Workflows

This leads to that, which leads to...

An alternative use case to the bundling of actions or action composition, where related data must be handled with explicit order of execution, is workflows. A workflow is the use case where the result of one action may result in the dispatching of another, perhaps with derived data, perhaps with otherwise unrelated data.

Let's face it. Sometimes we have to work with odd data structures in order to make things work. One of the key tenants of NgRx is denormalization of data, so we may have multiple collections that contain entities related to each other.

Use Case

A potential example use case here, however contrived it may be (although not necessarily unheard of, especially for those who work with existing, well-established and particularly older backends), might be tracking the things a user favorites with joiner records in a many-to-many table. Lets say we have a single-user app that needs to easily display this information. We may want to track an isFavorite flag on our things, and update said flag whenever our joiner entities are created or deleted.

Chaining Actions

Unlike our full order creation example, where we needed to track a group of entities together and manage the order of execution for their creation, this use case requires that we chain one action off of the success of another.

We do not need any new custom actions for this use case, however we do need some effects. Let's say we have Posts that a user may Like. The actual likes are tracked as a join between the postId and the userId and are created and deleted as independent entities in the database. For simplicity of processing this data in our UI and UI performance, we track an wasLiked flag on our Post entities in state.

like.effects.ts

export class LikeEffects {
    constructor(private actions$: Actions, private store: Store<AppState>) {}
    
    createSuccess$ = createEffect(
        () => this.actions$.pipe(
            ofEntityType(Like, CreateSuccess),
            withLatestFrom(this.store.select(allPosts)),
            map(([{entity: like}, allPosts]) => 
                allPosts.find(post => post.id === like.postId)
            ),
            filter(post => !!post),
            map(post => new LoadSuccess(Post, {...post, wasLiked: true}))
        )
    );
    
    deleteSuccess$ = createEffect(
        () => this.actions$.pipe(
            ofEntityType(Like, DeleteSuccess),
            withLatestFrom(this.store.select(allPosts)),
            map(([{entity: like}, allPosts]) => 
                allPosts.find(post => post.id === like.postId)
            ),
            filter(post => !!post),
            map(post => new LoadSuccess(Post, {...post, wasLiked: false}))
        )
    );
}

Don't forget register your new LikeEffects class with the NgRx EffectsModule!

Understanding Workflows

These two effects are fairly strait forward. They look for the success of prior Create or Delete actions dispatched for Like entities. They then grab all the posts currently loaded into state, find the post related to the like, and dispatch a LoadSuccess. This may seem odd, however we do not actually wish to load a Post from the database, we only wish to update the related Post in state...the goal is efficiency and performance.

The Auto-Entity meta reducer handles LoadSuccess in a specific way...it merges the specified entity into state. If it already exists, it is updated, if not it is added. Since creating a Like marks a post as liked for the specified user, we know we can set wasLiked to true when a Like is successfully created. Similarly, we know we can set wasLiked to false when a Like is successfully deleted.

In the event that you may need to update many entities at once, rather than just a single entity, a similar result may be achieved by dispatching LoadManySuccess by including the array of updated entities.

Do not dispatch LoadAllSuccess, as the semantics are different. The nature of LoadMany means that when its success result is dispatched, the entities will be MERGED into the existing state. In contrast, the nature of LoadAll means that when its success result is dispatched, the entities will REPLACE whatever is currently in state.

Further, if either the Create or Delete of a Like fails, due to network connectivity issues, a server outage, etc. then we do not update the Post in state, which might incorrectly reflect to the user their status for the given post.

Higher Order Facade

Ok, now that we have the necessary effects to handle these state updates, let's update our facades to support the new behavior here. In fact...let's create a new facade of a special type, what we like to call a Manager, to handle the composition of information from multiple entities into another.

This is just a matter of code organization, and for simpler applications adding the necessary method to an existing facade, say the PostFacade. For larger applications, segregating out behavior that relies on multiple entities can help keep facades small and manageable.

@Injectable({providedIn: 'root'})
export class PostManager  {
    constructor(private likeFacade: LikeFacade) {}
    
    likePost(post: readonly Post, user: readonly User): void {
        this.likeFacade.create({ postId: post.id, userId: user.id});
    }
}

We could try to simplify things here. We could, say, bring in the PostFacade here as well, and right after creating our like we update the Post and directly set the wasLiked flag to true. We could...

Remember that NgRx is asynchronous. Dispatching an initiating action such as Create or Delete does not imply success, or even completion. It simply means you have requested a particular activity be performed. Said action's status is unknown until the result action, such as CreateSuccess or DeleteFailure, has subsequently been dispatched.

When we create something, such as our Like joiner entity here, that dispatches the request to create the entity. This request may be fulfilled at a later time, but it is not guaranteed to be fulfilled by the time we get to the next line of code. It may in fact never be fulfilled depending on network or server conditions.

By creating a simple workflow with effects and subsequent action dispatches, we are able to enforce and maintain the integrity of our data in our application. We only update the wasLiked flag once the CreateSuccess or DeleteSuccess actions for the liked Post have been dispatched, something much harder to do in a facade or component.