More details on exception when having a nested field with date_range and include_in_root set

[ibotello]

Elasticsearch Version

7.14.2

Installed Plugins

No response

Java Version

bundled

OS Version

ESS

Problem Description

We get an exception when we have multiple nested documents all trying to add a binary doc values field to the root doc.
This exception appears since Lucene only lets you add a single instance of a binary doc values field to a document.
Would be nice to have a more detailed exception message to actually understand what is happening.

Steps to Reproduce

Creating a sample index with the nested type, and having a date_range field:

PUT my-index-00001
{
  "mappings": {
    "properties": {
      "NestedField": {
        "type": "nested",
        "include_in_root": true,
        "properties": {
          "DateField": {
            "type": "date_range"
          }
        }
      }
    }
  }
}

Trying to ingest one document:

PUT my-index-00001/_doc/1
{
  "NestedField": [
    {
      "DateField": {
        "gte": "2015-10-31",
        "lte": "2015-10-31"
      }
    },
    {
      "DateField": {
        "gte": "2016-10-31",
        "lte": "2016-10-31"
      }
    }
  ]
}

We get the next exception:

{
  "error" : {
    "root_cause" : [
      {
        "type" : "illegal_argument_exception",
        "reason" : "DocValuesField \"NestedField.DateField\" appears more than once in this document (only one value is allowed per field)"
      }
    ],
    "type" : "illegal_argument_exception",
    "reason" : "DocValuesField \"NestedField.DateField\" appears more than once in this document (only one value is allowed per field)"
  },
  "status" : 400
}

Logs (if relevant)

No response

[josefschiefer27]

I am getting running into the same issue using other data types (in my case with wildcard - interestingly keyword works - see below)

For instance, if I create an index with

PUT my-index-000001
{
  "mappings": {
    "properties": {
      "user": {
        "type": "nested",
        "include_in_root": true,
        "properties": {
          "first": {
            "type": "keyword"
          },
          "last": {
            "type": "keyword"
          }
        }
      }
    }
  }
}

With this index the following ingestion will work:

PUT my-index-000001/_doc/1
{
  "user": [
      { "first" : "Bob", "last" : "Kesler" },
      { "first" : "Robert", "last" : "Maxim" }
    ]
}

However, if I change the data type to wildcard

PUT my-index-000001
{
  "mappings": {
    "properties": {
      "user": {
        "type": "nested",
        "include_in_root": true,
        "properties": {
          "first": {
            "type": "wildcard"
          },
          "last": {
            "type": "wildcard"
          }
        }
      }
    }
  }
}

... the same ingestion command will fail with the error "DocValuesField "user.first" appears more than once in this document (only one value is allowed per field)".

PUT my-index-000001/_doc/1
{
  "user": [
      { "first" : "Bob", "last" : "Kesler" },
      { "first" : "Robert", "last" : "Maxim" }
    ]
}

The interesting thing is that the following command works:

PUT my-index-000001/_doc/1
{
  "user": [
      { "first" : "Bob", "last" : "Kesler" }
    ]
}

So it looks like there is only an issue when there is an array of nested objects using include_in_root or include_in_parent.

[josefschiefer27]

Any updates about this issue? It would be nice to understand the root cause of this issue. The described behavior is a bug that only surfaces with certain data types when using nested with include_in_root or include_in_parent.

[patrick-radius]

I'm having this issue too. it seems both include_in_root and include_in_parent behave the same.

[lmignon]

Any update about this issue? I'm facing the same one and it would be nice to know why and how we can fix it...

[elasticsearchmachine]

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

[romseygeek]

This is the same issue as #70261. Field types that store information in a lucene binary doc values field don't work with include_in_root or include_in_parent because each child will try and store its information on the parent document separately, and lucene only allows a single binary doc values field instance per doc.

We have a couple of options I think:

• detect this up-front in the mappings and throw an error at index creation / template validation
• update our indexing code to merge multiple binary fields together into a single value on the parent doc; this would obviously be preferable for end-users as it would mean everything would Just Work, but may not be supportable for all field types

[josefschiefer27]

Would lucene issue #11702 fix this problem?

[josefschiefer27]

@romseygeek any plans to fix this issue? As it stands right now this is broken functionality for include_in_root and include_in_parent for certain data types.

[elasticsearchmachine]

Pinging @elastic/es-search-foundations (Team:Search Foundations)