Home / Articles / Search Engine Optimization / Introduction to Elasticsearch: What Is It and How This Search Engine Works

Introduction to Elasticsearch: What Is It and How This Search Engine Works

Whether you’re in eCommerce or model risk management services, you could be one of the tens of thousands of enterprises that could benefit from using Elasticsearch every day. When you read the words “search engine” you think of something like Google, but Elasticsearch is for businesses that need to quickly search massive amounts of their own data.

In this article, we’ll cover what Elasticsearch is, how it works, and everything else you need to know to make it work for your business.

What is Elasticsearch?

Elasticsearch is the most popular enterprise search engine, but many have never heard of it. Elasticsearch allows you to store, query, and analyze huge datasets in real-time. It’s commonly used for sophisticated queries and high-performance applications like eBay and Netflix.

ELK stack
The ELK stack. (Source: logz.io)

Elasticsearch is one of the most popular database systems accessible today, mostly for search and logs analysis. It’s part of the ELK Stack – Elasticsearch, Logstash, Kibana – which is today's most popular log analytics platform. This is a suite of free and open tools for data intake, enrichment, storage, analysis, and visualization.

The ELK stack is just as useful for scraping the web as it is for combing through internal logs and data. ObjectRocket's Twitter integration uses Elasticsearch to pull in tweets from hashtags you’ve chosen to watch. This enables you to pull in vast amounts of data automatically and prepare them for complex searches at scale. The ELK stack is known for its easy REST APIs, distributed nature, speed, and scalability.

The other parts of the ELK stack are Logstash and Kibana. Logstash is an open-source tool that lets you take data from a number of sources, alter it, and forward it to the next step of any given process. With plugins and pre-built filters, it allows users to ingest data from nearly anywhere.

Kibana is a data visualization tool that provides easy-to-use interactive charts and pre-built filters.

As more of the world’s IT infrastructure moves to the cloud, the ELK stack offers a cost-effective log analysis solution that allows your developers and DevOps to acquire useful insights into system failure and application performance. As we’ll see, companies are also able to use this to enhance functions like customer service and fraud detection.

How Does Elasticsearch Work

ELK stack tutorial
ELK stack tutorial. (Source: HowToDoInJava)

Adding data to Elasticsearch is called “indexing”. If you’re running an eCommerce operation, you can set up automatic pipelines which will send data to Elasticsearch for an index in real-time. Elasticsearch is a kind of API, which means there are plenty of options for you to input data into it using either the POST or PUT methods.

As long as you have the API key set up, you can add data to Elasticsearch from any other application that can speak JSON. All the data is supplied as a JSON object, but you don’t need to press it into that shape by hand. Elasticsearch doesn’t need the data structure to be defined ahead of time.

It’s a lot to take in, so let’s break Elasticsearch down by some useful terms.


Documents are the simplest type of data in Elasticsearch and are expressed in JSON. A document is similar to a row in a database. A document in Elasticsearch can be any structured data encoded in JSON, not only text.


A collection of documents with comparable qualities is an “index”. In Elasticsearch, an index is the broadest category against which you can make a search query. Any documents in an index are related logically.

You can have an index for “customers”, one for “SKUs”, and one for “purchases” in the context of an eCommerce website. An index is given a name that is used to refer to it while carrying out operations on the documents it contains.

Inverted Index

In Elasticsearch, an index is actually an “inverted index”, which is the mechanism that all search engines use. It's a data structure that stores key-value pairs and their places in a document or series of documents. Instead of storing strings of text directly, an inverted index divides each document into individual search phrases (e.g. each word). It then maps each search term to the documents in which it appears.

If the term “Elasticsearch” appears 20 times in this article, one row in the index will have the key-value pair “Elasticsearch: 20”. This kind of thing is what allows Google to make decisions about which pages are most relevant to your search. And it’s how Elasticsearch is able to break documents down into more granular data.


A cluster is a collection of one or more connected nodes. Any given job is divided amongst the different “nodes”, allowing searches to happen faster than they would on one computer. The distribution of tasks, searching, and indexing among all nodes in an Elasticsearch cluster is what gives it its power.


A single server that is part of a cluster is known as a node. A node is a computer that stores data and helps the clustered index and search. Elasticsearch nodes can be set up in a variety of ways.

There’s the “master node”, which controls the rest of the cluster. The “data node”, stores data and runs operations like searching. And there’s the “client node”, which directs requests to and from the master and data nodes.


Elasticsearch allows you to split the index into “shards”, which are smaller portions of the index. Each shard is a completely functional and self-contained “index” that can be hosted on any cluster node.


Elasticsearch allows you to create “replica shards” or simply “replicas,” which are copies of your index's shards. A primary shard is assigned to each document in an index. Replicas provide redundant copies of your data to defend against hardware failure, which becomes a problem in large-scale operations like Google and Facebook.

What is Elasticsearch Used for

Elasticsearch and the ELK stack are as popular in cutting-edge artificial intelligence research as it in the day-to-day running of big companies.

For example, Netflix uses the ELK Stack to monitor and analyze customer service operations. They’re able to automatically classify and query huge amounts of data automatically. They also take advantage of Elasticsearch’s automatic sharding, replication, and large ecosystem of plugins.

Walmart uses ELK to obtain insights into customer habits and track store performance. In a special case, it’s also been used to fight fraud. By taking in over 4 billion metadata records from transactions, Walmart has been able to use that knowledge to identify fraud in real time with information like IP addresses, locations, and other system traffic.

Gift card schemes targeting the elderly have been a particular problem, and Walmart has been able to save customers millions of dollars by catching those transactions as they happen.

Adobe uses Elasticstack to manage huge applications that have to search millions of items like the images in Adobe Stock. Adobe’s Elasticsearch plugins work alongside their own image recognition AI, including a “similarity” plugin and a “search ranking” plugin.

This powers image recognition that allows users to find photos similar to their inputs. (Think of Google’s Reverse Image Search.) Elasticsearch also enables them to recognize faces, identify objects, and automatically tag images for user searches.

Elasticsearch is machine-learning ready. When data is ingested, the ELK stack analyses it, ensuring that you have the metadata you need to run searches on in your log. Across several types of neural network architecture, like convolutional neural networks, Elasticsearch is able to carry out machine learning queries to your data quickly. All of this happens in real-time, as images are uploaded to the platform.

As the value of convolutional neural networks becomes more and more apparent, it’s more urgent that businesses make use of the data they’re collecting on their operations already. What is a convolutional neural network? It’s a way to train AI that involves two AIs: the trainer and the trainee.

As the trainee makes guesses about the data in front of it, such as “these two images of faces are of the same person”, the trainer gives a yes-or-no answer, which was given by the human creators. If the answer was wrong, some “neurons” are randomly altered and the test runs again. This could happen several thousand times a second for many hours until the AI is reliably inferring the correct answers from the data it’s given.

Elasticsearch enables you to implement MLOps best practices like classification, regression, and outlier detection to your data. And Elasticsearch’s inference ingests processor will apply your machine learning models to your data as soon as it comes in. 

Why Use Elasticsearch

Elasticsearch is more than just a search engine. With plugins and the ability to use machine learning models that make Google so powerful, it allows businesses to build sophisticated data operations that make use of the customer data they’re generating every day.

Also Read

Photo of author

Article by Po Han Lin

Keep Reading