Well, as soon the solution doesn't scale - which will likely be within a couple of years - one needs to rearchitect and recode.
I suppose single node+db works fine for hobby SAAS and expert programmers who can squeeze every bit of performance from the machine. But for the average in-efficient programmer or teams, standard PostGres as DB and HA services fronted by a load-balancer work better.
> Well, as soon the solution doesn't scale - which will likely be within a couple of years - one needs to rearchitect and recode.
Your project is far more likely to die within a couple of years from having too few users than from having too many. If you find you're having scaling issues then that's a good time to solve them - your project is popular enough to justify the investment.
You'll likely be recoding your project quite a lot in the early phase of the development anyway as you learn more about the problem and (most importantly) more about what your users want/need. Anything that can reduce the cost of this early iteration will pay dividends later because it makes it more likely that you'll need to scale - and by the time you do need to scale you'll be scaling something that you understand and other people want.
I suppose single node+db works fine for hobby SAAS and expert programmers who can squeeze every bit of performance from the machine. But for the average in-efficient programmer or teams, standard PostGres as DB and HA services fronted by a load-balancer work better.