mapping for objects with an arbitrary amount of properties in elasticsearch

Question

i have a given document:

{
  "foo": {}
}

where foo can have an arbitrary amount of properties. lets assume i will import millions of documents into my index, in which each of foo's properties do have other values.

that means my mapping which will be built dynamically will grow enormous. is there any kind of way where i can tell elasticsearch something like

take everything you have in foo and just accept it as it is (or stringify foo) without having a resulting million-lines-mapping???

or do i have to care by myself, before indexing documents?

if so, there a 2 solutions i think

1. JSON.stringify foo

2. map every property in foo into key/value pairs, and create an array of objects:

   // object
   {
     "foo": [
       {"key": "bar1", "value": "bar1's value"},
       {"key": "bar2", "value": "bar2's value"}
     ]
   }
   
   // resulting mapping
   {
     "type": {
       "properties": {
         "foo": {
           "properties": {
             "key": {
               "type": "string" 
             },
             "value": {
               "type": "string" 
             }
           }
         }
       }
     }   
   }
   

would you prefer then solution 1 or 2, and why?

appreciate your help!

Answer 1

You cannot make Elasticsearch stringify it for you. You can make it ignore whatever's under "foo" and it'll be a part of "_source", but then it will not be searchable at all.

The second approach can make a lot of sense, depending on how you're going to query it, and what you can know about the kinds of values you will accept.

There is a related question at Dynamic Type with Mappings describing this approach, with a runnable example here: https://www.found.no/play/gist/7596633

The idea is that you'll have a nested document per value. This works well if the number of values per document is not huge. If you don't use nested documents, your document would be returned if you search for "key": "bar1" and "value": "bar2's value".