![Cover image for Simple state management in Angular with only Services and RxJS](https://res.cloudinary.com/practicaldev/image/fetch/s--EvdPjWMS--/c_imagga_scale,f_auto,fl_progressive,h_420,q_auto,w_1000/https://thepracticaldev.s3.amazonaws.com/i/0anf9cgj62ienccplamx.png)
One of the most challenging things in software development is state management. Currently there are several state management libraries for Angular ...
For further actions, you may consider blocking this person and/or reporting abuse
Exactly what I'm trying to do in my app. An advice: don't put HTTP, async services, etc. into the store. Keep it separate. State management has nothing to do with such services and business logic. Ngrx Effects is a terrible mix of two concepts.
You're right, side effects must always be separated from state management, this was a quick example, i'll try to clean it up when I have some free time.
I've almost made a career out of reducing the complexity and size of Angular 2+ apps by appropriately using services and RxJs instead of convoluted, home-grown solutions to state management.
My resume:
Very well put together, succinct article!
We don't use NgRX at my workplace, but we do have a couple of data source services.
They kind of just grew organically.
I see the benefits of following your rules of immutability, having a private
behaviourSubject
, and the getter and setter. Also the second readonly observable which pipes thebehaviourSubject
is very nice.Is the term
storeService
a widely used suffix?I currently use
dataSourceService
.Thanks!
nope it's up to you to name it, I usually name it something like TodosStore
I think I prefer the suffix store over dataSource.
Also the ngrx stuff talks about stores a lot. Seems to be a popular term.
This is exactly what I do in my project. I trialled ngrx/store but found it so overly bloated with a ridiculous amount of boilerplate for very little gain. This pattern is simple, elegant, easy to reason about.
Give ngrx/data a look for a more streamlined version of ngrx/store. It's excellent for managing collections of objects. It provides almost all of the functionality you might need out of the box and is based on configuration so you end up writing very little logic.
This looks really nice. After 4ish projects across React and Angular, I have yet to find the Redux pattern remotely worth it... But I'll say that has a lot to do with particular executions...
Regardless, this seems like this would make life a lot easier. :)
One question, tho: is the
shareReplay
incompletedTodos$
not redundant? It would callshareReplay
twice in a row in the pipe.To be honest I'm not sure, I've put it for good measure, but theoretically yes, i didn't have to use shareReplay on filtering considering that the original source is already multicast, I'll do some tests to confirm, and remove it later.
As a few others have pointed out, this is basically Akita. Akita gives you all the basic crud methods you need to start managing either an array of entities (todos) or a single entity, but with none of the boilerplate required by ngrx or ngxs. There's almost no setup, and in addition to getting a rxjs-based approach to state management, you also get all the other bells and whistles, like action tracking with the redux dev tools, time travelling, action tracing, etc. I started out with rxjs stores, and then tried ngrx, ngxs, went back to rxjs stores, then finally found akita, and I see no reason to ever not use it-
github.com/datorama/akita
Hi, some perf prob from Akita ? medium.com/@vpranskunas/deep-compa...
I figure out if you get these probs too ?
Honestly, unless your app is extremely complicated, anyone using a state management library for Angular has misunderstood the component life cycle and is pretty much trying to make Angular into a React application.
why use behavior subject with sharereplay?
get todos() is also overkill. It's not a good way to get data from observer, except in async pipe. For adding todo we should use scan operator, same as for mark as completed. Subject with shareReplay (or ReplaySubject) solves all the problems and BehaviorSubject is really redundant here.
Thanks for the feedback, updated.
Accessors for the same member name must specify the same accessibility
todos getter/setters must either both be public or both be private. My preference is they are private, so that publicly, only the observable is accessible.
Another alternative is make private set _todos and public get todos.
You probably already know better than me which options are best, but thought I throw some options out there for others that come across this.
Thanks for the excellent example. Do you by any chance have an example for todosGroup and todos, and how they would work with each other when navigating from group to single todo?
Thank you for sharing. I value your input as a developer and person. :D
Thanks for the article! Some points/questions for someone who is quite new to Angular:
todosTrackFn
. Didn't know that existed!Thanks for this post, it's help us very well.
When you define the "readonly completedTodos$" in the service, you use the getter (this.todos) in the map operator to filter the stream. What doesn't use directly the "todos" var entry from the map operator like this:
readonly completedTodos$ = this.todos$.pipe(
map(todos => todos.filter(todo => todo.isCompleted))
) ?
Hey, both do the same thing, but yes you are correct, using todos from the map makes more sense, i updated the article.
