On Fri, Mar 18, 2011 at 11:38 PM, Doug Cutting <[email protected]> wrote: > Is that what you're after? Why would you need this?
Probably a small case, in which I would require reading from multiple sources in my job (perhaps even process them differently until the Map phase), with special reader-schemas for each of my sources. This could be custom-built easily, but I just wondered if general use-cases of avro datafiles could benefit from such a thing. Right now AvroJob.setInputSchema(...) sets given schema as "avro.input.schema" in the Job, and my suggestion was to make it something like /path/1+avro.input.schema, /path/2+avro.input.schema so that each instantiated record reader for mappers (via MultipleInputs) can pick up its own special reader schema (since they get a /path/2 via FileSplit). -- Harsh J http://harshj.com
