Talent Plan Key-Value Store

Talent Plan, an open source training program initiated by PingCAPcourse, this project is "TP 201: Practical Networked Applications in Rust", implements a Key-Value Store in Rust.

Note

Multiple threads

1. Using BufReaderWithPos make performance a little bit worse, maybe file::seek is efficient enough, meantime if-branchs are extra price?
2. KvStore::index_map would be used both by readers and the writer(singleton), RwLock would be a little bit inefficient(writer is locked during index_map is writting-locked for updating), lock-free map is prefered. Using DashMap now.
3. Possible optimization: engine is cloned for each incoming connection, it leads to create ReaderMap for every request, which is not necessarily, considerating using syncpool.
   • before optimization, clones engine for each connection:

     for stream in listener.incoming() {
         let engine = self.engine.clone();
         self.pool.spawn(move || {
             serve(engine, stream);
             // ...
         })
     }

   • draft the optimization, replace KvsEngine::clone() with Arc<SyncPool>::clone(), 25% performance improved in thread-pool-size=1, no change in other size, interesting...

     pub struct KvsServer<E: KvsEngine, P: ThreadPool> {
         // ... other fields
         engine_pool: Arc<SyncPool<E>>,
     }
     for stream in listener.incoming() {
         let engine_pool = self.engine_pool.clone();
         self.pool.spawn(move || {
             serve(engine_pool, stream);
             // ...
         })
     }