According to the spec, the default value for a union is assumed to have
the type of the first element of the union.
http://avro.apache.org/docs/current/spec.html#schema_record
So some valid fields would be:
{"name":"x", "type":["long", "null"], "default": 0}
{"name":"y", "type":["null", "long"], "default": null}
The following are invalid fields, since the type of the default value
does not match that of the first union element.
{"name":"x", "type":["long", "null"], "default": null}
{"name":"y", "type":["null", "long"], "default": 0}
Python may not implement this strictly, but Java does.
This is a common point of confusion. We should probably document it
better. I'm not sure whether it's causing the problem you're seeing,
but perhaps it is.
Cheers,
Doug
On 06/06/2012 04:15 AM, François Kawala wrote:
Dear all,
Despite my desperate effort to get a working schema I can not manage to
specify that a field of a given record can be either : "a given type" or
"null". I've tried with unions but the back-end that I have to use seems
to be unhappy with it. More precisely : I'm trying to output the result
of a Streaming MR job within an AVRO container. This job is written in
python an executed through dumbo (http://www.dumbotics.com), and a
custom OutputFormat is used
(https://github.com/tomslabs/avro-utils/tree/master/src/main/java/com/tomslabs/grid/avro)
However since this custom OutputFormat relies on org.apache.avro
sources, I've thought this list could be a good spot to call for help.
Thanks for reading,
François.
------------------------------------------------------------------------
Here is some complementary elements :
Fragment of the schema that I think to be responsible of my troubles :
{"name": "in_reply_to", "type": [{"type": "long"},"null"], "default":"null"}
I've also unsuccessfully tried :
{"name": "in_reply_to", "type": [{"type": "long"},"null"]}
{"name": "in_reply_to", "type": ["null",{"type": "long"}]}
Each ending with the same error message :
org.apache.avro.AvroTypeException: Expected start-union. Got
VALUE_NUMBER_INT
Error Stack :
at org.apache.avro.io.JsonDecoder.error(JsonDecoder.java:460)
at org.apache.avro.io.JsonDecoder.readIndex(JsonDecoder.java:418)
at
org.apache.avro.io.ResolvingDecoder.doAction(ResolvingDecoder.java:229)
at org.apache.avro.io.parsing.Parser.advance(Parser.java:88)
at
org.apache.avro.io.ResolvingDecoder.readIndex(ResolvingDecoder.java:206)
at
org.apache.avro.generic.GenericDatumReader.read(GenericDatumReader.java:142)
at
org.apache.avro.generic.GenericDatumReader.readRecord(GenericDatumReader.java:166)
at
org.apache.avro.generic.GenericDatumReader.read(GenericDatumReader.java:138)
at
org.apache.avro.generic.GenericDatumReader.read(GenericDatumReader.java:129)
at
com.tomslabs.grid.avro.TextTypedBytesToAvroOutputFormat$AvroRecordWriter.write(TextTypedBytesToAvroOutputFormat.java:102)
at
com.tomslabs.grid.avro.TextTypedBytesToAvroOutputFormat$AvroRecordWriter.write(TextTypedBytesToAvroOutputFormat.java:88)
at org.apache.hadoop.mapred.ReduceTask$3.collect(ReduceTask.java:446)
at
org.apache.hadoop.streaming.PipeMapRed$MROutputThread.run(PipeMapRed.java:421)