Comment by nijave
3 days ago
I've also noticed AWS tends to have less "magic" global services and tends to favor cell architecture with partitions and isolation.
These super duper magic global services seem to be the cause of most outages since the blast radius is so huge.
On the other hand, the proposition of a magic, infinitely scaleable service endpoint is nice from a developer perspective.
Even on AWS, if you go for the managed magic version of the thing, they'll make you pay more, lose some flexibilitym and the relinquished control will change things in a way that benefits AWS (slower scaling, limitations, unnecessary overprovisioning, overhead).
An example - if you scale things manually by provisioning and starting EC2 instances via API - it will be more performant and cheaper than either Lambda or ECS Fargate (or Batch...). But those things at least work reliably.
With the other two cloud providers, you'll likely run into a bug you cannot fix yourself, and you will have no support to help you.