English | 中文

This is rotom, a high performance, low latency tiny Redis Server written in Go. It replicates the core event loop mechanism AeLoop in Redis based on I/O multiplexing.

• Implements the AeLoop single-threaded event loop from Redis using the epoll network model.
• Excellent optimization, achieving native Redis performance.
• Compatible with the Redis RESP protocol, allowing any Redis client to connected.
• Implements data structures String, Hash, Set, List, ZSet.
• Supports RDB & AOF.
• Supports 20+ commonly used commands.

AeLoop (Async Event Loop) is the core asynchronous event-driven mechanism in Redis, which mainly includes:

1. FileEvent: Uses I/O multiplexing to handle read and write events on network sockets, categorized into READABLE and WRITABLE.
2. TimeEvent: Handles tasks that need to be executed after a delay or periodically, such as expiring items every 100ms.
3. When events are ready, they are processed by callback functions bound to those events.

In rotom, the AeLoop event loop mechanism in Redis is replicated, specifically:

1. When a new TCP connection arrives, the AcceptHandler obtains the socket fd and adds it to the event loop, registering a read event.
2. When the read event is ready, ReadQueryFromClient reads buffered data into queryBuf.
3. ProcessQueryBuf parses and executes the corresponding commands from queryBuf.
4. The command execution result is saved, and the socket fd's write event is registered.
5. When the write event is ready, SendReplyToClient writes all results back to the client. A write event may return multiple read event results at once.
6. Resources are released, and the process continues until the service is shut down.

Rotom has made several optimizations in data structures:

• dict: Rotom uses stdmap as the db hash table, with built-in progressive rehashing.
• hash: Based on zipmap with higher memory efficiency.
• set: Uses zipset when the set is small and mapset when it is large.
• list: Uses a quicklist based on listpack for a doubly linked list.
• zset: Uses zipzset when small and hash + skiplist when it is large.

Notably, zipmap and zipset are space-efficient data structures based on listpack, which is a new compressed list proposed by Redis to replace ziplist, supporting both forward and reverse traversal and solving the cascading update issue in ziplist.

The test will run rotom on the same machine with appendonly disabled, and use redis-benchmark tool to test the latency of different commands.

goos: linux
goarch: amd64
pkg: github.com/xgzlucario/rotom
cpu: 13th Gen Intel(R) Core(TM) i5-13600KF

• Support for LRU cache and memory eviction.
• Gradually rehashing in dict.
• RDB and AOF Rewrite.

Running Locally

First, clone the project to your local machine:

git clone https://github.com/xgzlucario/rotom

Ensure your local Go environment is >= 1.22. In the project directory, run go run . to start the service, which listens on port 6379 by default:

$ go run .
2024-07-18 23:37:13 DBG 
 ________      _____                  
 ___  __ \_______  /_____________ ___   Rotom 64 bit (amd64/linux)
 __  /_/ /  __ \  __/  __ \_  __ '__ \  Port: 6379, Pid: 15817
 _  _, _// /_/ / /_ / /_/ /  / / / / /  Build: 
 /_/ |_| \____/\__/ \____//_/ /_/ /_/

2024-07-18 23:37:13 INF read config file config=config.json
2024-07-18 23:37:13 INF rotom server is ready to accept.

Running in a Container

Alternatively, you can run it in a container. First, build the Docker image by running make build-docker:

REPOSITORY       TAG           IMAGE ID       CREATED         SIZE
rotom            latest        0cd91943967a   5 seconds ago   20.9MB

Then, start the container:

docker run --rm -p 6379:6379 --name rotom rotom:latest