Skip to content

Production installation

Tracardi is a distributed system and as such should be installed on production servers. The installations described in the previous sections are mainly used for testing purposes. A production installation should be built with at least five Tracardi API instances, two instances of GUI, and three elasticsearch instances.

Setting up an elastic search cluster is beyond Tracardi configuration and can be handled by ready to use cloud instances that expose one IP but hide 3 and more instances of the database.

API and GUI

The production setup should have a clusters with Tracardi API and one cluster with Tracardi GUI (2 instances). GUI cluster should be accessible only from trusted network for security reasons. Although the cluster with Tracardi API may be set up as one cluster we recommend running 2 clusters with slightly reconfigured instances.

One cluster, the one available on the Internet, consisting of at least 3 instances should be configured with environment variable EXPOSE_GUI_API set to no. This way the publicly available API will have only /track endpoint available. This cluster will be used for event collection from the website on the Internet. All other endpoints that are needed for the GUI will be disabled.

Another cluster, the one available in the internal network, or available on the Internet but restricted to certain set of IPs, should have the environment variable EXPOSE_GUI_API set to yes. This cluster will be used by GUI to control Tracardi.

Tracardi API cluster

A single instance of Tracardi API (by default run from docker container) starts 25 workers. Each worker is able to handle asynchronous connections.

Each instance of Tracardi API should be run on a separate server (physical or virtual). The point is that in the event of failure, the other instances could take over the tasks of the failed instance. There is no obstacle to run more than three Tracardi instances.

Tip

We recommend installing Tracardi inside a k8s environment. Kubernetes automatically can bring back to life failed docker instances.

Load balancing

A cluster of Tracardi instances (GUI or API) should be hidden behind the load balancer that will expose one IP to the world and direct external traffic to individual instances. To avoid single point of failure create multiple instances of load balancer.
Tracardi do not require long-lasting sticky sessions, so you can configure your DNS servers to return multiple A records (IP addresses) for your domain.

This kind of architecture will allow you to scale Tracardi for small and high traffic.

If Tracardi is configured to run in HTTPS mode, the load balancer should accept encrypted traffic. Requests inside the cluster may stay not encoded. This way, you can bypass the certificate management problems for each Tracardi instance.

Tracardi GUI cluster

Tracardi GUI cluster should be created the same way the API cluster. If you have 2 separate clusters for internal and external traffic remember to configure GUI to connect to cluster handling internal traffic.

Q & A

Q: What is the minimal number of instances? A: This all depends on your traffic. If you do not have big traffic you could run one cluster of APIs with 2 instances and install GUI on your local machine.

Q: Do tracardi need any particular routing inside cluster? A: The internal routing from load balancer to Tracardi instances can be for example: round-robin. Tracardi do not require long-lasting sticky sessions.

Q: What load can take Tracardi? A: We tested single instance of Tracardi with 25 workers connected to one instance of elasticsearch, and it was capable of responding to 600 requests per second.