Add a UDF to extract the window start time from the message key

[apurvam]

We should add a WINDOWSTARTTIME() UDF which returns the start time of the window if the underlying topic is windowed, or else return some sentinel value (like LONG_MAX) if it isn't.

Getting access to the window start time is generally useful, and we can't depend on ROWTIME for this piece of information. See #1497

[codestoned1]

Hey, I'm new to contributing. Can I try adding this?

[rodesai]

@zlex7 go for it!

[apurvam]

yes. thanks @zlex7 . I am assigning this to you.

[apurvam]

Actually, I can't assign it to you for some reason. In any case, feel free to take a stab at it. You can follow the example from here to add a new UDF: https://github.com/confluentinc/ksql/pull/1611/files

[miguno]

Thanks for taking a stab at this, @zlex7 !

[rmoff]

would be good to include WINDOWENDTIME for completeness :)

[blueedgenick]

Thinking aloud, maybe a udf is not the best solution for this - how about we instead provide an additional pseudo-column that's automatically created on the output table, say WINDOWTIME, just like we do for the message key and timestamp already. I wonder if this would feel more natural?

[mjsax]

I would recommend to return NULL if the UDF is applied to a non-windowed column. A sentinel value as LONG_MAX might not make much sense IMHO and could be miss-leading downstream.

[rmoff]

@blueedgenick IMO a UDF would be a useful stop-gap solution for something that has in effect caused a regression in functionality. I think the discussion around UDF vs system column did take place, and TBH so long as we get it built sooner than later either would be fine :)

[apurvam]

@big-andy-coates said he would take this on.

[big-andy-coates]

Yep, I'm working on this... it's more involved than it might look.