Is the implementation of EventEmitter under your control? or it is just a mock implementation of external service? If they're under your control I think you can make the emitNextFailureHandler returns true when encountering FAIL_OVERFLOW or FAIL_TERMINATED so that the EventEmitter can retry emit.

Simple example:

Sinks.EmitFailureHandler emitNextFailureHandler = (signalType, emitResult) -> {
    return Sinks.EmitResult.FAIL_OVERFLOW.equals(emitResult) || Sinks.EmitResult.FAIL_NON_SERIALIZED.equals(emitResult);
};

Also, I'd like to discuss this in another aspect.

I think the repository::remove is a synchronized method that blocks your thread so that's why the queue is under pressure so hard. This might not be a best practice for reactive programming. A better way to do this might be to make the repository.remove() return a Mono or Flux (in practice you may need to change the dependencies to Reactive related dependencies), then use .flatMap(repository::remove) instead of .doOnNext()