Open lennartkoopmann opened 9 years ago
This is not that easy to achieve at the moment, or at least it would require some time to investigate and play with it. There is mainly one problem:
null
, undefined
, NaN
or Infinity
, which are some of the values that we may get when calculating a histogram and no data is available. As far as I know, all the libraries we use have that problem.The easiest solution would be to leave out all those values and have a fragmented line, but there are also problems with that:
Thanks @edmundoa :) What is your proposed solution for now? Maybe it already gets better when we unify search result and dashboard charts because it is less confusing then.
I think the main problem is exactly what you mentioned, @lennartkoopmann: at the moment we have three different libraries and I don't think it is possible to fix this issue on all of them. We need to allocate time to leave only one graphing library for dashboards and search results (or two, if we really need certain feature).
Leaving aside the complication of libraries, the right solution would be to add discontinuities to the graphs when values are null
or not present. We should investigate if this is possible and go with this solution.
Another solution would be to still represent null
values as zero in graphs, but use the real values in the hover. This has the advantage of being feasible with most graphing libraries, but it's also not really accurate.
Well, null
represented as 0 is very, very confusing. In my case, I've ended up changing scale etc. This is not acceptable.
Imo, @lennartkoopmann proposal of this being configurable is quite interesting. As for myself it could be great if there would be an option for graylog to replace null with previous not-null sample or so.
Dashboard chart widgets show missing/NULL values as a 0 on the scale. This should be configurable because sometimes you expect missing data but this does not mean that the value was 0.