Flipping one feature switch turned off an unrelated feature too

Symptom

A switch file was created to disable one collector. An unrelated restart job stopped as well. The two have nothing to do with each other.

Cause

Inside one script, unrelated features shared the same exit path.

[ -f .amz_off ] && exit 0      # switch for collector A
run_collector_a
run_restart_job                 # ← sits below, so it dies too

★★★ When unrelated features share an exit path, turning one off turns both off.

★★ And whoever flipped it believes only one thing was disabled

★★★ The intended thing was disabled exactly as expected, so verification passes. The unintended casualty emits no signal at all. The log says "disabled collector A"; nothing anywhere says "and the restart job."

★ Why it keeps happening

Switches are typically written as if off: exit at the top of the file. Everything below becomes subordinate to that switch, and the person adding the switch does not look at what is below.

★★ And new features get appended to the end of the file, so over time that one switch at the top disables more and more. It was genuinely accurate the day it was written.

Fix

★ Write a feature switch as a condition around its own block, not as an exit.

[ -f .amz_off ] || run_collector_a
[ -f .wc_off  ] || run_restart_job

How to verify

★★ Before flipping a switch, look at what lives below it. And ★★★ verify the result feature by feature — not "did the intended thing stop?" but "is the unintended thing still running?" Check only the former and it always passes.

★ The same shape is already in this collection — when the first stage of a fallback chain continues, every later stage dies. An early exit speaks for everything beneath it.

---