npawar commented on issue #3513: DATETIMECONVERT udf does not work for customized timezone and bucket size > 1 day URL: https://github.com/apache/incubator-pinot/issues/3513#issuecomment-440402832 If we were to support WEEKS in output granularity, where should the bucketing align from? In case of outputFormat in EPOCH, the bucketing can just be done in epoch weeks. But if output is in SDF, we have many options - If we start bucketing from epoch start, January 1st 1970 was a Thursday, so all buckets would be starting Thursday. - If we want a calendar alignment, should we start Monday/Sunday? - If we keep current behavior, days 1-7 would align with bucket yyyyMM01, days 8-14 with yyyyMM08 and so on. The number of week buckets in a year will not equal to 52 in this case, as we will have a half bucket created for every month's last few days. I prefer option 1, seems the most predictable and clean approach. We can also make all bucketing follow this. No resetting buckets at boundaries (DAYS should not reset at month, HOURS should not reset on days). But this would mean, you will not get a calendar alignment, and would have to offset your date if you need it aligned to the calendar. @kishoreg @mayankshriv please provide your suggestions if possible.
---------------------------------------------------------------- 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]
