Exponentially decaying reservoir update #1135
Conversation
WeightedSnapshot should behave after long inactive reservoir as newly created reservoir.
white spaces
white spaces
white spaces
white spaces
|
Thanks for the pull request! I will review today. |
|
Thank you! Let me know if you need anything to be changed . |
| @@ -38,6 +38,12 @@ public WeightedSample(long value, double weight) { | |||
| * @param values an unordered set of values in the reservoir | |||
| */ | |||
| public WeightedSnapshot(Collection<WeightedSample> values) { | |||
| if (values.size()==1 && Double.compare(values.iterator().next().weight,0D)==0) { | |||
There was a problem hiding this comment.
I think it's not very obvious why we perform this check in the constructor of the snapshot. I understand that the idea to avoid the situation when we have only one element in the sample with a 0 weight (because it provides bogus data), but I believe it should be performed in ExponentiallyDecayingReservoir, where the logic of rescaling resides. What do you think, if we add a guard condition:
if (scalingFactor == 0) {
values.clear();
}in the ExponentiallyDecayingReservoir#rescale method? I think it could be a more clear solution.
There was a problem hiding this comment.
Hi thanks, yes I will look into that today. I agree here it isn't the best to solve this problem.
|
Should be done, can you check please? |
|
Thank you very much for the contribution! |
Exponentially Decaying Reservoir was giving incorrect values in the snapshot if the inactive period was too long. This could result in wrong metrics aggregation.