How should client-side errors be handled for SLOs / alerts?

[emschwartz]

At Fiberplane, we currently have an issue where our success rate alert is firing because we had a small increase in client-side errors due to a specific endpoint being called when users are logged out. The Unauthorized error is part of an error enum in our Rust code and currently all error variants are treated as an "error" as far as SLOs are concerned.

One option is to add a way to specify whether certain errors should be ignored (autometrics-dev/autometrics-rs#34).

However, I think there's a more general question about what the right way to handle client-side errors is. Should all client side errors be ignored from the SLOs?

On the one hand, you don't exactly control the input that the client is giving so it might not be your fault that there's an error. On the other hand, if you suddenly had a big spike in client-side errors that could indicate that your sign-in flow is somehow broken or your front-end app is submitting invalid data to your API.