29
September, 2014 NoSQL – A Quick Tour

NoSQL Basics - A Quick Tour

Embed Size (px)

DESCRIPTION

The document talks about the overview behind the need and drive for NoSQL databases. It also mentions about some of the most popular NoSQL databases in the market.

Citation preview

Page 1: NoSQL Basics - A Quick Tour

September, 2014

NoSQL – A Quick Tour

Page 2: NoSQL Basics - A Quick Tour

Objectives

What is NoSQL

How is data growing ?

Challenges

What’s the solution ?

NoSQL Features and Types

Eventual Consistency

High-level Overview of some popular No-SQL DB

NoSQL – Not mandatory

Popular Jargons

Useful Links

Page 3: NoSQL Basics - A Quick Tour

What is NoSQL Abbreviation of “Not an SQL”, any data source which doesn’t come under SQL category

A new way to store and retrieve data (specially sparse and unstructured) in modern day high-volume real-time web traffic,

batch processing, and analytics

Not an alternative for RDBMS but a parallel concept

A “Shared Nothing” architecture as opposed to “Shared Architecture” in RDBMS (Philosophy of Shared Nothing (NA)

architecture - A shared nothing architecture (SN) is a distributed computing architecture in which each node is independent

and self-sufficient, and there is no single point of contention across the system. More specifically, none of the nodes share

memory or disk storage.). The term was first coined by Michael Stonebraker at University of California at Berkeley in his

1986 paper “The Case for Shared Nothing.”

Can be tagged with technologies which doesn’t use SQL and relational mapping of data

Promotes huge storage of data and efficient retrieval, supports normal CRUD operations

Not a stringent follower of ACID due to its inherent nature

Works with “Big Data” for supporting high volume processing of data

Comes with varied flavor of products often coined due to research at social networking giants like Twitter, Facebook,

LinkedIn, Google, Yahoo etc.

Data can be stored in four basic databases like key/value pair, column-family data, document, and graph

Page 4: NoSQL Basics - A Quick Tour

What is NoSQL Continued.. Mostly these stores or databases don’t expect pre-defined schemas or normalization and consistency

Primary purpose of NoSQL is to have fast and efficient storing and processing of constantly growing data without the

constraint of relation database and provide the scalable architecture to support future growth of data without compromising

the performance

First generation and second generation of NoSQL

Mantra of NoSQL : “Getting an answer quickly is more important than getting a

correct answer”

If you can’t split it, you can’t scale it

—Randy Shoup, Distinguished Architect, eBay

Page 5: NoSQL Basics - A Quick Tour

How is data growing ?

Page 6: NoSQL Basics - A Quick Tour

How is data growing ? Continued..

There are two concepts Big User and Big Data along with Cloud Computing Big User – Number of users accessing web is growing rapidly accessing several kind of data such Personal information, social data like tweets, likes, blogs, click streams, comments, follows, or geo location data, log files, system generated data, user generated data, sensor-generated data etc. The growing number of users can’t be predicted specially the advancement of mobile space Big Data – Source of data has increased tremendously but the actual data increased exponentially. Data is not in-terms of gigabyte but in more higher number (tera/penta/exa/zetta/yotta bytes) Cloud Computing – More and more Data is stored in the cloud and access to data should be fast The number of concurrent users skyrocketed as applications increasingly became accessible via the web (and later on mobile

devices)

The amount of data collected and processed soared as it became easier and increasingly valuable to capture all kinds of data

The amount of unstructured or semi-structured data exploded and its use became integral to the value and richness of applications

Page 7: NoSQL Basics - A Quick Tour

Challenges

Shared

Application Server Layer to Traditional RDBMS

Add CPU

Add RAM V

e

r

t

i

c

a

l

Shared Disk

Oh No! I am loaded

Application Server Layer to NoSQL DB

Commodity

Server

Commodity

Server Commodity

Server

Commodity

Server

Commodity

Server

Page 8: NoSQL Basics - A Quick Tour

Challenges Continued..

