FutureQ - v0

[hertzcodes]

hi ahmad we gotta talk about this.
the replication (raft) is vibe coded and i still haven't reviewed claude's code in details.

only the producers is implemented.

overall I could reach 30k/s write speed (with payloads varying between 0 to 2kb) on a single node with 200 concurrent requests. ( we must find good values for raft per workload though )
also the client is important ( the more concurrent requests the better speed ( because we will have less syncs on dragonboat's pipeline during the same time ))
on 3 node replicas it gives us way less ( around 8k ) with strong consistency.
it was all on localhost btw. so it's defo way less in production
i know it's pretty bad but this is the most durable thing with the least complexity. we might have to add batching, increase concurrency internally in the sdk we are going to write and also support no acks or async disk writes to increase the speed.

there's no batching. only single message acks (always acks) and syncs
we gotta decide how much performance and what features we want for V2.

I don't have any idea for the consumer part and the deletion of keys yet. also the guarantees we want to have on the delivery.

just take a look at it and tell me what you think.
I think using pebble is actually a good idea since dragonboat uses pebble itself underneath.

[hertzcodes]

@anvari1313 I think we can finally merge this :D
still a lot to go through but we can start testing the features and eventually get rid of the AI slop parts