Question

I have some data like so in elasticsearch:

account (http://localhost:9200/myapp/account/1)
========
name
state
groups //groups is an array containing objects like so {id: 1, starred: true}
       //the id is the id of the group type

email (http://localhost:9200/myapp/email/1?parent=1)
========
email
primary //just a boolean to denote whether this email is primary or not

group
========
name
description

emails are children of account.

Based on imotov’s excellent answer, I am able to run a search on account.name and email.email and return ALL accounts where the searched prefix matches the above 2 fields:

{
  "query": {
    "bool": {
      "must": [
        {
          "term": {
            "statuses": "active"
          }
        }
      ],
      "should": [
        {
          "prefix": {
            "name": "a"
          }
        },
        {
          "has_child": {
            "type": "email",
            "query": {
              "prefix": {
                "email": "a"
              }
            }
          }
        }
      ],
      "minimum_number_should_match" : 1
    }
  }
}

What I would like to do now is to return 2 custom fields for each result:

• For each result return a field called email, if the search was matched on the email type (which is a child of account), return that email, otherwise return the primary email linked to the account, if none, null can be returned.

• For each result return a field called group. The value of the field should contain the name of the starred group whose id is stored in the groups array. Essentially: Find group.id where group.starred is true in each account.groups, then return the matching group.name from the group type base on the id we found.

I have been looking at script fields, but I am not sure if it is able to return fields for each hit. I am also not sure if the above is actually accomplishable in ES.

Could someone provide some pointers as to whether this is possible and how to get started?

Answer 1

Currently, there is simply no way to get to access the data in a has_child or nested clause.

The only solution is to get some data, make some decisions on the client and then get more data.

Here’s what I did to achieve it:

• Run the query above and get back the data.

• To deal with showing the match emails or the primary email (run on the email type):

  {"query":
      {"bool":{
         "should":{
          {
             "prefix":{
                "email" : "the query"
             }
          },
          {
             "terms":{
                "_parent" : ["list", "of", "account", "ids"]
            }
         }
        }
      }
    }
  }
  

Base on the above query, we can get any email addresses that were matched by the search term. Remember to set the fields in the above query to include _parent.

We can then use array_diff() or a similiar function in languages other than PHP to diff the parent ids from above and and original query. This should then give us a list of accounts where no email was matched. Simply then send another request to get the primary emails for those accounts.

For the groups, send a query to type account and:

• Constrain _id to the list of account ids.
• Constrain group.starred to true.

This should get you a list of starred groups. To get more information (name etc), send a query to type group and:

• Constrain _id to the group ids above.

Finally, do some client side processing to put it together so that it is easier for the next part of your program to use.