Redis Clone in C++

A minimal Redis server built from scratch using C++. No frameworks, no interpreters, just raw socket programming, data structures, and systems-level engineering.

Supported Commands (so far)

SET key value
GET key
DEL key

Supports pipelined requests: fire off multiple commands without waiting for replies in between. Efficient and clean.

---

Event Loop based: Why No Threads?

Instead of multithreading, this Redis clone uses a custom event loop built with poll(). Here's why:

• Threaded IO doesn't scale well for high-connection servers: C10k problem
• Memory-heavy: Each thread = stack space.
• Short-lived clients (like PHP apps) = thread churn = CPU + latency hit.
• Multi-process/threading makes worst-case latency worse.
• Modern high-perf servers (like Nginx & Redis) use event loops.

Bottom Line:

Event loop > Threads when scaling to thousands of concurrent connections.

Side note:

A production implementation should replace poll() with epoll() as it stores the FDs in the kernel so that the FD args aren't copied in every itereation. Since this is a small scale implementation, I took the liberty to just use poll().

---

Why C++?

C++ gives low-level control like C, plus:

• std::string, std::vector = fewer memory bugs cuz we don't wanna cleanup the custom Dynamic buffers/arrays as they would be passed around a lot.
• Zero Python/Node-style abstraction overhead
• Built a custom dynamic buffer system inspired by vectors: minimal, efficient, tailored to RESP protocol parsing.

---

Hashmaps ≠ STL

I didn't use std::unordered_map. Here's why:

STL Isn't Built for Real-Time Systems

• STL is throughput-optimized, not latency-safe.
• Worst-case: an insert triggers O(N) resize, pausing everything cause who needs that :O

Solution: Progressive Resizing

• On resize, keys are migrated gradually, not all at once.
• Uses calloc() over malloc()+memset() to:
  • Leverage mmap-backed lazy-zeroing
  • Avoid O(N) latency when initializing large memory chunks

Collision Strategy: Chaining (Not Open Addressing)

• I used linked lists per bucket
• Benefits:
  • Stable references (no pointer invalidation during resize)
  • O(1) inserts and deletes
  • Easier to build using intrusive data structures

---

Intrusive Data Structures

Forget templates and void pointers.

I embeded structure nodes inside our data and use a container_of() macro (inspired by the Linux kernel) to access data from within a node.

Advantages:

• No heap allocations for structure nodes
• Better cache locality
• Enables multi-indexed data structures
• Shares nodes across collections like Sorted Sets (score + name)

---

In Progress

• AVL Trees (for balanced Sorted Sets)
• Timers & Timeouts
• TTL & Expiration (EXPIRE, PERSIST)

---

Run It

g++ -Wall -Wextra -O2 -g server.cpp hashtable.cpp murmurhash.cpp -o server
./server
g++ -Wall -Wextra -O2 -g 03_client.cpp -o client
./client

Resources

Redis Internals

Beej’s Guide to Network Programming Using Internet Sockets by Brian “Beej” Hall along with man7 pages for detailed reference.

Network programming, data structures, and low-level C by James Smith