A heartbeat is not the same as a pulse

A server can answer a ping and still be useless to you. Ping only proves the lights are on -- that a packet went out and a packet came back. It says nothing about whether the one process you actually depend on is running, listening, or has wedged itself into silence an hour ago. Treating "it's up" and "it works" as the same fact is the most common way a real outage gets missed for longer than it should.

The fix isn't to stop checking the easy signal -- a reachable host is still useful information, it rules out the worst case fast. The fix is to never let it stand in for the signal that actually matters. If the thing you're protecting is a checkout page, the only honest health check is hitting the checkout page, not the box underneath it. If it's a database, it's a real query, not a process list that shows the container as "running." Every layer between the easy signal and the one you care about is a place a failure can hide in plain sight, reporting green while the thing on top of it is already dark.

This gets worse, not better, when the easy signal and the real signal usually move together -- because then the gap only shows up on the one day they don't, and that's exactly the day you need the distinction to hold. A monitoring setup that only checks what's cheap to check isn't lying to you. It's just answering a question you didn't ask, confidently, in place of the one you did.

The discipline is boring and specific: name the actual dependency, check that dependency directly, and treat every proxy signal -- ping, process-up, container-healthy -- as a hint, never a verdict.