I notice that the StackBlitz project refers to an undefined trackBy function:
todosTrackFn
I can't see any problem that it is causing. The project compiles ok. TSLint flags the problem in my vscode edit buffer.
yep it's a typo, fixed it on stackblitz
Hey Aslan! I wrote a similar article on DEV. It is inspired by yours! thanks!
dev.to/angular/simple-yet-powerful...
TLDR Let’s create our own state management Class with just RxJS/BehaviorSubject (inspired by some well known state management libs).
Whats your opinion about using this method in a complex application that doesn't use any state management at all yet (due to lack of initial planning), but our team would like to change this in the future regarding modules one by one starting with the most complex one. Is it a good idea to go with this one, or would it be better idea to go with ngrx because of the complexity?
Depends on the scale, if you have a big app with a lot of components than need to share state, and a lot of moving parts, i strongly suggest ngrx, i used this method (the one described in the article) to build several libraries that have internal state, and it works very well, so no need to use ngrx in component libs.
Inspired by this article, i have made a library "ng-simple-state" based on simple state management in Angular with only Services and RxJS npmjs.com/package/ng-simple-state
I have also implemented the support to Redux DevTools browser extension and localstorage state persistence.
Thanks for your inspirational post.
Inspired by this article, i have made a library "ng-simple-state" based on simple state management in Angular with only Services and RxJS npmjs.com/package/ng-simple-state
I have also implemented the support to Redux DevTools browser extension and localstorage state persistence.
Thanks for your inspirational post.
Hello Aslan,
I found a critical problem with this solution (which btw I found while trying to solve the same critical problem with my very own nearly identical code).
After literally weeks of troubleshooting this I am backed up against the wall and considering alternatives to Angular because I need the features that (it appears) only BehaviorSubject can solve.
But for now let's look at your Stackblitz example. At first glance it works perfectly. Bravo. I said to myself the same thing on my own code.
Then, try these things and you will soon realize there is a very serious problem and it is difficult to solve: corruption in the two arrays.
Recreate the problems in two methods:
Method 1:
Open your stackblitz and click rapidly on items in the incomplete and completed lists. Mark items as complete and the others as incomplete.... keep doing it maybe 20 times. You will see the following corruption take place:
a) Items will appear more than once on each list
b) Clicking an item to complete it will result in another already-completed item becoming incomplete and vice versa.
c) Clicks being ignored by the UI
Method 2:
Leave your stackblitz open in chrome and untouched for an hour or so. Come back and notice the first click (lets say mark an item as complete) is dead (or it appears dead). But upon the next click you will be able to add the same item twice to the completed list.
I believe there is an async / await sort of sequence issue that because of template code executing faster than service code, the UI will show array values that differ from what is in state service behaviorsubject.
Your example behaves just like my own.
I am not certain why more people have not noticed? Any ideas as to the root cause? I would be happy to demonstrate this over a zoom call.
If you are reaching this level of complexity with more than 1-2 services then i would suggest start looking into using a library with some more restraints.
At least if you are more than 1-2 devs touching the codebase.
Otherwise long-term feature additions and bugfixes are likely to turn this into state soup eventually, once developers start to slack a bit or implement newer idea's.
Which is bound to happen sooner or later in a system without opinionated restraints.
I liked the demo service example, but would have liked to see some with less observable bootstrapping, and maybe just some general guidelines.
Since that's what i find smaller projects benefit more from in my experience. :)
Found the article a good read however, thanks!
This was my thought as well.
From looking around it seems like (as of today) Akita basically is this approach, plus some organizational patterns/practices that should help encourage consistency.
I do wish there was a library out there with a longer track record that was lighter than NGRX.
Thanks for this. Have been exploring NgRx and have been feeling "this isn't worth it". Did a google search for "simple Angular state management" and here I am.
depends on the case, when I work on large scale apps with a lot of moving parts, it's very worth it
Hey Aslan, I do the same thing, except I don't understand the need for sharedReplay. You're already sharing the same observable because it's assigned to a public readonly property. And you're also using a BehaviourSubject which should always return the current value to new subscribers. Also, why not just use a ReplaySubject if you need the previous value? You lose getValue, but it's not such a big deal because you probably want to subscribe anyway. What am I missing?
BehaviourSubject is not a simple observable, it's an observer AND an observable at the same time that can give you its latest state synchronously.
That doesn't mean the observer is shared, and that doesn't change the fact that when someone subscribes to todos$, a new dedicated observer will be created for it. A shared observable means that all subscribers are getting their data stream from a single observable source. (lookup rxjs multicasting)
More info here: learnrxjs.io/operators/multicastin...
