You intentionally break something just a little to force dependents to notice, before turning it off completely
I thought it was an analogy to the electrical problem: flickering lights due to high demand.
It refers to a situation where a system is deliberately designed to fail (usually for short periods of time), to still provide some level of service while alerting others that the system is soon to be turned off.
It's also used in utility power supplies to describe line voltage going below spec. It's considered a dangerous condition in that context too, as lots of non-smart equipment tend to run at higher amperage at lower voltage and/or fail to start/run and catch fire.
1: https://developerhelp.microchip.com/xwiki/bin/view/products/...
You do it when you have a bunch of automated integrations with you and you have to break them. The lights arent on at the client: their dev teams are focused on other things, so you have to wake them up to a change that's happening (either by causing their alerting to go off, or their customers to complain because their site is broken)
Later is was coopted to mean any problems with power supply not including outright drop to zero-zero/disconnections. cf microcontroller brown-out handling, also mentioned above.
Then later it seems it was generalized to mean sort-of-non-terminal problem with supply of most anything.
The OED reports that the disrupted-electrical supply sense of blackout was first used in 1934; the air-defence one (no light) in 1935. However, the OG use seems to be in the theatre, where the lights are shut off during set changes (1913, probably earlier).
Brownout does seem to be a WWII-era term, but more related to conservation/shortages than air defence.