gianm commented on PR #17912: URL: https://github.com/apache/druid/pull/17912#issuecomment-2808129100
This change makes things less clear and, to my recollection, is historically incorrect. Deprecation of the `approxHistogram` aggregator was discussed in #6869 and implemented in #7195. There wasn't discussion of deprecating `fixedBucketsHistogram` (originally developed in #6638), and I do not believe there was any intent to deprecate that one. Furthermore, the rationale for deprecating `approxHistogram` doesn't apply to `fixedBucketsHistogram`. We deprecated `approxHistogram` because it's an adaptive-style algorithm, but generally just worse than the datasketches quantiles algorithm. It's generally been observed to be both slower and less accurate, so there's really no point in using it. On the other hand, `fixedBucketsHistogram` is not an adaptive-style algorithm. It is very simple, very fast, but getting good accuracy requires some knowledge of the input data distribution, and in some cases good accuracy requires a lot of space. But, there are legitimate scenarios where you would prefer a fixed-bins approach like `fixedBucketsHistogram` over even a good adaptive-style algorithm, so there is a colorable argument that it's good to offer both. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
