How to ignore error and continue infinite stream?

问题:

I would like to know how to ignore exceptions and continue infinite stream (in my case stream of locations)?

I'm fetching current user position (using Android-ReactiveLocation) and then sending them to my API (using Retrofit).

In my case, when exception occurs during network call (e.g. timeout) onError method is invoked and stream stops itself. How to avoid it?

Activity:

private RestService mRestService;
private Subscription mSubscription;
private LocationRequest mLocationRequest = LocationRequest.create()
            .setPriority(LocationRequest.PRIORITY_HIGH_ACCURACY)
            .setInterval(100);
...
private void start() {
    mRestService = ...;
    ReactiveLocationProvider reactiveLocationProvider = new ReactiveLocationProvider(this);
    mSubscription = reactiveLocationProvider.getUpdatedLocation(mLocationRequest)
            .buffer(50)
            .flatMap(locations -> mRestService.postLocations(locations)) // can throw exception
            .subscribeOn(Schedulers.newThread())
            .observeOn(AndroidSchedulers.mainThread())
            .subscribe();
}

RestService:

public interface RestService {
    @POST("/.../")
    Observable<Response> postLocations(@Body List<Location> locations);
}

回答1:

mRestService.postLocations(locations) emit one item, then complete. If an error occur, then it emit the error, which complete the stream.

As you call this method in a flatMap, the error continue to your "main" stream, and then your stream stops.

What you can do is to transform your error into another item (as described here : https://stackoverflow.com/a/28971140/476690 ), but not on your main stream (as I presume you already tried) but on the mRestService.postLocations(locations).

This way, this call will emit an error, that will be transformed to an item/another observable and then complete. (without calling onError).

On a consumer view, mRestService.postLocations(locations) will emit one item, then complete, like if everything succeed.

mSubscription = reactiveLocationProvider.getUpdatedLocation(mLocationRequest)
        .buffer(50)
        .flatMap(locations -> mRestService.postLocations(locations).onErrorReturn((e) -> Collections.emptyList()) // can throw exception
        .subscribeOn(Schedulers.newThread())
        .observeOn(AndroidSchedulers.mainThread())
        .subscribe();

回答2:

You may want to use one of the error handling operators.

• onErrorResumeNext( ) — instructs an Observable to emit a sequence of items if it encounters an error
• onErrorReturn( ) — instructs an Observable to emit a particular item when it encounters an error
• onExceptionResumeNext( ) — instructs an Observable to continue emitting items after it encounters an exception (but not another variety of throwable)
• retry( ) — if a source Observable emits an error, resubscribe to it in the hopes that it will complete without error
• retryWhen( ) — if a source Observable emits an error, pass that error to another Observable to determine whether to resubscribe to the source

Especialy retry and onExceptionResumeNext look promising in your case.

回答3:

Just pasting the link info from @MikeN's answer incase it gets lost:

import rx.Observable.Operator;
import rx.functions.Action1;

public final class OperatorSuppressError<T> implements Operator<T, T> {
    final Action1<Throwable> onError;

    public OperatorSuppressError(Action1<Throwable> onError) {
        this.onError = onError;
    }

    @Override
    public Subscriber<? super T> call(final Subscriber<? super T> t1) {
        return new Subscriber<T>(t1) {

            @Override
            public void onNext(T t) {
                t1.onNext(t);
            }

            @Override
            public void onError(Throwable e) {
                onError.call(e);
            }

            @Override
            public void onCompleted() {
                t1.onCompleted();
            }

        };
    }
}

and use it close to the observable source because other operators may eagerly unsubscribe before that.

Observerable.create(connectToUnboundedStream()).lift(new OperatorSuppressError(log()).doOnNext(someStuff()).subscribe();

Note, however, that this suppresses the error delivery from the source. If any onNext in the chain after it throws an exception, it is still likely the source will be unsubscribed.

回答4:

If you just want to ignore the error inside the flatMap without returning an element do this:

flatMap(item -> 
    restService.getSomething(item).onErrorResumeNext(Observable.empty())
);

回答5:

Try calling the rest service in a Observable.defer call. That way for every call you'll get a chance to use its own 'onErrorResumeNext' and the errors won't cause your main stream to complete.

reactiveLocationProvider.getUpdatedLocation(mLocationRequest)
  .buffer(50)
  .flatMap(locations ->
    Observable.defer(() -> mRestService.postLocations(locations))
      .onErrorResumeNext(<SOME_DEFAULT_TO_REACT_TO>)
  )
........

That solution is originally from this thread -> RxJava Observable and Subscriber for skipping exception?, but I think it will work in your case too.

回答6:

Add my solution for this problem:

privider
    .compose(ignoreErrorsTransformer)
    .subscribe()

private final Observable.Transformer<ResultType, ResultType> ignoreErrorsTransformer =
        new Observable.Transformer<ResultType, ResultType>() {
            @Override
            public Observable<ResultType> call(Observable<ResultType> resultTypeObservable) {
                return resultTypeObservable
                        .materialize()
                        .filter(new Func1<Notification<ResultType>, Boolean>() {
                            @Override
                            public Boolean call(Notification<ResultType> resultTypeNotification) {
                                return !resultTypeNotification.isOnError();
                            }
                        })
                        .dematerialize();

            }
        };

回答7:

A slight modification of the solution (@MikeN) to enable finite streams to complete:

import rx.Observable.Operator;
import rx.functions.Action1;

public final class OperatorSuppressError<T> implements Operator<T, T> {
    final Action1<Throwable> onError;

    public OperatorSuppressError(Action1<Throwable> onError) {
        this.onError = onError;
    }

    @Override
    public Subscriber<? super T> call(final Subscriber<? super T> t1) {
        return new Subscriber<T>(t1) {

            @Override
            public void onNext(T t) {
                t1.onNext(t);
            }

            @Override
            public void onError(Throwable e) {
                onError.call(e);
                //this will allow finite streams to complete
                t1.onCompleted();
            }

            @Override
            public void onCompleted() {
                t1.onCompleted();
            }

        };
    }
}