After briefly reading about systemd’s tmpfiles.d, I have to ask why it was used to create home directories in the first place. The documentation I read said it was for volatile files. Is a users home directory considered volatile? Was this something the user set up, or the distro they were using. If the distro, this seems like a lot of ire at someone who really doesn’t deserve it.

I have a similar issue when I am visiting my parents. Despite having 30 mbps upload at my home, I cannot get anywhere near that when trying to access things from my parents house. Not just Plex either, I host a number of services. I’ve tested their wifi and download, and everything seems fine. I can also stream my Plex just fine from my friends places. I’ve chalked it up to poor (or throttled) peering between my parents ISP and my ISP. I’ve been meaning to test it through a VPN next time I go home.

Here’s a drawing of what I think might be happening to your private traffic: traffic diagram

One major benefit to this approach is CloudFlare does not need to revoke an entire public certificate authority (CA) if a singular private tunnel’s Certificate Authority is compromised.

I somewhat wonder if CloudFlare is issuing two different certs. An “internal” cert your servers use to serve to CloudFlare, which uses a private CA only valid for CloudFlare’s internal services. CloudFlare’s tunnel service validates against that internal CA, and then serves traffic using an actual public CA signed cert to public internet traffic.

Honestly though, I kinda think you should just go with serving everything entirely externally. Either you trust CloudFlare’s tunnels, or you don’t. If you don’t trust CloudFlare to protect your services, you shouldn’t be using it at all.

Just serve the CloudFlare certs. If the URL is the same, it won’t matter. Doesn’t matter if you’re talking to a local private address like 192.166.1.100 or a public IP. If you’re accessing it via a DNS name, that is what is validated, not the underlying IP.

PS. If you tried this and are having issues. We need more details about how things are set up, and how you are accessing them.

I think I misunderstood what exactly you wanted. I don’t think you’re getting remote GPU passthrough to virtual machines over ROCE without an absolute fuckton of custom work. The only people who can probably do this are Google or Microsoft. And they probably just use proprietary Nvidia implementations.

I believe what you’re looking for is ROCE: https://en.wikipedia.org/wiki/RDMA_over_Converged_Ethernet

But, I don’t know if there’s any FOSS/libre/etc hardware for it.

If you are fine with the slim: US amazon.

I’ve heard good things about used/refurb HP (elite desk and pro desk) and Lenovo (m700 and m900) mini-pcs. A quick search shows they’re going for ~120-140$ for a quad core with 16 gigs of memory.

Check out minisforum, for example this intel mini-pc. They have a ton of selection, not just that one example.

In the LastPass case, I believe it was a native Plex install with a remote code execution vulnerability. But still, even in a Linux container environment, I would not trust them for security isolation. Ultimately, they all share the same kernel. One misconfiguration on the container or an errant privilege escalation exploit and you’re in.

You are not being overly cautious. You should absolutely practice isolation. The LastPass hack happened because one of their engineers had a vulnerable Plex server hosted from his work machine. Honestly, next iteration of my home network is going to probably have 4 segments. Home/Users, IOT, Lab, and Work.

Keep in mind, RAID is fault tolerant, not fault proof. For critical data, keep in mind the 3-2-1 rule. Stored in 3 locations, 2 separate mediums, 1 offsite.