[Lens] Missing values - Fitting functions should support start/end values for fitting

[wylieconlon]

You can see this feature in the charts examples, but Lens does not use the endValue setting. As a result, we have charts that have a sudden start value even if the underlying data is intended to be continuous:

cc @markov00 I'm not sure that the fitting functions support a startValue?

[elasticmachine]

Pinging @elastic/kibana-app (Team:KibanaApp)

[markov00]

@wylieconlon actually the endValue is used for both edge of the chart, using the nearest it will draw a horizontal line at the height of each edge.
If this is not what you are looking for, can you specify better your use case? why are you looking for a startValue but you are not interested into an endValue?

I'm seeing also a weird artifact on the dashed line
have you seen this in other situations?

[wylieconlon]

@markov00 I think it's fine if both start and end values are the same, but I didn't realize that it would depend on the fitting function. Which fitting functions use endValues?

[markov00]

• carry uses this value if there is a missing point at the beginning
• lookahead uses this value if there is a missing point at the end
• average and linear use the value for beginning and end missing points

[flash1293]

@wylieconlon can we close this issue? It seems like there is nothing to do right now.

[wylieconlon]

@flash1293 Since all three of the fitting functions that have this behavior are supported in Lens, I think there is a change that should be made. I tested this out locally by setting endValue: 0 and using the linear fitting function. Here's how it looks:

Before:

After:

[flash1293]

@wylieconlon The first behavior seems preferable to me as it's using less assumptions about the data. It's also the behavior of vislib as far as I can tell. It seems like I'm missing the obvious improvement here, can you elaborate?

[wylieconlon]

The main improvement is that this draws a vertical line, instead of starting the points in the middle of the screen. This would be useful for metrics like uptime/downtime, where the transition between the "no data" and "has data" state is important to visualize. But like all of the fitting functions I think this is use-case dependent, and I'm suggesting that we provide users the choice instead of removing it.

[flash1293]

Makes sense, thanks for clarifying.