CAPtain.js

CAPtain.js is a first prototype of what I think the next generation of distributed programming languages should look like. The main idea behind CAPtain.js is to make the trade-off between availability and consistency (think CAP theorem) explicit from a programming perspective. Concretely, as a distributed programmer you explicitly state which pieces of your system's state should be available (and eventually consistent) or (strongly) consistent (but not always available).

If you are interested in how all of this works in practice please consider going through Spiders.js' tutorial first. CAPtain.js builds forth on Spiders.js and therefore heavily relies on its abstractions and concepts.

Usage

Install with npm:

npm install spiders.captain

Tutorial

This tutorail aims to provide a brief introduction to CAPtain.js' basic abstractions. This collaborative grocery list application provides a complete and running example.

Building Blocks: Availables,Eventuals and Consistents

CAPtain.js extends Spiders.js with three new classes: availables, eventuals and consistents. Each class represents a possible availability/consistency trade-off.

Objects of type Available ensure that field accesses and method calls always return a value (regardless of the caller's connectivity). The API offered by Availables is basically that of regular TypeScript/JavaScript objects:

import {Available, CAPActor, CAPplication} from "spiders.captain"
class AVCounter extends Available{
    value = 0
    
    inc(){
        return ++this.value
    }
}
let counter = new AVCounter()
counter.value //Returns 0
counter.inc() //Returns 1

Availables adhere to pass-by-copy semantics. In other words, whenever an Available is sent between two actors the receiving actor receives a deep copy of the original object (if this doesn't make sense to you, see the Spiders.js tutorial):

class TestActor extends CAPActor {
    rcvCounter(counter){
        counter.inc() //Returns 1
    }
}

let counter = new AVCounter()
let app = new CAPplication()
let act = app.spawnActor(TestActor)
act.rcvCounter(counter)
counter.inc() //Returns 1

Eventuals extend the behaviour of Availables by keeping all copies of a particular Eventual instance eventually consistent (i.e. not strongly eventually consistent). In a nutshell if two actors have a copy of the same Eventual object, CAPtain.js ensures that the state of these objects is synchronised. In other words, both actors can concurrently modify the state of their copy of the object (even while being disconnected from each other) without worrying about synchronising these modifications.

The @mutating annotation allows you to signal the CAPtain.js runtime that a particular method mutates an Eventual's state. You can install onTentative and onCommit listeners which are respectively triggered whenever the state of an Eventual is changed locally or globally.

import {CAPActor, CAPplication, Eventual, mutating} from "spiders.captain";

class EVCounter extends Eventual{
    value = 0

    @mutating
    inc(){
        return ++this.value
    }
}

let ev = new EVCounter()
ev.onTentative(()=>{
    ev.value //Triggered first, returns 1
})

ev.onCommit(()=>{
    ev.value //Triggered second, returns 1
})
ev.inc()

Eventuals are passed between actors using pass-by-replication semantics. In a nutshell, this entails that the passed object is deeply copied and that the copies are kept eventually consistent behind the scenes. The counter example bellow illustrated the workings of Eventuals across actors. The example can trivially be ported to work across machines given Spiders.js' inherent distribution mechanisms.

class TestActor extends CAPActor{
    rcvCounter(counter){
        counter.onTentative(()=>{
            //Triggered when "this" actor invokes "inc"
        })
        counter.onCommit(()=>{
            //Triggered whenever a new global value for counter is confirmed
            //This can either be the result of "this" actor invoking inc or the "other" actor invoking inc 
        })
        counter.inc()
    }
}

let app  = new CAPplication()
let counter   = new EVCounter()
let act1 =  app.spawnActor(TestActor)
let act2 = app.spawnActor(TestActor)
act1.rcvCounter(counter)
act2.rcvCounter(counter)

The example spawn two actors which both get a copy of a Counter Eventual. Subsequently, both actors install onTentative and onCommit listeners and concurrently increment the counter's value. Depending on the interleaving of messages the state change listeners might be triggered in different orders. However, the onCommit listeners will eventually trigger a final time for both actors. At that point in time CAPtain.js guarantees that the value of both counter copies is 2.

In the example above it can happen that both actors read different values for the counter (i.e. if this read happens in between synchronisation rounds). If you desire stronger consistency guarantees CAPtain.js offers Consistents which guarantee sequential consistency. In contrast to Availables and Eventuals, Consistents offer an asynchronous API. Accessing a Consistent's field or invoking one of its methods returns a promise which only resolves when the consistency of the result can be guaranteed by CAPtain.js:

import {Consistent,CAPplication,CAPActor} from "spiders.captain";

class CCounter extends Consistent{
    value = 0

    inc(){
        return ++this.value
    }
}

let counter = new CCounter()
counter.value.then((v)=>{
    //v will be bound to 0 when promise resolves
})
counter.inc().then((v)=>{
    //v will be bound to 1 when promise resolves
})

Consistents are passed between actors using pass-by-reference semantics. Conceptually, this entails that actors only ever have a "far reference" or proxy to a consistent:

class TestActor extends CAPActor{
    rcvCounter(counter){
        counter.value.then((v)=>{
            //Returns the same v for both actors unless someone invokes inc on the counter in between reads
        })
    }
}

let app  = new CAPplication()
let counter   = new CCounter()
let act1 =  app.spawnActor(TestActor)
let act2 = app.spawnActor(TestActor)
act1.rcvCounter(counter)
act2.rcvCounter(counter)

This difference between this example and the example using the Eventual counter above lies in the consistency guarantees the counter provides. In the previous example it could be that both actors read different values for the counter's value (e.g. due to one of the actors being disconnected from the network). In this example both actors will always read the same value for the counter, provided that no inc operation interleaves both reads.

From Eventual to Consistent and Back Again

CAPtain.js provides two built-in function which allow you to convert an Eventual to a Consistent and vice versa. On one hand freeze accepts an Eventual as argument and creates a new Consistent which represents a snapshot of the Eventual's state at freeze time. On the other hand, thaw accepts a Consistent as argument and returns a new Eventual which represents a snapshot of the Consistent's state at thaw time. Both functions are provided in the Actor/Application libraries:

class TestActor extends CAPActor{
    foo(someEventual,someConsistent){
        this.libs.freeze(someEventual) //Returns a consistent
        this.libs.thaw(someConsistent) //Returns an eventual
    }
}

Restrictions

The interactions between Availables, Eventuals and Consistents is restricted in order to guarantee their respective properties. By "interactions" I mean field assignment and method parameters. For example, CAPtain.js will trow a runtime exception if you try to assign the field of a Consistent to an Eventual value. The reason for this being that the Consistent would no longer be able to guarantee the strong consistency of that particular field. The following table summarises the interactions between all three data types:

| | Available | Eventual | Consistent | | |------------|:---------:|:--------:|:----------:|---| | Available | OK | OK | NOK | | | Eventual | OK | OK | NOK | | | Consistent | NOK | NOK | OK | |

Custom Consistency Requirements

TODO

Reading

In case you are interested in this work beyond this tutorial you are more than welcome to read our papers about CAPtain or Spiders.js.