simple_query_string returns nothing for "*" search

[pierrre]

My simple_query_string returns nothing with "*" search.

If I use a query_string, it works (it returns hits).
The fields and analyzer are the same.

If I wrap my simple_query_string with a bool+must (or should) query, it works. (???)

Is it a bug?

Elasticsearch version: 1.5.1

[dakrone]

I believe this may be related to #9866, which was fixed in Lucene 5.1 (see: https://issues.apache.org/jira/browse/LUCENE-6298). I'll look and see whether it is the same issue.

[pierrre]

My Lucene version is 4.10.4.

What should happen if I search "*" with a simple_query_string?
Return all docs or nothing?
The query_string returns all docs.

[clintongormley]

This has been fixed. * will return all docs

[pierrre]

Thank you!

[pierrre]

Actually, this search:

{
  "fields": [],
  "query": {
    "simple_query_string": {
      "fields": [
        "shortDescription"
      ],
      "query": "*"
    }
  }
}

returns nothing with Elasticsearch 2.2

@clintongormley is the fix for this issue released?

[clintongormley]

Initially I thought this might be related to #13214 but no, a query string of "*" doesn't match any docs, with or without custom fields:

PUT t/t/1
{}

GET t/_search
{
  "query": {
    "simple_query_string": {
      "query": "*"
    }
  }
}

GET _validate/query?explain
{
  "query": {
    "simple_query_string": {
      "query": "*"
    }
  }
}

The explanation returned by the last request is simply ""

[cbuescher]

I looked into this a bit by adding an IT test. It looks like Lucenes SimpleQueryParser only recognizes a terms with a closing '*' as prefix query. A single wildcard without leading charactes is treated as a regular term in SimpleQueryParser#consumeToken(). This might even be the intended behaviour. For query_string the situation is different since it uses MapperQueryParser internally which has an allowLeadingWildcard option. Maybe @jimferenczi or @dakrone have an opinion on whether this can or should be fixed in SimpleQueryParser or is even intended.

[dakrone]

@cbuescher personally I think a SQS query for * should be internally rewritten to a MatchAllDocsQuery, which I think means this would go into Lucene. If people agree I can work on a patch for the next Lucene release

[clintongormley]

@dakrone not so sure - should it be match all or should it match all docs that have a value for the specified fields? See #13214 for a similar discussion

[dakrone]

@clintongormley ahh okay, in light of that discussion:

"*" always means match all documents
"field:*" means find all documents that contain a non-null value in the field (incl empty string)

Since simple_query_string can either use the default_field or have fields
specified, I'm thinking we should do:

• If query is "*" and no fields are specified, treat it as match all
  documents
• If query is "*" and fields are specified, query is essentially field1:* OR field2:* OR field3:* which matches documents that have non-null values in
  the field

Do you think that'd be a better solution?

[clintongormley]

@dakrone Given that the end user can't specify fields in the SQS like you can in QS (user:kimchy) I've come around to thinking that * should always mean match all docs.

[dakrone]

I opened https://issues.apache.org/jira/browse/LUCENE-7490 for this with a patch

[dakrone]

Looks like this will be in Lucene 6.3, so I'm marking this as "stalled" until 6.3 is released and incorporated.

[weberhofer]

I think this issue has been solved with ES 5.3

[cbuescher]

@weberhofer thanks a lot, I just checked with 5.1 (which is based in Lucene 6.3) and 5.2 and it seems to be fixed. I will close this.