Possible problem with Fast Vector Highlighter + Multi-Fields between 7.10.2 and 7.16.3

[Mrodent]

Elasticsearch Version

7.16.3

Installed Plugins

No response

Java Version

bundled

OS Version

Linux (Mint 20), W10

Problem Description

The same mappings and query DSL dict which work fine to produce multi-word, multi-colour highlighting in v 7.10.2 don't seem to work with v 7.16.3: on both Linux and W10 I have had to revert to 7.10.2. I'd love it if an expert could check on what's going on: maybe there has been a substantive documented change between the two ES versions.

I had a look at the latest version (8.0) for this here, but nothing jumped out at me. However, in the section on fvh and multi-fields I noticed that nothing was said about using pre-tags and post-tags. The point about my example is that different terms are highlighted with different colours (in fact backgrounds).

I have posted this in the ES forum. No replies after a few days, hence my opening of an issue here.

Steps to Reproduce

Given this mapping:

       mappings = \
        {
          "mappings": {
            "properties": {
              "esdoc_text": {
                "type": "text",
                "term_vector": "with_positions_offsets",
                  "fields": {
                    "stemmed": {
                      "type": "text",
                      "analyzer": "english",
                      "term_vector": "with_positions_offsets",
                    }
                  }
                }
              }
            }
          }

and this query DSL dict:

       data = \
        {
          'query': {
            'simple_query_string': {
              'query': self.search_string,
              'fields': [
                self.text_field
                
              ]
            }
          },
          'highlight': {
            'fields': {
              self.text_field: {
               'type': 'fvh',
               'pre_tags': [
                    '<span style="background-color: yellow">',
                    '<span style="background-color: skyblue">', 
                    '<span style="background-color: lightgreen">', 
                    '<span style="background-color: plum">', 
                    '<span style="background-color: lightcoral">', 
                    '<span style="background-color: silver">',
                ],
               'post_tags': [
                    '</span>', '</span>', '</span>', 
                    '</span>', '</span>', '</span>', 
                ]
              }
            },
            'number_of_fragments': 0
          }
        }

... I get beautiful highlighted results using 7.10.2. But using 7.16.3, the hits are returned, but there is no "highlight" key at all in the returned response.json().

NB in the above, clearly and as one might surmise from the mapping, self.text_field is set to either "esdoc_text" (for standard analyser results) or "esdoc_text.stemmed" (for english stemmer analyser results).

Logs (if relevant)

No response

[elasticmachine]

Pinging @elastic/es-search (Team:Search)

[jtibshirani]

@Mrodent thank you for reporting this. We didn't make any intentional changes to the fvh highlighter behavior between 7.10 and 7.16. I was able to reproduce using the mappings you sent and confirmed it's a regression.

I think this is caused by #74348, where we no longer extract sub-paths in SourceLookup#extractRawValues. The fvh highlighter still uses this code path, which means it misses highlights on multi-fields. We will look into fixing this in an upcoming Elasticsearch version.

[jtibshirani]

@romseygeek would you be up for looking at this since you're very familiar with #74348 and the highlighting code?

[Mrodent]

Thanks for the confirm. Will keep an eye-out for progress.

Incidentally, as as side-note for @romseygeek:

If I change from "simple_query_string" to "query_string" (and this applies even in 7.10.2), I find that the colouring (i.e. the order in which the pre-tags are applied) seems to become haphazard. I.e. instead of term 1: yellow, term 2: skyblue, term 3: lightgreen, I might find the results highlighted: 1: lightcoral, 2: silver, 3: lightcoral (again).

This is clearly not my intention: I want the highlighting colours (i.e. tags) to apply in the order given.

Knowing less than nothing about what's going on under the hood here, I have no knowledge whether this can be addressed at the same time. But it'd be nice.