You'll need to put your leaf fields inside a map vector to refer to them in
the way you want.
On Mar 4, 2016 5:04 PM, "Jiang Wu" <[email protected]> wrote:
> We are working on a custom Drill storage plugin to retrieve data from a
> proprietary a JSON based storage. The drill version being used is 1.4.0.
> Assuming the data is a JSON looking like this:
>
> {
> ...
> "topping": {
> "id": "5001, 5002, ...",
> ...
> }
> ...
> }
>
> When we submit this query:
>
> select t.topping.id from meld.project1.event.table1 t;
>
> The plugin receives a SchemaPath object for "topping.id". The plugin
> then creates an output vector with the provided SchemaPath as the field
> name, e.g.
>
> final MajorType type = Types.optional(MinorType.VARCHAR); // assuming we
> always use this type
> final MaterializedField field = MaterializedField.create(schemaPath,
> type); // schemaPath is given to the plugin
>
> final Class<? extends ValueVector> clazz = (Class<? extends ValueVector>)
> TypeHelper.getValueVectorClass(type.getMinorType(), type.getMode());
>
> ValueVector vector = output.addField(field, clazz);
>
> The data is then added into the vector and returned. However, the returned
> field cannot be matched with what Drill expects. So we get something like
> this:
>
> 0: jdbc:drill:zk=local> select t.topping.id from
> meld.project1.event.table1 t;
> +---------+
> | EXPR$0 |
> +---------+
> | null |
> | null |
> | null |
> | null |
> | null |
> | null |
> +---------+
>
> When we are expecting to get this:
>
> 0: jdbc:drill:zk=local> select t.`topping.id` from
> meld.project1.event.table1 t;
> +-------------------------------------------+
> | topping.id |
> +-------------------------------------------+
> | 5001, 5002, 5003, 5004 |
> | 5001, 5002, 5005, 5007, 5006, 5003, 5004 |
> | 5001, 5002, 5003, 5004 |
> | 5001, 5002, 5005, 5007, 5006, 5003, 5004 |
> | 5001, 5002, 5005, 5003, 5004 |
> | 5001, 5002, 5005, 5003, 5004 |
> +-------------------------------------------+
>
> In the second query, the SchemaPath is a nested path. Our plugin accepts
> this specification and retrieve the same results. So what are we doing
> wrong here? How do we correctly return values for a nested JSON field?
>
> Thanks.
>
> -- Jiang
>