Cumulative sum count does not display correctly when there are no values for some intervals

[rayafratkina]

Kibana version: 6.2
Elasticsearch version: 6.2
Server OS version: hosted
Browser version: Chrome
Browser OS version: Windows
Original install method (e.g. download page, yum, from source, etc.):

I am looking at a chart showing cumulative count of PRs in v6.3 by area over time. In the recent time, not all areas had any new PRs added and it looks like for the time slots when no PRs are added, the chart does not show anything instead of showing the prior cumulative total count.

Steps to reproduce:

1. Y-axis:
   Aggregation=Cumulative sum
   Metric = Custom metric; Aggregation = Count
2. Buckets:
   Split series
   Aggreagation = terms
   Field = area. keyword
3. X-axis:
   Sub-aggregation = Date histogram
   Field = timestamp
   Interval = custom 8h

[rayafratkina]

Looks like data is missing for the later time intervals

[thomasneirynck]

This seems more like an Elasticsearch issue.

One thing to take into account that it makes intuitive sense that the cumulative sum should just carry along for missing values. But that really only works for counts. For example, you couldn't expect the same behavior for cumulative sum of a more complex metric (say, average of bytes).

I believe that's one of the reasons why ES doesn't have built-in fitting functions, because it really depends on the domain on how you'd want to fit (e.g. carry last value, average of preceding values, ...).

This seems more like an ER, for either pushing fitting functions inside ES, or have provide an alternative for that on the client. In that sense, this is a similar issue as #13111.

[rayafratkina]

From the user's pov it looks like a bug - there is a clear expectation for the behavior of cumulative sum and it's not happening. I think same would hold for other cumulative metrics, although the behavior may be different for each.

[timroes]

Indeed it seems, that ES doesn't fit that values, in which case we won't visualize them. So there are two potential fixes for that. a) ES could introduce fitting for functions, which I think they won't (but you can try to open a ticket for that :D) or b) we properly allow fitting null values to other values.

To have this in a clean separate new feature request, I opened #17717 and would close with reference to that new ticket.