would cost, a manual check may be entirely reasonable. But for services people depend on, one meaningful outcome check seems more valuable than a page full of indicators that only establish reachability.

I haven’t run a comparative test of these approaches; I’m proposing a way to think about them and would like to hear what has worked in practice.

* Which service appeared healthy while its actual function was broken?
* What check eventually caught it?
* Do you use test transactions, application metrics, or manual checks after upgrades?
* How do you handle legitimate periods of inactivity?
* Where has deeper monitoring become more trouble than it was worth?

https://redd.it/1x4lwfl
@r_SelfHosted