Posted: 2026-08-29
OpenBSD VPSs on Slow Servers hosts can have stability issues related to the clock. Debian VPSs don't seem to have a problem. This is under active investigation. It appears that the AMD hosts, possibly related to the switch from HDD to SSD, are impacted more than Intel. I have experienced no stability issues on Prometheus (7.8, Intel, HDD), where I have several servers.
/etc/sysctl.conf:
kern.timecounter.hardware=tsc
I am not 100% confident that this resolves the issue, but it may be worth trying.
vmctl pause myhostname vmctl unpause myhostname
These commands may hang for you and you may have to ctrl+c after each one -- that's a different issue, I think.
One of my first customers wrote to me about stability issues on their OpenBSD VPS.
This was very puzzling because I "dogfood" and host almost all of my servers on Slow Servers. Specifically, Prometheus. I haven't been having any stability issues.
I've also had a lot of customers on Loki and Odin, no complaints there. Most of those customers have been using Debian.
I wasn't able to reproduce the issue until I migrated a VPS of mine, "potato" from Prometheus to Odin. I migrated for a few reasons -- partly in hopes of phasing out Prometheus and Ullr, but also to see if I might be able to reproduce the issue. Sure enough, in a few days it appeared.
Basically, the clock gets stuck and "loops" over a 5 second period. This does all kinds of weird things. For me, Caddy was working, Prosody was not, and SSH handshakes would start and not finish. I was able to login to the serial console, though!
I wrote to the bugs@ OpenBSD mailing list and Dave Voutila quickly responded suggesting I try to "pause" and "unpause" the server. This immediately fixed the issue.
This investigation is ongoing. I wanted to post it in case anyone else has been having head scratching issues. If you do, please let me know! Feel free to try the pause/unpause as well and let me know if that took care of it.
I will be testing/patching as soon as I can. OpenBSD is a first class VPS for me (almost all of my servers are OpenBSD) and of course the hosts run OpenBSD as well. There's still some questions about impact and scope. There may be a patch that will fix this. It sounds like it's a fairly particular bug that others have seen, but isn't easy to reproduce reliably on any hardware.
There's also no guarantees, to my knowledge, that everyone will run into it. But I would say that my workload on "potato" was not a lot, and it was able to do it in a few days after being stable prior to migration.