MesosCon Europe 2017

Given the nature of its business, Criteo has a tremendous need for scalability. Historically based on bare-metal infrastructure and serving billions of user requests per day with the lowest latencies, we will present how Mesos became a first-class citizen on our platform.

In this talk, we will describe how we deploy and operate 600+ Mesos agents, hosting hundreds of applications and handling up to 200K requests per second, with up to 40Gbps of outgoing traffic.
We'll mostly focus on net/dev/ops concerns, like how we dealt with configuration management, app secrets, logging, service discovery, networking, metrology and SLAs, tests, etc.

The aim of this presentation is to share what we have learnt during this long journey of setting up a production-grade Mesos infrastructure from scratch and what we expect from it in a near future.