Load balancing across multiple frontends for fault tolerance and highload?

Question

Highload

I would like to know how the load is balanced between several front-end servers.

I know about the method with several A records in DNS with a small TTL. I know that some devices are used by the type of virtual IP, which, in the event of a failure of the main server, is tied to the backup.

This is all good, of course, but how do they balance between frontends in real HIGHLOAD projects, where fault tolerance is important?

Answer 1

CDN

Answer 2

magic boxes (balancers). How they are arranged I need to know.

Here are a dozen load balancers with source codes . Of these, the most popular is perhaps HAproxy, although the rest have a lot to learn. Manuals on the topic wagon and bogie.

Answer 3

Nowadays, most often Kubernetes scatters requests across pods, in which both web servers with front statics and backend microservices spin.

Answer 4

Various techniques in balancing are explained here.

Answer 5

If you do not have a cloud, but physical servers, then you need a couple of balancer servers with a floating VIP.
How to organize VIP migration - for example, keepalived, pacemaker,
how to organize balancing - you need to select for tasks from the good old L4 balancer LVS (Linux Virtual Server) in DR mode for maximum return, to ordinary nginx and haproxy which can both L4 and L7.

Answer 6

1) Uses a cluster(s) of load balancers.
They are built either on their own, or use ready-made cloud services.
An example of the first, except for open source products, is https://www.f5.com/products/big-ip-services/iserie... .
An example of cloud services is Network/Application Load Balancer on AWS.
2) When you need to give different servers for different countries (geolocation) - there are solutions for building your own and ready-made services
See "geolocation dns" in Google for details.