Highlighting offset exception on 1 shard (of 5). Bad chars?

Elastic Stack Elasticsearch
1

I have a query on parents and children that should perform the "highlighting" on some fields.

When I run the query (250 docs) there is no problem as long as the size is lower than a certain number (in this case 84).
When setting bigger than 84, or just setting "from":84 and "size":1, then I get an error on a shard that prevents the highlighting to be performed correctly.

======================================
{
"took": 45,
"timed_out": false,
"_shards": {
"total": 5,
"successful": 4,
"failed": 1,
"failures": [
{
"index": "index_parent_child",
"shard": 0,
"status": 500,
"reason": "FetchPhaseExecutionException[[index_parent_child][0]: query[filtered((((content:arbeit content:arbeit) | (description:arbeit description:arbeit) | (item_nr:arbeit item_nr:arbeit) | (code:arbeit code:arbeit)) child_filter[file/item](filtered(content:arbeit content:arbeit)->cache(_type:file)) parent_filter[item](filtered(((description:arbeit description:arbeit) | (item_nr:arbeit item_nr:arbeit) | (code:arbeit code:arbeit)))->cache(_type:item)))~1)->cache(_type:file _type:item)],from[84],size[1]: Fetch Failed [Failed to highlight field [description]]]; nested: InvalidTokenOffsetsException[Token l-ops exceeds length of provided text sized 93]; "
}
]
},
"hits": {
"total": 367,
"max_score": 2.6496673,
"hits": []
}
}

I think that the problem is cause the content into the fields is in German and that somewhere the multi bytes characters are not managed well, ending into the Exception.

I just tried to reduce the query just to the item, removing the highlighting (so that it could execute with success and show me the interested document) and I got that its "description" field (where it fails) has this content:

======================================
"Medizintechnik / Arbeitstisch- und Schrankanlagen /\r\nSanierung und Erneuerung des Zentral-OPs"

In this case it seems there is no multibyte character (umlaut oder scharfen-S). So I suspect the fault is of those "carriage return" and "new line".

Do you know:

    1. if this is a bug in Lucene?
    1. how could I solve the problem?

Thank you all