Comment by chx

Yeah, I've been saying this for a long long time now, an early blog post of mine http://drupal4hu.com/node/305.html and this madness just got worse because of Kubernetes et al. Kubernetes is a Google solution. Are you sure Google-sized solutions are right for your organization?

Also, an equally pseudo controversial viewpoint: it's almost always cheaper to be down than engineering a HA architecture. Take a realistic look at downtime causes outside of your control -- for example, your DDoS shield provider going down etc. etc. and then consider how much downtime a hardware failure adds and now think. Maybe a manual failover master-slave is enough or perhaps even that's overkill? How much money does the business lose by being down versus how much it costs to protect from it? And can you really protect from it? Are you going to have regular drills to practice the failover -- and absurdly, will the inevitable downtime from failing a few of those be larger than a single server downtime? I rarely see posts about weighing these while the general advice of avoiding single points of failure -- which is very hard -- is abundant.