#entity #sled-database #sled #bincode #rocksdb #embedded-database

macro reindeer-macros

A small entity-based embedded database with a minimal no-SQL relationnal model, written in pure Rust

1 unstable release

0.3.0 Mar 10, 2023

#24 in #sled-database


Used in reindeer

MIT license

33KB
299 lines

Reindeer 🦌

Reindeer 🦌 lifts your sled!

A small structural layer on top of sled, serde and bincode

Reindeer is a small embedded entity store built on top of sled, using serde and bincode for serialization, written entirely in rust.

It serves as a convenient middle ground to store, retreive and update structs in an embedded database with a bare-minimum relationnal model.

Getting Started

Create a sled database

use reindeer::Db;
let db = reindeer::open("./my-db")?;

💡 Since this is just a sled DB, this object can be copied and sent accross threads safely.

From there, you have two options :

  • Derive the Entity trait
  • Implement the trait manually.

Implementing the Entity trait

By using the derive macro:

Entities need to implement the Serialize and Deserialize traits from serde, which are conveniently re-exported from reindeer

use reindeer::{Serialize,Deserialize,Entity}

#[derive(Serialize,Deserialize,Entity)]
pub struct MyStruct {
    pub id : u32,
    pub prop1 : String,
    pub prop2 : u64,
}

If your struct already has an id field, then it will be used as the key for your store. Its type must either be an integral type, a String or a Vec<u8>, or a tuple of those types.

The name of your store in the database will be the name of the entity, with its original case. Make sure, in this case, that it's the only entity with this name.

Otherwise, you can use the entity helper attribute to specify a different key field and name :

#[derive(Serialize,Deserialize,Entity)]
#[entity(name = "user", id = "email")]
pub struct User {
    pub email : String,
    pub prop1 : String,
    pub prop2 : u64,
}

By Implementing the Entity trait manually

Entities need to implement the Serialize and Deserialize traits from serde, which are conveniently re-exported from reindeer:

use reindeer::{Serialize,Deserialize,Entity}

#[derive(Serialize,Deserialize)]
pub struct MyStruct {
    pub id : u32,
    pub prop1 : String,
    pub prop2 : u64,
}

Then you need to implement the Entity trait and implement three methods : get_key, set_key and store_name, as well as define an associated type, Key

  • Key is the type of the identifier for each instance of your entity ("primary key"). It must implement the AsBytes trait. 😌☝ It's already implemented for String, u32, i32, u64, i64 and Vec<u8>, as well as for any 2-elements tuple of those types, so you should not need to implement it yourself.

  • The key represents the unique key that will be used to identify each instance of your struct in the database, to retreive and update them, it is of type Key

  • The store_name is the name of the entity store. It should be unique for each Entity type (see it as the table name).

use reindeer::{Entity, Serialize,Deserialize};

#[derive(Serialize,Deserialize)]
struct MyStruct  { key : u32, prop1 : String }

impl Entity for MyStruct{
    type Key = u32;
    fn store_name() -> &'static str {
        "my-struct"
    }
    fn get_key(&self) -> &Self::Key {
        &self.key
    }
    fn set_key(&mut self, key : &Self::Key) {
        self.key = key.clone();
    }
 }

Register your entity with the system

Register the entity once, when you launch your application.

let db = reindeer::open("./my-db")?;
MyStruct::register(db)?;

💡 Registering the entity will make it possible for Reindeer to handle safe deletion of entity entries. Without this, trying to delete an unregistered entity entry will result in an error.

Save an instance to the database

You can now save an instance of your struct MyStruct to the database :

let db = reindeer::open("./")?;
let instance = MyStruct {
    id : 0,
    prop1 : String::from("Hello"),
    prop2 : 2335,
}
instance.save(&db)?;

💡 If id 0 already exists in the database, it will be overwritten!

