RX Java - Retry some code that throws exception

Question

I'm trying to use RX Java to consume some data coming from a source that keeps sending objects.

I'm wondering how to implement a retry policy for cases in which my own code throws an exception. For example a network exception should trigger a retry with exponential backoff policy.

Some code :

 message.map(this::processMessage)
                 .subscribe((message)->{
                     //do something after mapping
                 });

processMessage(message) is the method which contains the risky code that might fail and its the part of code that I want to retry but I dont want to stop the observable from consuming data from the source.

Any thoughts on this?

score 12 · Accepted Answer · edited Aug 15 '19 at 16:05

message
    .map(this::processMessage)
    .retryWhen(errors -> errors.flatMap(error -> {  
        if (error instanceof IOException) {
          return Observable.just(null);
        }
        // For anything else, don't retry
        return Observable.error(error);
     })
     .subscribe(
         System.out::println,
         error -> System.out.println("Error!")
     );

or catch the error

message.map(this::processMessage)
           .onErrorReturn(error -> "Empty result")
           .subscribe((message)->{})

or procses the error

message
    .map(this::processMessage)
    .doOnError(throwable -> Log.e(TAG, "Throwable " + throwable.getMessage()))
    .subscribe(
         System.out::println,
         error -> System.out.println("Error!")
     );

Untested, but retryWhen differs to repeatWhen that is not only called in onComplete.

http://blog.danlew.net/2016/01/25/rxjavas-repeatwhen-and-retrywhen-explained/ -> Each error is flatmapped so that we can either return onNext(null) (to trigger a resubscription) or onError(error) (to avoid resubscription).

Backoff Policy:

source.retryWhen(errors ->  
  errors
    .zipWith(Observable.range(1, 3), (n, i) -> i)
    .flatMap(retryCount -> Observable.timer((long) Math.pow(5, retryCount), TimeUnit.SECONDS))
);

flatMap + timer is preferable over delay in this case because it lets us modify the delay by the number of retries. The above retries three times and delays each retry by 5 ^ retryCount, giving you exponential backoff with just a handful of operators!

Your suggestions do not include any logic that repeats the failing code. Also, if you use onErrorReturn, the observable will stop consuming new values — Nick Tsitlakidis, Mar 06 '17 at 10:32
Modified. Should solve your question. If not just comment, then ill provide a working code. — Emanuel, Mar 06 '17 at 10:37

score 2 · Answer 2 · answered Mar 06 '17 at 10:25

2

Take an example from articles:

https://medium.com/@v.danylo/server-polling-and-retrying-failed-operations-with-retrofit-and-rxjava-8bcc7e641a5a#.a6ll8d5bt

http://kevinmarlow.me/better-networking-with-rxjava-and-retrofit-on-android/

They helped me oneday.

answered Mar 06 '17 at 10:25

Dmitry Gorkovets

2,208
1
10
19

score 0 · Answer 3 · answered Mar 06 '17 at 12:10

0

Recently I developed the library that exactly suits your needs.

RetrofitRxErrorHandler

If you combine Exponential strategy with backupObservable you will get the expected result.

answered Mar 06 '17 at 12:10

R. Zagórski

20,020
5
65
90

cool project but I'm not using Retrofit or http. I have a constant feed coming from RabbitMQ and the observable responds to that. – Nick Tsitlakidis Mar 06 '17 at 12:16
So the code may be the basis for what you are trying to achieve. – R. Zagórski Mar 06 '17 at 12:36

RX Java - Retry some code that throws exception

3 Answers3