"query_string" dosen't analyze wildcard queries

Elastic Stack Elasticsearch
1

Hi,

after upgrade from ES 2.4.4 to ES 5.6.1 the following query dosen't get analyzed by our analyzer

{
  "analyzer": {
    "default": {
      "type": "custom",
      "tokenizer": "fesadTokenizer",
      "char_filter": [
        "german_special",
        "strip_multiple_chars"
      ],
      "filter": [
        "asciifolding",
        "lowercase"
      ]
    }
  }

  "query": {
    "bool": {
      "must": [
        {
          "query_string": {
            "query": "*chifffahrt",
            "default_field": "_all",
            "allow_leading_wildcard": true,
            "analyze_wildcard": true
          }
        }
      ]
    }
  }
}

I think the reason is https://github.com/elastic/elasticsearch/commit/14d2b122a1b91ba14fc13fbb103e553cfddd5778#diff-3d1f1c3586dc5fe67e7615030feaecec

After this change the getPossiblyAnalyzedWildcardQuery(indexedNameField, termStr) was removed and the org.apache.lucene.analysis.Analyzer#normalize(java.lang.String, java.lang.String) is used instead.

The defined char_filter's and token_filter's dosn't get called which is a problem for us cause we dosn't receive any hits.

Is this behaviour as expected in ES 5.6.1 or is it a bug or configuration error?

2

At least the lowercase and asciifolding filters should be applied. I can't tell about the char filters and you tokenizer since I don't know how they are implemented.

Can you share the output of the following request with both versions?

GET your_index_name/_validate/query?rewrite=true
{
  "query" : {
    "query_string": {
      "query": "*chifffahrt",
      "default_field": "_all",
      "allow_leading_wildcard": true,
      "analyze_wildcard": true
    }
  }
}
3

The solution is that our own CharFilterFactory which extends org.elasticsearch.index.analysis.AbstractCharFilterFactory also must implement the interface org.elasticsearch.index.analysis.MultiTermAwareComponent which was not the case so far.

Cause org.apache.lucene.analysis.Analyzer#normalize(java.lang.String, java.lang.String) calls org.elasticsearch.index.analysis.CustomAnalyzer#initReaderForNormalization where only CharFilterFactory's which are instanceof MultiTermAwareComponent will be used

        @Override
        protected Reader initReaderForNormalization(String fieldName, Reader reader) {
          for (CharFilterFactory charFilter : charFilters) {
            if (charFilter instanceof MultiTermAwareComponent) {
              charFilter = (CharFilterFactory) ((MultiTermAwareComponent) charFilter).getMultiTermComponent();
              reader = charFilter.create(reader);
            }
          }
          return reader;
        }

The same applies to the TokenFilterFactory's

        @Override
        protected TokenStream normalize(String fieldName, TokenStream in) {
          TokenStream result = in;
          for (TokenFilterFactory filter : tokenFilters) {
            if (filter instanceof MultiTermAwareComponent) {
              filter = (TokenFilterFactory) ((MultiTermAwareComponent) filter).getMultiTermComponent();
              result = filter.create(result);
            }
          }
          return result;
        }
4

This is correct. Do I get it right that it fixed your issue?

5

Yes this fixed my issue.

Thanks

6

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.