git.net

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [DISCUSS] Support Higher-order functions in Flink sql


Hi everyone,

thanks for starting the discussion. In general, I like the idea of making Flink SQL queries more concise.

However, I don't like to diverge from standard SQL. So far, we managed to add a lot of operators and functionality while being standard compliant. Personally, I don't see a good reason for forking the Calcite parser just for little helper functions that could also be expressed as subqueries.

Instead, we could think about some user-defined functions. Instead of:

TRANSFORM(arrays, element -> element + 1)

we could do:

TRANSFORM(arrays, "element -> element + 1")

The second argument could either be SQL or some more domain-specific standard language.

Similar efforts have been done for querying JSON data in the new SQL JSON standard [1] (they are using XQuery or XPath syntax).

Just some ideas from my side.

Regards,
Timo

[1] https://docs.oracle.com/en/database/oracle/oracle-database/12.2/adjsn/query-json-data.html#GUID-119E5069-77F2-45DC-B6F0-A1B312945590


Am 05.12.18 um 09:54 schrieb TANG Wen-hui:
Hi XueFu, Jark,
Thanks for your feedback. That's really helpful.
Since Flink has already supported some complex types like MAP and ARRAY,
it would be possible to add some higher-order functions to deal with MAP and ARRAY, like Presto[1,2] and Spark have done.
As for "syntax for the lambda function ", I have started a discussion in Calcite's mail list to look forward some feedbacks.
I am willing to follow up the topic and come up with a design doc later.
Best,
Wen-hui



winifred.wenhui.tang@xxxxxxxxx
From: Jark Wu
Date: 2018-12-05 10:27
To: dev; xuefu.z
Subject: Re: [DISCUSS] Support Higher-order functions in Flink sql
Hi Wenhui,
This is a meaningful direction to improve the functionality for Flink SQL.
As Xuefu suggested, you can come up with a design doc covering the
functions you'd like to support and the improvements.
IMO, the main obstacle might be the syntax for the lambda function which is
not supported in Calcite currently, such as: "TRANSFORM(arrays, element ->
element + 1)". In order to support this syntax,
we might need to discuss it in Calcite community. It is not like DDL
parser, the DDL parser is easy to extend in a plugin way which is Calcite
suggests.
It would be great if you can share more thoughts or works on this. Best,
Jark
On Mon, 3 Dec 2018 at 17:20, Zhang, Xuefu <xuefu.z@xxxxxxxxxxxxxxx> wrote:
Hi Wenhui,

Thanks for bringing the topics up. Both make sense to me. For higher-order
functions, I'd suggest you come up with a list of things you'd like to add.
Overall, Flink SQL is weak in handling complex types. Ideally we should
have a doc covering the gaps and provide a roadmap for enhancement. It
would be great if you can broaden the topic a bit.

Thanks,
Xuefu


------------------------------------------------------------------
Sender:winifred.wenhui.tang@xxxxxxxxx <winifred.wenhui.tang@xxxxxxxxx>
Sent at:2018 Dec 3 (Mon) 16:13
Recipient:dev <dev@xxxxxxxxxxxxxxxx>
Subject:[DISCUSS] Support Higher-order functions in Flink sql

Hello all,

Spark 2.4.0 was released last month. I noticed that Spark 2.4
“Add a lot of new built-in functions, including higher-order functions, to
deal with complex data types easier.”[1]
I wonder if it's necessary for Flink to add higher-order functions to
enhance it's ability.

By the way, I found that if we wants to enhance the functionality of Flink
sql, we often need to modify Calcite. It may be a little inconvenient,so
may be we can extend Calcite core parser in Flink to deal with some
non-standard SQL syntax, as mentioned in Flink SQL DDL Design[2].

Look forward to your feedback.

Best,
Wen-hui Tang

[1] https://issues.apache.org/jira/browse/SPARK-23899
[2]
https://docs.google.com/document/d/1TTP-GCC8wSsibJaSUyFZ_5NBAHYEB1FVmPpP7RgDGBA/edit#



Winifred-wenhui Tang