/MiniKV

A high-performance minimalist KV database.

Primary LanguageC++

A KV high-performance mini-database based on memory and C++17

This project is inspired by Redis source code.

Command line tools

Developed command line tool kvctl. value type:string

yunfei@ubuntu:~/MiniKV/build$ ./kvctl --key qjx --operate set --value world --ip 43.143.229.22
+--------+
| Status |
+--------+
| OK     |
+--------+
yunfei@ubuntu:~/MiniKV/build$ ./kvctl --key qjx --operate get --ip 43.143.229.22
+-------+
| Value |
+-------+
| world |
+-------+

value type:list

yunfei@ubuntu:~/MiniKV/build$ ./kvctl --key zyf --operate set --value hello --encoding list --ip 43.143.229.22
+--------+
| Status |
+--------+
| OK     |
+--------+
yunfei@ubuntu:~/MiniKV/build$ ./kvctl --key zyf --operate set --value world --encoding list --ip 43.143.229.22
+--------+
| Status |
+--------+
| OK     |
+--------+
yunfei@ubuntu:~/MiniKV/build$ ./kvctl --key zyf --operate get --ip 43.143.229.22
+-------+
| Value |
+-------+
| hello |
| world |
+-------+

search key using regex:

yunfei@ubuntu:~/MiniKV/build$ ./kvctl --keyRex .* --operate findkey --ip 43.143.229.22
+--------+
| Key    |
+--------+
| zyf    |
| yunfei |
| qjx    |
+--------+

build

Dependencies: grpc, protobuf, gflags In the project dir, do:

cd build && cmake .. && make

then you can get kvserver and kvclient.

run

./kvserver

Persistent data frame format

avatar

Performance

Currently, no performance optimization has been done. TPS 10000+.

About

  • This project is based on the gRPC framework to implement the memory-based KV cache middleware, which realizes the client and server respectively. Support list, string type KV cache, data snapshot and progressive Rehash.

  • Communication between client and server is realized based on gRPC, and concurrency security is realized based on read-write lock; The client supports the addition, deletion, modification and query of keys.

  • Support data snapshot, background thread timing persistence, based on user-defined data frame format; Service start automaticly to read snapshot.

  • The underlying storage structure mimics the design of Redis hash table, solves hash conflicts with zipper method, and realizes automatic memory management with intelligent pointer.

  • Imitate Redis to implement the expired key based on the expired hash table, and clean up the expired key through lazy deletion.

  • Simulate the design of Redis double hash table, and implement progressive rehash after the background thread calculates the load factor regularly.