Getting Started with Elasticsearch, Kibana, and Vega-Lite graphs

Vega offer a way to create custom visualizations, beyond the ones that come standard with Kibana. In this post we will use Vega-Lite (simplified Vega) syntax to show data from Elasticsearch.

  • First, lets create some sample data. In Kibana, on the left hand side, click Dev Tools Console, copy this text, click in the middle of it, and hit the green “play” button. This creates a new cars index, and defines the fields structure. You should see "acknowledged": true, "shards_acknowledged": true on the right.
PUT /cars
 "mappings" : {
  "_default_" : {
   "properties" : {
      "Name": {"type": "keyword" },
      "Miles_per_Gallon": {"type": "float" },
      "Cylinders": {"type": "integer" },
      "Horsepower": {"type": "float" },
      "Weight_in_lbs": {"type": "float" },
      "Year": {"type": "date" }
  • Next, clear the dev console, open this link, copy its entire content to the dev console, and also hit the green play button on it. You should see a long response with "errors": false,.

  • Now, open Visualizers, create a new Vega visualization (it should be at the bottom), and copy/paste this Vega-Lite graph. You should immediately see all of the imported data as a scatter plot.

  "$schema": "",
  "data": {
    "url": {
      "index": "cars",
      "%context_query%": true,
      "body": {"size": 10000}
    "format": {"property": "hits.hits"}
  "mark": "circle",
  "encoding": {
    "tooltip": {"field": "_source.Name", "type": "ordinal"},
    "y": {"field": "_source.Horsepower", "type": "quantitative"},
    "x": {"field": "_source.Miles_per_Gallon", "type": "quantitative"}

Let’s analyse this graph piece by piece.

“version”: "...”

With version, we specify that this is a Vega-Lite graph, as opposed to a Vega graph. Vega offers much more flexibility, at the expense of being slightly more verbose.


Data could be either a static URL, or an object that describes ElasticSearch query. For our example, we simply get the maximum number of the original documents (10,000) to keep things simple. Unlike Vega, Vega-Lite data can only have a single data source. The data will be returned as:

  "hits": {
    "hits": [
        "_id": "AV9P65tJNuzjRfBwoDV5",
        "_source": {
          "Name": "ford mustang boss 302",
          "Miles_per_Gallon": null,
          "Cylinders": 8,
          "Horsepower": 140,
          "Weight_in_lbs": 3353,
          "Year": "1970-01-01"
        "_id": "AV9P65tKNuzjRfBwoDV8",
        "_score": 1,
        "_source": {
          "Name": "toyota corona mark ii",

Some values were removed for brevity. For our graph, we are only interested in the list of values inside the hits.hits, so property parameter tells Vega-Lite to ignore everything else. Note that the actual values are stored inside the _source subfield, so later we need to use _source.<field> to access them.

Hint: you can use browser debug tools to view the raw data responses from ES - hit F12, and make a minor graph modification (e.g. add a space). You can also view Vega debugging state with VEGA_DEBUG variable, e.g.'source_0')[0] would show the first data element.

marks is what determines how data should be shown. It can also be a bar, line, and many other.

encoding channels set the mapping between our data fields, and how they should be drawn, e.g. that certain fields should represent the X or Y coordinate, the color of the line or the shape of the symbol. Each field needs to indicate the type of data it contains - quantitative, temporal, ordinal, or nominal.

Written on October 25, 2017
License: CC-BY