mce: [Hardware Error]: CPU 0: Machine Check: 0 Bank 4:

[GeNe64]

The server was ordered as a dedicated server, I have remote access only.

I think I would be communicating with the provider of this server, if it is hardware, and it sounds like it is, then someone with physical access will need to do the "mechanic" part.

They've tested it and said it's ok.

[GeNe64]

Only a random thought, same challenges faced by Debian stable users with newer hardware on (desktop), certainly has to hold true in any other use(servers.) Perhaps consider installing a newer kernel and firmware versions etc. Rather than downgrading would likely be going the other way, hopefully providing improved support for the chosen hardware. Should install some monitoring software onto a server anyway. Dead simple way to rule out hardware and determine if it is Os misconfig, install a gnu/nix distro like Ubuntu onto it, does that install show the same quirks and problems ?

If runs smoother/better w/o displaying similar negative behavior, hardware's fine, Debian's not setup right for that system.

I need only Debian to install Proxmox on it. When I run it unloaded (OS, Proxmox and a 10 unloaded VMs) it's ok and works for 6+ days. If I start to load VMs (CPU, SSD, Network) then it crashes or something else in 1/2 days on both servers in different datacenters.

I can't find any useful in standard logs. Any suggestions regarding monitoring software?

[Deb-fan]

Nope .. only vaguely aware of wth proxmox even is. A virtualization container type thing. Badly lacking in learning about virtualization all around. So what/which tools or how to approach trouble shooting it(proxmox) is beyond me ... Sorry, though from what those techs apparently said to you and what you're describing, not hardware but software problems. If they have help forums and likely do would spend time and ask the people there, who use that software for help n pointers on running down issues and possible fixes.

[LE_746F6D617A7A69]

The server was ordered as a dedicated server, I have remote access only.

I think I would be communicating with the provider of this server, if it is hardware, and it sounds like it is, then someone with physical access will need to do the "mechanic" part.

They've tested it and said it's ok.

Thermal throttling alerts are saying something else - the machine has a problem with cooling, what suggest that f.e. it's too hot in the server room.
Temperatures from SMART could show a more clear picture of what is happening in that Data Center.

[Deb-fan]

Come on now. The server room is too hot in a couple datacenters? Errr or, incorrectly configged software w processes hammering hell out of cpus causing heat and crash issues ? Which of these more likely? Still again ... asking this in a mostly desktop oriented Debian gnu/nix community vs asking in proxmox or kvm ones?

[LE_746F6D617A7A69]

Thermal throttling means insufficient cooling - this is a fact, not a guess.

[CwF]

Any tips?

...get your money back.
First off, that isn't exactly server grade hardware.
Most server grade stuff would never throttle due to core temp. The socket temp would be the trigger in a proper setup.

[LE_746F6D617A7A69]

Any tips?

...get your money back.
First off, that isn't exactly server grade hardware.
Most server grade stuff would never throttle due to core temp. The socket temp would be the trigger in a proper setup.

+1

[GeNe64]

Finally, I've solved the issue by adding intel_idle.max_cstate=1 to the file /etc/default/grub

Code: Select all

GRUB_CMDLINE_LINUX_DEFAULT="consoleblank=0 intel_idle.max_cstate=1"

Code: Select all

# update-grub

and rebooting.

[LE_746F6D617A7A69]

Finally, I've solved the issue by adding intel_idle.max_cstate=1

This makes completely no sense.
In Your previous post You said that the connection was lost after a series of warnings saying that the critical temperature has been reached. Practically this means, that the CPU has nearly melted down (max Tjunction is 100deg.C, and the treshold is 95deg.C).
The kernel parameter intel_idle.max_cstate=1 completely disables power saving - how could it help in this situation?
Yes, there was a problem with hard lookups in BayTrail CPUs, where this option was used as a workaround - but this not the case here.

I think that some other factors could have come into play here - like f.e. someone have "fixed" a problem with air conditioning system by opening all the doors and windows in that not-so-cold room

[GeNe64]

This makes completely no sense.
In Your previous post You said that the connection was lost after a series of warnings saying that the critical temperature has been reached. Practically this means, that the CPU has nearly melted down (max Tjunction is 100deg.C, and the treshold is 95deg.C).
The kernel parameter intel_idle.max_cstate=1 completely disables power saving - how could it help in this situation?
Yes, there was a problem with hard lookups in BayTrail CPUs, where this option was used as a workaround - but this not the case here.

I think that some other factors could have come into play here - like f.e. someone have "fixed" a problem with air conditioning system by opening all the doors and windows in that not-so-cold room

Yep, that's the problem. The bug is very strange and described here https://forum.proxmox.com/threads/rando ... 597/page-3
It's not possible to find anything useful in logs but server crashes all the time.
I was trying to link any weird messages in logs (temp, mce, etc) and crashing but couldn't resolve it anyway.
It's a bug of Intel CPUs that can be fixed by adding intel_idle.max_cstate=1