🎓 Doc Freemo 🇳🇱 @freemo@qoto.org

@alexa@fedi.absturztau.be
Thats the one :)

@snder

@snder
I have the bulk of it saved as a gist right now that actually has all the relevant steps in it just with no text to accompany it. your welcome to take a peek, just not sure how helpful it will be in its current form

@alexa@fedi.absturztau.be

@alexa@fedi.absturztau.be
Thanks I probably will eventually. I documented every step in detail with that in mind.

@snder

@snder You are doing great :)

@snder Been running production servers for many decades now :) It pays to do it right.

@snder Since we will be working together soon I'll show you around the system when we get together.

@snder It can be, particularly if you use it like a traditional system and just bring up a server and ssh into it and customize it and leave it up 24/7 as most old-school admins are use to. If that is all you intend to do AWS will be a waste of money and more expensive, and potentially by a pretty big margin.

That is not the power of AWS nor how it should be used.

The power of AWS is you have programatic control over requisitioning and configuring every level of the hardware from network switches, to instances, to firewalls, to DNS. Stuff like autoscaling with a few scripts to requisition new hardware becomes trivial.

As such your intended to use it that way. There are basically three types of isntances you can get: Reserved, On-Demand, Spot. An On-Demand instance is the most expensive but its also what people use when they are using AWS like a normal system. With on-demand you rent a server for as long as you like, it never goes down, then you terminate it when you wish and only pay for the time you use. It is very expensive and only intended for short periods of time where you need a reliable server. Reserve instances are half the price, but you get locked into buying them for a year or more up front. So it only really saves you money if you know you will use it for long-term. They can be half the price of on-demand or less. Spot instances are the cheapest of them all and even cheaper than most options from other hosts (by a big margin). These are instances you bring up like on-demand, but you bid for them and basically the price fluctuates. But they are always cheaper than the above two options, often by as much as 90% cheaper. The caveat with spot instances is that if the owner of the instance wants it back to use it at any time the server might get terminated int he middle of what ever it is doing, so these are NOT intended for reliable uptime.

By using a scaling cluster as I defined though instances can go up and down freely. As long as at least one instance is up than another instance going down doesnt hurt anything. So all my instances are spot instances.

this is how AWS was intended to be used. Its a fraction the cost of traditional hosts, significantly more stable (due to redundancy), and auto-scales for load.

So long story short its only more expensive if you dont know what your doing or you arent doing it right :)

@snder It also autoscales so if traffic on any of our servers start to max out it automatically runs additional servers to handle. This idea also applies to the gitlab runners.

Everything is elastic!

@snder I just used other peoples existing work and adapted it to aws. But yea its really crazy cool setup that is insanely easy to admin

@snder Neither exactly. All the qoto services I engineered my own AWS pipelines for.

We use ECS with some custom hacks that bring down cost as well as make it portable to other non-aws servers.

Basically i use a nginx reverse proxy container along with a companion container that automates load balaning and SSL certificates. The way it works is whenever i add any container of any kind to the cluster I simply set two environment variables telling it the domain name it will be hosted on and whihc port it exposes the web server on. At that point the load balancing container automatically detects the new docker container, reads the variables, and creates a new reverse proxy link. It then automatically goes and obtains/applies a new SSL certificate from lets encrypt and applies it to the reverse proxy link.

So basically to get gitlab to work i just brought up a container with the proper settings and everything magically worked, just like my other services.

@snder I'm less confident of that.

@snder Regardless it will be fixed eventually

@snder Thats not an invalid cert your seeing. The cert is valid. We just pull in javascript from http sources not just https which chrome is weird about and reports as insecure. If you ook at the cert itself its actually completely valid.

@snder on https? certs valid what do you mean?

If there is one thing about America since 9/11 that is clear to me as an American who fled the country for a better life in Europe... the terrorists won.

#uspol #terrorism

#QOTO update.

I recently added auto-scaling high-power CI runners to our gitlab instance which is free and open to all open-source projects. They even autoscale to keep up with demand.

We include three types of CI runners now, Default compute, GPGPU, and FPGA capable runners.

https://git.qoto.com

@mngrif sorry ill send you link now in prvate message