Highload 2014, full HA
As part of the new HL, I will talk about the most perfect HA. If you think that HA comes down to balancing on the front, ipvs, outputting and entering nodes within roles - you are very much mistaken. A full-fledged HA is the work of a huge number of people, these are huge resources, and, more importantly, attention in every detail. This post was created to make it clear how people’s influence is projected on projects. I went through many projects. am.ru was the finale of my concept of cool and the final of "how not to."
I really want to tell you everything is very accessible and simple, to be honest, the words “You have done so much ...” torment me a little - I was just given this opportunity by chance — it would be much more correct. By the way, this is a really huge problem when applying for a job after such projects, your resume is not answered, not because you are many years old, but because you simply don’t know (as they think) how to hammer in your nails with such a microscope.
So: 8 years ago, I wondered about fault tolerance. It was at the time of the permanent transition from 2.4 kernel to 2.6, disputes about the essence of being and sorts-based gentoo, slackware and production-ready caps, debian and other tasty, but partially framework systems, why the “framework” is not today.
What were our options? There was a holy NAT, there was haproxy, there was nginx, which until now can only proxy the web and mail and was ipvs on which I spent a lot of time and have no regrets about it, I had the opportunity to catch Igor and mail him with questions.
')
The first project in the framework of HL, which turned out and went into production - it was WebIndex in the framework of TNS Russia (Gallup) - it was a very interesting project, a year and a half long, with a very good test bench, synthetics on tests, an awesome team of people who work there, it was a real highload, with bigdata. This experience became the basis of the beginning of the study of HA.
And they called me from scratch to build a whole data center. Well, it turned out. You will laugh, but only a few people built such a huge data center. I ask to consider that I speak about IT. And do not judge strictly.
I really want to tell you everything is very accessible and simple, to be honest, the words “You have done so much ...” torment me a little - I was just given this opportunity by chance — it would be much more correct. By the way, this is a really huge problem when applying for a job after such projects, your resume is not answered, not because you are many years old, but because you simply don’t know (as they think) how to hammer in your nails with such a microscope.
So: 8 years ago, I wondered about fault tolerance. It was at the time of the permanent transition from 2.4 kernel to 2.6, disputes about the essence of being and sorts-based gentoo, slackware and production-ready caps, debian and other tasty, but partially framework systems, why the “framework” is not today.
What were our options? There was a holy NAT, there was haproxy, there was nginx, which until now can only proxy the web and mail and was ipvs on which I spent a lot of time and have no regrets about it, I had the opportunity to catch Igor and mail him with questions.
')
The first project in the framework of HL, which turned out and went into production - it was WebIndex in the framework of TNS Russia (Gallup) - it was a very interesting project, a year and a half long, with a very good test bench, synthetics on tests, an awesome team of people who work there, it was a real highload, with bigdata. This experience became the basis of the beginning of the study of HA.
And they called me from scratch to build a whole data center. Well, it turned out. You will laugh, but only a few people built such a huge data center. I ask to consider that I speak about IT. And do not judge strictly.
Source: https://habr.com/ru/post/239595/
All Articles