API should support filling in zero values for group by intervals with no data

[pauldix]

Heard this request and I don't think it's uncommon. Maybe something like this:

select count(val) from some_series group by time(5m) fill(0)

select count(state) from some_series group by time(5m), state fill(0)

The second query should fill in zero values for every unique state it has seen so far.

[FGRibreau]

Nice work! I can confirm you that I was always doing this kind of thing client-side, that's a really helpful feature!

[consense]

This is indeed a nice feature, but currently it only fills zero values between min(time) and max(time) even if the query was looking at a larger time interval.

i.e.:
datapoints at July 1st 2012 12:00 and August 1st 2012 12:00
Query:
SELECT sum(someintcolumn) FROM series WHERE time > '2012-01-01' AND time < '2012-12-31' GROUP BY time(1d) fill(0);

What happens is that the query returns 32 results (days between July 1st and AUgust 1st) instead of 365 as I would expect given the WHERE clause.

[jvshahid]

We can't make this the default behavior. The default start time of the query is the lowest time that could be represented by an int64. If the user issued select count(value) from foo group by time(10m) fill(0) it will return an insane number of points. What you're saying doesn't sound crazy, but we need to tweak it so we don't accidentally generate an insance number of points.

[consense]

Thanks for your detailed explanation John - makes sense.

Filling non-existing buckets with zeros on the client side works fine as well - was just surprised when I tested the fill(0) function.

[pauldix]

This is a thing we should support. See #426 to track progress.