From MySQL to MongoDB at Wordnik (Tony Tam)

MongoSF 4/30/2010From MySQL to MongoDB

Migrating a Live Application

Tony Tam

WHAT IS WORDNIK

Project to track language like GPS for English

Dictionary is a road block to the languageRoughly 200 new words created dailyLanguage is not static

Capture information about all wordsMeaning is often undefined in traditional senseMachines can determine meaning through

analysisNeeds LOTS of data

WHY SHOULD YOU CARE

Every Developer can use a Robust Language API!

Wordnik migrated to MongoDB> 5 Billion documents> 1.2 TBZero application downtime

Learn from our Experience

WORDNIK

Not just a website!But we have one

Launched Wordnik entirely on MySQLHit road bumps with insert speed ~4B rows

on MyISAM tablesTables locked for 10’s of seconds during insertsBut we need more data!

Created elaborate update schemes to work around itLost lots of sleep babysitting servers while

researching LT solution

WORDNIK + MONGODB

What are our storage needs?Database vs. Application LogicNo PK/FK constraintsNo Stored ProceduresConsistency?

Lots of R&DTried most all noSQL solutions

MIGRATING STORAGE ENGINES

Many parts to this effortSetup & AdministrationSoftware DesignOptimization

Many types of data at WordnikCorpusStructured Hierarchical DataUser Data

Migrated #1 & #2

SERVER INFRASTRUCTURE

Wordnik is Heavily Read-onlyMaster / Slave deployment

Looking at replica pairsMongoDB loves system resources

Wordnik runs dedicated boxes to avoid other apps being sent to disk (aka time-out)

Memory + Disk = Happy MongoMany X the disk space of MySQLEasy pill to swallow until…

SERVER INFRASTRUCTURE

Physical Hardware2 x 4 core CPU, 32gb RAM, FC SANHad bad luck on VMs

(you might not)Disk speed => performance

SOFTWARE DESIGN

Two distinct use cases for MongoDBIdentical structure, different storage

engineSame underlying objects, same storage fidelity

(largely key/value)Hierarchical data structure

Same underlying objects, document-oriented storage

SOFTWARE DESIGN

Create BasicDBObjects from POJOs and used collection methodsBasicDBObject dbo = new BasicDBObject("sentence",s.getSentence()) .append("rating",s.getRating()).append(...);

ID Generation to manage unique _ID valuesAnalogous to MySQL AutoIncrement behaviorCompatible with MySQL Ids (more later)dbo.append("_ID", getId());collection.save(dbo);

Implemented all CRUD methods in DAO Swappable between MongoDB and MySQL at

runtime

SOFTWARE DESIGN

Key-Value storage use caseEasy as implementing new DAOs

SentenceHandler h = new MongoDBSentenceHandler();

Save methods construct BasicDBObject and call save() on collection

Implement same interfaceSame methods against DAO between MySQL

and MongoDB versionsData Abstraction 101

SOFTWARE DESIGN

What about bulk inserts?FAF Queued approach

Add objects to queue, return to callerEvery X seconds, process queueAll objects from same collection are appended

to a single List<DBObject>Call collection.insert(…) before 2M

charactersReduces network overheadVery fast inserts

SOFTWARE DESIGN

Hierarchical Data done more elegantlyWordnik Dictionary ModelJava POJOs already had JAXB annotations

Part of public REST apiUsed Mysql

12+ tables13 DAOs2500 lines of code50 requests/second uncachedMemcache needed to maintain reasonable speed

SOFTWARE DESIGN

TMGO

SOFTWARE DESIGN

MongoDB’s Document Storage let us…Turn the Objects into JSON via Jackson

Mapper (fasterxml.com)Call saveSupport all fetch types, enhanced filters1000 requests / secondNo explicit cachingNo less scary code

SOFTWARE DESIGN

Saving a complex objectString rawJSON = getMapper().writeValueAsString(veryComplexObject);

collection.save(new BasicDBOBject(getId(),JSON.parse(rawJSON));

Fetching complex objectBasicDBObject dbo = cursor.next();ComplexObject obj = getMapper().readValue(dbo.toString(), ComplexObject.class);

No joins, 20x faster

MIGRATING DATA

Migrating => existing data logicUse logic to select DAOs appropriatelyRead from old, write with newGreat system test for MongoDB

SentenceHandler mysqlSh = new MySQLSentenceHandler();SentenceHandler mongoSh = new MongoDbSentenceHandler();while(hasMoreData){ mongoSh.asyncWrite(mysqlSh.next()); ...}

MIGRATING DATA

Wordnik moved 5 billion rows from MySQLSustained 100,000 inserts/secondMigration tool was CPU bound

ID generation logic, among other

Wordnik reads MongoDB fastRead + create java objects @ 250k/second (!)

GOING LIVE TO PRODUCTION

Choose your use case carefully if migrating incrementally

Scary no matter whatTest your perf monitoring system first!

Use your DAOs from migrationTurn on MongoDB on one server,

monitor, tune (rollback, repeat)Full switch over when comfortable

GOING LIVE TO PRODUCTION

Really?SentenceHandler h = null;if(useMongoDb){ h = new MongoDbSentenceHandler();}else{ h = new MySQLDbSentenceHandler();}return h.find(...);

OPTIMIZING PERFORMANCE

Home-grown connection poolingMaster onlyConnectionManager.getReadWriteConnection()

Slave onlyConnectionManager.getReadOnlyConnection()

Round-robin all servers, bias on slaves

ConnectionManager.getConnection()

OPTIMIZING PERFORMANCE

CachingHad complex logic to handle cache

invalidationOut-of-process caches are not freeMongoDB loves your RAMLet it do your LRU cache (it will anyway)

HardwareDo not skimp on your disk or RAM

IndexesSchema-less design

Even if no values in any document, needs to read document schema to check

OPTIMIZING PERFORMANCE

Disk spaceSchemaless => schema per document

(row)Choose your mappings wisely({veryLongAttributeName:true}) =>

more disk space than ({vlan:true})

OPTIMIZING PERFORMANCE

A Typical Day at the Office for MongoDBAPI call rate: 47.7 calls/sec

OTHER TIPS

Data TypesUse caution when changingDBObject obj = cur.next();long id = (Long) obj.get(“IWasAnIntOnce”)

Attribute namesDon’t change w/o migrating existing data!

WTFDMDG????

WHAT’S NEXT?

GridFSStore audio files on disk

Requires clustered file system for shared access

Capped Collections (rolling out this week)

UGC from MySQL => MongoDBBeg/Bribe 10gen for some Features

Questions?