Thanks, Matthias, for the feedback! > Not sure if "bound" is the right term? My understanding from your comment is that just the name "*bound*" is not the right one, while having this query type makes sense. Right? If *bound* is a confusing word, we can simply change it to a better one. I do not have a better idea at the moment. We can change it to "*whatever query type* with untilTimestamp"? and change the corresponding method name from " withWhateverWithTimestampBound" to "withWhateverWithUntilTimestamp"?
> Should we also add `withLowerTimeBound` and `withUpperTimeBound` (similar to what `RangeQuery` has)? In none of the range queries, we have a lower bound for time. If you look at them, they have either both time bounds, no bounds (all-versions), or just upper bound (UntilTimestamp as an input parameter). Do you think we should add those types of queries as well? We can add them for both key and range queries. I agree with you. > Btw: I think we should not pass `long` for timestamps, but `Instance` types. Good point. Making the user convert the timestamp to a long does not make sense. > For time-range queries, do we iterate over the values in timestamp ascending order? That was my assumption, at least. > If yes, the interface should specify it? Mmmm. You mean having that in the method input parameters? > Also, would it make sense to add reverse order I think yes. > Also, for time-range query, what are the exact bound for stuff we include? I 'm not sure if I get what you mean. Your point is that we have the following options and should choose one?! 1. *(*asOfTimestamp, untilTimestamp*)* 2. *(*asOfTimestamp, untilTimestamp*]* 3. *[*asOfTimestamp, untilTimestamp*)* 4. *[*asOfTimestamp, untilTimestamp*]* > For key-range / time-range queries: do we return the result in `<k,ts>` order or `<ts,k>` order? Also, what about reverse iterators? I suggest having both. The user can specify that. About reverse iterators, again, my idea is that it makes sense to have them. > About ` ValueIterator` ......... Thanks. I got that and corrected the KIP. On Wed, Jul 26, 2023 at 11:40 PM Matthias J. Sax <mj...@apache.org> wrote: > Thanks for the KIP Alieh. Glad to see that we can add IQ to the new > versioned stores! > > > > Couple of questions: > > > single-key lookup with timestamp (upper) bound > > Not sure if "bound" is the right term? In the end, it's a point lookup > for a key plus timestamps, so it's an as-of timestamp (not a bound)? Of > course, the returned record would most likely have a different (smaller) > timestamp, but that's expected but does not make the passed in timestamp > a "bound" IMHO? > > > single-key query with timestamp range > > single-key all versions query > > Should we also add `withLowerTimeBound` and `withUpperTimeBound` > (similar to what `RangeQuery` has)? > > Btw: I think we should not pass `long` for timestamps, but `Instance` > types. > > For time-range queries, do we iterate over the values in timestamp > ascending order? If yes, the interface should specify it? Also, would it > make sense to add reverse order (also ok to exclude and only do if there > is demand in a follow up KIP; if not, please add to "Rejected > alternatives" section). > > Also, for time-range query, what are the exact bound for stuff we > include? In the end, a value was a "valid range" (conceptually), so do > we include a record if it's valid range overlaps the search time-range, > or must it be fully included? Or would we only say, that the `validFrom` > timestamp that is stored must be in the search range (what implies that > the lower end would be a non-overlapping but "fully included" bound, > while the upper end would be a overlapping bound). > > For key-range / time-range queries: do we return the result in `<k,ts>` > order or `<ts,k>` order? Also, what about reverse iterators? > > About ` ValueIterator` -- think the JavaDocs have c&p error in it for > `peekNextRecord` (also, should it be called `peekNextValue`? (Also some > other JavaDocs seem to be incomplete and not describe all parameters?) > > > Thanks. > > > > -Matthias > > > > On 7/26/23 7:24 AM, Alieh Saeedi wrote: > > Hi all, > > > > I would like to propose a KIP to support IQv2 for versioned state stores. > > > > > https://cwiki.apache.org/confluence/display/KAFKA/KIP-960%3A+Support+interactive+queries+%28IQv2%29+for+versioned+state+stores > > > > Looking forward to your feedback! > > > > Cheers, > > Alieh > > >
