Add ignore_above option for ICUCollationKeywordFieldMapper

[clement-tourriere]

ICUCollationKeywordFieldMapper is not derived from KeywordFieldMapper, but from FieldMapper, so it doesn't have the ignore_above configuration.

The use case is to be able to add a sort field for specific language, ignoring long values.
For the moment the only solution is to perform this check in the client and separate the sort field from the keyword field.

I can provide a PR for this feature.

[elasticmachine]

Pinging @elastic/es-search

[jtibshirani]

@clement-tourriere to check my understanding of your use case: would you expect that ignore_above be applied to the input value, before collation is run?

[clement-tourriere]

In fact my use case is to be able to configure a max length on text field indexed with a sort "sub field" https://www.elastic.co/guide/en/elasticsearch/reference/current/multi-fields.html. The use case is to ignore long field (considering that they are not sortable) and in the worst case to protect against lucene byte-ref limit https://www.elastic.co/guide/en/elasticsearch/reference/current/ignore-above.html.

It might be more logical to apply the ignore_above check after collation instead of the input value.

[jtibshirani]

Thanks for the additional details. I could also see an argument for applying ignore_above before collation -- this matches the behavior for keyword fields, where we check ignore_above against the input before any normalizer is applied. I've asked other team members for their thoughts on this issue as well.

[jpountz]

+1 to support ignore_above like keyword and make it behave similarly, ie. check the length of the original string rather than the collation. @clement-tourriere you suggested the opposite but I suspect you would be fine with this behavior, wouldn't you?

[clement-tourriere]

Yes, it would be absolutely fine for our use case to check the length before collation ;)

[jtibshirani]

Great, now that we are clear on that I will take a look at your PR.