Is it worth using a similar balancing method?

Question

Load balancing

Hello.
I would like to learn from professionals in designing architectures of highly loaded projects in a similar approach:
1. A request from a user goes to a cluster balancer.
2. The cluster determines the type of request (for example, that this is a search request - I will consider this example).
3. The cluster selects the closest geographically and least loaded cluster for search queries, after which it sends a request to it and waits for a response from it.
4. The search cluster, in turn, feeds the request first to the cluster for request stemming (selects the closest and least loaded).
5. The search cluster feeds the query to the closest and least loaded cluster for the classifying neural network.
6. The search cluster feeds the query to the closest and least loaded cluster with meta files, where N results are selected and they are sorted by the inverse frequency method.
7. The search cluster collects the answer and sends it to the balancing cluster.
8. Cluster-balancer gives a response to the user.
A string of small length is transmitted - from 1 to 10 words, the response is a JSON string with the specified number (20-100) of elements and the identifier of the last processed meta-file (it also serves as an ID for writing to the DBMS).
The servers themselves (cluster members) are written in Python using raw sockets.
I would like to know: is this method suitable with a focus on 10000+ RPS?
I also think that the balancing cluster itself plays the role of a bottleneck, is it so?

Answer 1

too many redirects
, it is obvious that the balancer and all cluster nodes must be in the same place, otherwise it completely loses its meaning, then send the search request directly to the nearest search server. why the hell is there some kind of balancer and redirection? just another point of failure and an additional node to increase the response time or what?