K filters behavior? about heroic HOT 9 CLOSED

Check at the bottom of http://spotify.github.io/heroic/#!/docs/aggregations

Filtering aggregations return an entire time series, if it matches the given criteria. That is, if it has any data point witch matches it.

from heroic.

@udoprog Would you consider adding another constructor property to the k filters such as "on", which takes either "series" or "point", thereby allowing to either filter out entire series matching the criteria or specific points within a series?

from heroic.

@juruen what do you say?

from heroic.

The use case I have right now is needing to sum only the positive deltas of a time series. So basically was hoping to use a chain, e.g. delta | abovek(0, "point") | sum(1h).

from heroic.

My only consideration to add this feature would be naming. Should we just have different aggregation group names that indicate that they filter time-series or points instead of using an argument?

from heroic.

@juruen Yeah, that might be better. E.g. {"type": "pointsabovek", "k": 100} vs. {"type": "seriesabovek", "k": 100}.

I tried to take a stab at this using an "on" param, but couldn't get Optionals to work with the JsonProperty. Then I made it a required argument, but couldn't get the tests to pass due to an invalid argument type in LombokDataTest. If this seems like something you could implement rather quickly, would you please consider doing so?

from heroic.

@juruen @udoprog we are going to hire a contractor to work on this. Do you guys have any preference on my question above? In effect, should it be its own separate aggregation or an optional argument on the existing K-filters?

from heroic.

We currently don't have a good infrastructure to do complex decoding of the arguments in a consistent way. So my preference would currently be to just call them pointsabove, and pointsbelow. The k suffix doesn't add much.

from heroic.

We've devised a solution for this. Pull request incoming...

from heroic.