嵌套可枚举对象（i18n）的ElasticSearch映射

Question

I'm at a loss as to how to map a document for search with the following structure:

{
  "_id": "007ff234cb2248",
  "ids": {
    "source1": "123",
    "source2": "456",
    "source3": "789"
  }
  "names": [
    {"en":"Example"}, 
    {"fr":"exemple"}, 
    {"es":"ejemplo"},
    {"de":"Beispiel"}
  ],
  "children" : [
    {
      "ids": {
        "source1": "CXXIII",
        "source2": "CDLVI",
        "source3": "DCCLXXXIX",
      }
      names: [
        {"en":"Example Child"}, 
        {"fr":"exemple enfant"}, 
        {"es":"Ejemplo niño"},
        {"de":"Beispiel Kindes"}
      ]
    }
  ],
  "relatives": {
    // Typically no "ids" at this level.
    "relation": 'uncle',
    "children": [
      {
        "ids": {
          "source1": "0x7B",
          "source2": "0x1C8",
          "source3": "0x315"
        },
        "names": [
          {"en":"Example Cousin"}, 
          {"fr":"exemple cousine"}, 
          {"es":"Ejemplo primo"},
          {"de":"Beispiel Cousin"}
        ]
      }
    ]
  }
}

The child object may appear in the children section directly, or further nested in my document as uncle.children (cousins, in this case). The IDs field is common to levels one (the root), level two (the children and the uncle), and to level three (the cousins), the naming structure is also common to levels one and three.

My use-case is to be able to search for IDs (nested objects) by prefix, and by the whole ID. And also to be able to search for child names, following an ( as yet undefined ) set of analyzer rules.

I haven't been able to find a way to map these in any useful way. I don't believe I'll have much success using the same technique for ids and names , as there's an extra level of mapping between names and the document root.

I'm not even certain that it is even mappable . I believe at least in principle that the ids should be mappable as terms, and perhaps that if I index the names as terms in some way, too.

I'm simply at a loss, and the documentation doesn't seem to cover anything like this level of complex mapping.

I have limited (read: no) control of the document as it's coming from the CouchDB river, and the upstream application already relies on this format, so I can't really change it.

I'm looking for being able to search by the following pseudo conditions, all of which should match:

• ID: "123"
• ID by source (I don't know how best to mark this up in pseudo language)
• ID prefix: "CDL"
• Name: "Example" , "Example Child"
• Localized name (I don't even know how best to pseudo-mark this up!

The specifics of tokenising and analysis I can figure out for myself, when I at least know how to map

• Objects when both the key and the value of the object properties are important
• Enumerable objects when the key and value are important.

Answer 1

If the mapping from an ID to its children is 1-to-many, then you could store the children's names in a child field, as a field can have multiple values. Each document would then have an ID field, possibly a relation field, and zero or more child fields.