TermQueryBuilder and WildcardQueryBuilder should rewrite queries on `_index` so that shards get skipped

[jpountz]

term and wildcard queries already create either a MatchAllDocsQuery or a MatchNoDocsQuery at the shard level depending on the index name and its aliases. See IndexFieldType#termQuery and IndexFieldType#wildcardQuery for details.

It would be nice to go further and handle these in TermQueryBuilder#rewrite and WildcardQueryBuilder#rewrite. The benefit is that if someone queries index foo with a filter on { "term": { "_index": "bar" } }, then the top-level query will rewrite to a MatchNoDocsQuery, which is exactly the information that the can_match phase leverages in order to skip shards.

It would be a convenient and efficient way for Kibana to filter on a subset of data streams that are configured in the index patterns.

[elasticmachine]

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

[jpountz]

Handling TermsQueryBuilder (note the s) would be helpful too.

[markharwood]

@jpountz I expect the required rewrite logic would be something like this:

    if (fieldName.equals("_index") && context.indexMatches(value.toString()) == false) {
        return Queries.newMatchNoDocsQuery(
                "The index [" + context.getFullyQualifiedIndex().getName() + "] doesn't match the provided value [" + value + "].");
    }

But I added this example is in the AbstractQueryBuilder.doToQuery method override where "context" arg is a QueryShardContext with that helpful index-name matching method.

In your suggested placement of AbstractQueryBuilder.doRewrite the context arg passed is a less useful class - QueryRewriteContext without the index-name checking logic.

Would it matter if the MatchNoDocs rewrite happens in the doToQuery method?

[jpountz]

Indeed it would probably look something like that. You can call queryRewriteContext.convertToShardContext(); to get a shard context - with some caveats. See for instance how RangeQueryBuilder#doToRewrite works.

Would it matter if the MatchNoDocs rewrite happens in the doToQuery method?

Unfortunately yes. We already do this in doToQuery today actually via IndexFieldType#termQuery. However the heuristic that the can_match phase uses to know whether a shard might have any documents is by looking at whether the query rewrites to a MatchNoneQueryBuilder on that shard.