@jpwilliams/distributed-promise v0.2.4

@jpwilliams/distributed-promise

Distribute a promise across multiple processes connected via Redis. If two seperate processes make the same call, only one will actually do the work, but both promises will return simultaneously with the same data.

Has pretty types and works with any data compatible with JSON.stringify and JSON.parse.

npm install --save @jpwilliams/distributed-promise

import { DistributedPromiseWrapper } from '@jpwilliams/distributed-promise'

const wrapper = new DistributedPromiseWrapper({
	redis: myRedisClient
})

// any function
// can be synchronous or asynchronous
function joinStr (...strs) {
	return strs.join()
}

const sharedJoinStr = wrapper.wrap(joinStr)
const result = await sharedJoinStr('foo', 'bar')
// result = 'foobar'

Why?

This facilitates the memoisation of expensive function calls in a distributed system and helps combat the issue of a cache stampede.

It does this by only allowing a single service per job to actually do the work and makes all other services wait for that result and not compute their own.

How

First, after creating a wrapper via new DistributedPromiseWrapper, you can wrap any functions for use with the library, regardless of what they return. The wrapping process will always return a function that returns a Promise, even if the original function did not.

// (...objs: object[]): object
const combine = (...objs: object[]): object => Object.assign(...objs)

// (...objs: object[]): Promise<object>
const sharedCombine = wrapper.wrap(combine)

Internally, when a wrapped function is called, a simple route is followed:

1. Get relevant data from cache based on input args. Found data? Return it.
2. Try to attain a lock to get permission to do the work locally.
3. If we get the lock, perform work locally, push the result to the cache, publish via Redis and return.
4. If we didn't get the lock, wait for the data to be published via Redis and return it when it arrives.

In basic terms, if

API

new DistributedPromiseWrapper(config: DistributedPromiseConfig)

Creates a new wrapper to use to wrap functions.

config DistributedPromiseConfig

• redis: RedisClient The RedisClient instance to use to connect.
• lockTimeout?: number The amount of time in milliseconds to hold the Redis lock for when doing work locally. Defaults to 30000 (30 seconds).
• ttl?: number The amount of time in milliseconds before items expire from the cache. Defaults to 1800000 (30 minutes).
• keyPrefix?: string The prefix to use for all keys the library uses in Redis. Defaults to distributed-promise.
• lockPrefix?: string The prefix to use for locks in Redis. Defaults to lock.
• notifPrefix?: string The prefix to use for notifications in Redis. Defaults to notif.
• keySeperator?: string The seperator to use between the segments of key data in Redis. Defaults to :.

Returns DistributedPromiseWrapper.

DistributedPromiseWrapper.wrap(work: InputFn, config?: WrapConfig | string)

Wraps a function, ready to share. An internal key is needed for caching and data retrieval. If just work is passed, this key will be the name of the function. If it does not have one, the library will throw.

work InputFn

A function. (...args: any) => any - any arguments and any return.

config WrapConfig | string

If undefined, the internal key required will be grabbed from the name of the work function. If it does not have a name, the library will throw.

If a string, the internal key will be set to that string.

To customise your wrap further, you can send a WrapConfig:

• key: string The internal key to use.
• timeout?: number The timeout to wait for an external process to do the work before giving up and rejecting the promise.

Returns your wrapped function.

Caveats

Keep in mind that this library can only deal with raw data; you can't make a database connection and magically share it with everyone. ;)

Todo

• Detect if work is already happening locally and tap in to the local Promise rather than going to Redis.
• If a TTL of 0 is set, still look to Redis for locking and the receipt of data via pubsub, but never actually cache the data.