happn: The app where immediacy and mobility meet
About happn
Founded in Paris in 2014, happn created a dating app based on two key concepts: real time and hyper-geolocation. The app helps users to find people that they've crossed paths with during the day. happn offers free access with some paid services (freemium). Its revenues are based on advertising and the purchase of additional features within the app.
Tell us your challenge. We're here to help.
Contact usBy relying on the Google Cloud, happn has better secured its international app development in more than 40 countries on 5 continents.
Google Cloud results
- Supports the entire happn information system (design, production, backup, analytics)
- Frees teams from administrative constraints, thanks to managed services
- Provides the technical and financial flexibility needed to extend happn services
Nearly 40M registered users making 10B monthly requests
Simplicity and efficiency are the hallmarks of a successful app, and happn is no exception. It has built a mobile system that facilitates, in real time, meetings between people crossing paths by chance, in the same place, at the same time.
Available in about 15 languages, happn operates in more than 40 countries and 50 major cities in Europe, Asia, Oceania, and North and South America. happn allows users to find people that they've crossed paths with and would like to meet. The app is hyper-geolocalized and works in real time: each time a user crosses paths with another member, their profile is displayed in the timeline.
happn had originally chosen a hosting offer that placed certain limitations on the company's growth and expansion needs. Billing was based on a monthly fee calculated by the number of servers determined beforehand. This was not very flexible when faced with fluctuating consumption. Functionally, there was a significant delay between the request for activation of new servers and their operation, which meant computing power couldn't be adjusted on the fly.
“We need power availability to fulfill the promise of our product: real-time updates when our users cross paths. The business model proposed by Google, which did not require server instances to be reserved, best met our expectations."
—Sébastien Preneta, Vice President of Engineering, happnA full-scale selection process
By 2015, the company had 10 million users in several countries, but its international development was being hampered by these constraints. happn decided to change from a fixed infrastructure to a flexible platform, by way of platform as a service (PaaS).
In early 2016, the technical team drew up specifications and prototypes in order to perform a full-scale test of the behavior of its software components: Java (microservices architecture), Cassandra (for persistent data), RabbitMQ (for message threads), and Elasticsearch (for indexing and searching for data). The platforms of three global cloud specialists were featured in the company's call for tenders.
Based on samples of the various services provided by happn (user profile creation, location, interactions between users), the prototypes were stress tested during the design, deployment, and production phases. Managed services were also evaluated for stability and ease of use.
Multiple gains
At the end of the process, Google Cloud emerged victorious. "Our essential criterion was based on processing performance not only at one given time, but over a period of time too. Our audience varies according to the time of day in each time zone. We need power availability to fulfill the promise of our product: real-time updates when our users cross paths," explains Sébastien Preneta, Vice President of Engineering at happn. "The business model proposed by Google, which did not require server instances to be reserved, best met our expectations. The attitude and commitment demonstrated by Google were also determining factors."
Carefully prepared by happn and Google technical teams, the export, backup, and switchover of data and programs was carried out seamlessly in mid-2016. This transfer involved tens of terabytes of data. For a few weeks, both platforms were interconnected with continuous synchronization. When the time came, switching to Google Cloud caused the app to be unavailable for less than 30 minutes. happn now has about 500 servers distributed among Google Cloud data centers.
"The infrastructure for these analytics services is managed directly by Google Cloud. It would have taken us much longer to build an equivalent in an internal infrastructure. This way, we can focus on our business challenges, rather than technical ones."
—Sébastien Preneta, Vice President of Engineering, happnWell-located data
Aside from images, all happn data is stored and backed up in Cassandra and MariaDB instances, operated by happn's technical teams, as well as in Google Cloud Datastore, and Google BigQuery for Google managed services.
Google Compute Engine is at the heart of the virtual server architecture required by happn, providing multiple functions:
- Domain name management
- Operation and performance monitoring
- Orchestration (via Salt)
- Gateway to the programming interfaces (Nginx, PHP, and Java)
- User searches
- Cache management (memcached and Redis)
- Asynchronous messaging operation (RabbitMQ and Google Cloud Pub/Sub)
- Load balancing (HAProxy and Google Cloud Load Balancing)
"This model is fundamental for facilitating our design testing of new architectures: we create an unlimited number of new virtual servers instantly, establish links between them, and test software versions and operating systems. We turn them on, test them, and shut them down on demand, without constraints," explains Preneta.
"Launching the app in countries the size of continents represents a real commercial and technical challenge. Google Cloud helps ensure a consistent quality of service and the availability of the service anywhere in the world."
—Sébastien Preneta, Vice President of Engineering, happnDifferent perspectives
happn also uses a number of analytical tools offered by Google Cloud: Google BigQuery (data repository), Google Cloud Dataflow (real-time processing), and Google Cloud Dataproc (Spark and Hadoop). "The infrastructure for these analytics services is managed directly by Google Cloud. It would have taken us much longer to build an equivalent in an internal infrastructure. This way, we can focus on our business challenges, rather than technical ones," explains Preneta.
Custom built by happn's teams, dozens of dashboards are used daily by the company's general management, marketing, financial, commercial, and technical departments. Events are collected through Google Cloud Pub/Sub, cross-referenced with external sources, then analyzed by Google Dataproc and Google BigQuery.
Flexibility
The flexibility of Google Cloud is invaluable when happn deploys its app in new markets. As a precaution, the technical team chose to double the available computing power when launching in India in early 2017. Two weeks after launch, the company adjusted this capacity based on the average number of downloads and daily connections.
"Launching the app in countries the size of continents represents a real commercial and technical challenge. Google Cloud helps ensure a consistent quality of service and the availability of the service anywhere in the world," adds Preneta. "Beyond the computing power and the adapted pricing, Google managed services greatly simplify the task at hand. The time saved on updates, patches, and maintenance means our engineers can focus on analytics and functional enhancements to our app, rather than on problems with the plumbing. We look forward to enjoying even more managed services from Google teams in the future."
Tell us your challenge. We're here to help.
Contact us
About happn
Founded in Paris in 2014, happn created a dating app based on two key concepts: real time and hyper-geolocation. The app helps users to find people that they've crossed paths with during the day. happn offers free access with some paid services (freemium). Its revenues are based on advertising and the purchase of additional features within the app.