The first step in getting started with building a custom project for ReactiveSearch is to add your own data. In this guide, we explain how you can import your data with the correct schema.

Creating an App

ReactiveSearch uses Elasticsearch as its underlying database system.

Note

An app within ReactiveSearch's context refers to an index in Elasticsearch.

Here's a short gif walking through the app creation process in appbase.io, a hosted realtime Elasticsearch service.

If you are using an Elasticsearch cluster, you would be able to create an index with the following command:

Copy
curl -XPUT elasticsearch:9200/my_app/

Note

ReactiveSearch is compatible with Elasticsearch v2, v5 and v6.

Overview of the App Data Model

Elasticsearch's data model is JSON based, and data within an app is organized as JSON objects that belong to a type (or not if you are using v6). Types provide a logical namespace to the JSON data which can be used while querying data. You can read more about the data model behind Elasticsearch over here.

Data Store

Note

Elasticsearch v6 removes the concepts of types, where data is simply stored as JSON objects within an index.

Importing Custom Data

In this section, we will cover how to add data using three different approaches. We will use Dejavu - a GUI for Elasticsearch for showing the process.

via Importer

Importer is the most recommended way to index data into an Elasticsearch index. It works with JSON and CSV file types, and also supports cloning an existing app.

Importer also allows you to configure mappings prior to indexing the data.

via Dejavu GUI

Let's say you have your data organized as a CSV or XLS file.

  1. Use an online tool like http://www.csvjson.com/csv2json to convert your CSV data into a JSON format.
  2. Go to your dejavu's Data Browser section and click "Add Data".

  1. Add the JSON formatted data here, a single object should be added as a JSON object while multiple records can be added using the Array semantics.

Tip

Add up to a hundred records at a time for best results. Ideal when your data set is small.

via API

Elasticsearch uses a RESTful API for both indexing and retrieving data. Whether you are using Python or Java or Javascript, you make a HTTP API call there.

This is how a REST call looks to insert a single object into an app inside a type called books.

Copy
curl -XPUT https://API_CREDENTIAL@scalr.api.appbase.io/$app/books/1 --data-binary '{
   "department_name": "Books",
   "department_id": 1,
   "name": "A Sample Book on Network Routing",
   "price": 5595
}'

scalr.api.appbase.io is the domain for the Elasticsearch cluster, you can replace this with the location of your cluster domain.

Data Mapping

Data mapping is the process of specifying a schema for your data, which determines how fields are indexed and stored. While Elasticsearch auto-detects the schema based on the kind of JSON value through a process known as dynamic mapping, it is also possible to set this mapping statically. You can read more about mappings over here.

Dejavu provides a GUI for setting the mapping of a new field, as well as viewing existing mappings.

Note

If you are coming from a SQL background, there are two things to keep in mind regarding Elasticsearch mappings:

  1. They are immutable. Once specified (or dynamically set), they can't be changed.
  2. Being a full-text search engine as well as an aggregations engine, Elasticsearch supports specifying multiple mappings for a field. This allows it to store the same field in different ways, thus allowing for versatile querying.

Custom Analyzers

We recommend adding the following custom analyzers to your Elasticsearch index. We will use them when specifying mappings for a $dataField to be used for searching.

The autosuggest_analyzer indexes the field by breaking each word into tokens of length [1, 20] always beginning with the first character of the word. This allows for fast suggestions retrieval in a DataSearch UI. It is ideally suited for creating autocompletion based searches.

The ngram_analyzer indexes the field by breaking each word into tokens of length [2, 9] but beginning with any character of the word. This allows for fast suggestions retrieval in a DataSearch UI. It is ideally suited for finding partial (but non-fuzzy) matches.

Note

If you are using dejavu/importer for indexing data, it already supports these custom analyzers with the Text and SearchableText datatypes.

Adding custom analyzers requires you to first close the index with the following command:

Copy
POST /:index/_close

followed by the actual addition of analyzers with:

Copy
PUT /:index/_settings
{
  "analysis" : {
    "analyzer":{
        "autosuggest_analyzer": {
            "filter": [
                "lowercase",
                "asciifolding",
                "autosuggest_filter"
            ],
            "tokenizer": "standard",
            "type": "custom"
        },
        "ngram_analyzer": {
            "filter": [
                "lowercase",
                "asciifolding",
                "ngram_filter"
            ],
            "tokenizer": "standard",
            "type": "custom"
        }
    },
    "filter": {
        "autosuggest_filter": {
            "max_gram": "20",
            "min_gram": "1",
            "token_chars": [
                "letter",
                "digit",
                "punctuation",
                "symbol"
            ],
            "type": "edge_ngram"
        },
        "ngram_filter": {
            "max_gram": "9",
            "min_gram": "2",
            "token_chars": [
                "letter",
                "digit",
                "punctuation",
                "symbol"
            ],
            "type": "ngram"
        }
    }
  }
}

followed by opening of the index. It is important to open the index up for any indexing and search operations to occur.

Copy
POST /:index/_open

You can read more about this over here.

Text Datatype in Search v/s Aggregations

There are two different types of use-cases that Elasticsearch supports:

  1. Search - We talked about this use-case in the above section. Here, we typically apply a search specific analyzer to get fast and accurate search results back. The field to be applied in the dataField prop for search specific components is of Text type (v5 and above) or of String analyzed type (v2).
  2. Aggregations - Aggregations and sorting are operations that require using a non-analyzed string. The field to be applied in the dataField prop for aggregation specific components such as Lists is of Keyword type (v5 and above) or of String not-analyzed type (v2).

Note

If a $dataField is to be used both for searching and aggregating, it is recommended to create a .raw sub-field for a Text datatype field whose datatype is set as Keyword. And when aggregating (or sorting) is required, use $dataField.raw instead of $dataField.

Starting with v5, Elasticsearch supports this as a default.