We got an issue with avro serialization/de-serialization when running a map-reduce job against avro data. The input schema of the job is union of two schemas. One of them is generated with Sqoop. When avro calls to resolveUnion(Schema union, Object datum) in GenericData.java, it compares the names of component schemas of the union with that of the schema of datum. However, the comparison alway fails because one of the name prepends with the corresponding namespace and the other not. We have a workaround for this. But we think this is a bug. Can somebody verify this and fix it? Thanks.
Harsh J 2012-01-31, 05:53