Thanks for quick reply. I appreciate it.
It sounds reasonable but this is not a case unfortunately. All fields and their types belong to pre-defined mapping json-object. According to nature of documents each document contains subset of these fields but each field of any document belongs to this mapping json-object. I use Node,js (elasticsearch module) to create mapping, indices and insert documents to Elasticsearch. I checked in Kibana that types of all fields correspond to types I set in mapping for each field but fields with a keyword type sometimes (every 1-3 days) may be changed and have extra field - field.keyword or vise versa field.keyword may disappear and original field became aggregatable. As a result index pattern in Kibana get yellow status and visualizations have lack of data.