Chapter 9 – Best Practices

1. HTTP traces, potentially combined with some application-specific attributes, can help answer most questions about tiny RESTful service behavior. We can aggregate metrics from traces using OpenTelemetry Collector or at query time on the backend. We still need metrics for resource utilization though. The right questions to ask here are how much this solution costs us and whether there is the potential to reduce costs with sampling and how much we must spend to keep alerts running based on queries over traces. If it’s a lot, then we should look into adding metrics. So, the answer is – yes, but it can be more cost-efficient to add other signals.
2. In an application under heavy load, every bug will happen again and again. No matter how small of a sampling rate we choose, we’ll record at least some occurrences of such an issue. A high sampling rate would likely have some performance impact, but more importantly, it’ll be...