Error handling

There are two major approaches how to handle errors in streams. You can retry your stream and how it eventually will work or you can take the error and transform it.

Retry - how bout now?

This approach makes sense when you believe the error is temporary for some reason. Usually shaky connections is a good candidate for this. With a shaky connection the endpoint might be there to answer like for example every 5th time you try. Point is the first time you try it might fail, but retrying x times, with a certain time between attempts, will lead to the endpoint finally answering.

retry

The retry() operator lets us retry the whole stream, value for value x number of times having a signature like this :

retry([times])
1

The important thing to note with the retry() operator is that it delays when the error callback is being called. Given the following code the error callback is being hit straight away:

import { of } from 'rxjs';
import { map } from 'rxjs/operators';

let stream$ = of(1,2,3).pipe(
  map(value => {
    if(value > 2) { throw 'error' }
  })
);

stream$.subscribe(
  data => console.log(data),
  err => console.log(err)
)
1
2
3
4
5
6
7
8
9
10
11
12
13

The stream effectively dies when the error callback is being hit and this is where the rety() operator comes in. By appending it like so:

import { of } from 'rxjs';

let stream$ = of(1,2,3).pipe(
  map(value => {
    if(value > 2) { throw 'error' }
  }),
  retry(5)
)

1
2
3
4
5
6
7
8
9

This will run the sequence of values 5 more times before finally giving up and hitting the error callback. However in this case, the way the code is written, it will just generate 1,2 five times. So our code isn't really utilizing the operator to its fullest potential. What you probably want is to be able to change something between attempts. Imagine your observable looked like this instead:

import { of } from 'rxjs';

let urlsToHit$ = of(url, url2, url3);
1
2
3

In this its clearly so that an endpoint might have answered badly or not at all on your first attempt and it makes sense to retry them x number of times.

However in the case of ajax calls, and imagining our business case is shaky connections it makes no sense to do the retry immediately so we have to look elsewhere for a better operator, we need to look to retryWhen()

retryWhen

A retryWhen() operator gives us the chance to operate on our stream and handle it appropriately

retryWhen( stream => {
  // return it in a better condition, hopefully
})
1
2
3

Lets' write a piece of naive code for a second

import { of } from 'rxjs';
import { map } from 'rxjs/operators';

let values$ = of( 1,2,3,4 ).pipe(
  map(val => {
    if(val === 2) {
      throw 'err';
    } else return val;
  }),
  retryWhen( stream => {
    return stream;
  })
);

values$.subscribe(
  data => console.log('Retry when - data',data),
  err => console.error('Retry when - Err',err)
)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

The way it's written it will return 1 until we run out of memory cause the algorithm will always crash on the value 2 and will keep retrying the stream forever, due to our lack of end condition. What we need to do is to somehow say that the error is fixed. If the stream were trying to hit urls instead of emitting numbers a responding endpoint would be the fix but in this case we have to write something like this:

import { interval } from 'rxjs';
import { take, map } from 'rxjs/operators';

let values$ = interval(1000).pipe(
  take(5)
)

let errorFixed = false;

values$.pipe(
  map((val) => {
    if(errorFixed) {
      return val;
    } else if( val > 0 && val % 2 === 0) {
      errorFixed = true;
      throw { error : 'error' };
    } else {
      return val;
    }
  }),
  retryWhen((err) => {
    console.log('retrying the entire sequence');
    return err;
  })
)
  .subscribe((val) => { console.log('value',val) })

// 0 1 'wait 200ms' retrying the whole sequence 0 1 2 3 4
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

This however resembles a lot of what we did with the retry() operator, the code above will just retry once. The real benefit is being to change the stream we return inside the retryWhen() namely to involve a delay like this:

.retryWhen((err) => {
  console.log('retrying the entire sequence');
  return err.delay(200)
})

1
2
3
4
5

This ensures there is a 200ms delay before sequence is retried, which in an ajax scenario could be enough for our endpoint to get it's shit together and start responding.

