Glad to see you post this here. I’ve been experimenting with selfhosted analytics for a while now and have attempted your project here a couple times. The thing that kills me is the Clickhouse requirement. It makes it impossible to host on a lightweight VPS. Like why should my analytics platform require so much more compute than my simple static site? Am I missing something?
Okay this is excellent content, thank you!
I went through and fiddled with some more stuff to try and get this working to no avail. However, it inspired me to take apart netboot.xyz a bit more, and I was able to grab an efi and get next boot to load the efi file. It took me too long to realize you need the console tty arguments as part of the boot cmdline to get it working interactively, but after I got there I got it netbooted. Sadly though, it almost immediately runs into an OOM condition and thus isn’t practical on a free tier x86 asset. It would probably work on an aarch64 node, but I already have my allotted arm node spun up and working so I don’t have a free one to practice with.
Solid write-up though, thank you for putting that together!
The “gotcha” with Oracle free tier is that you can’t install from arbitrary media, so the typical netboot.xyz or any iPXE workflow is out. No console access, no pre-bootloader access, nothing.
I’ve been fiddling with kexec, but it doesn’t seem like a supported method of loading the lkrn file from netboot…
This is super interesting to me, so by all means, if you have the kung-fu to show how this works I would happily read through that!
Interesting. I’ve had two instances running for over 2 years and haven’t noticed that. It might be that I just don’t notice it though. I’m not scrutinizing it much.
I am not well versed with kexec but I always understood it to be a kernel reboot without power cycling the “metal.” Please enlighten us with an example! I don’t see how you’d replace the entire userspace (and possibly filesystem) with simply kexec.
Certainly! As others have said, don’t hang anything worth value on it without an out of band backup strategy, they’re famous for unscrupulously deleting things with no warning. Oracle is a miserable company.
Free is free though!
I would recommend giscus over discus, but yes, certainly a valid approach!
I have been pleased with giscus on my blog (http://roguesecurity.dev/ ) Its powered via github discussions.
Agreed, prosody is great! I’ve been doing some experimenting with ejabberd and it seems more enterprise-ready, but I haven’t found anything that is discernable as far as feature advantages.
Sounds like a great opportunity to breath some life into it! If you really have the itch for IRC, there’s a slidge bridge to connect IRC to XMPP!
Agreed! Runtime environment management is so much nicer with modern containerization. You or ally can’t overstate how much better it is to have app stack state be entirely divorced from OS state. I’m very pleased they’re back on the bandwagon as well.
Stand up a server and come join our MUC!
UPDATE: For anyone who comes back to this, or any new readers – I have added a MUC (chat room) on my XMPP server for discussion of any tech-related things, akin to the subject-matter of this blog. Hope to see you there!
I have experimented with Simplex, but it feels less tuned toward hosting federated infrastructure and more tuned toward participation with the greater network in a pseudo-anonymous fashion.
Adoption is also always a hurdle with any ecosystem like this, and XMPP is certainly ahead of Simplex in that avenue.
It has a long healthy life ahead! Come join the party, the proof is in the pudding.
😆 +1 for reading enough to see that! Thank you!
I’m one of those people that ends up using the vocabulary I once learned to get the most value out of it. Would hate to waste all that. Haha.
This is also a great article! Thanks for the link.
One cool point in favor of XMPP is that in a public setting (MUCs), there’s community. Moparisbest is an active participant in several of the MUCs that I’m in. Very cool!
Yeah they just redid their container image pipeline and these containers are the result!
Super true. I think this was best exemplified by SignalGate
This is great, I have not seen this post before. Thank you for sharing.
You make an excellent point here, that the burden of security and privacy is put on the user, and that means that the other party in which you’re engaged in conversation with can mess it up for the both of you. It’s far from perfect, absolutely. Ideally you can educate those that are willing to chat with you on XMPP and kill two birds with one stone, good E2EE, and security and privacy training for a friend. XMPP doesn’t tick the same box as Signal though, certainly. I still rely heavily on Signal, but that data resides on and transits a lot of things that I don’t control. There’s a time and a place for concerns with both, but I wanted to share my strategy for an internal chat server that also meets some of those privacy and security wickets.
Hey thanks so much for the engagement. I was trying to run it on a VPS that cost $35/year. 2GiB of RAM wasn’t quite enough to make it work for me, granted that was with the webserver and ancillary supporting services.
I’ll find an opportunity to test it out though, as rybbit looks great. I appreciate the mention on the other FOSS products, that’s a good look for you. I have plenty of experience with umami already. Cheers!