Agregate

A "missing piece" of DB clients for Node.JS, accenting on familiar JS experience, developer's freedom and simplicity of usage

Programmers waste enormous amounts of time thinking about, or worrying about, the speed of noncritical parts of their programs, and these attempts at efficiency actually have a strong negative impact when debugging and maintenance are considered. Donald Knuth

disclaimer: this software is alpha stage and is not intended to be used in heavy production projects. I am is not responsible for data loss and corruption, lunar eclipses and dead kittens.

Enviroment and preparations

Agregate's only back-end for now is neo4j (v2 and v3 beta are supported) for now. You can install it or request a free SaaS sandbox trial.

This (es6, es2015, plus decorators, static class properties and bind operator) babel preset is recommended to be used for the best experience. However, library is shipped with compiled to ES5 files by default. Note: bind instance property expressions work only in babel 6.5+ plugins, older versions will throw an error due to already removed bug in babel

Familiar JS experience

So, you need User class reflecting DB "table". You simply write

class User extends Record {}
 
const user = new User({name: 'foo'})
user.surname = 'bar'
user.save()
.then(() => User.where({name: 'foo'}))
.then(([user]) => console.log(user))  //=> User {name: 'foo', surname: 'bar'}

No factories, complex configs and CLI tools. Every enumerable property (except relations, but it will be explained later) is reflectable into DB, back and forth.

Developer's freedom

Common DB lib usually requires you to keep specific file structure and/or using CLI tools and/or remember hundreds of methods, properties and signatures. Agregate was aimed to keep Minimal API Surface Area. Agregate API is fully promise-based, Relation is trying to mimic Set API, and Record instance has just 2 core methods - Record#save and Record#delete, whose API is obvious.

Simplicity of usage

The whole declaration of class would be something like:

const {Connection, Record} = require('agregate')
//class name will be used as "table name". You can overload it with static "label" property
class Entry extends Record {
    //indexes are optional static properties 
    //which are used only for making DB query 'CREATE INDEX' during register() call. 
    static indexes = new Set('foo', 'bar')
    //for now agregate is backed by npmjs.com/package/neo4j,
    //so Connection constructor is just proxying everything up to that package. 
    //You can usually just use URL string syntax.
    //static properties are inheritable, so you only need to declare in once in parent class
    static connection = new Connection('http://neo4j:password@localhost:7474');
}
//As we cannot have a callbacks on class constructor, (at least without crazy hacks) 
//explicit .register() call is required for any concrete class
Entry.register() 

Wait, but I need relations

OK, let's add relations.

const {Connection, Record, Relation} = require('agregate')
//we assume that we already made something like Record.connection = ...
 
class RecordObject extends Record {
    //signature of constructor is 
    //(source: Record|Relation, label: string[, {target?: RecordClass, direction?: number = 1}])
    subjects = new Relation(this, 'relation');
}
 
class RecordSubject extends Record {
    //target is limitation of relation to one record group, and direction is, well, direction. 
    //Direction is 1 by default, which is '->' relation. -1 relation is '<-'. 
    //It means there can be 2 relations with same label in different directions. 
    //0-relation is plain '-', there can be only 1 relation of this type.
    subjects = new Relation(this, 'relation', {target: Object, direction: -1});
}
 
RecordObject.register()
RecordSubject.register()
 
async function main() {
    const object = await new RecordObject({foo: 'bar'}).save()
    const subject = await new RecordSubject().save()
    await object.subjects.add(subject)
 
    console.log(await subject.objects.size()) // => 1
    const objects = await subject.objects.entries()
    console.log(objects[0].foo) => //bar
}

Deep relations are extremely simple:

import Role from './role'
import Permission from './permission'
export default class User extends ConnectedRecord {
    roles = new Relation(this, 'has_role', {target: Role});
    permissions = new Relation(this.roles, 'has_permission', {target: Permission});
    hasPermission = ::this.permissions.has
}

Relation instances have bunch of pretty methods to use:

class Relation {
    //overloaded method to implement one-to-one relation
    async only(): Record
    async only(null | Record): void
    //just for you to know: signature of Record.where and Relation#where are 100% same
    async where(params?: WhereParams, opts?: WhereOpts): Array<Record>
    
    //only non-familiar method. Returs intersection of relation and passed set
    async intersect(records: Array<Record>): Array<Record>
    
    //this part mimics es6 Set class
    async add(records: Array<Record>): void
    async clear(): void
    async delete(records: Array<Record>): void
    async entries(): Array<Record>
    async has(records: Array<Record>): bool
    //note: size is not property, but async method
    async size(): number
}

Auto-generated record properties

all of the properties are non-enumerable, non-configurable and exists only for reflected record.

• uuid - automatically generated on creation
• createdAt - automatically generated on creation
• updatedAt - automatically generated on creation and update

OK, but how can I make complex queries?

Record.where and Relation#where methods are provided for querying.

All details are provided in API page, in brief - order, limit, offset can be used for filtering. Equality, existence, numeric (greater/less), string (starts/ends with, contains), array (contains/includes) queries are provided.

Examples:

Entry.where({foo: 1000}, {limit: 10, offset: 10, order: 'createdAt'})
Entry.where({updatedAt: {$gte: Date.now() - 1000}}, {order: ['createdAt DESC']})
Entry.where({foo: {$exists: true, $startsWith: ['b', 'ba', 'baz'], $endsWith: 'bar', $contains: 'z'}})
 
// here e.g. {foo: [1,2,3,4,5], bar: 3} will be reflected.
// $has stands for "db record has fields", $in - for "db record is in list of possible fields"
Entry.where({foo: {$has: [1,2,3]}, bar: {$in: [1,2,3]}})
 
//$in can also work with array
Entry.where({foo: {$in: [[0], [1,2,3,4,5]]}}})

Hooks?

beforeCreate, afterCreate, beforeUpdate, afterUpdate, beforeDestroy, afterDestroy are available hooks

class Entry extends Record {
    async beforeCreate() {
        //this.connection points to transaction during the transaction, so you have to pass it if calling other classes
        const test = await Test.where({id: this.id}, this.connection)
        this.testId = test.testId
    }
}

Transactions and atomicity?

Yes, Agregate has transactions.

Hooks (see above) are always ran inside a transaction (transaction is same for pre-hook, operation itself and post-hook).

Decorator @acceptsTransaction({force: true}) can be used (with babel-plugin-transform-decorators-legacy, will be changed to new syntax when new spec will become stable), or transaction can be constructed explicitly by connection.transaction()

All transactions should be committed or rolled back. Decorator commits everything automatically on success, rolls back on error.

On SIGINT Agregate will attempt to rollback all not closed yet transactions. By default Neo4j rolls back transactions in 60 seconds after last query.

Good example of transaction usage is provided Record.firstOrCreate sugar-ish method:

class Record {
    @acceptsTransaction
    static async firstOrInitialize(params) {
        const tx = this.connection.transaction()
        let [result] = await this.where(params, {limit: 1}, tx.transaction())
        if (!result)
            result = await new this().save(params, tx.transaction())
        await tx.commit()
        return result
    }
}

Roadmap

• sort
• offset and limit
• has
• deep relations
• indexes
• rich Record.where query syntax ($lt, $gt, $lte, $gte, $has, $in and so on)
• one-to-one relations
• relations validation
• total test coverage
• performance optimisations
• optimistic and pessimistic locks?
• tests for transaction usage
• tests for eventEmitter