bhushanmahajan25/distributed-cache

mini redis like server from scratch using C/C++

Distributed Cache

This project attempts to implement Redis like cache server from scratch using C/C++. This cache server runs on 6379.

How to run

1. Clone the repository and make sure you have bin and obj directories after cloning. If not, then create both of them.
2. Run make compile.
3. Open a terminal for the server and run ./bin/server.
4. Open another terminal for the client and run ./bin/client [command in lowercase].

Supported Commands:

1. SET: set a key-value pair. Ex. set k v
2. GET: get a value for key. Ex. get k
3. DEL: deletes a key-value pair. Ex. del k
4. KEYS: lists all the keys. Ex. keys

Note: Provide all the commands in lowercase.

Following are the brief features of the cache server:

• Client/Server Communication:

  • Client-Server connection is made using TCP/IP sockets.

• Protocol Parsing:

  • The protocol or command is a simple list of strings, ex. set key val.

• Data Serialization:

  • This project implements serialization protocol of 5 data types:
    • Null
    • String
    • Error code and message
    • Int
    • Array

• Event Loop:

  • To handle multiple clients in non-blocking IO mode, event loop is implemented using poll() syscall.

• Data Structure:

  • The server implements a chaining hashtable. A chaining hashtable is easy to code; it doesn’t require much choice-making.

• The sorted set and AVL tree:

  • To implement sorted set we need AVL tree. It offers the ability to sort your data in order, but also has the unique feature of querying ordered data by rank.
  • The real Redis project uses skiplist.

• TTL using Heap:

  • Heap is used to implement TTL to kick out idle TCP connections.

• Thread Pool and Asynchronous tasks

  • If the size of a sorted set is huge, it can take a long time to free its nodes while deleting the keys and the server is stalled during the destruction of the key. This can be easily fixed by using multi-threading to move the destructor away from the main thread.

Future upgradation

• Using epoll() instead of poll() since, the argument for the poll can become too large as the number of fds increases.
• Implementation of RAFT consensus algorithm for leader election and log replication.