timeseries query doesnt match groupby

[pdeva]

reference discussion:
https://groups.google.com/forum/#!topic/druid-development/BSpDmAq-7Jk

i dont have the exact query, but i remember doing just a doubleSum.
there was only 1 row with 1 metric column in druid. minute level granularity was asked for.

thing to notice:
the row is stored at timestamp between 2 minute boundaries (04:33:25).
the query on the other hand, asks for minute level granularity, so values at 04:33 and 04:34 are returned by timeseries (though not so by groupby).

[xvrl]

What version of Druid were you using?

On Aug 26, 2014, at 3:06, pdeva notifications@github.com wrote:

reference discussion:
https://groups.google.com/forum/#!topic/druid-development/BSpDmAq-7Jk

i dont have the exact query, but i remember doing just a doubleSum.
there was only 1 row with 1 metric column in druid.

—
Reply to this email directly or view it on GitHub.

[pdeva]

0.6.145

[gianm]

PR #705 should address this for points at the end of the data interval, like your case, but interior points will still be zero-filled.

By "zero-filled" I mean that if you have data in minute 0 and minute 2, but not minute 1, and you do a minutely granularity query, then minute 1 will be filled in with a zero even though there are no data points for that minute. By "interior points" I mean time buckets that are contained within minTime/maxTime, even if they don't have any actual data in them.

I think that groupBy queries do not do any zero-filling at all, but timeseries queries do for interior data points. I guess whether they should or not is something of a philosophical question.

[fjy]

I believe this has been fixed. Please reopen if it is not the case.

[gianm]

I think timeseries still zero fills and groupBy still does not.

[xvrl]

Timeseries should not be zero-filling anymore. We've fixed issues around that a while back.

[gianm]

Ok, it does zero-fill by default, but I forgot about the skipEmptyBuckets thing that turns it off. I think I can be forgiven for that since it's not documented :)

I'll write some docs for it, but also wonder if it would make sense to change the default behavior in the future. I think zero-filling would be better off by default.

[gianm]

/cc #1698

[vogievetsky]

I think that with DSQL this ticket can be closed, I doubt you would ever change skipEmptyBuckets defaults. @gianm thoughts?