Inconsistency in multi_match query

[bittusarkar]

There is a discrepancy in the behaviour of multi_match query when it comes to resolving field names containing wildcard characters.

The following query will return zero results if the field subject does not exists.

{
  "multi_match" : {
    "query":    "this is a test", 
    "fields": [ "subject" ] 
  }
}

But the following query will fail with an exception saying "No fields specified for multi_match query" if there is no field whose name starts with "ms" in the index being queried. Ideally it should return zero results too instead of failing right?

{
  "multi_match" : {
    "query":    "this is a test", 
    "fields": [ "ms*" ] 
  }
}

I have a use case where fields are dynamically added to the mapping and users can provide the field name to match search text on. Fields added to the mapping also have a metadata string appended to their names which the user may not be aware about. Hence I need to use wildcards in the field name in a multi_match query and there are chances that no field name matches too.

[clintongormley]

I agree. Related to #12016

[michel-kraemer]

I bumped into this too. Is there a workaround that I can apply until this issue has been solved?

[0xbok]

Is this fixed now?
I just pulled the code from master and the second query is returning zero results (I ensured that there is no field starting with 'ms').

[qwerty4030]

@clintongormley Confirmed this is not reproducible on latest master with the following:

POST test-index/test-type/test-doc1
{
  "test-key":"test-value"
}

POST test-index/_search
{
  "query": {
    "multi_match": {
      "query": "test-query",
      "fields": [
        "doesnt-match*"
      ]
    }
  }
}
results in 0 hits

However the same query fails on 2.3.2. Which release will the changes from #12016 land in?

[clintongormley]

This is fixed in master thanks to a massive refactoring of how searches are parsed - this obviously can't be backported. If somebody wants to contribute a quick fix for 2.4, then please submit a PR.

[andrewla]

One workaround this that I've found is to add a dummy field to the fields specification, like "fields": ["dummydummydummy", "ms*"], which will trigger the behavior on missing fields without causing the wildcard misfiring. Not sure if this makes the query significantly more expensive.

[michel-kraemer]

This issue seems to be fixed in 5.0.0. As far as I am concerned it can be closed.

[clintongormley]

Fixed in 5.0, closing