Highly available infrastructure for learning management system

The client was interested to get the ability to automate horizontal scaling and achieve highly available and fault-tolerant infrastructure. At the same time the application was implemented on mare-metal Hetzner servers that grew to be an issue with the growth of popularity for e-learning solutions. Since the platform's amount of traffic expanded, the maintenance cost and the need for horizontal scaling arose too.

The Kubernetes-based application allowed us to implement an autoscaling solution that prevented the infrastructure from failing at high traffic due to the load balancing. Also, it ensured fault tolerance for better availability and, therefore, a positive user experience.

With an instanced approach available on Hatzner, we were able to make the system repetitive. It means that scaling the application required much less time and only one utility.

The current monitoring solution is based on Prometheus and Grafana for virtualization as the most efficient combination for monitoring and alerting among modern DevOps practices.

The new logging system for the application was based on Locky, unlike the previous one on ELK Stack, to be centralized for better manageability. Despite being more comfortable in use, it is lightweight. Changing ELK to Locky made instances smaller, therefore, boosted the application performance and allowed us to reduce expenses on infrastructure maintenance.

To ensure clocklike work for the application, we used two types of databases: MySQL DB and Elasticsearch for fast data navigation. Later we implemented ClickHouse, the latest database technology, for quick data computing. A cluster database improved application availability since the solution protected nodes from failing when one of them went down. So if one of the nodes fails, data would be protected.

We backup Elasticsearch and ClickHouse. The backup system was updated with a policy that deletes unnecessary or old backups that save space and money. This allowed the system to perform faster and more efficiently.

Two factors ensured infrastructure maintenance cost reduction:

• The system automatically deletes unnecessary backups;
• The infrastructure uses only the resources it needs, no less due to the implemented auto-scaling solution.

‍