A system is only as available as the parts it leans on. Chain dependencies in seriesA chain where every part must be up for the whole to work β so a failure anywhere takes the system down, and availabilities multiply. and their availabilities multiply down β three 99.9% services in a row are just 99.7% together. Add redundancy in parallelRedundant copies where any one staying up keeps the system up β so you're only down when all of them fail at once. and the gaps multiply away instead. List your parts and read off the system number.
β 1d 2h of downtime a year Β· 2h 9m a month
Chaining 3 parts multiplies their availability down to 99.7% β about 1d 2h of downtime a year. Your weakest link, Frontend at 99.9%, sets the ceiling.