gianm commented on issue #6869: [Proposal] Deprecating "approximate histogram" in favor of new sketches URL: https://github.com/apache/incubator-druid/issues/6869#issuecomment-454665859 This proposal looks good to me. approxHistogram had a good run. > Provide advice on how to choose between the quantiles sketch or the moments sketch Possibly a loaded question, but, do you have thoughts on what kind of advice we'd include here? > Is it possible/valid for the newer sketch aggs to be able to operate on an old "approximate histogram"? If so, this would make migration easier for users. My guess is no, since the algorithms are all different. > Change the APPROX_QUANTILE Druid SQL function to use whatever aggregator type is stored in a segment, and use a default option when used on a numeric column This will require a bit of an enhancement to the SQL metadata layer. Check out RowSignature, in particular this part: ```java case COMPLEX: // Loses information about exactly what kind of complex column this is. type = Calcites.createSqlTypeWithNullability(typeFactory, SqlTypeName.OTHER, true); break; ``` For `APPROX_QUANTILE` to be able to behave differently based on the type of sketch stored in the underlying column, we'll need to figure out some way to encode that information in the type. Or some way to make it accessible to the `SqlAggregator` conversion out-of-band. Encoding it in the type somehow is preferable, though, if possible, because avoiding the need for out-of-band info is always nice. It would also make it easier to expose it through the INFORMATION_SCHEMA.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
