Microservices and general “everything in the cloud” sentiment is stupid, it has ridiculous oerformance overheads and adds single points of failure that can easily prevent half the world from functioning.
I will disagree on microservices. I think it gets used for things that don’t really require it like anything that gets hype, it gets overused. But I will give you example of exactly why and how it can be used correctly.
My company was early on the Kubernetes bandwagon. Before Amazon and Google provided hosted Kubernetes solutions, I had to write and bootstrap clusters automatically.
We had all our microservices deployed on k8s when covid hit. We went from about 4000k visitors/hour to 400,000k visitors an hour. This happened over a single week. k8s scaled out perfectly, our services scaled out perfectly and we had zero hiccups or downtime. We didn’t have to do anything from an engineering perspective except increase our hard limits on the compute scale outs. During evening hours when load went down, the cluster would scale down.
If we were using a monolith application, we would have been really fucked. No way we could have scaled like that.
Microservices and general “everything in the cloud” sentiment is stupid, it has ridiculous oerformance overheads and adds single points of failure that can easily prevent half the world from functioning.
I agree and I hate it. As much as possible I want to work with files locally.
I will disagree on microservices. I think it gets used for things that don’t really require it like anything that gets hype, it gets overused. But I will give you example of exactly why and how it can be used correctly.
My company was early on the Kubernetes bandwagon. Before Amazon and Google provided hosted Kubernetes solutions, I had to write and bootstrap clusters automatically.
We had all our microservices deployed on k8s when covid hit. We went from about 4000k visitors/hour to 400,000k visitors an hour. This happened over a single week. k8s scaled out perfectly, our services scaled out perfectly and we had zero hiccups or downtime. We didn’t have to do anything from an engineering perspective except increase our hard limits on the compute scale outs. During evening hours when load went down, the cluster would scale down.
If we were using a monolith application, we would have been really fucked. No way we could have scaled like that.