Prometheus: A Guide to Scalable Architecture Construction

Prometheus: A Guide to Scalable Architecture Construction

Cloud-native, distributed, scalable applications with tens or hundreds of micro services need reliable monitoring and alerting. You probably use Prometheus to monitor cloud-native apps. Prometheus is best for monitoring containerized micro services and Kubernetes infrastructure.

Prometheus uses a local time-series database to save the metrics it collects. Metrics from rapidly expanding cloud-native applications might quickly overwhelm a Prometheus server’s storage capacity, which presents a significant challenge. If you can’t figure out how to scale Prometheus, you’ll have to either collect less data or make your measurements less precise. Both of these options reduce the value of the metrics you’re gathering, so you should avoid them. If you want to keep up with the growing pains of your scalable cloud application, you’ll need monitoring prometheus scaling guide.

The Scalability of Prometheus

The development of the application it monitors can be accommodated by a number of approaches to Prometheus back-end architecture. The term “back end” refers to the centralized location where all the data collected by the Prometheus instances that are keeping an eye on your application may be stored and queried.

The question of how many instances of Prometheus to run and where to run them is a different scalability concern that is significantly influenced by your application’s design. It is simple for a single Prometheus server to collect metrics from a cluster of thousands of machines.

While monitoring physical servers, you might deploy one Prometheus instance per datacentre, while monitoring cloud VMs, you might deploy one instance per region, and when monitoring containers, you might deploy one instance per Kubernetes cluster.

There are two main ways to construct a scalable Prometheus back end: manually through Prometheus federation, or automatically through shardable systems supported by long-term storage (LTS). Our Prometheus introductory course will provide you a high-level overview of the system.

The Federation’s Prometheus at Full Size

Federation is the simplest method for scaling Prometheus. Federation, in its simplest form, is the ability for one Prometheus server to harvest time series data from another. Prometheus offers two types of federation: hierarchical and cross-service.

Confederation of Ranks

Each Prometheus server in a hierarchical federation collects and stores data from numerous Prometheus servers below it in the tree.

Cross-Service Federation

Federation among services is quite similar but less complex. With cross-service federation, one Prometheus server can collect time-series metrics from other Prometheus servers instead of collecting everything into a central location. All of your data may be queried from any of the Prometheus servers because each server replicates the information it receives from its siblings.


Both forms of federations reduce distributed queries. We should query a federated Prometheus server for data. This aids Prometheus sharding. One Prometheus instance can scrape 1,000 services. If that’s not enough, you can deploy numerous Prometheus servers to collect metrics from different service subsets. One Prometheus server would collect data from all shards. It’s also useful for rolling up metrics when multiple Prometheus servers monitor the same micro service but you only want to query one. This configuration is easy with a recording rule that provides a low-cardinality time series the parent Prometheus instance can scrape.

Clare Louise