Process management in HAProxy has evolved since the beginning of the project; With the advent of new init systems and process managers, managing HAProxy processes became complicated due to the multi-processes architecture of the project. The Master Worker mode simplifies the process management and introduces interesting features. This talk will briefly review the history of the Master Worker mode and the new features it brought.

This presentation will provide an overview of the new Data Plane API, which you can use to configure your load balancer using HTTP REST commands. You will learn how to get started and how to build tools to work with it. You will also learn how to configure a frontend/backend and modify a configuration using the API.

In 2018, we migrated several video-on-demand/replay platforms from on-premise to the AWS cloud. HAProxy was the key to this migration, allowing us to move safely and without any downtime. We’ll take a deep dive through the configuration used to secure our migration, highlighting aspects such as our use of the “observe layer7” keyword. I’ll also cover how we use GOReplay to replicate traffic and our use of the HAProxy Ingress Controller in front of our Kubernetes clusters.

At Criteo, we have been working on building a tight integration between HashiCorp Consul and HAProxy. In this talk, we will explain how we provision our HAProxy instances dynamically using Consul Connect, a new service mesh technology that allows HAProxy to talk to its peers from machine-to-machine without a traditional load balancer. We will detail how we are able to create a service with DNS, add load balancing, and configure SSL certificates in mere seconds. Since Criteo is working actively on enabling HAProxy with Consul Connect, we will explain the challenges of scaling Service Mesh architectures for large infrastructures.

Oui.sncf sells tickets and passes for rail travel around Europe. We operate the #1 French e-commerce website with more than 83 million travel products sold and more than 12 million unique visitors per month. We’ve been using HAProxy since 2009. When we decided to build our own CDN solution in 2015, we knew we’d include HAProxy as a main component. In this talk, we will show you how HAProxy is integrated into our CDN infrastructure and how we use it daily to manage, update, configure, and troubleshoot our infrastructure.

CenterDevice offers secure document storage and sharing with OCR, full-text indexing and automatic versioning. Our SREs use HAProxy to gain insights into the usage of our services, diagnose issues, and throttle or reject requests based on HAProxy ACLs. In this talk, you will see how we generate HAProxy configuration files using templates. We will describe our custom log format and demonstrate how we integrate HAProxy data with Kibana.

Keeping your service configuration aligned over hundreds of hosts is never a simple task. This talk will illustrate how the University of Paderborn automated the integration of HAProxy into our infrastructure. As our current generation of load balancer appliances approached the end of life and we thought about improving how we managed our services, our goal was clear: we needed a scalable, consistent, active-active setup of load balancers that could be easily automated with open-source tools. We achieve scalability with Anycast but needed to make sure the configurations could keep up with application changes.

At Willhaben, we run Austria’s largest classified advertising marketplace. We started using HAProxy due to our need to move to an all-HTTPS environment. Since then, we’ve leveraged HAProxy for geo-redundancy, HTTP/2, integration with Kubernetes, and blocking suspicious activity. In this talk, I will explain how we gradually shifted from basic HAProxy functionality to our current deployment, and will also describe where we encountered trouble in our production environment and how we overcame it. We are currently serving 5-6 gigabits per second of peak traffic via HAProxy, with about 20k requests per second.

At Criteo, we work at the cutting edge of commerce marketing, using Machine Learning and Artificial Intelligence to help our customers grow their businesses through hyper-relevant advertising. We run tens of thousands of servers, host containers that continuously move across data centers, and scale services through our managed APIs, with HAProxy playing a critical role across our fast-paced, event-driven infrastructure. This presentation will describe our journey to achieve load balancing served via a user-centric API in such a large and complex environment. We will share tricks and design considerations that helped us to go from a user intent expressed through an API to a scalable service running globally.

DoubleVerify authenticates the quality of digital media for the world’s largest brands. We use HAProxy to accomplish our goals of processing billions of requests daily, increasing and decreasing load balancer pools without affecting customers, and avoiding downtime during upgrades and maintenance periods. We will share our criteria we used when we set out to move from hardware to software load balancers, how we arrived at the decision to use HAProxy, and the results we’ve seen before and afterwards.