Obtainium seems to have a very interesting take on this. Thanks for the link! I will check it out 👍

Hi,

Just to put things into perspective.

Well, this example dates from some years ago, before LLMs and ChatGPT. But I agree that the principle is the same. (an that was exactly my point).

If you analyse this. The error the person made was that he assumed an arduino to be like a PC, .. while it is not. An arduino is a microcontroller. The difference is that a microcontroller has resources that are limited: pins, hardware interrups, timers, .. An addition, pins can be reconfigured for different functions (GPIO, UART, SPI, I2C, PWM, ...) Also, a microcontroller of the arduino-class does not run a RTOS, so is coded in "baremetal". And as there is no operating-system that does resource-management for you, you have to do it the application.

And that was the problem: Although resource-management is responsability of the application-programmer, the arduino environment has largly pushed that off the libraries. The libraries configure the ports in the correct mode, set up timers and interrupts, configure I/O devices, ...And in the end, this is where things went wrong. So, in essence, what happened is the programmer made assumption based on the illusion created by the libraries: writing application on arduino is just like using a library on a unix-box. (which is not correct)

That is why I have become carefull to promote tools that make things to easy, that are to good at hiding the complexity of things. Unless they are really dummy-proof after years and decades of use, you have to be very carefull not to create assumptions that are simply not true.

I am not saying LLMs are by definition bad. I am just careful about the assumptions they can create.

Yes, that was indeed the question.

If I read it correct, you need a specialised distro for this. You cannot do this on a off-the-shelf Debian or Ubuntu?

I'll do some searching on 'unmutable Linux'. Thanks for the (very quick) answer! 😀

Well, based on advice of Samsy, take a backup of home-server network to a NAS on your home-network. (I do home that your server-segment and your home-segment are two seperated networks, no?) Or better, set up your NAS at a friend's house (and require MFA or a hardware security-key to access it remotely)

I have been thinking the same thing.

I have been looking into a way to copy files from our servers to our S3 backup-storage, without having the access-keys stored on the server. (as I think we can assume that will be one of the first thing the ransomware toolkits will be looking for).

Perhaps a script on a remote machine that initiate a ssh to the server and does a "s3cmd cp" with the keys entered from stdin ? Sofar, I have not found how to do this.

Does anybody know if this is possible?

In this case, it is not you -as a customer- that gets hacked, but it was the cloud-company itself. The randomware-gang encrypted the disks on server level, which impacted all the customers on every server of the cloud-provider.

First of all, thanks to all who replied! I didn't think there would have been that many people who self-host a SSO-server, so I am happy to see these replies.

As a side-note, I have also been looking into making the setup more robust, i.e. add redundancy. For a "light redundant" senario (not fully automatic, but -say- where I have a 2nd instance ready to run, so I just need to adapt the DNS-record if it is needed), can I conclude from the "makeing a backup" question, that I just need to run a 2nd instance of postgres and do streaming-replication from the main instance to the backup-instance ?

Or are there other caviats I haven't thought about?

For me, the first goal is to simply understand the setup. I now have been able to create a setup with two frontend jvb-instances and one backend. In the end, the architecture setup of a jitsi-server is quite nicely explained, and -by delving a little bit into the startup scripts of the docker-based jitsi setup, you do get some idea of how things fit together.

From a practicle point of view, I think I'll go for the basic setup (1 backend, 2 frontends) natively on two servers, and -if the backend server would go down- just have a dockerised backup-setup ready to go if it would be needed.

Thanks!

for the nextcloud instance on my local LAN , I use the .local domain (multicast DNS). Just enable avahi on your server and you can use hostname.local on your network without having to deal with local DNS on your router and so on.

Hi, I have it running as of today. apache reverse-proxy native on the server and "stable-8922" in docker.

I have been wondering if it makes sense to move the jvb from docker to the server. I guess that is the part of the system that pulls most of the traffic. I don't know if this make any real difference for performance or not.

Anycase. All, thanks again for the help. Appriciate it. :-)

Kr.

Hi Neutrom, I don't know this one. I'll check it out. Thx! 👍

Hi,

I have also been thinking about selfhoating a jisti-meet server. Just how easy / difficult is it to selfhost it? Do you run it in docker or natively? Linux or some other OS (FreeBSD)?

Kr.