This repository the implementation:

1. Generic optimistic transaction manger, which supports concurrent execution of transactions, providing serializable snapshot isolation, avoiding write skews.
2. An atomicity, consistency, isolation, MVCC and WASM friendly in-memory database based on the transaction manager.

English | 简体中文

Introduction

This repository contains a transaction framework (async-txn and txn) based on optimistic concurrency control, which is inspired by foundationdb's paper and badger.

This repository contains two kinds of in-memory key-value database which supports both async (async-skipdb) and sync (skipdb):

1. SerializableDb

   Supports both concurrent execution of full serializable snapshot isolation transactions and optimistic concurrency control transactions.

   Transactions are created by SerializableDb::serializable_write can handle all kinds of write skew correctly.

   Transactions are created by SerializableDb::optimistic_write can handle all kinds of direct dependent write skew, but cannot handle all kinds of indirect dependent write skew e.g. https://wiki.postgresql.org/wiki/SSI#Intersecting_Data.

2. OptimisticDb

   Only support oncurrent execution of optimistic concurrency control, which means the write transaction cannot detect all kinds of write skew.

   All kinds of direct dependent write skew can be handled correctly, but cannot handle all kinds of indirect dependent write skew e.g. https://wiki.postgresql.org/wiki/SSI#Intersecting_Data.

Features

• Atomicity, Consistency, Isolation, MVCC, concurrent safe and almost lock-free.
• No extra allocation and copy, there is no Arc wrapper for both key and value stored in the database, which means that users provide K and V, and database store K and V directly.
• Zero-copy and in-place compaction, which means there is no copy, no extra allocation when compacting.
• Concurrent execution of transactions, providing serializable snapshot isolation, avoiding write skews.
• Both read transaction and write transaction are Send + Sync + 'static, which means you do not need to handle annoying lifetime problem anymore.
• Lock-free and concurrent safe read transaction: the read transaction is totally concurrent safe and can be shared in multiple threads, there is no lock in read transaction.
• BTreeMap like user friendly API and all iterators implement Iterator trait, which means users use Rust powerful conbinators when iterating over the database.
• Async version is runtime agnostic, tokio, async-std, smol, wasm-bindgen-futures and any other async runtime.
• 100% safe, sets [forbid(unsafe_code)].

Installation

• For sync

  [dependencies]
  skipdb = "0.2"
  

• For async

  • tokio

    [dependencies]
    async-skipdb = { version = "0.2", features = ["tokio"] }
    

  • async-std

    [dependencies]
    async-skipdb = { version = "0.2", features = ["async-std"] }
    

  • smol

    [dependencies]
    async-skipdb = { version = "0.2", features = ["smol"] }
    

  • wasm-bindgen-futures

    [dependencies]
    async-skipdb = { version = "0.2", features = ["wasm"] }
    

Examples

Please see skipdb or async-skipdb.

License

skipdb is under the terms of both the MIT license and the Apache License (Version 2.0).

See LICENSE-APACHE, LICENSE-MIT for details.

Copyright (c) 2024 Al Liu.

lib.rs:

Watermark and Closer implementation for implementing transaction.