In the previous chapters, we used data from cAdvisor to scale services automatically. Specifically, Prometheus was firing alerts if memory limits were reached. When memory utilization was over the limit, we were scaling the service associated with the data. While that approach is a good start, it is far from enough for the type of the system we're building. As a minimum, we need to measure response times of our services. Should we look for an exporter that would provide that information?
The chances are that your first thought would be to use haproxy_exporter (https://github.com/prometheus/haproxy_exporter). If all public requests are going through it, it makes sense to scrape response times and set some alerts based on collected data. That model would be in line with the most of the other monitoring systems. The only problem with that approach is that...