Need to scale out (i.e. sharding) and without compromising latency and performance (keeping low latency and high

throughout)

Need to avoid scale up with costly and complex high-end servers

Should be easily scaled out with low cost commodity servers without any application downtime

Data is not only structured but majority of the social networking data is unstructured. So, there is a need to support schema less data structures

There are many cases where transaction is not of prime concern and costly and heavy writes are not needed

Should support easy and quick replication as well as failover

Page 9: NoSQL Basics - A Quick Tour

What’s the solution ?

Real time and batch processing for analytical and operational data by second generation of NoSQL like Couchbase The picture depict a scenario where both batch processing as well as real-time data is being handled by a combination of Hadoop and Couchbase.

Page 10: NoSQL Basics - A Quick Tour

NoSQL Features and Types

Early adopter: Google’s Bigtable is used for throughput sensitive batch processing to latency based online queries. Used in Google Earth, Finance, Orkut, Analytics etc. This is based on the concept of column family data store. There are mainly 3 types of NoSQL databases as mentioned below. Document Database MongoDB and Couchbase (An amalgamated version of CouchDB and Membase) Couchebase started as Apache incubator project and then continued by Couchbase Inc Implemented in Erlang and C, and Javascript execution environment Used by Apple, BBC, and many others.

Page 11: NoSQL Basics - A Quick Tour

NoSQL Features and Types Continued..

Key/Value pair and Eventual Consistency datastore Redis, Membase, Voldemort, and Cassandra Cassandra supports key/value pair and eventual consistency (based on Amazon’s DynamoDB). Developed by Facebook and implemented in Java Clients are available in Java, PHP, Python, Grails, Ruby, .NET. Used by Facebook, Twitter, Paddy Power, GitHub, and many others. Redis supports key-value pair and a distributed in-memory as well as persistent storage system Started by Salvatore Sanfilippo in 2009 as an independent project Implemented in C Clients are available in PHP, Java, Ruby, Python, C++ etc. Used by Craigslist Sorted Column Family datastore HBase (developed based on Google's bigtable) Created by Powerset and donated to Apache. Implemented in Java. Used by Facebook, Yahoo! and many others. Access method is JRuby, Java, Thrift, REST, ProtoBuf etc. Graph Database Neo4J, FlockDB Neo4J was developed in 2003 and implemented in Java Accessed through REST and Gremlin interfaces. Used by Box.com, ThoughtWorks There are other types of NOSQL databases available as well. XML Database Object Database Grid and Cloud Database Multimodel Database

Page 12: NoSQL Basics - A Quick Tour

Eventual Consistency

Consistency – Data should be consistent across user transactions and each client can have the same view of the data Availability – System should be available always and user can read as well as write always Partition – System should be partition-tolerant and work in a distributed environment Eventual Consistency implies BASE (Basically Available, Soft state, Eventual Consistency)

Brewer’s CAP Theorem Succinctly put, Brewer’s Theorem states that in systems that are distributed or scaled out it’s impossible to achieve all three (Consistency, Availability, and Partition Tolerance) at the same time. You must make trade-offs and sacrifice at least one in favor of the other two.

ACID BASE

Strong consistency Weak consistency – stale data OK

Isolation Availability first

Focus on “commit” Best effort

Nested transactions Approximate answers OK

Availability? Aggressive (optimistic)

Conservative (pessimistic) Simpler!

Difficult evolution (e. g. schema) Faster

Easier evolution

Page 13: NoSQL Basics - A Quick Tour

Eventual Consistency – HAPPY GO LUCKY

An example of booking flights for two close friends for attending a conference

There is only one ticket left

Data Center

(Asia)

Data Center

(US)

Anand Scott

DATA SYNC-UP

GOOD FRIENDS

bookflighttickets.com

SYNC

DOWN

Will the tickets be booked ?

