What type of bonding to choose for a fault-tolerant network?

Question

linux

Task: to combine groups of servers on Linux located in the same data center into a fault-tolerant network so that the failure of any link or switch does not lead to a loss of connectivity. As a bonus, get an increase in bandwidth where possible. Each server has 2 network cards.

For such tasks, Linux has bonding. Here is the most complete documentation I have found.

Question: what mode of bonding'a and what network organization between switches to choose?

If you follow the documentation, it turns out that only active-backup is suitable for fault tolerance:

11.2.1 HA
Bonding Mode Selection for Multiple Switch
Topology the other modes require all links to terminate on the
same peer for them to behave rationally.

That is, it is argued that when working in other modes, all links from hosts must end in one switch, otherwise they will work "irrationally". But it is not clear why this is so.

Wouldn't a scheme like this work, where each host binds to 2 switches in balance-alb mode, the switches for bonding are not specially configured in any way, but at the same time each switch is connected to 2 others using STP? Then, in theory, we will get both an increase in throughput and fault tolerance.

  +-------+         +--------------+      +--------------+         +-------+
  | HostA |.........|      Switch1 |......|      Switch3 |.........| HostX |
  +-------+ .     . +--------------+      +--------------+ .     . +-------+
             .   .         .                      .         .   .  
              . .          .                      .          . .   
   bonding     x           .         STP          .           x   bonding
              . .          .                      .          . .  
             .   .         .                      .         .   .  
  +-------+ .     . +--------------+      +--------------+ .     . +-------+
  | HostB |.........|      Switch2 |......|      Switch4 |.........| HostY |
  +-------+         +--------------+      +--------------+         +-------+

Or is there a more correct solution for this task that I did not see?

Answer 1

In your scheme, it all depends on how the switch reacts to the fact that the same MAC in its switching table will appear behind two different ports. For example, on switch4, MAC HostA will be on ports to Switch1 and Switch3, provided that STP has disabled the link between Switch1 and Switch2

Answer 2

If the switches are able to mLACP, then you can active / active with links to different switches.