3 millions users ?

No !

5 millions users !

The story

How Blablacar is born ?

A man

who wants to see his family...

without tickets...

seeing plenty empty cars...

HAD AN IDEA !

The concept

How Blablacar works ?

Driver: Empty car ?

Passenger: Looking for a vehicle ?

Travel together !

V1

Quick & dirty !

"Development" start in 2004

U know Joomla ?

But have you met Mambo ?

Don't be jealous !

OK the prototype works !

Now let's move !

V2

Plain old school PHP

Development start in January 2008

Context

From scratch

Use the same database than the V1

September 2008

Launch in France

January 2010

Launch in Spain

June 2011

Launch in UK

May 2012

Launch in Italy

1

million users use website every month

8 680

Average rpm during last month

329

Average response time during last month (in ms)

4 683

commits on Git (in the 3 last years)

285 588

lines of code...

V2 works fine

France (with booking)

Spain

UK

Italy

But it's not good enough.

No code convention

Home made code

Need more developers

And still need ! =)

Not scalable

To search trip ? SQL !

Hard to maintain

Imagine if YOU have to update the lib.trip.php file !

lib.trip.php

3770 lines

69 functions

926 lines... for 1 function !

It look like this

if (...) {
    if () {
        for () {
            if (){
            }else{
            }
        }
        if () {
            for () {
                if () ;
                else
            }
        }
    }
    if(){
    }
}
                        

No tests !

Better than v1 right ?

V3

Let's do beautiful code !

Context

BlaBlaCar in Europe

• Less than 2 millions members
• Online booking in France only
• 5 countries
• From 5 (at the begin) to 20 people in tech team today

Business requirements

• We can't stop the site (yeah, really !)
• We can't stop improving v2

Technical needs

Well known framework

• To hire developers
• To avoid long learning curve
• To not reinvent the wheel (i18n, forms, templates, etc.)
• To code quickly

Symphony

Good code

• Readabability
• Fiability
• Speed
• Scalability

Readability ?

PSR-2

Code reviews

Fiability ?

• Unit tests with PHPUnit
• Functionnal tests with Behat
• Continous integration with Bamboo

Speed ?

• Asynchronous job
• Cache (thanks sf2)

Scalibility ?

• ElasticSearch
• RabbitMQ

Development start in March 2012

And after 6 months...

October 2012

Launch in Belgium

October 2012

Launch in Netherlands

August 2012

Launch in Portugal

October 2012

Launch in Luxembourg

October 2012

Launch in Poland

Migration to v3

October 2012

UK

November 2012

Italy

November 2012

Spain

And our last country...

Released in 1 week only

April 2013

Launch in Germany

What do you think ? :)

Not only a PHP refactoring !

For search ?

No more SQL !

For asynchronous tasks ?

No more SQL !

For locks ?

No more SQL !

For logs ?

No more tail on servers !

For servers ?

No more manual installation !

And also

Between 5 and 15 deployments / day

New Relic for application monitoring

S3 for photo

EC2 (logstash, newsletters, ...)

30 slots for dev / preprod env

So now ?

Is v3 better than v2 ?

Average rpm during last month

V2: 8 680

V3: 2 790

Average response time during last month (in ms)

V2: 329

V3: 278

Lines of code

V2: 285 588

V3: 199 774

Not really relevant right ?

What do we think about sf2 ?

• To hire developers
• To avoid long learning curve
• To not reinvent the wheel (i18n, forms, templates, etc.)
• To code quickly

What do we think about our code

• Readabability
• Fiability
• Speed
• Scalability

Readability

Too many events

Inside sf2 !

We made this

It made this

Workers, workers, workers !

(you're going to hear about it in the next slides)

Fiability

Shared DB between v2 & v3

Very hard to maintain

Are you DBA ?

Speed... in workers

Effectively, v3 is faster than v2 but...

Workers inside application ?

• We have to boot the whole sf2 kernel !
• Lot of code to deploy just for a worker
• Lot of code = memory leaks

Now we use Cilex

• Still have a DIC (Pimple)
• Clone a skeleton, write your own logic and you're down
• Fast to deploy
• blablacar/worker-lib

Scalability

1 - Direct access to DB

The front should not access directly to DB

The front should not access directly to DB

We're written Java...

• Simple REST API
• Multiple backends (MySQL, CouchBase, ElasticSearch, ...)
• No more direct access to data
• Normalize data for v3 (and all applications ?)
• Wait & See. :)

2 - V3 is too monolithic.

It's not ONLY about workers !

SOA !

Decoupled applications

Workers, workers, workers !

• mails
• sms
• user actions
• user ips
• ...

But also...

• geoproxy
• phonenumber
• waypagg

So... you're down ?

Migrate France to V3

Migrate Mobile to V3

Explode our application

(I really want do to that ! \o/)

Became an international company ?