Hacker News Reader

Comment by godot | Hacker News Reader

These stories almost always boil down to this fundamental conflict of what you want for a personal project vs a business. (though in this case yes, Troy Hunt's HIBP is larger than a lot of startup businesses)

In a business setting, you want your service to stay up, at the cost of spike in costs if accidents or mistakes happen.

In a personal project, you want there to be hard limit on cost, and your service to go down if spikes call for it. (I'm relatively sure that no one wants their personal projects to incur a bill of thousands of dollars by accident.)

View on HN

> In a business setting, you want your service to stay up, at the cost of spike in costs if accidents or mistakes happen.

No you don’t. This is absolutely not a given. Being a “business” doesn’t mean you suddenly have unlimited budget.

The vast majority of businesses are not “web scale” and are better off taking an availability outage than suddenly handling 1,000,000x the normal volume of traffic.

I'd say that it definitely depends on the business.

If you are selling you product via your web site and you're suddenly on TV with millions watching and accessing your site, you definitely don't want the server to go downand autoscaling + a bit higher cost would be great.

Only if you can actually fulfil those orders. If your production can't be increased, and the business is small, the cost could far outweigh the potential profit.

Not if you’re selling a real product. The only world where people can scale from 1000 orders a day to a million orders a day overnight is one where you’re not selling anything physical.

I don't agree with this - even for businesses there is always a limit over which there is serious trouble for bottom line. I think cloud providers should allow one to set a hard cost limit over which everything shuts down. For personal projects the limit might be $100 and for small businesses $100k, but even rich companies have it (not the same reason, but Knight Capital comes to mind).

Certainly the cloud providers probably make money by not having hard limits.

But it's also the case that if they did implement hard limits of some sort, you'd be reading blog posts about how AWS destroyed my project just when it was going big because someone stuck a circuit breaker foot gun in some corner and everything stopped working properly when usage spiked.

I do think there should probably be a hard circuit breaker. It should be simple and therefore inflexible. And it should come with a big warning sign. Still people will get burned because someone will set it, a project grows, and one day it goes off.

But... they do implement limits.

As far as I know, they implement billing alerts but, aside from some student and some other limited account types, they're alerts. You'll get an email that you've hit your limit but your bill will continue to go up until you shut things down.

And do note that these alerts are not instant. With Azure, if their backend reports and alerts are timed wrong, you're still on the hook for 2-3 days' worth of costs.

Aws doesn't, there are only alerts. Don't know about others.

They have limits on things like how many EC2 instances you can have but not on things like bandwidth.

While you can raise those limits by request I'm also not sure whether you can actually reduce them again later.

Azure (and I'm sure other cloud providers do) allow you to set email notifications for when your bill goes over a set amount so you can stop it before it happens.

If you're using a cloud provider I'd highly recommend setting one of those up.

In Azure it's under your Subscription and then Budgets

AWS can send you alerts when it looks like you will go over your budget for the period, before you're way over your projections, which is a nice feature.

I truly believe they want you to use a lot of their resources on a consistent, long-term basis; they don't get long-term value from people having short, one-off anomalies, so budgets and monitoring are aligned with their customers - just not total cost of ownership calculations :)

Is it really that black and white? I think there is a continuum in hosting service. Not just A) very low end VPS, and B) unlimited cloud.

The fact is that there are low end VPS, middle end VPS, high end VPS, and dedicated servers. If you started from a low end VPS, it is very easy to gradually upgrade your VPS.

A $5/month VPS can be used to play for tons of things. I just don't get people who use free tier cloud, unless you just want to learn about the cloud hosting per se.

The problem is that the cloud alone will not allow you to scale infinitely. Slapping a standard DB-based application on the cloud will not magically make it scalable. RDS will still be a bottleneck and can't be scaled online for example (Aurora might change that, but that's a very recent development).

Making your application scalable is a significant effort that may involve different trade-offs. Your typical Prestashop or Magento e-commerce site will still max out the DB and go down, cloud or not, but with the cloud you'll end up with a huge bill in addition to your downtime.

Engineering your application to be scalable is an option that's often not made for cost/time to market reasons which is fine, but in this case the cloud will give you much less scalability than a lot of people believe.

Isn’t that why services such as AWS lightsail and digital ocean exist?

Reply on news.ycombinator.com