I suggest you use Mapper<AvroKey<Object>,NullWritable> or even Mapper<AvroKey<SpecificRecord>,NullWritable>
You can than decide on sorting key by checking the datum() of the avro key with instanceof or with SpecificRecord#getSchema()#getName in the latter case. The schema for the map output (AvroJob#setMapOutputKeySchema()) will be the union again and the key depends on your use case. The sorting by default should use the provided key for sorting the (key,[values]) On Mon, Feb 27, 2017 at 6:48 AM, Zephod <zep...@tlen.pl> wrote: > Dear Johannes, > Thank you for your suggestion :) I know the types in advance (data model is > specific). Where could I read how the union type works? Are there any > examples e.g.: > -what kind of types should I use for type parameters of the mapper/reducer? > -how the values will be sorted in the shuffle phase? > -how do I check what type are the values I'm iterating over in the reducer? > > > > > > -- > View this message in context: http://apache-avro.679487.n3. > nabble.com/Avro-with-MultipleInputs-in-the-new-API-tp4036886p4036890.html > Sent from the Avro - Users mailing list archive at Nabble.com. >