VSDB

VSDB is a 'Git' in the form of KV-database.

Based on the powerful version control function of VSDB, you can easily give your data structure the ability to version management.

Make everything versioned !!

To view the change log check here.

Highlights

• Support Git-like verison operations, such as:
  • Create countless branches and merge them to their parents
  • Rolling back a 'branch' to a specified historical 'version'
  • Querying the historical value of a key on the specified 'branch'
• Most APIs is similar as the coresponding data structures in the standard library
  • Use Vecx just like Vec
  • Use Mapx just like HashMap
  • Use MapxOrd just like BTreeMap
• ...

Examples

Suppose you have a great algorithm like this:

struct GreatAlgo {
    a: Vec<...>,
    b: BTreeMap<...>,
    c: u128,
    d: HashMap<...>,
    e: ...
}

Simply replace the original structure with the corresponding VSDB data structure, and your algorithm get the powerful version control ability at once!

#[dervive(Vs, Default)]
struct GreatAlgo {
    a: VecxVs<...>,
    b: MapxOrdVs<...>,
    c: OrphanVs<u128>,
    d: MapxVs<...>,
    e: ...
}

let algo = GreatAlgo::default();

algo.get_by_branch_version(...);
algo.branch_create(...);
algo.branch_create_by_base_branch(...);
algo.branch_create_by_base_branch_version(...);
algo.branch_remove(...);
algo.version_pop(...);
algo.prune();

NOTE !!

the #[derive(Vs)] macro can be applied to structures whose internal fields are all types defined in VSDB (primitive types and their collections are also supported), but can not be applied to nesting wrapper among VSDB-types, we recommend you to use the multi-key APIs if you indeed require these functions(better performance also), or you will have to implement the VsMgmt trait manually.

This data structure can be handled correctly by #[derive(Vs)]:

#[derive(Vs)]
struct GoodCase<K, T> {
    a: VecxVs<i64>,
    b: SubItem0,
    c: SubItem1,
    d: SubItem2,
    e: u8,
    f: Vec<i16>,
    g: VecDeque<i64>,
    h: BTreeSet<u16>,
    i: HashMap<K, AtomicU64>,
    j: HashSet<i32>,
    k: LinkedList<()>,
    l: Box<dyn AsRef<bool>,
    m: Box<dyn AsRef<[Vec<u128>]>>,
    n: PhantomData<T>,
}

#[derive(Vs)]
struct SubItem0(MapxVs<u8, u8>, VecxVs<u8>);

#[derive(Vs)]
struct SubItem1 {
    a: OrphanVs<i16>,
    b: MapxOrdVs<String, u8>
}

#[derive(Vs)]
struct SubItem2 {
    a: i8,
    b: u128
}

// // A nope implementation of `VsMgmt` for custom stateless types.
// // the `#[derive(Vs)]` on 'SubItem2' is same as this implementation.
// impl VsMgmt for SubItem2 {
//     impl_vs_methods_nope!();
// }

But this one can NOT be handled correctly by #[derive(Vs)]:

// It can be compiled, but the result is wrong !
// The versioned methods of the inner 'MapxVs<u8, u8>' will missing,
// We recommend you to use the 'multi-key' APIs of VSDB, or
// you will have to implement the 'VsMgmt' trait manually.
#[derive(Vs)]
struct BadCase {
    a: VecxVs<MapxVs<u8, u8>>,
}

Please check the multi-key functions if you have requirements of the above or similar scenes.

Some complete examples:

• Versioned examples:
  • Simple blockchain
  • Operations one by one
• Unversioned examples:
  • Simple web server
  • Operations one by one

Compilation features

• [default] sled_engine, use sled as the backend database
  • Faster compilation speed
  • Faster running speed in the versioned functions
  • Support for compiling into a statically linked object
• rocks_engine, use rocksdb as the backend database
  • Faster running speed in the unversioned functions
  • Can not be compiled into a statically linked object
• [default] msgpack_codec, use msgpack as the codec
  • Faster running speed
• bcs_codec, use bcs as the codec
  • Created by the 'Libre' project of Facebook
  • Security reinforcement for blockchain scenarios
• [default] derive, enable the Vs procedural macro
• merkle, enable an optional mekle-tree implementation
• compress, enable compression in the backend database
• hash, enable an optional hash function
  • Based on the 'blake3' crate

Low-level design

Based on the underlying one-dimensional linear storage structure (native kv-database, such as sled/rocksdb, etc.), multiple different namespaces are divided, and then abstract each dimension in the multi-dimensional logical structure based on these divided namespaces.

In the category of kv-database, namespaces can be expressed as different key ranges, or different key prefix.

This is the same as expressing complex data structures in computer memory(the memory itself is just a one-dimensional linear structure).

User data will be divided into two dimensions: 'branch' and 'version', the functions of the 'basic' category are stateless, and the functions of the 'versioned' category are stateful. In the internal implementation, each stateful function is implemented based on its corresponding stateless function, all stateful data has two additional identification dimensions ('branch' and 'version'), somewhat like the logic in Git. Stateless functions do not have the feature of 'version' management, but they have higher performance.

NOTE

• The serialized result of a VSDB instance can not be used as the basis for distributed consensus
  • The serialized result only contains some meta-information(storage paths, etc.)
  • These meta-information are likely to be different in different environments
  • The correct way is to read what you need from it, and then process the real content
• Version names must be globally unique
  • Using a same version name on different branches is also not allowed