How does it work?

A typical Kong setup is made of two main components:

  • Kong's server, based on the widely adopted NGINX HTTP server, which is a reverse proxy processing your clients' requests to your upstream services.
  • Kong's datastore, in which the configuration is stored to allow you to horizontally scale Kong nodes. Apache Cassandra and PostgreSQL can be used to fulfill this role.

Kong needs to have both these components set up and operational. A typical Kong installation can be summed up with the following picture:

Kong server

The Kong Server, built on top of NGINX, is the server that will actually process the API requests and execute the configured plugins to provide additional functionalities to the underlying APIs before proxying the request upstream.

Kong listens on several ports that must allow external traffic and are by default:

Additionally, those ports are used internally and should be firewalled in production usage:

  • 8001 provides Kong's Admin API that you can use to operate Kong. See admin_api_listen.
  • 7946 which Kong uses for inter-nodes communication with other Kong nodes. Both UDP and TCP traffic must be allowed. See cluster_listen.
  • 7373 used by Kong to communicate with the local clustering agent. See cluster_listen_rpc.

You can use the Admin API to configure Kong, create new users, enable or disable plugins, and a handful of other operations. Since you will be using this RESTful API to operate Kong, it is also extremely easy to integrate Kong with existing systems.

Kong datastore

Kong uses an external datastore to store its configuration such as registered APIs, Consumers and Plugins. Plugins themselves can store every bit of information they need to be persisted, for example rate-limiting data or Consumer credentials.

Kong maintains a cache of this data so that there is no need for a database roundtrip while proxying requests, which would critically impact performance. This cache is invalidated by the inter-node communication when calls to the Admin API are made. As such, it is discouraged to manipulate Kong's datastore directly, since your nodes cache won't be properly invalidated.

This architecture allows Kong to scale horizontally by simply adding new nodes that will connect to the same datastore and maintain their own cache.

Which datastores are supported?

Apache Cassandra

Apache Cassandra ( is a popular, solid and reliable datastore used at major companies like Netflix and Facebook. It excels in securely storing data in both single-datacenter or multi-datacenter setups with a good performance and a fail-tolerant architecture.

Kong can use Cassandra as its primary datastore if you are aiming at a distributed, high-availability Kong setup. The two main reasons why one would chose Cassandra as Kong's datastore are: - An ease to create a distributed setup (ideal for multi-region). - Ease to scale. Since Kong maintains its own cache, only plugins such as Rate-Limiting or Response-Rate-Limiting will require a highly responsive datastore. Cassandra is a good fit for such setups.

We recommend putting Cassandra on performant machines with a generous amount of CPU and Memory, like AWS m4.xlarge instances. If you are aiming at Cassandra for your production infrastructure, make sure to go through a few important reads:

Note: If you don't want to manage/scale your own Cassandra cluster, we suggest using Instaclustr for Cassandra in the cloud.


PostgreSQL is one of the most established SQL databases out there, meaning your team might already use it or have experience with it.

It is a good candidate if the setup you are aiming at is not distributed, or if you feel comfortable scaling PostgreSQL yourself. It is also worth pointing out that many cloud providers can host and scale PostgreSQL instances for you, most notably Amazon RDS.

Again, since Kong maintain its own cache, performance should be of concern for most use-cases, making PostgreSQL a good candidate for your Kong cluster too.

How does it scale?

When it comes to scaling Kong, you need to keep in mind that you will mostly need to scale Kong's server and eventually ensure its datastore is not a single point of failure in your infrastructure.

Kong server

Scaling the Kong Server up or down is fairly easy. Each server is stateless meaning you can add or remove as many nodes under the load balancer as you want as long as they point to the same datastore.

Be aware that terminating a node might interrupt any ongoing HTTP requests on that server, so you want to make sure that before terminating the node, all HTTP requests have been processed.

Kong datastore

Scaling the datastore should not be your main concern, mostly because as mentioned before, Kong maintains its own cache, so expect your datastore's traffic to be relatively quiet.

However, keep it mind that it is always a good practise to ensure your infrastructure does not contain single points of failure (SPOF). As such, closely monitor your datastore, and ensure replication of your datastore.

If you use Cassandra, one of its main advantages is its easy-to-use replication capabilities due to its distributed nature. Make sure to read the documentation pointed out by the Cassandra section of this FAQ.

What are plugins?

Plugins are one of the most important features of Kong. All the functionalities provided by Kong actually are plugins. Authentication, rate-limiting, transformation, logging etc, are all implemented independantly. Plugins can be installed and configured via the Admin API running alongside Kong.

Almost all plugins can be customized not only to target a specific proxied service, but also to target specific Consumers.

From a technical perspective, a plugin is Lua code that's being executed during the life-cycle of a proxied request and response. Through plugins, Kong can be extended to fit any custom need or integration challenge. For example, if you need to integrate the you API's user authentication with a third-party enterprise security system, that would be implemented in a dedicated plugin that is run on every request targetting that given API.

Feel free to explore the Plugins Gallery and the Plugin development guide. Learn how to enable plugins with the plugin configuration API.

How many microservices/APIs can I add on Kong?

You can add as many microservices or APIs as you like, and use Kong to process all of them. Kong currently supports RESTful services that run over HTTP or HTTPs. Learn how to add a new service on Kong.

You can scale Kong horizontally if you are processing lots of requests, just by adding more Kong servers to your cluster.

How can I add authentication to a microservice/API?

To add an authentication layer on top of a service you can choose between the authentication plugins currently available in the Plugins Gallery, like the Basic Authentication, Key Authentication and OAuth 2.0 plugins.

Where can I get help?

You can read the official documentation or ask any question to the community and the core mantainers on our official chat on Gitter. We are also on Freenode at #kong.

You can also have a face-to-face talk with us at one of our meetups.

Were you looking for a question that you did't find? Open an issue!