Practical Ruby Projects with MongoDB - MongoSF

Who am I?

Alex Sharp
Lead Developer at OptimisDev

@ajsharp
alexjsharp.tumblr.com
github.com/ajsharp

We’re going to talk about
two things.

This talk is not about shiny
objects

It’s not about the new
hotness

This talk is about creating
software.

Migrating to another
technology is inefﬁcient

Most people are reluctant to
move off RDBMS-es

Relational DBs are both
1.Antiquated
2.Inefﬁcient

A brief history of SQL
Developed at IBM in early 1970’s.
Designed to manipulate and retrieve data stored in
relational databases

A brief history of SQL
Developed at IBM in early 1970’s.
Designed to manipulate and retrieve data stored in
relational databases

this is a problem

We don’t care about storing
data

We care about persisting
state

Relational DBs are an
antiquated tool for our needs

Ok, so what?
SQL schemas are designed for storing
and querying data, not persisting objects.

To reconcile this mismatch, we
have ORM’s

Ok, so what?
We need ORMs to bridge the gap (i.e. map)
between SQL and native objects (in our
case, Ruby objects)

We’re forced to create relationships for data
when what we really want is properties for our
objects.

@alex = Person.new(
:name => "alex",
:stalkings => [Friend.new("Jim"), Friend.new("Bob")]
)

Native ruby object

<Person:0x10017d030 @name="alex",
@stalkings=
[#<Friend:0x10017d0a8 @name="Jim">,
#<Friend:0x10017d058 @name="Bob">
]>

JSON/Mongo Representation
@alex.to_json
{ name: "alex",
stalkings: [{ name: "Jim" }, { name: "Bob" }]
}

SQL Schema Representation
# in a SQL schema
people:
- name

stalkings:
- name
- stalkee_id

# in a SQL schema
people:
- name

stalkings:
Wha Wha!?
- name
- stalkee_id

# in a SQL schema
people:
- name

stalkings: stalkings ain’t got
- name no stalkee_id
- stalkee_id

# in a SQL schema
people:
- name

stalkings: stalkee_id is how we map our
- name object to ﬁt in a SQL schema
- stalkee_id

# in a SQL schema
people:
- name

stalkings:
i.e. the “pointless join”
- name
- stalkee_id

Ruby -> JSON -> SQL
@stalkings=
Ruby [#<Friend:0x10017d0a8 @name="Jim">,
]>

@alex.to_json
JSON { name: "alex",
}

people:
- name
SQL
stalkings:
- name
- stalkee_id

Ruby -> JSON -> SQL
@stalkings=
Ruby [#<Friend:0x10017d0a8 @name="Jim">,
]>

@alex.to_json
JSON { name: "alex",
}

people:
- name Feels like we’re having
SQL to work too hard here
stalkings:
- name
- stalkee_id

This may seem trivial for
simple object models

But consider a tree-type
object graph

Ok, so what?
You’re probably thinking...
“Listen GUY, SQL isn’t that bad”

Ok, so what?
But we can do much, much better.

Summary of Problems
mysql is both antiquated and inefﬁcient

Needs for a persistence layer:

1.To persist the native state of our objects

2.Should NOT interfere w/ application development

2.Should NOT interfere w/ application development
3.Provides features necessary to build modern web apps

Mongo was made for web
apps

Mongo stores everything in
binary JSON

Simplifying schema design
Maps/hashes/associative arrays are arguably
among the best object serialization formats

Mongo Representation
{ name: "alex",
}


{ name: "alex",
}
In Mongo, stalkings is an “embedded document”


{ name: "alex",
}
Great for one-to-many associations


{ name: "alex",
}
No JOINS required.


{ name: "alex",
}
This is a good thing for scalability, because
JOINS limit our ability to horizontally scale.


{ name: "alex",
}
But more importantly, this is more intuitive than
messing with foreign keys

Mongo allows us to store
objects in a near-native format

Lots of 1-to-many relationships
can be implemented as an
embedded doc

This results in way fewer
“pointless JOINs”

The ability to easily scale out
is an important part of the
philosophy behind Mongo

Also has implications for in
how we store our objects

If scaling out is easy, who
cares about the DB getting
“too large”?

Just ﬁre up another EC2
instance and be done with it.

Auto-sharding is going to
make this stupid easy.

Other features necessary for
building web apps

Four Examples

1.Accounting Application
2.Capped Collection Logging
3.Blogging Application
4.Storing binary data w/ GridFS

A simple general ledger
accounting application

The object model
ledger

*
transactions

*
entries

The object model

# Credits Debits

Transaction { 1
{ :account => “Cash”,
:amount => 100.00 }
{ :account => “Notes Pay.”,
:amount => 100.00 }

Ledger Entries
Transaction { 2 { :account => “A/R”,
:amount => 25.00 }
{ :account => “Gross Revenue”,
:amount => 25.00 }

Object model summary
Each ledger transaction belongs to a ledger.
Each ledger transaction has two ledger entries which
must balance.

