Monitoring Theory
teddziuba.com
teddziuba.com
To this I assign many of the resource utilization alerts, such as disk or memory full. They're certainly actionable, but, in very many cases, they don't tell you if a critical service is down because of it. Arguably, these are another kind of "cool story, bro" alert.
Regardless, they're a common frustration because it's something that's so easy to monitor, it's even an example/default in Nagios, along with the oh so uninformative load average[1]. It's a specific giving in to the general temptation of monitoring something just because it's easy.
[1] I routinely remove this from all monitoring systems I touch, even if it means modifying source and custom compiling, since, invariably, someone will point to it, without even knowing what it means to be on the run queue.
For memory, depending on the definition of "full," it, too, can be just fine. Even having something like individual app servers die and get restarted by Apache isn't such a bad thing if the cause is a slow enough memory leak.
And even with a free load balancer like HAproxy, you can have a set of fallback machines usable by any backend that's over-capacity, with the spin-up being on-demand, like, say, with Apache workers.