The Registry Fish Pattern

Want to keep track of and work with all fish of a specific type? Meet the RegistryFish pattern.


The problem

Something that has come up quite a few times is the need to track all fish of a certain kind in someplace. You may, for example, have a MaterialRequestFish representing a specific material request triggered by a worker or a machine. There are going to be many instances of this fish, one for each material request. Now, what if you wanted to show a list of all these available material requests somewhere?

One thing you might want to do is to implement one huge fish that tracks all instances internally, something like an AllOpenMaterialRequestsFish . This fish would then contain all material requests and logic for dealing with all these material requests at once.

But this will make things unnecessarily complicated. Instead of just implementing logic for one instance, you will always have to deal with all the irrelevant instances. For example: even if you wanted to use just one material request to schedule a forklift or AGV, your logic would need to deal with all the irrelevant material requests as well.

There is a better way…

The solution

A fish should always be as small as possible (think of it as a digital twin). This pattern will lead you to a data model that is scalable , reusable , composable , and maintainable . Keeping in line with this, we split the problem into two parts:

  1. Implement state and logic of a single entity
  • Write a fish responsible for a single instance (e.g.: MaterialRequestFish.of(id) )
  1. Track and access many instances of an entity
  • Write a registry fish that tracks all instances (e.g.: MaterialRequestFish.registry )

This is the Registry Fish Pattern . It allows you to cleanly separate the concerns of the logic of an individual entity and keeping track of many instances thereof. Let’s jump in with an example.


Let’s take a fish representing a material request as an example. Here is how you could write this fish:

import { Fish, FishId, Pond } from '@actyx/pond'

// Very simple material request state
type State = {
  status: 'undefined' | 'created' | 'done' | 'canceled',
  id: string
// Our fish has three kind of events create, complete, canceled
type Event = {
  eventType: 'MaterialRequestCreated' | 'MaterialRequestCompleted' | 'MaterialRequestCanceled',
  id: string
const materialRequestTag = Tag<Event>('com.example.materialRequest')

// The actual fish
export const MaterialRequestFish = {
  of: (id: string): Fish<State, Event> => ({
    fishId: FishId.of('com.example.materialRequest', id, 0),
    initialState: {
      status: 'undefined',
    // subscribe to the events of tagged with the materialRequestTag + id
    where: materialRequestTag.withId(id),
    // handle the create, complete, canceled event and return the new state
    onEvent: (state, event) => {
      switch (event.eventType) {
        case 'MaterialRequestCreated':
          return { status: 'created', id }
        case 'MaterialRequestCompleted':
          return { status: 'done', id }
        case 'MaterialRequestCanceled':
          return { status: 'canceled', id }
          return state

If we now wanted to somehow deal with all material requests, we could write a registry fish as follows:

// A map as state for better performance and de-duplication
type IdMap = Record<string, boolean>

export const MaterialRequestFish = {
  registry: ({
    fishId: FishId.of('com.example.materialRequest.registry', 'registry', 0),
    // start with an empty map
    initialState: {},
    // in this example, we require all events in this tag.
    // You have to select/tag them more carefully in your advanced event-streams
    // check out the docs about tags:
    where: materialRequest,
    // handle the create, complete, canceled event and return the new state
    onEvent: (state, event) => {
      switch (event.eventType) {
        case 'MaterialRequestCreated':
          // Add the id to the registry
          state[] = true
        case 'MaterialRequestCompleted':
        case 'MaterialRequestCanceled':
          // Drop the from the registry
          delete state[]
      return state
  } as Fish<IdMap, Event>),

Here is how you could now, for example, use the MaterialRequestFish.registry to show a list of all existing material request names:

import { FishName, Pond } from '@actyx/pond'
import { MaterialRequestFish } from './materialRequestFish'

Pond.default().then(pond => {
  // Observe the registry fish and log the state to the console
  pond.observe(MaterialRequestFish.registry, state => console.log(state))
}).catch(() => console.error('Is ActyxOS running?'))

What if you want to observe the state of the actual MaterialRequestFish? You can do so using the RxPond. Here is how:

import { Pond } from '@actyx/pond'
import { RxPond } from '@actyx-contrib/rx-pond'
import { MaterialRequestFish } from './materialRequestFish'
import { combineLatest } from 'rxjs'
import { switchMap } from 'rxjs/operators'

// assuming that we already have a pond instance in your application
Pond.default().then(pond => {
    // Observe the registry fish to get the materialRequest ids
    // we will map the list of Ids to the entities with the RxJS pipeline
    // find mor information about RxJS here:
      // switch over to the entity fish
      switchMap((idMap) =>
        // Use RxJS's combineLatest to get one stream with all material requests as an array
          // map the id of the state to an EntityFish
          Object.keys(idMap).map(id =>
            // observe a fish of each entry in the ids array
    // subscribe to the stream to get all entity states
    .subscribe(allEntityStates = console.log(allEntityStates))
}).catch(() => console.error('Is ActyxOS running?'))

Hopefully, this snippet gives you an idea of the Registry Fish Pattern !

One thing that you may have noticed is that the registry is pretty generic and could be used all over different projects. Let’s see how we can pack that into an npm package.

By the way, this pattern is not only useful for a registry and its entities. We can use it to resolve references in one fish and wake up other fish according to a given field.


  • forklift → current job
  • material request → production order.
    ForkliftFish.of('RoadRunner 1'),
    state => [state.currentMaterialRequest],
    currentMaterialRequest => console.log(currentMaterialRequest),