Avoid duplicated response about lolraft HOT 2 CLOSED

My initial consideration as to this problem is to solve by TTL map (moka can be used).

Let's consider a request (request_id=K) that will yield a response B. Then it should work like this.

1. Initially, there is no entry in the map.
2. When the app sees the request first time, it stores (K, B, ttl=T1) in the map.
3. When the app sees the completion for command with request_id=K, it completes to client and then append CompleteRequest entry. Now the entry is (K, empty, ttl=T2)
4. When app sees a request with request_id=K, it ignores execution because they is an entry for K. Since there is a trace of K has been completed, then drop the completion if exists.
5. When the entry reaches (K, empty, ttl=0), it will be evicted from the map.

from lolraft.

Done.

pub struct ResponseCache {
    responses: spin::Mutex<HashMap<String, Bytes>>,
    completes: TTLSet,
}

ResponseCache keeps the binding of request_id and the response data and request_id for only 5 seconds after completion to the client.

Let's consider this log with 3 entries.

[Exec][Exec (completion)][Exec]

This could happen in this scenario:

1. Send 1st to node A
2. node A down
3. Send 2nd to the node B
4. Retry 3rd to the node B

In this case, the request is executed at 1st and complete at 2nd but 3rd does nothing because responses still has the binding.

After 2nd, new entry [Complete] is queued and replicated. In executing this entry, the node moves the binding to responses to completes to prevent further execution of the requests in same request_id.

from lolraft.