Nested search with ArangoSearch

You can search for nested objects in arrays that satisfy multiple conditions each, and define how often these conditions should be fulfilled for the entire array

Nested search is only available in the Enterprise Edition, including the ArangoGraph Insights Platform.

By default, arangosearch Views index arrays as if the parent attribute had multiple values at once. This is also supported for search-alias Views by enabling the searchField option. With trackListPositions set to true, every array element is indexed individually and can be queried separately using the respective array index. With the nested search feature, you get another option for indexing arrays, in particular nested objects in arrays.

You can let the View index the sub-objects in a way that lets you query for co-occurring values. For example, you can search the sub-objects and all the conditions need to be met by a single sub-object instead of across all of them.

Consider the following document:

{
  "dimensions": [
    { "type": "height", "value": 35 },
    { "type": "width", "value": 60 }
  ]
}

You would normally index the dimensions field and its sub-fields with an arangosearch View definition like the following:

{
  "links": {
    "<collection>": {
      "fields": {
        "dimensions": {
          "fields": {
            "type": {},
            "value": {}
          }
        }
      }
    }
  },
  ...
}

Or using an inverted index via a search-alias View, in arangosh:

db.<collection>.ensureIndex({
  name: "inv-idx",
  type: "inverted",
  searchField: true,
  fields: [
    "dimensions.type",
    "dimensions.value"
  ]
});

db._createView("viewName", "search-alias", { indexes: [
  { collection: "<collection>", index: "inv-idx" }
]});

You might then write a query like the following to find documents where the height is greater than 40:

FOR doc IN viewName
  SEARCH doc.dimensions.type == "height" AND doc.dimensions.value > 40
  RETURN doc

This query matches the above document despite the height only being 35. The reason is that each condition is true for at least one of the nested objects. There is no check whether both conditions are true for the same object, however. You could add a FILTER statement to remove false positive matches from the search results, but it is cumbersome to check the conditions again, for every sub-object:

FOR doc IN viewName
  SEARCH doc.dimensions.type == "height" AND doc.dimensions.value > 40
  FILTER LENGTH(doc.dimensions[* FILTER CURRENT.type == "height" AND CURRENT.value > 40]) > 0
  RETURN doc

The nested search feature allows you to condense the query while utilizing the View index:

FOR doc IN viewName
  SEARCH doc.dimensions[? FILTER CURRENT.type == "height" AND CURRENT.value > 40]
  RETURN doc

The required arangosearch View definition for this to work is as follows:

{
  "links": {
    "<collection>": {
      "fields": {
        "dimensions": {
          "nested": {
            "type": {},
            "value": {}
          }
        }
      }
    }
  }
}

Note the usage of a nested property instead of a fields property, configuring the View to index the objects in the dimensions array so that you can use the Question mark operator to query the nested objects. The default identity Analyzer is used for the fields because none is specified explicitly.

Similarly, the required inverted index definition for using a search-alias View to perform nested searches needs to index the parent dimensions field, as well as the nested attributes using the nested property under the fields property:

db.<collection>.ensureIndex({
  name: "inv-nest",
  type: "inverted",
  fields: [
    {
      name: "dimensions",
      nested: [
        { name: "type" },
        { name: "value" }
      ]
    }
  ]
});

db._createView("viewName", "search-alias", { indexes: [
  { collection: "<collection>", index: "inv-nest" }
]});

Defining how often the conditions need to be true

You can optionally specify a quantifier to define how often the conditions need to be true for the entire array. The following query matches documents that have one or two nested objects with a height greater than 40:

FOR doc IN viewName
  SEARCH doc.dimensions[? 1..2 FILTER CURRENT.type == "height" AND CURRENT.value > 40]
  RETURN doc

If you leave out the quantifier, it defaults to ANY. The conditions need to be fulfilled by at least one sub-object, but more than one sub-object may meet the conditions. With a quantity of 1, it would need to be one match exactly. Similarly, ranges require an exact match between the minimum and maximum number, including the specified boundaries. To require two or more sub-objects to fulfill the conditions, you can use AT LEAST (2), and so on.

  • To use the question mark operator with the ALL quantifier in SEARCH queries against arangosearch Views, you need at least ArangoDB v3.10.1 and set the storeValues property of the View to "id".
  • The expression of the AT LEAST quantifier needs to evaluate to a number before the search is performed. It can therefore not reference the document emitted by FOR doc IN viewName, nor the CURRENT pseudo-variable.
  • Using the question mark operator without quantifier and filter conditions ([?]) is possible but cannot utilize indexes.

Searching deeply nested data

You can index and search for multiple levels of objects in arrays. Consider the following document:

{
  "dimensions": [
    {
      "part": "frame",
      "measurements": [
        { "type": "height", "value": 47 },
        { "type": "width", "value": 72 }
      ],
      "comments": "Slightly damaged at the bottom right corner."
    },
    {
      "part": "canvas",
      "measurements": [
        { "type": "height", "value": 35 },
        { "type": "width", "value": 60 }
      ]
    }
  ]
}

To index the array of dimension objects and the nested array of measurement objects, you can use an arangosearch View definition like the following:

{
  "links": {
    "<collection>": {
      "fields": {
        "dimensions": {
          "nested": {
            "measurements": {
              "nested": {
                "type": {},
                "value": {}
              }
            },
            "part": {},
            "comments": {
              "analyzers": [
                "text_en"
              ]
            }
          }
        }
      }
    }
  }
}

The equivalent search-alias View and inverted index definition is as follows, using arangosh:

db.<collection>.ensureIndex({
  name: "inv-nest-deep",
  type: "inverted",
  fields: [
    {
      name: "dimensions",
      nested: [
        {
          name: "measurements",
          nested: [
            { name: "type" },
            { name: "value" }
          ]
        },
        "part",
        {
          name: "comments",
          analyzer: "text_en"
        }
      ]
    }
  ]
});

db._createView("viewName", "search-alias", { indexes: [
  { collection: "<collection>", index: "inv-nest-deep" }
]});

The default identity Analyzer is used for the type, value, and part attributes, and the built-in text_en is used for the comments.

A possible query is to search for frames with damaged corners that are not wider than 80, using a question mark operator to check the part and comments, and a nested question mark operator to check the type and value:

FOR doc IN viewName
  SEARCH doc.dimensions[? FILTER CURRENT.part == "frame" AND
         ANALYZER(TOKENS("corner damage", "text_en") ALL == CURRENT.comments, "text_en") AND
         CURRENT.measurements[? FILTER CURRENT.type == "width" AND CURRENT.value <= 80]]
  RETURN doc

The conditions of the inner question mark operator need to be satisfied by a single measurement object. The conditions of the outer question mark operator need to be satisfied by a single dimension object, including the measurement conditions of the inner operator. The example document does match these conditions.