What do you need to create a failover cluster?

Question

NAS

Good afternoon!
(I surfed the toaster, but did not find answers to the questions)
There was a need for an inexpensive, fault-tolerant, scalable solution on Windows to use 1C. I won’t set it up, I’m asking for an understanding of the general structure, operation and licensing,
but, unfortunately, there were a lot of ambiguities.
Input data:
Users from 1C will work on the server via RDP. Bases both file, and SQL. The size and number of databases, the number of users will constantly grow.
It is necessary:
​​To create a fault-tolerant scalable solution with a price / quality ratio with uninterrupted (or minimal downtime) user experience with 1C. Load balancing between servers.
Questions arose (if I write something wrong, I’m mistaken, then feel free to correct me, I’m just glad to find the truth).
Hardware questions:
1. I understand that for a minimally fault-tolerant (highly available) cluster, we will need 3 pieces of iron? 2 for working servers with load balancing between them (so that there is no iron downtime) and 1 for NAS, where 1c databases will be stored (file and SQL. SQL server will be here?), backups (along with storage on remote storage), user profiles, user mounted drives, and shares.
2. From point 1 we see that the weakest point is NAS. By disk for reliability and speed RAID 10, 2 power supplies. But what to do in case of a hypothetical NAS failure? (the motherboard burned out, etc.). Keep exactly the same piece of iron and replicate everything to it? Given that the NAS will rarely fail, it's just an idle server, maybe for years ...
As an option, use 3 identical pieces of iron. On the NAS, respectively, the processors are slower, there is not much RAM. All data from the NAS is replicated to one of the working servers (add disks in the mirror for the replica). In the event of a NAS failure, we poke the disks into a working server, and we will have it as a temporary NAS and 1 working server remains for the duration of the NAS repair. Downtime = time to transfer disks from one server to another.
There may be other more rational options, tell me as experts.
3. When resources are exhausted, how will new physical servers be added to the cluster?
Do we purchase a Win license and programmatically include a piece of hardware in a cluster? Or is it not so simple?
4. On working servers, we use SAS disks in the mirror. On NAS - SSD in RAID 10 (and is it better to have 8 SSDs by 256 than 4 by 512, so that performance is better?).
Or not?
Questions on software and licensing:
1. Got confused in clusters. 1C as software has a cluster implementation, windows has a cluster implementation. What to use? Given the licensing of all this.
https://infostart.ru/public/307973/- i.e. on win it is possible not to use a cluster, but only with the help of 1s? Cons of this? Or do you need to use both? Or one of? Confused…
2. How many licenses will be required for a Windows cluster? Windows, SQL, 1C (excluding licenses depending on the number of users on the server).
I understand that it is more profitable to use the free Microsoft Hyper-V Server with Failover Clustering (failover + load balancing). And further, licenses for the number of virtual machines (Windows Stand gives 2 licenses). How many licenses and virtual machines will there be in the end?
With SQL how to be? How to implement its fault tolerance? What version and license of SQL will be needed for this?
3. Has anyone bought licenses from auctions? (companies are bankrupt, their property was put up for auction, as a result, a license can be bought decently cheaper). Is it possible to do so? Or only from resellers of official companies?
4. To make it cheaper, it is not necessary to use 2016 licenses, can you buy older and cheaper 2012 ones? (win, SQL, terminal, user)
5. How to deal with licenses - transfer them to the adjuster or enter the owner himself? (for greater safety of the safety of licenses)
Thank you for your time. Thanks for sharing your experience.

Answer 1

This is all a finger in the air, a description of beautiful technologies. The result of such aimless activities is wasted time and money. First you need to decide what availability is now and what you need, based on this, reserve the nodes.

Answer 2

1 Reconcile.
2 Cluster - a group of standard servers connected to each other via a high-speed network and equipped with special software that allows you to direct their resources to solve common problems /canonical definition/.
3 It follows from the definition that there are at least three sources of failure: servers, network, special software. The probabilities of failures, which are multiplied with all the consequences.
4 CAP theorem
5 From all this it follows that you need to choose one thing:
CA in all nodes, the data is consistent and available, while it sacrifices resistance to decay into sections;
CP provides a holistic result at every moment and is able to function in conditions of decay, but achieves this at the expense of accessibility;
AP is not guaranteed integrity, but the conditions of availability and resistance to decay into sections are met.
6 By experience. If you decide to make a cluster FS to get rid of many pitfalls, be prepared that it will someday fall apart. This is not a panacea.
7 Quite a few cluster builders forget or give up on the high performance network requirement. There should not just be 1 GB / s, there should be at least 1 GB / s, interface aggregation and a normal hard switch with a large and high-speed cache, minimal delays. But 10 GB/s is better. This is quite an expensive pleasure.