wmware nosql

© 2011 VMware Inc. All rights reserved

NoSQL / Spring Data

Polyglot Persistence – An introduction to Spring Data

Pronam Chatterjee

pronamc@vmware.com

2

Presentation goal

How Spring Data simplifies the

development of NoSQL

applications

3

Agenda

• Why NoSQL?• Overview of NoSQL databases• Introduction to Spring Data• Database APIs

- MongoDB- HyperSQL- Neo4J

4

Relational databases are great

• SQL = Rich, declarative query language• Database enforces referential integrity• ACID semantics• Well understood by developers• Well supported by frameworks and tools, e.g. Spring JDBC,

Hibernate, JPA• Well understood by operations• Configuration • Care and feeding• Backups• Tuning• Failure and recovery• Performance characteristics

• But….

5

The trouble with relational databases

• Object/relational impedance mismatch- Complicated to map rich domain model to relational schema

• Relational schema is rigid- Difficult to handle semi-structured data, e.g. varying attributes- Schema changes = downtime or $$

• Extremely difficult/impossible to scale writes:- Vertical scaling is limited/requires $$- Horizontal scaling is limited or requires $$

• Performance can be suboptimal for some use cases

6

NoSQL databases have emerged…

Each one offers some combination of:

•High performance

•High scalability

•Rich data-model

•Schema less

In return for:

•Limited transactions

•Relaxed consistency

•…

7

… but there are few commonalities

• Everyone and their dog has written one

• Different data models- Key-value- Column- Document- Graph

• Different APIs – No JDBC, Hibernate, JPA (generally)

• “Same sorry state as the database market in the 1970s before SQL was invented” http://queue.acm.org/detail.cfm?id=1961297

8

• NoSQL usage small by comparison…

• But growing…

NoSQL databases have emerged…

10

Agenda

• Why NoSQL?• Overview of NoSQL databases• Introduction to Spring Data• Database APIs

- MongoDB- HyperSQL- Neo4J

11

Redis

• Advanced key-value store- Think memcached on steroids (the good kind)- Values can be binary strings, Lists, Sets, Ordered Sets, Hash maps, ..- Operations for each data type, e.g. appending to a list, adding to a

set, retrieving a slice of a list, …- Provides pub/sub-based messaging

• Very fast:- In-memory operations- ~100K operations/second on entry-level hardware

• Persistent- Periodic snapshots of memory OR append commands to log file- Limits are size of keys retained in memory.

• Has “transactions”- Commands can be batched and executed atomically

K1

K2

K3

V1

V2

V2

14

Redis use cases

• Use in conjunction with another database as the SOR

• Drop-in replacement for Memcached- Session state- Cache of data retrieved from SOR- Denormalized datastore for high-performance queries

• Hit counts using INCR command

• Randomly selecting an item – SRANDMEMBER

• Queuing – Lists with LPOP, RPUSH, ….

• High score tables – Sorted sets

Notable users: github, guardian.co.uk, ….

14

vFabric Gemfire - Elastic data fabric

• High performance data grid

• Enhanced parallel disk persistence

• Non Disruptive up/down scalability

• Session state- Cache of data retrieved from SOR

- Denormalized datastore for high-performance queries

• Heterogenous data sharing

•Java

• .net

•C++

• Co-located Transactions

14

Gemfire - Use Cases

• Ultra low latency high throughput application

• As an L2 cache in hibernate

• Distributed Batch process

• Session state- Tomcat

- tcServer

• Wide Area replication

15

Neo4j

•Graph data model

- Collection of graph nodes- Typed relationships between nodes- Nodes and relationships have properties

•High performance traversal API from roots

- Breadth first/depth first

•Query to find root nodes

- Indexes on node/relationship properties- Pluggable - Lucene is the default

•Graph algorithms: shortest path, …

•Transactional (ACID) including 2PC

•Deployment modes

- Embedded – written in Java- Server with REST API

16

Neo4j Data Model

19

Neo4j Use Cases

• Use Cases

- Anything social- Cloud/Network management, i.e. tracking/managing physical/virtual resources- Any kind of geospatial data- Master data management- Bioinformatics- Fraud detection- Metadata management

• Who is using it?

- StudiVZ (the largest social network in Europe)- Fanbox- The Swedish military- And big organizations in datacom, intelligence, and finance that wish to remain anonymous

20

MongoDB

• Document-oriented database- JSON-style documents: Lists, Maps, primitives- Documents organized into collections (~table)

• Full or partial document updates- Transactional update in place on one document- Atomic Modifiers

• Rich query language for dynamic queries

• Index support – secondary and compound

• GridFS for efficiently storing large files

• Map/Reduce

21

Data Model = Binary JSON documents

• Sequence of bytes on disk = fast I/O- No joins/seeks

- In-place updates when possible => no index updates

• Transaction = update of single document

{

"name" : "Ajanta",

"type" : "Indian",

"serviceArea" : [

"94619",

"94618"

],

"openingHours" : [

{

"dayOfWeek" : Monday,

"open" : 1730,

"close" : 2130

}

],

"_id" : ObjectId("4bddc2f49d1505567c6220a0")

}

{

"name" : "Ajanta",

"type" : "Indian",

"serviceArea" : [

"94619",

"94618"

