Use existing DocumentMapper when creating new MapperService

[nielsbauman]

When we create a new MapperService from an existing index for validation purposes, we often merge the mappings of the existing index into the temporary mapper service right after. This requires parsing the mappings and "merging" them (there's not much merging to do with an empty service, but we still need to traverse the mappings object). This can take up considerable time for larger mappings.

By passing the existing DocumentMapper instance of the index (if available) to the new MapperService, the aforementioned mapping merge becomes a no-op and saves all that redundant processing.

[elasticsearchmachine]

Hi @nielsbauman, I've created a changelog YAML for you.

[nielsbauman]

These changes are related to #138367, but can be merged/viewed separately. I performed similar benchmarks for this PR as well. See #138367 (comment) for a brief explanation of what experiments I performed.

Here is the same flamegraph on main (the first flamegraph in the comment I linked):

flamegraph-mappings-main-noop-pr.html

And here is the flamegraph on this PR:

flamegraph-mappings-only-mapper-copy-noop.html

The rightmost merge/parse path completely disappeared, saving a lot of processing.

[elasticsearchmachine]

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

[spinscale]

Thinking how this could be tested, I played around a little but my IDE seems not to like me too much today. Would a test like this (in IndicesServiceTests make sense to you?

    public void testCheckIndicesServiceMapper() throws IOException {
        IndicesService indicesService = getIndicesService();

        IndexMetadata indexMetadata = new IndexMetadata.Builder("test").settings(
            indexSettings(1, 1).put("index.version.created", IndexVersions.V_8_10_0)
                .put(SETTING_INDEX_UUID, randomUUID())
        ).build();
        IndexService indexService = indicesService.createIndex(indexMetadata, List.of(), true);

        IndexMetadata newIndexMetadata = IndexMetadata.builder(indexMetadata)
            .numberOfReplicas(2)
            .mappingVersion(indexMetadata.getMappingVersion()+1)
            .putMapping("""
            {
              "_doc":{
                "properties": {
                  "@timestamp": {
                    "type": "date"
                  }
                }
              }
            }""")
            .build();
        indexService.updateMapping(indexMetadata, newIndexMetadata);

        DocumentMapper documentMapper = indicesService.createIndexMapperServiceForValidation(indexMetadata).documentMapper();
        DocumentMapper cachedDocumentMapper = indicesService.createIndexMapperServiceForValidation(indexMetadata).documentMapper();

        assertNotNull(documentMapper);
        assertSame(documentMapper, cachedDocumentMapper);
    }

[nielsbauman]

Thanks for the suggestion! I pushed b91c6a6 with some minor changes to your suggestion. Let me know if that works for you.

[spinscale]

left one minor nit

[nielsbauman]

Thanks a lot for the review, @spinscale! This will help a great deal in reducing the performance impact of mapping updates.