Agreed this would absolutely be a win. Dont see need for a flag either. On Tue, Oct 1, 2024 at 1:31 PM Caleb Rackliffe <calebrackli...@gmail.com> wrote:
> Alrighty, with what looks like a fair amount of support, I'll declare > CASSANDRA-19968 <https://issues.apache.org/jira/browse/CASSANDRA-19968> ready > for some preliminary review. > > On Tue, Oct 1, 2024 at 2:41 PM Caleb Rackliffe <calebrackli...@gmail.com> > wrote: > >> We did add CASSANDRA-18940 >> <https://issues.apache.org/jira/browse/CASSANDRA-18940> to make sure >> local SAI post-filtering reads got picked up somewhere, but you're right >> that StorageProxy#readRegular() would start recording some index queries >> in the normal read metrics. >> >> On Tue, Oct 1, 2024 at 2:11 PM Jeremiah Jordan <jeremiah.jor...@gmail.com> >> wrote: >> >>> Did we add new metrics for index queries? The only issue I see is that >>> this change will mix index queries into the regular read metrics, where >>> before they were in the range metrics, so maybe some changes to metrics >>> should go with it. But I think this is a good change over all. >>> >>> On Oct 1, 2024 at 1:51:10 PM, Jon Haddad <j...@rustyrazorblade.com> >>> wrote: >>> >>>> This seems like it's strictly a win. Doesn't sound to me like a flag >>>> is needed. >>>> >>>> On Tue, Oct 1, 2024 at 2:44 PM Caleb Rackliffe < >>>> calebrackli...@gmail.com> wrote: >>>> >>>>> > (Higher rate of mismatches requiring a second full read? Why would >>>>> 2i be more likely?) >>>>> >>>>> Right, I don't see any reason they should be more likely to actuate >>>>> read-repair than slice queries are today... >>>>> >>>>> Didn't mention this above, but I'd obviously be open to having a >>>>> system property that switches this behavior. >>>>> >>>>> On Tue, Oct 1, 2024 at 12:43 PM Jeff Jirsa <jji...@gmail.com> wrote: >>>>> >>>>>> >>>>>> >>>>>> > On Oct 1, 2024, at 10:28 AM, Caleb Rackliffe < >>>>>> calebrackli...@gmail.com> wrote: >>>>>> > >>>>>> > Hello fellow secondary index enjoyers! >>>>>> > >>>>>> > If you're familiar with index queries, you probably know that they >>>>>> are treated as range reads no matter what. This is true even if the user >>>>>> query restricts results to a single partition. This means that they >>>>>> bypass >>>>>> the digest read process that normal single-partition reads do. >>>>>> >>>>>> TIL. >>>>>> >>>>>> > >>>>>> > While I don't think this is something that we need to consider for >>>>>> 5.0, I would be very interested in the next major release being able to >>>>>> use >>>>>> proper single-partition reads for partition-restricted index queries, >>>>>> allowing them to take advantage of digest reads. (If single partition >>>>>> slice >>>>>> queries do it, why not index queries?) >>>>>> >>>>>> This seems like an obvious yes, so reverse the question - is there >>>>>> any reason why we WOULDNT want to do this? >>>>>> >>>>>> (Higher rate of mismatches requiring a second full read? Why would 2i >>>>>> be more likely?) >>>>>> >>>>>>