Retreive an instance from the database

let instance = MyStruct::get(0,&db)?;

Retreive all instances

let instances = MyStruct::get_all(&db)?;

Get All entities respecting a condition

let instances = MyStruct::get_with_filter(|m_struct| {mstruct.prop1.len > 20},&db)?;

Delete an instance from the database

MyStruct::remove(0,&db)?;

Defining Relations

reindeer has three types of relations :

  • sibling : An entity that has the same key in another store (one to one relation)
  • parent-child : An entity which key is composed of its parent's key and a u32 (as a two-element tuple) for efficient one-to-many relations
  • free-relation you freely connect two instances of two separate Entities together. This can be used to achieve many-to-many relationships, but is less efficient than sibling and parent-child relationships in regard to querying the database. Use when sibling and parent-child are not possible.

Sibling relationships

To create a sibling Entity, you need to link the Entity structs together by overriding the specifying sibling stores and what happens when we delete one of them.

Sibling stores must share the same key type (and thus matching entities will have the same id).

💡 DeletionBehaviour determines what happens to the sibbling when the current entity is removed :

  • DeletionBehaviour::Cascade also deletes sibling entity
  • DeletionBehaviour::Error causes an Error if a sibling still exists and does not delete the source element
  • DeletionBehaviour::BreakLink just removes the entity without removing its sibling.

With the derive macro

You can specify siblings using the siblings helper attribute :

#[derive(Serialize,Deserialize,Entity)]
#[entity(name = "user", id = "email")]
#[siblings(("user_data",Cascade),("user_data2",Cascade))]
pub struct User {
    pub email : String,
    pub prop1 : String,
    pub prop2 : u64,
}

#[derive(Serialize,Deserialize,Entity)]
#[entity(name = "user_data", id = "email")]
#[siblings(("user",Error),("user_data2",Cascade))]
pub struct UserData {
    pub email : String,
    pub prop3 : String,
    pub prop4 : String,
    pub prop5 : i64
}

In the above example, deleting a User instance also deletes its sibling UserData instance, but deleting the UserData instance causes an error and deletes neither.

Manually

use reindeer::{Entity,DeletionBehaviour};
impl Entity for MyStruct1{
    /* ... */
    fn store_name() -> &'static str {
        "my_struct_1"
    }
    fn get_sibling_stores() -> Vec<(&'static str,DeletionBehaviour)> {
        return vec![("my_struct_2",DeletionBehaviour::Cascade)]
    }
}

impl Entity for MyStruct2{
    /* ... */
    fn store_name() -> &'static str {
        "my_struct_2"
    }
    fn get_sibling_stores() -> Vec<(&'static str,DeletionBehaviour)> {
        return vec![("my_struct_1",DeletionBehaviour::BreakLink)]
    }
}

💡 if sibling trees are defined, an entity instance might or might not have a sibling of the other Sibling store! Siblings are optionnal by default.

In the above example, deleting a MyStruct1 instance also deletes its sibling MyStruct2 instance, but deleting the MyStruct2 instance leaves its sibling MyStruct1 instance intact.

💡 Sibling Entities must have the same Key type.

Creating a sibling entity

let m_struct_1 = MyStruct1 {
    /* ... */
};
let mut m_struct_2 = MyStruct2 {
    /* ... */
};
m_struct_1.save(&db)?;
m_struct_1.save_sibling(m_struct_2,&db)?;

💡 this will update m_struct_2's key to m_struct_1's key using the set_key method, so it does not matter which key you initially provide before calling save_child.

⚠️ Note that if you create an entity in MyStruct2's store with the same key as an entity in MyStruct1's store without using save_sibling, the result is the same, and the two entities will be considered siblings all the same.

Retrieving a sibling entity

if let Some(sibling) = m_struct_1.get_sibling::<MyStruct2>(&db)? {
    /* ... */
}

💡 Note that a sibling may or may not be present, thus the Option type.

Parent-child relationship

For a parent-child relationship between entities to exist, the child entity must have a Key type being a tuple of :

  • The parent Key type
  • u32

💡 Children entities will be auto-incremeted and easily retreived through their parent key.

