Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Plain View
Avro, mail # user - Union in AvroMapper.map: Not in Union??


+
nir_zamir 2013-03-24, 10:23
+
nir_zamir 2013-04-07, 14:48
Copy link to this message
-
Re: Union in AvroMapper.map: Not in Union??
Martin Kleppmann 2013-04-10, 03:55
Hi Nir,

I'm not an expert with the avro.mapred APIs, but as far as I know,
AvroJob does not perform schema evolution, so the schema you provide
to AvroJob.setInputSchema has to be the exact same schema with which
your input files to the mappers are encoded. So if your input isn't
actually a union type, but your schema says that it is, decoding of
the input file will fail.

If you want to consume several different input directories with
different schemas in the same MapReduce job, you might be able to
build that using Hadoop's MultipleInputs, though I don't know how well
that plays with Avro. Alternatively, you could take a look at
http://crunch.apache.org/ which provides higher-level MapReduce APIs
and is designed to play nicely with Avro.

Hope that helps,
Martin

On 24 March 2013 03:23, nir_zamir <[EMAIL PROTECTED]> wrote:
> Hi,
>
> I have a small M/R which, for the sake of troubleshooting, just reads a
> single Avro file and sends it to an AvroMapper. In order to handle different
> schemas, I set the mapper to work with a Union Schema.
> For the simplicity, the union is a union of only one schema, taken from the
> Avro file. This is my run function:
>
>         public int run2(String[] args) throws Exception {
>                 List<Schema> schemas= new ArrayList<Schema>();
>                 Schema schema = readSchema(new Path(inputDir), conf);
>                 schemas.add(schema);
>         Schema unionSchema = Schema.createUnion(schemas);
>
>                 AvroJob.setInputSchema(conf, unionSchema);
>                 AvroJob.setMapOutputSchema(conf,
> Pair.getPairSchema(Schema.create(Type.INT),
>                                 unionSchema));
>
>                 JobClient.runJob(conf);
>                 return 0;
>         }
>
> And my mapper is:
>
>         public static class MyMap extends AvroMapper <GenericRecord,
> Pair<Integer, GenericRecord>> {
>                 @Override
>                 public void map( GenericRecord datum, AvroCollector<Pair<Integer,
> GenericRecord>> collector, Reporter reporter) throws IOException {
>                         collector.collect(new Pair<Integer, GenericRecord>(conversion_id,
> datum));
>                 }
>         }
>
> *Side note:* when I extract the schema manually from the Avro file and call
> GenericData.get().resolveUnion - the schema is correctly resolved and found
> in the union schema (the methods returns 0).
>
>
> *THE PROBLEM:*
> the 'collect' method throws an exception saying my datum's schema is not in
> the union:
>
> org.apache.avro.UnresolvedUnionException: Not in union
> [{"type":"record","name":"path_to_conversions_raw","namespace":"org.apache.avro.mapred","doc":"Sqoop
> import of
> path_to_conversions_raw","fields":[{"name":"conversion_id","type":["long","null"],"columnName":"conversion_id","sqlType":"-5"},{"name":"conversion_time","type":["long","null"],"columnName":"conversion_time","sqlType":"93"},{"name":"derived_time","type":["long","null"],"columnName":"derived_time","sqlType":"93"},{"name":"position","type":["int","null"],"columnName":"position","sqlType":"5"},{"name":"event_time","type":["long","null"],"columnName":"event_time","sqlType":"93"},{"name":"event_type_id","type":["int","null"],"columnName":"event_type_id","sqlType":"5"},{"name":"entity_id","type":["int","null"],"columnName":"entity_id","sqlType":"4"},{"name":"entity_type_id","type":["int","null"],"columnName":"entity_type_id","sqlType":"4"},{"name":"sv_click_type","type":["int","null"],"columnName":"sv_click_type","sqlType":"5"},{"name":"placement_id","type":["int","null"],"columnName":"placement_id","sqlType":"4"},{"name":"site_id","type":["int","null"],"columnName":"site_id","sqlType":"4"},{"name":"campaign_id","type":["int","null"],"columnName":"campaign_id","sqlType":"4"},{"name":"days_before_conversion","type":["int","null"],"columnName":"days_before_conversion","sqlType":"5"},{"name":"version_id","type":["int","null"],"columnName":"version_id","sqlType":"4"},{"name":"target_audience_id","type":["int","null"],"columnName":"target_audience_id","sqlType":"4"},{"name":"ip_attributes_id","type":["long","null"],"columnName":"ip_attributes_id","sqlType":"-5"},{"name":"batch_id","type":["long","null"],"columnName":"batch_id","sqlType":"-5"},{"name":"ccs_ad_id","type":["long","null"],"columnName":"ccs_ad_id","sqlType":"-5"}],"tableName":"path_to_conversions_raw"}]:
+
nir_zamir 2013-04-10, 10:43
+
Doug Cutting 2013-04-10, 18:47
+
Martin Kleppmann 2013-04-10, 20:30
+
nir_zamir 2013-04-14, 08:39