In a sub-document, the keys may be common across most documents in the collection, or they may contain a set of random values. For example, using hostnames as the keys in the "hosts" sub-document.

{
  ...
  "hosts": {
    "server1a": { ... }
    "server2a": { ... }
    "server3a": { ... }
  }
} 

 {
   ...
   "hosts": {
     "server1b": { ... }
     "server2b": { ... }
     "server3b": { ... }
  }
} 

The schema analysis should recognize the above situation to avoid creating thousands of sub-fields in the schema.