loading...
Cover image for Resolve or Die - Error Handling Strategies for Loading Data in Angular

Resolve or Die - Error Handling Strategies for Loading Data in Angular

walkingriver profile image Michael D. Callaghan Originally published at walkingriver.com ・7 min read

Crossposted from WalkingRiver Blog

Angular Route Guards are great. Is the user authenticated? No? The guarded page won't load. Is the user authenticated, but not a member of the appropriate authorization group? Also denied. Route Guards make this easy. They follow the Single Responsibility Principle, can be mocked for unit testing, and all is right with the world. What about Route Resolvers? Where do they fit in?

Route Resolvers

Route Resolvers are similar to the Route Guards, with a subtle difference. Instead of keeping the user from rendering the page, a resolver ensures that certain conditions are met before the route completes. Resolvers are typically used to preload data that a given page requires, before the page loads. The general thinking seems to be that there is no reason to show the user a page that is not complete, or still has data to be loaded.

Imagine a scenario where you have a page that displays order status for the person viewing your site. You might have a route that looks like this...

/orderstatus/:orderId

Of course you would have a Route Guard to ensure that the logged-in user matches the order being requested. That is the easy part, but do you also need a Route Resolver, to ensure the order details are loaded before showing the page? I think the answer is, "it depends."

What Data Are You Loading?

In the above scenario, you need to load the order details: what items are on the order, which of those items have been shipped or on back-order, and the tracking information for the items that have been shipped. There may be other details your system has to load. Maybe the item descriptions and payment information come from other systems, and you want to display those, too. The questions you need to ask yourself are which data items are so important that you will not even want to display the page without them? And what if you cannot load those items?

The answer to the first question is really an application design decision. Do you want to ensure that all of the order details have been successfully loaded before the Order Status Page appears? Or is it acceptable to render the page partially, and allow the order details to appear as its data is returned?

If the order status typically takes less than a second to load completely, waiting for it might be acceptable.

What If There Are Errors?

Depending on which pattern you choose, you then need to decide what to do in the event that one or more pieces of data cannot be loaded. A service could be down, the network could be flaky, etc. It really does not matter. What do you do?

Give up on error

If you are using a Route Resolver, and it encounters an unhandled exception, the route will not resolve, and your page will simply not render. This is obviously not ideal, and you should throw out this idea immediately.

You could catch and log the error, setting the data being loaded to an empty or default object, and then letting the resolver complete successfully. If you do that, you need to think of what the rendered page would look like. Is it still useful for the person trying to load the page? If the answer is yes, then perhaps this is an acceptable solution. You can actually make this choice a case-by-case decision, based on the importance of data being loaded. If it is not "mission critical", go ahead and let the page load.

Redirect to an error page

On the other hand, maybe your data is so critical that not a smidgen of the page should appear until it is all available. If any missing piece of data causes the page to be useless, consider redirecting to an error page with an explanation of what went wrong. You could optionally provide a link on the error page, allowing the user to retry. After all, it may have been a temporary blip.

Here is some sample code that illustrates one way to redirect to an error page if any errors are encountered in the Resolver. The primary advantage to this method is that the page itself need not be modified, or even know about the resolver. If the component loads, you can be sure that the data is there. Thus, the code I am showing here is the resolver itself, and I am not showing the page component.

The Resolver implements the Resolve interface, containing a single resolve() function. This function gets called before the page component is rendered. Inside of this function, we make a call to the data service to retrieve the data. My sample service waits 3.5 seconds, and then throws an error. For simplicity, my resolver only needs to know how to catch the error. At that point, the resolve function calls the Angular router to navigate to the /error page by its URL, and returns an EMPTY observable to the caller. I am not sure this last step was strictly necessary, but it satisfies the TypeScript compiler.

export class ErrorResolverService implements Resolve<any> {
  constructor(private data: DataService, private router: Router) { }

  resolve() {
    return this.data.getData()
      .pipe(
        catchError((e) => {
          this.router.navigateByUrl('/error');
          return EMPTY;
        }));
  }
}

Maybe Not Use a Resolver

Going back to the Order Status scenario, imagine that there is really only one piece of data that would make the entire page useless. And even then, if the page renders without this data, and instead displays an inline error, what is the ultimate harm?

Allow me to suggest another option. Consider not using a Route Resolver at all.

Loading Indicator

A common pattern I have seen is to render the complete page, but with no data. The page displays a "Loading" indicator of some sort while the order details are retrieved, with the page in a non-interactive state. Once the details are available, the indicator disappears and the user can interact with the page at that point. This is a pretty common pattern, which we have all seen.

Here is some sample code that implements this pattern, using the Ionic Framework 4 and Angular 8.

