1. Brian O‟Neill, Lead Architect, Health Market Science
bone@alumni.brown.edu
@boneill42

2.  Background
 Setup
 Data Model / Schema
 Naughty List (Astyanax)
 Toy List (CQL)

3. Our Problem
 Good, bad doctors? Dead doctors?
 Prescriber eligibility and remediation.

4. The World-Wide
Globally Scalable
Naughty List!
 How about a Naughty and
Nice list for Santa?
 1.9 billion children
 That will fit in a single row!
 Queries to support:
 Children can login and check
their standing.
 Santa can find nice children
by country, state or zip.

6. Installation
 As easy as…
 Download
http://cassandra.apache.org/download/
 Uncompress
tar -xvzf apache-cassandra-1.2.0-beta3-bin.tar.gz
 Run
bin/cassandra –f
(-f puts it in foreground)

7. Configuration
 conf/cassandra.yaml
start_native_transport: true // CHANGE THIS TO TRUE
commitlog_directory: /var/lib/cassandra/commitlog
 conf/log4j-server.properties
log4j.appender.R.File=/var/log/cassandra/system.log

8. Data Model
 Schema (a.k.a. Keyspace)
 Table (a.k.a. Column Family)
 Row
 Have arbitrary #‟s of columns
 Validator for keys (e.g. UTF8Type)
 Column
 Validator for values and keys
 Comparator for keys (e.g. DateType or BYOC)
(http://www.youtube.com/watch?v=bKfND4woylw)

9. Distributed Architecture
 Nodes form a token ring.
 Nodes partition the ring by initial token
 initial_token: (in cassandra.yaml)
 Partitioners map row keys to tokens.
 Usually randomly, to evenly distribute the data
 All columns for a row are stored together on disk
in sorted order.

10. Visually
Row Hash Token/Hash Range : 0-99
Alice 50
Bob 3
Eve 15
(1-33)

11. Java Interpretation
 Each table is a Distributed HashMap
 Each row is a SortedMap.
Cassandra provides a massively scalable version of:
HashMap<rowKey, SortedMap<columnKey, columnValue>
 Implications:
 Direct row fetch is fast.
 Searching a range of rows can be costly.
 Searching a range of columns is cheap.

13. Two Tables
 Children Table
 Store all the children in the world.
 One row per child.
 One column per attribute.
 NaughtyOrNice Table
 Supports the queries we anticipate
 Wide-Row Strategy

14. Details of the NaughtyOrNice
List
 One row per standing:country
 Ensures all children in a country are grouped together
on disk.
 One column per child using a compound key
 Ensures the columns are sorted to support our search
at varying levels of granularity
○ e.g. All nice children in the US.
○ e.g. All naughty children in PA.

15. Visually Nice:USA
Node 1 CA:94333:johny.b.good
(1) Go to the row. CA:94333:richie.rich
(2) Get the column slice
Nice:IRL
Node 2 D:EI33:collin.oneill
Watch out for: D:EI33:owen.oneill
• Hot spotting
• Unbalanced Clusters
Nice:USA
CA:94111:bart.simpson
Node 3
CA:94222:dennis.menace
PA:18964:michael.myers

16. Our Schema
 bin/cqlsh -3
 CREATE KEYSPACE northpole WITH replication = {'class':'SimpleStrategy',
'replication_factor':1};
 create table children ( childId varchar, firstName varchar, lastName varchar, timezone varchar,
country varchar, state varchar, zip varchar, primary key (childId ) ) WITH COMPACT STORAGE;
 create table naughtyOrNiceList ( standingByZone varchar, country varchar, state varchar, zip
varchar, childId varchar, primary key (standingByZone, country, state, zip, childId) );
 bin/cassandra-cli
 (the “old school” interface)

17. The CQL->Data Model
Rules
 First primary key becomes the rowkey.
 Subsequent components of the primary key
form a composite column name.
 One column is then written for each non-
primary key column.

