Hacker News Reader

I have to completely disagree. Application level metrics can also be emitted at the log level, but are much easier to work with as a metric. We have total request count, average request latency per route, and per HTTP response. This is extremely useful for finding performance regressions when new code is released.

This is misguided.

Think about, eg. Ceph. From the side of the user, it gives user the storage, from the side of Ceph admin, it's an application.

Now, imagine Ceph is used in Kubernetes to create PVC. Now, Kubernetes is your storage provider... and so on.

There are some basic performance metrics that come to mind that are also important:

Latency Uptime Throughput

And if you're working with a distributed system, you should really be using metrics to track the health of your system at the application level.

Latency, uptime and throughput for an application are non-functional requirements that should be built into your application. People like to generalize that and pawn that off to logging dashboard but I personally think that's a mistake.

Usually it’s input + state to repro - good luck logging that