Not serverless, not edge, not Kubernetes, not auto scaling | Hacker News Reader

Not serverless, not edge, not Kubernetes, not auto scaling

twitter.com

3 pointstosh5 comments

I am not really sure who this argument is for, who is arguing to run a blog or small site on K8s? Your $44k a month business is chump change in the world of tech, that is about $1 per minute. You can have a 2 hour outage and only lose $120.

Sure, for a personal blog or small business its fine, but having done SRE work for years I will tell you right now any one server will absolutely fail and it will happen at 4 in the morning.

At a previous job, we had revenue in the hundreds of thousands of dollars per minute, especially when we ran large ad campaigns during football games and such. Would you risk running that on one server?

What about a EMR / EHR? I sure as hell am not running an in-patient system on one VPS...

Once you commit to 99.99% it always gets more complicated, but that is part of growing up as a company.

having moved from owned and operated to aws, even products that went live without proper redundancy (literary one rh6 box in a data center) had less outages than our fully redundant aws deploys, which all have been far from advertised uptime all thanks to their outages.

but i don't know what the"article" talks about as i won't click xitter links

I get that, I have had more server failures in AWS than bare metal for sure. Thankfully with ECS / Aurora they are all seamless, but running EC2 alone has to be something like 99.6% uptime max. Although I question how you are fully redundant and having downtime :)

I am an advocate for rolling your own hosting, we are getting ripped off by hosting providers, even DO is terribly expensive now. You get a much better deal going with OVH (and their other brands) than any other host. Not to mention you get to avoid noisy neighbor problems. But you have to understand what its limits are, and when to switch off that mindset.

Also running all your sites on one VPS is madness, the blast radius of a problem is 100%, when you should be able to easily segment by application.

oh we don't have downtime when one ecs instance goes down... we had it when the entire feature they offered went down. but i don't deal with aws for over two years. i hear it got better. we always had from day one all blessed architecture from their consultants-salesman, dual region, quad vps, green blue ecs deploy, the whole shebang

...i forget that but your point is very good. cloud is marginally better than DO and such. sorry i forgot people use these things since a pipe is so cheap in the last decades.

I went to Costco yesterday and bought three cases of milk and four cases of soda, and it all fit into the back of my Honda civic hatchback! Doesn’t this prove that 18 wheelers are useless once and for all.

But really, I’ve never met an engineer who said that you shouldn’t try to keep things as simple as possible as long as possible.

Reply on news.ycombinator.com