Booking done

Page 14: NoSQL Basics - A Quick Tour

Brewer’s Theorem was conjectured by Eric Brewer and presented by him (www.cs.berkeley.edu/ ~brewer/cs262b-2004/PODC-keynote.pdf) as a keynote address at the ACM Symposium on the Principles of Distributed Computing (PODC) in 2000. Brewer’s ideas on CAP developed as a part of his work at UC Berkeley and at Inktomi. A look at Distributed Update and Replication Replication

Eventual Consistency – Distributed Environment

A V0 (V1)

B V0(V1)

Writes

Reads

Page 15: NoSQL Basics - A Quick Tour

Theorem of Two options and Three Alternatives

SYNCHRONOUS SYNCHRONIZATION

A Single Transaction, Consistency is of prime importance

ASYNCHRONOUS SYNCHRONIZATION

Can be achieved but if synchronization fails, no way to know when it happened

Option 1 : Let the Consistency take the precedence and Availability may be compromised considering the system supports

partition-tolerance (CP)

Option 2: Let the Availability take the precedence and Consistency may be compromised

considering the system supports partition-tolerance (AP)

Option 3: Let both Consistency and Availability take the precedence and the system is not partition-tolerant (AC)

Page 16: NoSQL Basics - A Quick Tour

A SMALL TALK

COUCHBASE (DOCUMENT)

REDIS (KEY-VALUE PAIR)

CASSANDRA (COLUMN-FAMILY AND EVENTUAL CONSISTENCY)

NEO4J (GRAPH)

HOW WILL WE GO ?

A SMALL OVERVIEW

BASIC FEATURES

Page 17: NoSQL Basics - A Quick Tour

Features of Couchbase

A high level overview of Couchbase is mentioned below. 1. Stores data in a JSON or binary format in the data store (called a document)

2. Supports basic CRUD operations like get, set, delete etc. Uses MVCC to continue with non-blocking IO for read/writes. 3. Provides a strong layer of caching of data in memory and automatically persists data in file system to support strong failover mechanism 4. Uses concepts like Buckets to group physical resources in a cluster logically with options like setting memory for each bucket as well as replication rule. 5. Each buckets divided in 24 logical partitions called vBuckets and used a cluster map to locate document in a cluster 6. vBuckets the lowest denominator to locate a document in a cluster through hash identifier for each document 7. Asynchronous storing of data in disk and replication data to other servers in a cluster as well as across data-center through XDCR feature 8. Very efficient and easy management of a distributed cluster (horizontal scaling), also known as “Scale Out” 9. Supports integration with Memcache protocol seamlessly 10. Rebalancing of data (documents) through change in the clustering (adding or removing nodes from a cluster and updating cluster map with updated location of documents) 11. Database index like feature through Views for faster access of indexed data 12. Highly secured access to Couchbase server through SASL mechanism 13. Provides the support for both optimistic (through compare and swap - CAS mechanism) and pessimistic locking (through explicit locking) 14. An asynchronous listener based approach for manipulating data through Future interface

Page 18: NoSQL Basics - A Quick Tour

How does Couchbase work ?

Page 19: NoSQL Basics - A Quick Tour

More Overview

Replication Process

Smart Client writes the data in server object-managed Cache

Documents is submitted to intra-cluster replication queue for replicating to other servers

The document is persisted to disk write queue asynchronously to write in the disk. The data is written to disk once disk queue flushes off

The data is replicated to other clusters through XDCR once the data in persisted in the disk and eventually indexed for searching.

Major Components Data Manager Object Managed Cache

server warm-up, checkpoint, TAP replicator, backfill, resident item ratio, NRU, ejection, item pager) Storage Engine

compaction) Query Engine

Index can be created and queried for JSON documents Secondary indexes are created through View and Design documents

Cluster Manager (orchestration node)

The Hearbeat watchdog The process Manager The configuration manager

Page 20: NoSQL Basics - A Quick Tour

Features of REDIS

A high level overview of Redis is mentioned below. 1. Started in 2009, REDIS (Remote Dictionary Service) is a distributed key-value pair database. A shared memory system

for very fast read and write capabilities. Fundamentally, an advanced version of Memcache database.

