[
https://issues.apache.org/jira/browse/CALCITE-1581?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15827594#comment-15827594
]
Xiaoyong Deng commented on CALCITE-1581:
----------------------------------------
Thanks, Julian.
My request is a stream compute scenario:
Binary stream datas with meta are generated by camera continuously, and we want
to get the car informations from the video stream, query like this:
{code}
select stream
udtf_analyze(binary_data, meta_data) as (car_color, car_speed, car_plate, ...)
from video_stream;
{code}
There could be no or many cars for one record(binary_data, meta_data) in, and a
set of records out. So we call it UDTF.
We tried to use TableFunction to realize it, but found that the query is used
like the following:
{code}
select
*
from table("s"."GenerateStrings"(5)) as t(n, c);
{code}
And we have no idea how to use the columns in our table "video_stream".
In our design, the udtf function is the only item of "select" clause, and the
following query is invalid.
{code}
select
c2,
func(c0, c1) as (f0, f1, f2)
from table_name;
{code}
{quote}
Can you give an example of an implementation of such a function? Would it be a
Java method returning a class?
{quote}
Do you mean how to define the function? The following code could be a reference:
{code}
public abstract class UDTF {
// in this method we specify input and output parameters: input
ObjectInspector and an output struct
public abstract StructObjectInspector initialize(ObjectInspector[] var1);
// here we process an input record and write out any resulting records
public abstract void process(Object[] var1);
// this function is Called to notify the UDTF that there are no more rows
to process. Clean up code or additional output can be produced here.
public abstract void close();
protected final void forward(Object o) {
// collect the result
}
}
public class ExampleUDTF extends UDTF {
public StructObjectInspector initialize(ObjectInspector[] var1) {
// initialize
}
public void process(Object[] var1) {
// List results = do some thing;
for(Object obj : results) {
forward(obj);
}
}
public void close() {
// close
}
// other functions
}
{code}
> UDTF like in hive
> -----------------
>
> Key: CALCITE-1581
> URL: https://issues.apache.org/jira/browse/CALCITE-1581
> Project: Calcite
> Issue Type: New Feature
> Reporter: Xiaoyong Deng
> Assignee: Julian Hyde
> Labels: udtf
>
> Support one row in and multi-column/multi-row out(one-to-many mapping), just
> like udtf in hive.
> The query would like this:
> {code}
> select
> func(c0, c1) as (f0, f1, f2)
> from table_name;
> {code}
> c0 and c1 are 'table_name' columns. f0, f1 and f2 are new generated columns.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
