Disable analysis for some strings when mapping

[LosD]

It turns out that the ES analyzers breaks up for easier searching. That makes a whole lot of sense in many cases, like searching description for someone mentioning a certain problem or things like that.

However, for things like labels, it is a problem. In our case, we are trying to aggregate on specific labels (used for priority), but there is a space in one of them, "Prio:Not now". This means that the terms filter will split the result into two terms, "Prio:Not" and "Now".

I haven't really analyzed (heh ;)) further, but there are probably other fields that is better left alone.

ES provides two approaches to fixing it: Disable the analyzer completely for the property, or make it into a multi_field, which can both have the original field analyzed, and a sub-field which is not analyzed, usually named raw. So in the labels case that would be labels.name (analyzed) and labels.name.raw (not analyzed).

It might be a problem to actually do something about it, as we rely on the automatic mapping right now, and changing that is not a small change. Does anyone know how ES reacts to partially mapped properties when storing documents? If that is not a problem, we could set up the mapping for the special fields and let the rest be mapped automatically.

[mihneadb]

I avoided this because of the reason you mentioned - I left it to the automatic mapping. Let's try specifying the special fields and see what happens?

Indeed, providing explicit mapping for all the GH response types is much work.