2. Creator of Redis - Salvatore Sanfilippo termed as “Data Structure Store” capable of storing complex data structure as keys like Set, List, Hash, Sorted Set, bitmaps etc. apart from normal strings. 3. Apart from a data structure store, it also works as blocking queue (Stack) and Publish-Subscribe system 4. A powerful command-line-interface (CLI) and rich API for the clients. 5. An expired based policy can be set for each key-value pair in order to let the list grow unbounded 6. Provides an option to save data in the disk, an unusual case for any key-value system which primarily operates over memory. There is the facility to take the snapshot in an interval based on some criteria like number of changes for keys etc. 7. An additional protection of data through Append-only file for each writes to save from crashing of the server 8. By default, Redis doesn’t provide a good way to handle security of its own. So, its better to use firewall on SSH to protect the secure data. 9. Supports a Master-Slave replication mechanism but not an multi-master scaling and fail-over intelligent system 10. A client managed cluster support through consistent-hashing rather than server side 11. Provides a probabilistic determination of non-existence of data through Bloom filters (managing sequence of bits) 12. Uses a special data structure called “Dynamic String Structure” (SDS) to store all the data internally 13. Uses its own Virtual Memory management to locate data in the disk

Page 21: NoSQL Basics - A Quick Tour

Replication in REDIS

Master Node

Slave Node (Read-Write)

Slave Node (Read-Write)

Slave Node (Read-only)

R

e

p

l

i

c

a

t

i

o

n

R

e

p

l

i

c

a

t

i

o

n

R

e

p

l

i

c

a

t

i

o

n

Disk

Non-Blocking Synchronization

Redis Server

Redis Client/Smart Client

Pe

rio

dic

Sa

ve

Redis uses hash slots to bucketing data elements across nodes, so that data is sharded across nodes for fault-tolerant. If any new node has been added or removed from the cluster, Redis maintains the linkage of data from old to new node through its internal node-to-node communication (ping-pong in Redis’s term) based on binary protocol. Under the hood, Redis also uses a Gossip-based protocol among the nodes to track the status of each node and take necessary actions in case some node went down or not responding. Redis has a smart client who can decide to connect to the right node in the cluster to find the data instead of client to any node arbitrarily.

Gossip Gossip

A B D C E

Page 22: NoSQL Basics - A Quick Tour

An Example of REDIS

Redis can be used for managing data where caching based simple key-value pair along with complex querying facility be given based on the keys. Analytics Caching Search Engine Messaging Broker 1. Get a list of cities under a zip code around the world 2. Get a list of books based on ISBN code where each book is associated with multiple “tagging” words 3. Build a sub-system which can browse through a catalog system to find the product data 4. Use a broker to collect data for multiple sources like managing centralized log content

Not so good use cases 1. Every bit of data is very precious 2. Multiple master-master setup and failover needed 3. ACID transaction is highly desired 4. Relational data is of prime importance

Page 23: NoSQL Basics - A Quick Tour

Features of APACHE CASSANDRA

A high level overview of Cassandra is mentioned below. If I had asked people what they wanted, they would have said faster horses. – Henry Ford 1. Influenced by Amazon’s DynamoDB for its distributed design and Google’s Bigtable for the data model, Cassandra is a hybrid

datastore supporting both column-family as well as key-value data with Eventual Consistency

2. Cassandra was developed by Facebook and it’s a sparse multi-dimensional hashtable 3. Supports secondary indexes apart from the index on the row-key 4. Supports powerful command-line-interface (CLI) as well as Thrift based multi-lingual drive type communication techniques for the clients. 5. Runs on decentralized mode of keeping each node identical, not like a master-slave topology 6. A tunable consistent system instead of eventual consistency (A always writeble system) 7. Uses Gossip protocol with hinted hands-off to perform peer-to-peer communication across nodes

