I always wonder if going all-in on cloud services actually saves most people time, after hearing about all the weird corner cases that need debugging. Because leased dedicated hardware really doesn’t take much time to manage, it’s entirely fungible and easy to switch off of if you’re not happy with support, it can support an enormous amount of traffic, and it does it for relatively low cost (especially if you need eg high bandwidth).
> I always wonder if going all-in on cloud services actually saves most people time...
Agreed. The apps and services I manage aren't very big compared to a lot of posters here, and my company is certainly not going to ever pay the big bucks for real talent.
So instead, I have to design everything knowing that my ops team is going to be mostly $100K / year "devops" guys with a few "cloud" certs but no real CS, dev, or even Linux knowledge (yes it's that hard to hire good people now (at the low rates my employer wants...)).
I've gotten to the point where I absolutely mandate that they don't try to use Terraform or CloudFormation scripts, because in the end, there are so many edge cases or glitches that it's easier to just write an install guide that shows them which buttons to click in the AWS or Azure console. <sigh>
And when I look at the costs we spend per month - including all the unanticipated charges like NAT Gateways and $20 / day "managed" Postgres instances, I assume we'd be better off dumping the cloud and reverting to our 2008 setup: Spending $10K on some Dell servers in a managed data center and hiring an old-school Linux admin to manually install and manage it all.
It doesn't seem to. My current place is all in on AWS and we have about 6 "DevOps" people dedicated to dealing with it all.
Last place we did pretty similar stuff with 3 data centers, all self hosted, self managed, mostly OSS stuff with about 6 sysadmins and way less hassle.
