123

4

Get g = new Get(ROW_KEY_BYTES); Result r= table.get(g);byte[] byteArray =

r.getValue(COLFAM_BYTS,COLDESC_BYTS);

String columnValue = Bytes.toString(byteArray);

17 ©2014 Cloudera, Inc. All rights reserved.17

Puts

12

3

4

Put p = new Put(ROW_KEY_BYTES);p.add(COLFAM_BYTES, COLDESC_BYTES, Bytes.toBytes("value"));

table.put(p);

18

HBase Schema DesignHow to design

19 ©2014 Cloudera, Inc. All rights reserved.19

No SQL Means No SQL

• Designing schemas for HBase requires an in-depth knowledge• Schema Design is ‘data-centric’ not ‘relationship-

centric’• You design around how data is accessed• Row keys are engineered

20

Treating HBase like a traditional RDBMS will lead to abject failure!Captain Picard

21 ©2014 Cloudera, Inc. All rights reserved.21

Row Keys

• A row key is more than the glue between two tables• Engineering time is spent just on constructing a row

key• Contents of a row key vary by access pattern• Often made up of several pieces of data:<group_id><email>

22 ©2014 Cloudera, Inc. All rights reserved.22

Schema Design

• Schema design does not start in an ERD• Access pattern must be known and ascertained• Denormalize to improve performance

• Fewer, bigger tables

23 ©2014 Cloudera, Inc. All rights reserved.

Jesse Anderson@jessetanderson