The getData() function creates and presents a loading indicator using the Ionic's LoadingController, then subscribes to the data service. It registers a finalize call to dismiss the loading indicator whether or not the service returns data or throws an error. My service only throws an error, so that's somewhat moot.

export class LoadingPage implements OnInit {
  sub;
  errorText = '';

  constructor(private data: DataService, private loading: LoadingController) { }

  ngOnInit() {
    this.getData();
  }

  async getData() {
    // Create and present the loading indicator 
    const loader = await this.loading.create({ message: 'Loading...' });
    await loader.present();

    // Ensure error message is cleared
    this.errorText = '';

    // Subscribe to the data service
    this.sub = this.data.getData()
      .pipe(finalize(() => loader.dismiss())) // always dismiss loader
      .subscribe({
        // Sample service only throws an error
        error: (e => this.errorText = e.text),
      });
  }
}

And this is some markup that would go with it. The error message <p> tag, including the retry link, is only displayed if the string is not empty..

<ion-content padding>
  <p>Data from service: &nbsp;</p>
  <p *ngIf="errorText">Error: {{errorText}}. 
    You can <a (click)="getData()">retry</a>, 
    or <a href="/home">return home</a>.
  </p>
</ion-content>

Skeleton Text

Yet another pattern renders the page with skeleton text, gray text bars in place of where the details will eventually appear. The page is fully interactive, giving the impression of faster performance, while the order details load. As the data is loaded successfully, it replaces the skeleton text. Any piece of data that fails to load could be replaced with an error message and a retry link.

Here is some sample code to implement this pattern. It is mostly the same, with the obvious elimination of the LoadingController code. I find this pattern to be a lot cleaner, because the error message and skeleton text visibility are both managed entirely by the view layer (i.e., the HTML).

export class SkeletonPage implements OnInit {
  sub;
  dataText = '';
  errorText = '';

  constructor(private data: DataService) { }

  ngOnInit() {
    this.getData();
  }

  async getData() {
    // Ensure error and data message is cleared
    this.dataText = '';
    this.errorText = '';

    // Subscribe to the data service
    this.sub = this.data.getData()
      .subscribe({
        error: (e => this.errorText = e.text),
      });
  }

Here is its corresponding HTML. There is more markup than in the prior example, primarily due to the inclusion of the skeleton text. The method of displaying the error and retry link is identical. However, the skeleton text is hidden as soon as either the data or the error message has a value.

<ion-content padding>
  <p>Data from service</p>
  <div *ngIf="dataText"></div>
  <div *ngIf="!dataText && !errorText">
    <ion-skeleton-text animated style="width:40%"></ion-skeleton-text>
    <ion-skeleton-text animated style="width:15%"></ion-skeleton-text>
    <ion-skeleton-text animated style="width:25%"></ion-skeleton-text>
  </div>
  <p *ngIf="errorText">
    Error: {{errorText}}. You can <a (click)="getData()">retry</a>, 
    or <a href="/home">return home</a>.
  </p>
</ion-content>

Either of these alternative patterns can be handled by kicking off the data load during ngOnInit, calling into the same data service the Route Resolver would call. This strategy maintains a proper separation of concerns, and allows the service to be mocked during testing.

Demo

Here is a demo I created with the above code to show all three of these options side-by-side. The demo consists of three Angular page components, each attempting to load some data from a common service. The service simply waits 3.5 seconds and throws an error.

Options Demo

If you wish to see the code in its entirety, here is the repo on GitHub.

Summary

I wrote this post in response to a real situation I had in a recent project. The main page needed four different pieces of data, all loaded from separate sources. The implementation used a Route Resolver to ensure all of the data loaded before the page appeared. During testing, one of the services was down and was timing out after 30 seconds. This means the user was kept staring at a white screen until the service gave up. Obviously this experience is not ideal, and led me to research other (better?) options.

As part of that research, I came up with some rules guidelines for deciding how best to load asynchronous data:

  1. If your page absolutely cannot be shown unless/until all data is loaded, use a resolver.
    • Do not ignore errors in your resolver, or your page will not render and the user will be stuck in limbo.
    • Redirect the user to an error page, optionally with a retry button.
  2. If your page is still somewhat useful with partial data, load the data asynchronously in ngOnInit.
    • Use skeleton text or a loading indicator.

Review the demo video to see how each one behaves, and choose the option that works best for your projects.

Feedback Appreciated

Do you have any comments or questions? Did I miss your favorite pattern? Did I make any mistakes in this post? Let me know on Twitter. I'm @walkingriver.

I also author courses for Pluralsight, exclusively targeting the Ionic Framework.

Posted on by:

walkingriver profile

Michael D. Callaghan

@walkingriver

Husband, father, developer at Walt Disney World; part-time author and video course producer, mostly about the Ionic Framework

Discussion

markdown guide