Technical metrics that are application-specific, such as response times, throughput, uptime, or error rates can be gathered at the system boundaries. This can happen via the interceptors or filters, depending on the situation. HTTP-relevant monitoring can be collected via a servlet filter for any technology that builds upon servlets, such as JAX-RS.

The following code snippet shows a servlet filter that gathers the response time and throughput in a Prometheus histogram metrics:

import javax.servlet.*;
import javax.servlet.annotation.WebFilter;
import javax.servlet.http.HttpServletRequest;

@WebFilter(urlPatterns = "/*")
public class MetricsCollectorFilter implements Filter {

    private Histogram requestDuration;

    @Override
    public void init(FilterConfig filterConfig) throws ServletException {
        requestDuration = Histogram.build("request_duration_seconds",
                "Duration of HTTP requests in seconds")
                .buckets(0.1, 0.4, 1.0)
                .labelNames("request_uri")
                .register();
    }

    public void doFilter(ServletRequest req, ServletResponse res,
            FilterChain chain) throws IOException, ServletException {
        if (!(req instanceof HttpServletRequest)) {
            chain.doFilter(req, res);
            return;
        }

        String url = ((HttpServletRequest) req).getRequestURI();
        try (Histogram.Timer ignored = requestDuration
                .labels(url).startTimer()) {
            chain.doFilter(req, res);
        }
    }

    @Override
    public void destroy() {
        // nothing to do
    }
}

This metric is registered similarly to the business-related example previously, and emitted via the Prometheus output format. The histogram buckets collect the time in four buckets, with the specified times from 0.1, 0.4, or 1.0 seconds, and everything above. These bucket configurations need to be adapted to the SLAs.

The servlet filter is active on all resource paths and will collect the statistics, qualified by each path.