Dispatch - A selective multicast operator for Most.js

This operator works like a mail room operation. If only a small number of packages and letters are received in a given time period and there are only a small number of potential recipients, it's usually ok to just stick the letters and packages on a shelf and let people come by and pick things up themselves. The same is true of a primary event stream being filtered by a number of downstream consumers.

As the number of inbound items and the number of potential recipients increases however, a more efficient approach starts to become necessary. Instead of having thousands of people crowd into the mail room and independently sift through all of the mail to find envelopes and packages for which they are the intended recipient- a process which would be horribly inefficient- a mail room operation sorts through the mail efficiently and dispatches each item to the intended recipients.

Installation

1# Using NPM:
2npm install --save most-dispatch
3
4# Using Yarn
5yarn add most-dispatch

Most.js is a wildcard-versioned peer dependency, which means you don't have to fight whatever version most-dispatch would otherwise be using.

Usage

The dispatch operator is initialized with a selector function, which takes an event as input and returns a key identifying an event recipient. The dispatch function then returns a function that takes a stream and returns a stream, as is the case for most other operators.

1// Call dispatch, passing in a function that identifies an event's recipient
2const dispatchEvents = dispatch(event => event.xyz);
3
4// The returned function takes a stream and returns a primary dispatch stream,
5const dispatcherStream = dispatchEvents(eventStream);
6
7// Alternatively, use most's `thru` operator for fluent stream composition
8const dispatcherStream = eventStream.thru(dispatchEvents);

Consuming the dispatcher stream

The stream returned by the dispatcher function emits events that it receives as-is, but as a tuple coupled with a select function. The select function takes a key as input and returns a stream of events pre-filtered to those which match the specified key.

1// The dispatcher stream emits the original event as a tuple with a select
2// function, as demonstrated by the following assertions:
3
4dispatcherStream.tap(([event, select]) => {
5  assert('xyz' in event);
6  assert(typeof select === 'function');
7  assert(select('foo') instanceof most.Stream);
8})

When a select stream is run, the dispatch source adds the resulting sink to a register of keys mapped to lists of sinks. When new events arrive at the dispatch source, in addition to re-emitting them as a tuple, as described above, the key selector function is run to map the event to a specific key. The event is then passed directly to any matching sinks that originated via a call to the select function using that same key.

In this way, instead of having a thousand streams filtering a single multicasted source stream potentially hundreds of times per second, a single pre-emptive filtering operation occurs within the dispatch source, and consuming streams need only wait for the events for which they have registered interest via their specified key. Consuming streams can be created and processed by watching the primary dispatch stream, usually via an accumulating operator such as scan or loop. When an event is received that does not match any known recipient, the accompanying select function can be called, passing in the related key, and the returned stream can then be applied as needed to efficiently process those events from then on.

Direct Selection

There are cases when you know in advance that a recipient will exist, and thus you'll likely want to be able to set up a dispatch target without having to first wait for an event associated with that target to arrive via the base dispatch stream. The stream returned by the dispatch() function includes the bound select() function as a property of the stream itself, allowing you to consume the dispatch stream using only the select() function if desired. In this case, consumption of the base dispatch stream itself is optional.

1const filteredStream = dispatcherStream.select('abc');

The select function is pre-bound to the dispatch source, so it can be passed around independently of the dispatch stream, and does not need to retain this as context. This makes it easier to use with functional techniques.

Note also that the select function is attached only to the stream object returned by the dispatch() function, and will thus be lost when applying any other standard stream operators to the dispatch stream, seeing as stream operators always discard the old stream reference and return a new one.

Example

This example is contrived, but demonstrates splitting an inbound stream into a set of derived streams using the dispatch function. The streams are kept in an immutable map (as an example that will often be applicable to other use cases) and all streams are then merged in order to take a look at their output.

1import Immutable from 'immutable';
2import {dispatch} from 'most-dispatch';
3import {periodic, scan, loop, filter, map} from 'most';
4
5function getDispatchKey(n) {
6  return String.fromCharCode(n % 3 + 65);
7}
8
9function updateStreamsMap({counter, streams}, [num, select]) {
10  const key = getDispatchKey(num);
11  if(streams.has(key)) {
12    return {seed: {counter, map}, value: null};
13  }
14  const multiplier = (counter + 1) * 2;
15  const stream = map(n => `Emitted by stream ${key}: ${n} * ${multiplier} = ${n * multiplier}`, select(key));
16  streams = streams.set(key, stream);
17  return {
18    seed: {counter: counter + 1, streams},
19    value: {key, streams}
20  };
21}
22
23function delegateEvents(num$) {
24  const dispatcher$ = dispatch(getDispatchKey, num$);
25  const initial = {counter: 0, map: Immutable.OrderedMap()};
26  const streams$ = loop(updateStreamsMap, initial, dispatcher$);
27  return filter(latest => latest !== null, streams$);
28}
29
30const number$ = scan(n => n + 1, periodic(100), 500); // ==> 500, 501, 502, ...
31
32delegateEvents(number$)
33  .flatMap(({key, streams}) => streams.get(key))
34  .take(10)
35  .tap(event => console.log('event:', event))
36  .drain();

Output:

event: Emitted by stream A: 500 * 2 = 1000
event: Emitted by stream B: 501 * 4 = 2004
event: Emitted by stream C: 502 * 6 = 3012
event: Emitted by stream A: 503 * 2 = 1006
event: Emitted by stream B: 504 * 4 = 2016
event: Emitted by stream C: 505 * 6 = 3030
event: Emitted by stream A: 506 * 2 = 1012
event: Emitted by stream B: 507 * 4 = 2028
event: Emitted by stream C: 508 * 6 = 3048
event: Emitted by stream A: 509 * 2 = 1018