Using the derive macro

You can define child stores the same way you define sibling stores, but using the children helper attribute:

#[derive(Serialize,Deserialize,Entity)]
#[entity(name = "user", id = "email")]
#[children(("document",Cascade))]
pub struct User {
    pub email : String,
    pub prop1 : String,
    pub prop2 : u64,
}

#[derive(Serialize,Deserialize,Entity)]
#[entity(name = "document")]
pub struct Document {
    pub id : (String,u32),
    pub prop3 : String,
    pub prop4 : String,
    pub prop5 : i64
}

Manual implementation

impl Entity for Parent{
    type Key = String;
    /* ... */
    fn store_name() -> &'static str {
        "parent"
    }
    fn get_child_stores() -> Vec<(&'static str)> {
        return vec![("child", DeletionBehaviour::Cascade)]
    }
}

impl Entity for Child{
    type Key = (String, u32);
    /* ... */
    fn store_name() -> &'static str {
        "child"
    }
}

In the above example, deleting the parent entity will remove all child entities automatically (thanks to the Cascade deletion behaviour).

For database integrity, it is strongly advised not to use DeletionBehaviour::BreakLink on parent/child relations, and instead use either Error of Cascade

Adding a child entity

let parent = Parent {
    /* ... */
};

let mut child = Child {
    /* ... */
}

parent.save_child(child,&db)?;

💡 this will update child's key to parent's key and an auto-incremented index using the set_key method, so it does not matter which key you initially provide before calling save_child.

Getting Children

let children = parent.get_children::<Child>(&db)?;

Free relations

Free relations follow the same pattern as other relation types, except they are freely created between any two entities. This can be used to achieve many to many relationships.

💡 Creating a free relation will automatically create its opposite relation, making it two-way.

Linking two entities

let e1 = Entity1 {
    /* ... */
};

let mut e2 = Entity2 {
    /* ... */
}

e1.create_relation(e2,DeletionBehaviour::Cascade, DeletionBehaviour::BreakLink,None,&db)?;

In the above example, deletion behaviour in both ways are provided : deleting e1 will automatically delete e2, but deleting e2 will leave e1 untouched and break the link between them.

DeletionBehaviour::Error is also an option here.

let related_entities = e1.get_related::<Entity2>(db)?;

To get only the first related entity from the other tree, use

let related_entity = e1.get_single_related::<Entity2>(db)?;

A name must have been supplied when creating the relation :

e1.create_relation(e2,DeletionBehaviour::Cascade, DeletionBehaviour::BreakLink,Some("main"),&db)?;
let related_entities = e1.get_related_with_name::<Entity2>("secondary",db)?;

To get only the first related entity from the other tree, use

let related_entity = e1.get_single_related_with_name::<Entity2>("main",db)?;

If needed, you can remove an existing link between entities:

e1.remove_relation(other,db)?;

or

e1.remove_relation_with_key::<OtherEntity>(otherKey,db)?;

Deadlocks 🔒

When defining DeletionBehaviour for your relations, be careful not to create deadlocks.

For instance, if two siblings mutually define a DeletionBehaviour::Error link, then none of them can ever be removed...

Also, be aware of the cycles you create in databases. While you can create relation cycles safely, the same deadlock rules as above apply, and the library will not detect them until you try to delete something.

Performance

While Sibling and Parent-child relations are performant by default, Free relations are less performant and rely on hidden object stores to work, forcing reads and writes to the database on relation creation and entity deletion. Be aware of this pitfall.

Also, defining cascading relations will run through relations reccursively when deleting entities, making the operation heavier than relation-less entities.

Auto-incrementing entities

If your entity Key type is u32, you can auto-increment new entities using

use reindeer::AutoIncrementEntity;
let mut new_entity = Entity {
    id : 0 // if you setup id with any key, saving will update it
    /* ... */
};
new_entity.save_next(db)?;
// new_entity's key is now the auto-incremente value

You entitie's key will be automatically updated with set_key to match the last found entry's ID, incremented by 1.

💡 Note that the AutoIncrementEntity trait needs to be in scope.

Dependencies

~1.8–2.4MB
~55K SLoC