The application is a good place to gather the technical metrics, ideally at the system boundaries. It's equally possible to collect these in a potential proxy server.

The application server already emits technically-relevant metrics such as information about resource consumption, threading, pooling, transactions, or sessions. Some solutions also provide Java agents that sample and emit technically-relevant information.

Traditionally, application servers are required to make technically relevant metrics available via JMX. This functionality is part of the Management API, but has never been used much in projects. One of the reasons for this is that the model and API are quite cumbersome.

However, it's helpful to mention that Java EE application servers are required to gather and provide data about its resources. The container emits this information via JMX. There are several ways to scrape this information.

There are so-called exporters available, applications that either run standalone or as Java agents, that access the JMX information and emit it via HTTP. The Prometheus JMX exporter, which exports the information in a similar format as shown previously, is an example of this. The benefit of this approach is that it doesn't add dependencies into the application.

The installation and configuration of Java agents is done in the application server, in a base container image layer. This once again emphasizes the principle that containers should not couple the application's artifact with implementation details.