Object Model with ActiveRecord
@credit_entry = LedgerEntry.new :account => "Cash",
:amount => 100.00, :type => "credit"
@debit_entry = LedgerEntry.new :account => "Notes Pay.",
:amount => 100.00, :type => "debit"

@ledger_transaction = LedgerTransaction.new
:ledger_id => 1,
:ledger_entries => [@credit_entry, @debit_entry]

Object Model with ActiveRecord
@credit_entry = LedgerEntry.new :account => "Cash",
:amount => 100.00, :type => "credit"
@debit_entry = LedgerEntry.new :account => "Notes Pay.",
:amount => 100.00, :type => "debit"

@ledger_transaction = LedgerTransaction.new
:ledger_id => 1,
:ledger_entries => [@credit_entry, @debit_entry]

In a SQL schema, we need a database
transaction to ensure atomicity

Object Model with Mongo
@ledger_transaction = LedgerTransaction.new :ledger_id => 1,
:ledger_entries => [
{ :account => 'Cash', :type => "credit", :amount => 100.00 },
{ :account => 'Notes Pay.', :type => "debit", :amount => 100.00 }
]

]

This is the perfect case for embedded documents.

]

We would never have a ledger entry w/o a
ledger transaction.

]

In this case, we don’t even need database
transactions.

]

Sweet.

Logging with Capped
Collections

Baseless statistic
95% of the time logs are NEVER used.

Most logs are only used when
something goes wrong.

Capped collections
Fixed-sized, limited operation, auto age-out
collections (kinda like memcached)
Fixed insertion order
Super fast (faster than normal writes)
Ideal for logging and caching

Great. What’s that mean?
We can log additional pieces of arbitrary data
effortlessly due to Mongo’s schemaless nature.

Now we have logs we can
query, analyze and use!

Also a really handy
troubleshooting tool

Scenario
User experiences weird, hard to
reproduce error.

Scenario
Wouldn’t it be useful to see the complete
click-path?

Bunyan
http://github.com/ajsharp/bunyan

Thin ruby layer around a
MongoDB capped collection

require 'bunyan'

# put in config/initializers for rails
Bunyan::Logger.configure do |c|
c.database 'my_bunyan_db'
c.collection 'development_log'
# == 100.megabytes if using rails
c.size 104857600
end

class BunyanMiddleware
def initialize(app)
@app = app
end

def call(env)
@status, @headers, @response = @app.call(env)
Bunyan::Logger.insert(prepare_extra_fields)
[@status, @headers, @response]
end
end

Bunyan::Logger.find('user.email' => 'ajsharp@gmail.com')

Stay tuned for a Sinatra-based
client front-end.

Mongolytics

Drop-in to a rails app
http://github.com/tpitale/mongolytics.git

Blogging Application
Much easier to model with Mongo than a relational
database

A post has an author

A post has many tags

A post has many comments

A post has many comments

Instead of JOINing separate tables,
we can use embedded documents.

require 'mongo'

conn = Mongo::Connection.new.db('bloggery')
posts = conn.collection('posts')
authors = conn.collection('authors')

# returns a Mongo::ObjectID object
alex = authors.save :name => "Alex"
post = posts.save(
:title => 'Post title',
:body => 'Massive pontification...',
:tags => ['mongosf', 'omg', 'lolcats'],
:comments => [
{ :name => "Loudmouth McGee",
:email => 'loud@mouth.edu',
:body => "Something really ranty..."
}
],
:author_id => alex
)

# returns a Mongo::ObjectID object
alex = authors.save :name => "Alex"
post = posts.save(
:title => 'Post title',
:body => 'Massive pontification...',
:tags => ['mongosf', 'omg', 'lolcats'],
:comments => [
{ :name => "Loudmouth McGee",
:email => 'loud@mouth.edu',
:body => "Something really ranty..."
}
],
Joins not necessary. Sweet.
:author_id => alex
)

MongoMapper
• MongoDB “ORM” developed by John Nunemaker

MongoMapper
• author of HttpParty

MongoMapper
• Very similar syntax to DataMapper

MongoMapper
• Declarative rather than inheritance-based

MongoMapper
• Declarative rather than inheritance-based
• Very easy to drop into rails

MongoMapper
class Post
include MongoMapper::Document

belongs_to :author, :class_name => "User"

key :title, String, :required => true
key :body, String
key :author_id, Integer, :required => true
key :published_at, Time
key :published, Boolean, :default => false
timestamps!

many :tags
end

class Tag
include MongoMapper::EmbeddedDocument

key :name, String, :required => true
end

Storing binary data w/
GridFS and Joint

Joint
1.MongoMapper plugin for accessing GridFS
2.Built on top of the Ruby driver GridFS API

class Bio
plugin Joint

key :title, String
key :description, String
timestamps!

attachment :headshot
attachment :video_bio
end

class Bio
plugin Joint

key :title, String
key :description, String
timestamps!
Can be served directly
attachment :headshot
from Mongo
attachment :video_bio
end

Using mongo w/ Ruby
Ruby mongo driver
MongoMapper
MongoID
MongoRecord (http://github.com/mongodb/mongo-
record)
MongoDoc (http://github.com/leshill/mongodoc)
Many other ORM/ODM’s under active development

Practical Ruby Projects with MongoDB - MongoSF

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to Practical Ruby Projects with MongoDB - MongoSF

Similar to Practical Ruby Projects with MongoDB - MongoSF (20)

More from Alex Sharp

More from Alex Sharp (9)

Recently uploaded

Recently uploaded (20)

Practical Ruby Projects with MongoDB - MongoSF

Editor's Notes