Hi Alan,
thanks for giving us some context. It looks like FLIP-313 has not seen
any progress for 2 years. We can still revive FLIP-313 for the hint
discussion but start with config options. It makes sense that there is
consistency between AsycScalarFunction and AsyncTableFunction with
respect to configuration. Overall FLIP-498 is in a good shape and its
time that this feature will land in master regardless of the contributor.
So I'm +1 to continue with this FLIP, unless there are objections from
others?
Cheers,
Timo
On 13.01.25 18:35, Alan Sheinberg wrote:
I just wanted to add a little bit to my response as I considered the
previous FLIP.
I support FLIP-313 and the proposal of the original author. I wasn't trying
to take it over, but wanted to show renewed interest.
I'm also focused on a smaller MVP, which is why I omitted hint
support, which seemed to be a previous point of discussion. This is in
line with what I implemented for AsycScalarFunction and has direct parity
with the configurations exposed. I just wanted to highlight that
difference.
Thanks,
Alan
On Mon, Jan 13, 2025 at 8:55 AM Alan Sheinberg <asheinb...@confluent.io>
wrote:
Thanks for the responses Timo, Dawid.
However, why have you decided to resubmit the FLIP with a newer number
rather than following up on FLIP-313? The only difference I see between
the
two FLIPs is the options/hints design. Could you elaborate where the
differences come from?
I think you're right about the hints being the only real difference in
public interface -- I'm not proposing to support them just yet. Since I
was thinking about the implementation details, I wanted to fill some of
those details in more than they were in the previous FLIP.
Just for my own knowledge, what's the community's policy for addressing
topics covered by previous FLIPs? I figured since it had been 1.5 years
and I had dug into the problem myself, I would create a new one rather than
voice support for the old one.
Thanks,
Alan
On Mon, Jan 13, 2025 at 7:25 AM Dawid Wysakowicz <dwysakow...@apache.org>
wrote:
Generally what is suggested here makes sense to me.
However, why have you decided to resubmit the FLIP with a newer number
rather than following up on FLIP-313? The only difference I see between
the
two FLIPs is the options/hints design. Could you elaborate where the
differences come from?
Best,
Dawid
On Thu, 2 Jan 2025 at 20:48, Alan Sheinberg <asheinb...@confluent.io
.invalid>
wrote:
I'd like to start a discussion of FLIP-498: AsyncTableFunction for async
table function support [1]
This feature proposes exposing AsyncTableFunction as a proper user
defined
function. The type already exists for Lookup Joins, but isn't usable as
other UDFs. This FLIP would bring it up to parity with others.
The motivation is similar to other async cases, namely improving
performance while issuing long-latency calls to external systems. This
is
similar to AsyncScalarFunction, which exists in Flink.
I realize this was effectively proposed in the past in FLIP 313. [2]
This
re-proposes that general feature and gives an update with additional
details informed by work on AsyncScalarFunction.
Looking forward to your feedback and suggestions.
[1]
https://cwiki.apache.org/confluence/display/FLINK/FLIP-498%3A+AsyncTableFunction+for+async+table+function+support
[2]
https://cwiki.apache.org/confluence/display/FLINK/FLIP-313%3A+Add+support+of+User+Defined+AsyncTableFunction
Thanks,
Alan
