Looks like it's failing because records must have at least one field.
On Tue, May 24, 2016 at 9:00 AM, christopher palm <[email protected]> wrote:
> Hi Ryan,
>
> Thanks for Clarifying, here is the section of the avro schema that I'm
> trying to write.
> {
> "name": "tireSlippage",
> "type": ["null", {
> "type": "record",
> "name": "TireSlippageEvent",
> "fields": []
> }],
> "default": null
> },
>
> This Schema was created using the ProtobufDatumReader from the original
> Protobuf class.
>
> Is this failing because the type is a record?
>
> Thanks,
> Chris
>
> On Tue, May 24, 2016 at 10:02 AM, Ryan Blue <[email protected]>
> wrote:
>
> > Chris,
> >
> > That change allows empty groups when reading, not when writing. For
> > reading, it allows you to do operations like count without projecting any
> > data columns but there's no corresponding use for writing.
> >
> > Optional fields are a different feature, which should work just fine. How
> > are you getting an empty group from an optional field? Avro's pattern for
> > optional is a union with null, so you should have something like this:
> >
> > ```json
> > {
> > "name" : "some_field",
> > "type" : [ "null", "string" ],
> > "default" : null
> > }
> > ```
> >
> > What is the schema you're trying to write?
> >
> > rb
> >
> > On Mon, May 23, 2016 at 8:32 PM, christopher palm <[email protected]>
> > wrote:
> >
> > > Hi All,
> > > I see that Parquet-363 made some changes to allow empty schema
> groups,
> > > however I am testing with this patch and still get the error below on
> an
> > > optional field.
> > > Cannot write a schema with an empty group: optional group
> > >
> > > Is this fix supposed to allow a parquet file to be created when the
> Avro
> > > schema has optional fields?
> > >
> > > Thanks,
> > >
> > > Chris
> > >
> >
> >
> >
> > --
> > Ryan Blue
> > Software Engineer
> > Netflix
> >
>
--
Ryan Blue
Software Engineer
Netflix