I need to be able to get the previous state synchronously in the reducers, as I already explained in my other comment, reducers are synchronous functions that take previous state and reduce it to a new state in an immutable way.
The reason we need the replay part is because some components might be rendered asynchronously (after fetching some data from api for ex), and might miss the emits in todos$ observable, so when they subscribe to it they won't know what happened in the source before component came to life, the replay will emit the latest value to all observers upon subscription.
On the other hand, the second share replay is not necessary, I updated the code.
Hope I answered your questions.
First of all, thanks for the article.
Its really great, and certainly expanded my horizons on the use of RxJS.
Just one question, on stack blitz you use toPromise() on the calls to the rest api. This could not also be achieved using observables?
this.todosService.setCompleted(id, isCompleted)
.pipe(take(1))
.subscribe({
next: value => {},
error: err => {
console.error(e);
this.todos[index] = {
...todo,
isCompleted: !isCompleted
}
}
});
I did see that you use it because of async/wait but not understand why.
I'm just trying to understand the difference between the two approaches.
Thanks!
Absolutely brilliant! Always know your actual toolset before buying into new ones!
I really liked this approach and I'm definitively going to use it! But I wonder how could it be used when we actually have to fetch data from an API?
It's easy, i'll make an example soon
Aslan, I like this approach quite a bit, but I have the same questions. What would it look like when fetching data from an API?
Hey @klouddy @gabrielaraujof , I've updated the stackblitz example with a real REST API and some interesting techniques on how to do optimistic updates and rollbacks: stackblitz.com/edit/angular-rxjs-s...
I've updated the stackblitz example with a rest api
Is there a reason why you solved it with EventEmitter in TodoComponent? Services are all about to solve that.
You could have done it directly like:
...(change)="todosStore.setCompleted(todo.id, $event.target.checked)"...
... (click)=todosStore.removeTodo(todo.id)...
stackblitz.com/edit/angular-rxjs-s...
A great example of clean and organized code.
How do you organize your stores files on the project structure? Do you keep in a separate folder or within the component itself? Thanks!
separate modules for each store, inside the module: reducers, effects, facades.
Hi, in the user's code I've noticed, that you pass 'event' to the remove method:
(remove)="todosStore.removeTodo($event)"
Shouldn't it be just an id?:
(remove)="todosStore.removeTodo(todo.id)"
both are correct, if the app-todo-component outputs the todo id in the event, it'll still work
depends on the implementation of the @Ouptut
Yea, I completely agree, they are the same, I guess there is a real small advantage to use todo.id instead of $event - if we use todo.id we explicitly say that we wanna to use id property from list of todos, otherwise if we use $event we delegate the responsibility to the child component to use 'remove' event properly (pass the correct value).
Great article. Thank you.
I'm trying to adapt this kind of state managment to my app.
In my scenario I'm subscribing to some events (adding, removing) and do some other logic (for example showing message "you have removed something").
How would you modify your todo example to add this kind of logic in a separate file/ or in other components?
What difference is there in using BehaviorSubject in a state manager class instead of sharing a service class with properties passed by reference?
onPush change detection, combining/manipulating data streams with RxJS, and it's really fast.
Thanks!
If this works out for me, and I implement it in 4 hours on my very crazy codebase... Im buying u coffee or a beer...
good luck haha, 4 hours is quite a challenge, let me know how it went
If possible, it would be great if you could update the article to match the StackBlitz code. Thanks
I'd like to keep the article as simple and lightweight as possible, and i commented the code on stackblitz, but if anything isn't clear enough feel free to ask
What do you think about using Subjects as observables in your app state service, I find the very convenient.
Thanks for sharing, this is exactly Im looking for.
Many thanks! I will use it in my current project!
It is easy to understand
Thank you for sharing Aslan!
Would you recommend this solution for large Enterprise Angular applications with +200 views?
I get two errors
*ngmetadataName
*Circular services
Any tips?
Useful and clean implementation of a simple state management.
Awesome post, do you have any ideas to better handle business logic and reactive forms? I find it's not that easy to make them well decoupled.
For me Redux has no interest at all and must add at least 50% more work (for nothing).
I notice that the StackBlitz project differs from your article somewhat. In particular, you convert observables to promises in several places. Can you explain why you do that? Thanks
I convert them to Promises so I can use async/await
It doesn't feel very clean that the todoStore in the rest api example is now depend on todoStoreService which itself depends on httpClient.