GOTCHA

The delay() operator is used within the retryWhen() to ensure that the retry happens a while later to in this case give the network a chance to recover.

retryWhen with delay and no of times

So far retry() operator has been used when we wanted to retry the sequence x times and retryWhen() has been used when we wanted to delay the time between attempts, but what if we want both. Can we do that? We can. We need to think about us somehow remembering the number of attempts we have made so far. It's very tempting to introduce an external variable and keep that count, but that's not how we do things the functional way, remember side effects are forbidden. So how do we solve it? There is an operator called scan() that will allow us to accumulate values for every iteration. So if you use scan inside of the retryWhen() we can track our attempts that way:

import { of } from 'rxjs';
import { map, retryWhen } from 'rxjs/operators';

let ATTEMPT_COUNT = 3;
let DELAY = 1000;
let delayWithTimes$ = of(1,2,3).pipe(
  map( val => {
    if(val === 2) {
      throw 'err'
    } else {
      return val;
    }
  }),
  retryWhen(e => e.scan((errorCount, err) => {
    if (errorCount >= ATTEMPT_COUNT) {
      throw err;
    }
    return errorCount + 1;
  }, 0).delay(DELAY));
);

delayWithTimes$.subscribe(
  val => console.log('delay and times - val',val),
  err => console.error('delay and times - err',err)
)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

Transform - nothing to see here folks

This approach is when you get an error and you choose to remake it into a valid Observable.

So lets exemplify this by creating an Observable who's mission in life is to fail miserably

import { throwError } from 'rxjs/operators';

let error$ = Rx.Observable.throwError('crash');

error$.subscribe(
  data => console.log( data ),
  err => console.log( err ),
  () => console.log('complete')
)
1
2
3
4
5
6
7
8
9

This code will only execute the error callback and NOT reach the complete callback.

Patching it

We can patch this by introducing the catch() operator. It is used like this:

import { of } from 'rxjs';

let errorPatched$ = error$.catch(
err => { return of('Patched' + err) }
);
errorPatched$.subscribe((data) => console.log(data) );
1
2
3
4
5
6

As you can see patching it with .catch() and returning a new Observable fixes the stream. Question is if that is what you want. Sure the stream survives and reaches completion and can emit any values that happened after the point of crash.

If this is not what you want then maybe the Retry approach above suits you better, you be the judge.

What about multiple streams?

You didn't think it would be that easy did you? Usually when coding Rxjs code you deal with more than one stream and using catch() operator approach is great if you know where to place your operator.

import { of, merge } from 'rxjs';
import { throwError } from 'rxjs/operators';

let badStream$ = throwError('crash');
let goodStream$ = of(1,2,3,);

let merged$ = merge(
  badStream$,
  goodStream$
);

merged$.subscribe(
  data => console.log(data),
  err => console.error(err),
  () => console.log('merge completed')
)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

Care to guess what happened?

  1. crash + values is emitted + complete
  2. crash + values is emitted
  3. crash only is emitted

Sadly 3) is what happens. Which means we have virtually no handling of the error.

Lets patch it

S we need to patch the error. We do patching with catch() operator. Question is where?

Let's try this?

import { merge, of } from 'rxjs';

let mergedPatched$ = merge(
  badStream$,
  goodStream$
)
  .catch(err => of(err));

mergedPatched$.subscribe(
  data => console.log(data),
  err => console.error(err),
  () => console.log('patchedMerged completed')
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14

In this case we get 'crash' and 'patchedMerged completed'. Ok so we reach complete but it still doesn't give us the values from goodStream$. So better approach but still not good enough.

Patch it better

So adding the catch() operator after the merge() ensured the stream completed but it wasn't good enough. Let's try to change the placement of catch(), pre merge.

import { merge, of } from 'rxjs';

let preMergedPatched$ = merge(
  badStream$.catch(err => of(err)),
  goodStream$
).catch(err => of(err));

preMergedPatched$.subscribe(
  data => console.log(data),
  err => console.error(err),
  () => console.log('pre patched merge completed')
)
1
2
3
4
5
6
7
8
9
10
11
12

And voila, we get values, our error emits its error message as a new nice Observable and we get completion.

GOTCHA It matters where the catch() is placed.

Survival of the fittest

There is another scenario that might be of interest. The above scenario assumes you want everything emitted, error messages, values, everything.

What if that is not the case, what if you only care about values from streams that behave? Let's say thats your case, there is an operator for that onErrorResumeNext()

import { throwError, of } from 'rxjs';
import { onErrorResumeNext } from 'rxjs/operators';

let secondBadStream$ = throwError('bam');
let gloriaGaynorStream$ = of('I will survive');

let emitSurviving = onErrorResumeNext(
  badStream$,
  secondBadStream$,
  gloriaGaynorStream$
);

emitSurviving.subscribe(
  data => console.log(data),
  err => console.error(err),
  () => console.log('Survival of the fittest, completed')
)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17

The only thing emitted here is 'I will survive' and 'Survival of the fittest, completed'.