18. CQL View
cqlsh:northpole> select * from naughtyornicelist ;
standingbycountry | state | zip | childid
-------------------+-------+-------+---------------
naughty:USA | CA | 94111 | bart.simpson
naughty:USA | CA | 94222 | dennis.menace
nice:IRL | D | EI33 | collin.oneill
nice:IRL | D | EI33 | owen.oneill
nice:USA | CA | 94333 | johny.b.good
nice:USA | CA | 94333 | richie.rich

19. CLI View
[default@northpole] list naughtyornicelist;
Using default limit of 100
Using default column limit of 100
-------------------
RowKey: naughty:USA
=> (column=CA:94111:bart.simpson:, value=, timestamp=1355168971612000)
=> (column=CA:94222:dennis.menace:, value=, timestamp=1355168971614000)
-------------------
RowKey: nice:IRL
=> (column=D:EI33:collin.oneill:, value=, timestamp=1355168971604000)
=> (column=D:EI33:owen.oneill:, value=, timestamp=1355168971601000)
-------------------
RowKey: nice:USA
=> (column=CA:94333:johny.b.good:, value=, timestamp=1355168971610000)
=> (column=CA:94333:richie.rich:, value=, timestamp=1355168971606000)

20. Data Model Implications
select * from children where childid='owen.oneill';
select * from naughtyornicelist where childid='owen.oneill';
Bad Request:
select * from naughtyornicelist where
standingbycountry='nice:IRL' and state='D' and zip='EI33'
and childid='owen.oneill';

22. No, seriously. Let‟s code!
 What API should we use?
Production- Potential Momentum
Readiness
Thrift 10 -1 -1
Hector 10 8 8
Astyanax 8 9 10
Kundera (JPA) 6 9 9
Pelops 7 6 7
Firebrand 8 10 8
PlayORM 5 8 7
GORA 6 9 7
CQL Driver ? ? ?
Asytanax FTW!

23. Connect
this.astyanaxContext = new AstyanaxContext.Builder()
.forCluster("ClusterName")
.forKeyspace(keyspace)
.withAstyanaxConfiguration(…)
.withConnectionPoolConfiguration(…)
.buildKeyspace(ThriftFamilyFactory.getInstance());
 Specify:
 Cluster Name (arbitrary identifier)
 Keyspace
 Node Discovery Method
 Connection Pool Information



24. Write/Update
MutationBatch mutation = keyspace.prepareMutationBatch();
columnFamily = new ColumnFamily<String, String>(columnFamilyName,
StringSerializer.get(), StringSerializer.get());
mutation.withRow(columnFamily, rowKey)
.putColumn(entry.getKey(), entry.getValue(), null);
mutation.execute();
 Process:
 Create a mutation
 Specify the Column Family with Serializers
 Put your columns.
 Execute

25. Composite Types
 Composite (a.k.a. Compound)
public class ListEntry {
@Component(ordinal = 0)
public String state;
@Component(ordinal = 1)
public String zip;
@Component(ordinal = 2)
public String childId;
}

26. Range Builders
range = entitySerializer.buildRange()
.withPrefix(state)
.greaterThanEquals("")
.lessThanEquals("99999");
Then...
.withColumnRange(range).execute();

28. CQL Collections!
http://www.datastax.com/dev/blog/cql3_collections
 Set
 UPDATE users SET emails = emails + {'fb@friendsofmordor.org'} WHERE
user_id = 'frodo';
 List
 UPDATE users SET top_places = [ 'the shire' ] + top_places WHERE
user_id = 'frodo';
 Maps
 UPDATE users SET todo['2012-10-2 12:10'] = 'die' WHERE user_id =
'frodo';

29. CQL vs. Thrift
http://www.datastax.com/dev/blog/thrift-to-cql3
 Thrift is legacy API on which all of the Java
APIs are built.
 CQL is the new native protocol and driver.

30. Let‟s get back to cranking…
 Recreate the schema (to be CQL friendly)
 UPDATE children SET toys = toys + [ „legos' ] WHERE childId = ‟owen.oneill‟;
 Crank out a Dao layer to use CQL collections
operations.

31. Shameless Shoutout(s)
 Virgil
 https://github.com/boneill42/virgil
 REST interface for Cassandra
 https://github.com/boneill42/storm-cassandra
 Distributed Processing on Cassandra
 (Webinar in January)