8. Uses Anti-entropy to manage data synchronization (replication) across multiple nodes with the updated version

9. Uses compaction to merge large datafiles for better management of spaces

10. Uses Bloom filter to find if any element is available in map

11. Uses a concept called “Tombstone” for soft delete. The data is physically deleted during compaction.

12. Uses “Staged Event-Driven Architecture” (SEDA) for highly efficient parallel processing

13. Uses the three separate processes (commit log, memtable, and SSTable) to store and manage data during write operation

14. Uses a concept called “Read Repair” to update outdated values in any node

Page 24: NoSQL Basics - A Quick Tour

CASSANDRA – Column Family

Suppose A customer has a personal information as well as address. So, two column families can be created, one for personal

data and the other for address.

In column family store, data is identified by the row-key. The difference from a RDBMS is that each row can have its own

column family data. In case, there are null values for some columns data is not stored there unlike RDBMS tables which

consume additional space.

Page 25: NoSQL Basics - A Quick Tour

CASSANDRA – An Example

Cassandra is a popular datastore for many popular large scale web applications. The following criteria describes some of the important use cases for Cassandra. High volume writes like tweets from Twitter or comments from Facebook Don’t need strong consistency High throughput for Writes Consistency can be controlled

Page 26: NoSQL Basics - A Quick Tour

Features of Neo4J

A high level overview of Neo4J is mentioned below. 1. A Graph database (mathematical modeling) for supporting relational information (relationship ) across multiple entities.

Developed by Neo technologies.

2. It is built on the concept of nodes, relationship, parameters (key/value pair), and labels

3. A proprietary query named “Cypher” for performing CRUD operations

4. A very high-performing NoSQL database for storing and retrieving connected data

5. Graph is a way to maintain multi-dimensional relation among entities

6. Highly applicable in social networking applications like social graphs, recommendation etc.

7. The Neo4J site provides a rich REPL (Read-eval- print loop) web interface for running queries as well as performing administrative works.

8. It is ACID compliant as well as provides high-availability and master-slave replication across multiple nodes

9. Provides easy client interface through REST and Gremlin

10. Provides fast-look up through Lucene

Sachin

Grapes

Gaurav

Java

Bikram friend

likes

friend

likes

eats

Page 27: NoSQL Basics - A Quick Tour

NoSQL – Not mandatory

NoSQL is not a replacement of SQL Generally NoSQL is not fit for applications which need strong consistency

Correctness of data is more important than availability of data

Transactional context is important than analytical processing

Data is structured and maintained through object relational hierarchy

Need to support Legacy database

Future of Databases The future of databases lies with the amalgamation of relational and NoSQL databases based on the need. Pramod J. Sadalage and Martin Fowler mentioned the concept of Polygot Persistence in their famous book “NoSQL Distilled”.

Page 28: NoSQL Basics - A Quick Tour

Popular Jargons

The following terms are some of the most common and important jargons used in the NoSQL world.

Sharding or Horizontal Scaling Quorum Gossip Protocol Hinted Hands-off Read Repair Consistent Hashing Merkle Tree

Page 29: NoSQL Basics - A Quick Tour

Useful Links http://static.googleusercontent.com/media/research.google.com/en//archive/bigtable-osdi06.pdf

Professional NoSQL by Shashank Tiwari

NoSQL Databases by Christof Strauch

Couchbase Server Under the Hood from Couchbase Inc.

http://www.slideshare.net/Muratakal/rdbms-vs-nosql-15797058

http://www.nosql-database.org/

http://www.youtube.com/watch?v=MmL9Lq6WbSY

http://incubator.apache.org/thrift/

http://www.allthingsdistributed.com/2007/10/amazons_dynamo.html

http://www.youtube.com/watch?v=uMxZ4RI6sCQ

http://planetcassandra.org/apache-cassandra-use-cases/

http://www.allthingsdistributed.com/files/amazon-dynamo-sosp2007.pdf

Seven Databases in Seven Weeks by Eric Redmond and Jim R. Wilson