Problem

Currently, Geode supports the inserting of JSON documents using the JSONFormatter util. Even though this is util is designed for performance it lacks the capability and finesse to determine whether a JSON document partially matches an existing PDXType definition. This leads to the problem that the following documents, logically represent the same object but would generat different PDXType definitions.

{
  "name": {
    "firstName": "John",
    "surname": "Doe"
  }
}

{
  "name": {
    "surname": "Doe",
    "firstName": "John"
  }
}

{
  "name": {
    "firstName": "John",
    "surname": "Doe"
  },
  "address": {
    "addressLine1":"Suite 200",
    "addressLine2":"1235 South Rd",
    "addressLine3":null,
    "state":"OH",
    "zipCode":"287233",
    "country":"USA"
  }
}

In addition to this, the ordering of the fields is also a factor when it comes to determining the PdxType definition. Changing the order of and field within a JSON document will cause the potential generation of a new PdxType definition.

The reason for this behavior is that JSON is not a strongly structured format, whereas a Java Object and PdxType have strongly defined structures.

Proposed Solution

In order to avoid the possible side effect that each JSON document has the potential to generate a new PdxType definition, I suggest the following:

1. Provide a format generic document format which can be used to describe the definition of a data type
2. Register a type definition for a JSON document, which contains the superset of fields for a given JSON document
3. Provide the type definition id within the JSON document, which would provide a hint to the PdxType registry which PdxType definition to use for the incoming JSON document.
4. Provide a service to add new type definitions to a running cluster
5. Export the defined type definitions into the generic type definition format

Advantages

The benefits of this solution are:

• Reduced PdxType definitions for JSON documents representing the same logical domain object
• Improved performance in processing of JSON documents, as the PdxType, is known and does not have to be defined and generated for each submitted document
• Ease of type definition management
• Ability to ingest JSON documents representing the same type with a variance of populated fields
• Ability to export existing type definitions into a human readable format
• Ability to externally register data types

Disadvantages

The drawbacks to this solution:

• The superset of all the fields for a JSON structure needs to be known
• The type definition id needs to be provided for each sub-element within the JSON structure
• If the type definition id is not provided the type definition generator will behave the same way it currently does. For every provided JSON document the system might potentially generate a new type definition.
• All clients need to know the registered definitions and id's in order to provide the correct id for any given structure within the JSON document

Type Registry Service

Types can be defined in the following manner:

• Using the Java API to load type definition from an external file
• Using GFSH to load the type definitions from an external file
  • Upon the server start up
  • At runtime whilst the cluster is already running

Types can only be registered after the servers have started and the current type registry has started and validated itself. This is because consistency is valued above the load time preferences.

Whilst the type definitions are being loaded from the file, the following process is followed:

1. Read definition from the file.
2. Check if definition already defined by checking if typeId already exist
3. If typeId already exist, validate that the structure being loaded is the same as the already defined type
   1. If type definitions are different throw TypeDefinitionAlreadyDefinedException
   2. If type not defined confirm that all referenced subTypes exist within existing type registry or within the current type definition file. If all types are correctly defined and validated, then load type definition into the type registry.

Type Registry Service API

The type registry service has to provide the following services:

• addTypeDefinitions(List<TypeDefinition>)
• exportTypeDefinitons():List<TypeDefinitions>
• removeTypeDefinitions(List<TypeDefinition>)
• lookupTypeDefinition(long typeDefId):TypeDefinitions
• lookupTypeIdByTypeDefinition(TypeDefinition):long

Proposed Type Definition Format

In the below section the type definition format is described. 

{
  "@type" : "JSON_Document" | Fully qualified className,  <-- The name of the object type being defined
  "@typeId" : ,  <-- A numberic field which would uniquely identify the type
  "fields":[   <-- A collection of fields
    {
      "fieldName" :  ,  <-- The name of the field
      "dataType" : "String" | "Integer" | "Double" | "Date" | "Float" | "Boolean" | "Long" | Object (default),  <-- The data type of the field (Not required when using @refTypeId)
	  "format" : "MM/dd/yyyy hh:mm:ss:SSS " | "#0.00" <-- Optional field to provide a format when dealing with dates of doubles	
      "@refTypeId" : ,  <-- The reference to an already defined type
    },
    {
      "fieldName" :  ,  <-- The name of the field
      "dataType" : "List",  <-- Indicates data type is a list/array
      "@refTypeId" : ,  <-- Numeric reference to an already defined type, which will be populated in the list
    },
    {
      "fieldName": , <-- The name of the field
      "dataType": "List", <-- Indicates data type is a list/array
      "subType": "String" | "Integer" | "Double" | "Date" | "Float" | "Boolean" | "Long" | Object (default) <-- The data type of the elements of a List
      "format": "MM/dd/yyyy hh:mm:ss:SSS " | "#0.00" <-- Optional field to provide a format when dealing with dates of doubles
    }
}

In the below example it shows how to define the type definition for a domain object.

[
  {
    "@type": "JSON_Document",
    "@typeId": 1,
    "fields": [
      {
        "fieldName": "firstName",
        "dataType": "String"
      },
      {
        "fieldName": "lastName",
        "dataType": "String"
      },
      {
        "fieldName": "age",
        "dataType": "Integer"
      },
      {
        "fieldName": "currentAddress",
        "@refTypeId": 2
      },
      {
        "fieldName": "previousAddresses",
        "dataType": "List",
        "@refTypeId": 2
      },
      {
        "fieldName": "luckyNumber",
        "dataType": "List"
      },
      {
        "fieldName": "dateOfBirth",
        "dataType": "Date",
        "format": "MM/dd/yyyy"
      },
      {
        "fieldName": "lastUpdated",
        "dataType": "Date",
        "format": "MM/dd/yyyy hh:mm:ss:SSS"
      }
    ]
  },
  {
    "@type": "JSON_Document",
    "@typeId": 2,
    "fields": [
      {
        "fieldName": "addressLine1",
        "dataType": "String"
      },
      {
        "fieldName": "addressLine2",
        "dataType": "String"
      },
      {
        "fieldName": "addressLine3",
        "dataType": "String"
      },
      {
        "fieldName": "state",
        "dataType": "String"
      },
      {
        "fieldName": "zipCode",
        "dataType": "String"
      },
      {
        "fieldName": "country",
        "dataType": "String"
      }
    ]
  }
]

Example JSON document with typeIds

{
  "@typeId": 1,
  "firstName": "John",
  "surname": "Doe",
  "currentAddress": {
    "@typeId": 2,
    "addressLine1": "Suite 200",
    "addressLine2": "1235 South Rd",
    "addressLine3": null,
    "state": "OH",
    "zipCode": "287233",
    "country": "USA"
  }
}

In the above example, the type definitions from the Custom External Type Definition Proposal for JSON above are used. Not all the fields have been populated that are defined for the type for typeId=1. In the previous implementation, the above example JSON document would have caused a new type to have been generated. In the proposed solution, the type registry was informed that this JSON object belonged to the type defined by typeId=1 and was mapped accordingly.