Hi,
We have a record format defined in the avro avdl. One of the field in the avdl
is of the type union {map<map<bytes>>, null}. The avro file with this avdl
schema is used as input and output of our two map/reduce jobs, based on avro
java api, respectively. We process records of the file in the map/reduce jobs
using avro generic record, where the type string is actually the Utf8 object.
We never encounter any issue with this approach. However, recently, we try to
use AvroStorage() of Pig to read in the avro file and, unlike the avro
map/reduce jobs, the value of the field with the above type definition appears
to be not correct. AvroStorage() uses the avro class DataFileReader to process
data. Is there anybody knows the difference in handling avro data between
DataFileReader and Avro Map/Reduce API with the Generic Record ? Is this a bug
in the class DataFileReader? Thanks.
Ey-Chih Chow
