Rewrite more queries to match_none when they target an unmapped field #132987
Conversation
javanna
force-pushed
the
enhancement/unmapped_fields_rewrite
branch
from
dfd6605 to
51b6f9f
Compare
javanna
added
:Search Relevance/Search
elasticsearchmachine
added
the
Team:Search Relevance
|
Pinging @elastic/es-search-relevance (Team:Search Relevance) |
|
Hi @javanna, I've created a changelog YAML for you. |
|
|
||
| QueryBuilder queryBuilderForProcessing = queryBuilder.rewrite(new SearchExecutionContext(executionContext)); | ||
| SearchExecutionContext copy = new SearchExecutionContext(executionContext); | ||
| configureContext(copy, isMapUnmappedFieldAsText()); |
There was a problem hiding this comment.
configuring the context was missing, hence the configuration values were not available to the underlying query rewrite. This fixes it.
| public void testTooLongRegexInRegexpQuery() throws Exception { | ||
| createIndex("idx"); | ||
| indexRandom(true, prepareIndex("idx").setSource("{}", XContentType.JSON)); | ||
| indexRandom(true, prepareIndex("idx").setSource("num", "value")); |
There was a problem hiding this comment.
no more failure when targeting an unmapped field, the test needs fixing to target a mapped field.
There was a problem hiding this comment.
What makes this confusing is that we have a default index.query.parse.allow_unmapped_fields which defaults to true, and then individual queries have ignore_unmapped (are all of them deprecated?).
With this change, I think ignore_unmapped: false is broken for every where that accepts an ignore_unmapped.
|
Hi @javanna, I've updated the changelog YAML for you. |
| protected QueryBuilder doIndexMetadataRewrite(QueryRewriteContext context) throws IOException { | ||
| MappedFieldType fieldType = context.getFieldType(fieldName); | ||
| if (fieldType == null) { | ||
| return new MatchNoneQueryBuilder("The \"" + getName() + "\" query was rewritten to a \"match_none\" query."); | ||
| } | ||
| return this; | ||
| } |
There was a problem hiding this comment.
ignoreUnmapped defined in the query should take precedence right? It has a default value of false for this query type.
There are a bunch of these (HasChildQueryBuilder, HasParentQueryBuilder, NestedQueryBuilder, GeoGridQueryBuilder,ParentIdQueryBuilder, and maybe more..but these are the ones I found quickly).
4 participants
We have a specific metadata rewrite that allows us to optionally rewrite a query builder based on mappings. It is used in some places to handle unmapped fields, but not consistently. This PR adds the rewrite to match_none to more queries when they target an unmapped field.
Closes #97129
Closes #131037