How to bring the site uptime to 100% as much as possible?

Question

System administration

I started developing a promising project, but so far it doesn’t fit in my head, how to make the site work smoothly? This is very important and will work on the credibility of the project. Obviously, one dedicated server is not enough for such purposes ...

Maybe there are links, or please explain the basic principles of serious sites?

Answer 1

Google keywords: cluster, dns-balancer, load-balancer. The notorious clouds partially solve this problem. But keep in mind that cloud hosting is usually virtual machines with all the ensuing consequences: first of all, a slow disk.
The second is well-written code. If 100500 database queries are made to generate one page, then there will not be enough clusters :)

Answer 2

Principle? Avoid single points of failure in the system.

Answer 3

1. Own / rented server in a reliable DC.
2. Straight-handed admin + automatic monitoring of key indicators (ping, port availability, running processes, etc.).
3. Not Windows.
I don’t know what kind of project you have so tricky, but before spreading about clusters, clouds and other trends, it would be nice to have some details.

Answer 4

Go to the clouds, to the Amazon for example, or there to oversun. Only cloud hosting can really guarantee almost 100% online. At first, it will cost you about the same as regular hosting, then, of course, with some load it becomes more expensive than keeping your own server, but cheaper than keeping two of your servers and fooling yourself with data synchronization.
When you clearly go beyond the capacity of a pair of physical servers in terms of load, then you will hire normal engineers and begin to equip your data center.

Answer 5

1) clean code - do not use old engines for 38 requests for the formation of 1-page
2) a good admin
3) clouds are not a panacea, but at the first stage you can get into them
4) a good comment in the post above, I will supplement it: nginx upstream, mysql- proxy, mysql clauster (there are similar solutions for both pg and mongo - depending on what you use)

Answer 6

You need to configure High-Availability software. Like Heartbeat or Pacemaker . Or use High-Availability technologies offered by modern hypervisors (VmWare VSphere, MS HyperV).
The difference is that HA software works at the hardware level and the hardware must be configured in exactly the same way. In case of failure, migrates the service to an available node. And in HA virtualization technologies, for example, you set up a virtual machine and it runs on a hypervisor. In the event of a failure, the entire virtual machine migrates to an available server.
I have worked with Heartbeat and Pacemaker. I will say that Heartbeat is certainly a crutch. Supports up to two nodes, does not know how to independently determine the failure of a resource and work it out. All these shortcomings are deprived of Pacemaker. Mega convenient crap, but there is still little documentation on it, since this is a leading-edge technology.