],

"openingHours" : [

{

"dayOfWeek" : Monday,

"open" : 1730,

"close" : 2130

}

],

"_id" : ObjectId("4bddc2f49d1505567c6220a0")

}

One document =

one DDD aggregate

One document =

one DDD aggregate

23

MongoDB query by example

• Find a restaurant that serves the 94619 zip code and is open at 6pm on a Monday

{

serviceArea:"94619",

openingHours: {

$elemMatch : {

"dayOfWeek" : "Monday",

"open": {$lte: 1800},

"close": {$gte: 1800}

}

}

}

{

serviceArea:"94619",

openingHours: {

$elemMatch : {

"dayOfWeek" : "Monday",

"open": {$lte: 1800},

"close": {$gte: 1800}

}

}

} DBCursor cursor = collection.find(qbeObject);

while (cursor.hasNext()) {

DBObject o = cursor.next();

…

}

DBCursor cursor = collection.find(qbeObject);

while (cursor.hasNext()) {

DBObject o = cursor.next();

…

}

25

MongoDB use cases

• Use cases

- Real-time analytics- Content management systems - Single document partial update- Caching- High volume writes

• Who is using it?

- Shutterfly, Foursquare- Bit.ly Intuit- SourceForge, NY Times- GILT Groupe, Evite, - SugarCRM

Copyright (c) 2011 Chris Richardson. All rights reserved.

26

Other NoSQL databases

• SimpleDB – “key-value”

• Cassandra – column oriented database

• CouchDB – document-oriented

• Membase – key-value

• Riak – key-value + links

• Hbase – column-oriented…

http://nosql-database.org/ has a list of 122 NoSQL databaseshttp://nosql-database.org/ has a list of 122 NoSQL databases

27

Agenda

• Why NoSQL?

• Overview of NoSQL databases

• Introduction to Spring Data

• Database APIs

- MongoDB

- HyperSQL

- Neo4J

28

NoSQL Java APIs

But

•Usage patterns

•Tedious configuration

•Repetitive code

•Error prone code

•…

Database Libraries

Redis Jedis, JRedis, JDBC-Redis, RJC

Neo4j Vendor-provided

MongoDB Vendor-provided Java driver

Gemfire Pure Java map API, Spring-Gemfire templates

30

Spring Data Project Goals

• Bring classic Spring value propositions to a wide range of NoSQL databases:

- Productivity- Programming model consistency: E.g. <NoSQL>Template classes- “Portability”

31

Spring Data sub-projects

• Commons: Polyglot persistence

• Key-Value: Redis, Riak

• Document: MongoDB, CouchDB

• Graph: Neo4j

• GORM for NoSQL

http://www.springsource.org/spring-datahttp://www.springsource.org/spring-data

32

Many entry points to use

• Auto-generated repository implementations

• Opinionated APIs (Think JdbcTemplate)

• Object Mapping (Java and GORM)

• Cross Store Persistence Programming model

• Productivity support in Roo and Grails

33

Cloud Foundry supports NoSQL

MongoDB and Redis are provided as services

è Deploy your MongoDB and Redis applications in seconds

34

Agenda

• Why NoSQL?

• Overview of NoSQL databases

• Introduction to Spring Data

• Database APIs

- MongoDB

- HyperSQL

- Neo4J

35

Three databases for today’s talk

Document database

Relational database

Graph database

36

Three persistence strategies for today’s talk

• Lower level template approach

• Conventions based persistence (Hades)

• Cross-Store persistence using JPA and a NoSQL datastore

37

Spring Template Patterns

• Resource Management

• Callback methods

• Exception Translation

• Simple Query API

38

Repository Implementation

39

• Also known as HSQLDB or Hypersonic SQL

• Relational Database

• Table oriented data model

• SQL used for for queries

• … you know the rest…

40

Spring Data Repository Support

• Eliminate bolierplate code – only finder methods

• findByLastName – Specifications for type safe queries

• JPA CrietriaBuilder integration QueryDSL

41

•Type safe queries for multiple backends including JPA, SQL and MongoDB in Java

•Generate Query classes using Java APT

•Code completion in IDE

•Domain types and properties can be referenced safely

•Adopts better to refactoring changes in domain types

http://www.querydsl.com

42

QueryDSL

• Repository Support

• Spring Data JPA

• Spring data Mongo

• Spring Data JDBC extensions

• QueryDslJdbcTemplate

43

Spring Data Neo4J

• Using AspectJ support providing a new programming model

• Use annotations to define POJO entities

• Constructor advice automatically handles entity creation

• Entity field state persisted to graph using aspects

• Leverage graph database APIs from POJO model

• Annotation-driven indexing of entities for search

44

Spring Data Graph Neo4J cross-store

• JPA data and “NOSQL” data can share a data model

• Separate the persistence provider by using annotations

– could be the entire Entity

– or, some of the fields of an Entity

• We call this cross-store persistence

– One transaction manager to coordinate the “NOSQL” store with the JPA relational database

– AspectJ support to manage the “NOSQL” entities and fields

• holds on to changed values in “change sets” until the transaction commits for non-transactional data stores

45

A cross-store scenario ...

You have a traditional web app using JPA to persist data to a relational database ...

46

JPA Data Model

8/3/11 Slide 46

47

Cross-Store Data Model