[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-users] Dom0 crashes without logging lately on Debian Stretch with Xen 4.8



Thanks to both of you for your detailed information. So as you both do not have the intel-microcode package installed it can't be that the issue. I do not make use of that package either myself. So what is left? Well it looks like I am running on older hardware, at least 5 years old hardware and who knows if this has some kind of influence. It might be interesting to get in touch with the hardware manufacturer (DELL?) and ask them if they have other customers with this issue. The only problem here is that as soon as you mention Debian they will stop listening to you :( If I remember correctly they only take support cases for supported commercial Linux distributions which basically boils down to RHEL and SLES... Maybe the DELL forums would be a better alternative. I would definitely recommend filling a bug issue with Debian and maybe even Xen... If you have some kind of stack trace that would also be interesting to see.

J.

On Tue, Nov 6, 2018 at 9:37 AM Roalt Zijlstra | webpower <roalt.zijlstra@xxxxxxxxxxx> wrote:
Hi John,

Yes, we are using PV only and we only run Debian Linux on the servers. We still have some DomU Jessie servers running with the stock kernel. We did update our Dells to the latest firmware so it does include more recent intel microcode with that. But on Debian we did not yet enable the intel-firmware yet, since we had so much instability and so much parameters that could be the culprit, we did not want to add another.
If your server is very busy, I think the chance to have a crash is higher. We have seen crashes on our active MySQL databases whereas the slave MySQL database server did not crash that quickly, however after using the slave MySQL database as primary database for a while (because we were debugging the crashed master database) it could very well happen that the slave would crash too.

We have done tests with downgrading firmware of Dell (which also means using an older intel microcode) but that did not help. So having the latest firmware is okay. 
We are now testing a few scenarios:
  •  one server with an older kernel (4.9.0-4-amd64), with DomU 3.16 kernel, which runs for 16 days now
  •  one server with the updated -kernel (4.9.0-8-amd64), with DomU 3.16 kernel, which runs for 28 days now surprisingly
  •  one server with the updated -kernel (4.9.0-8-amd64), and all DomUs on the backported 4.9 kernel.
It all doesn't really make much sense. We do have the expectation that the older kernel will keep on running and that the 4.9 DomUs will help to keep the servers alive. 
We have tested with 4.14 and 4.16 kernels (from backports) but that did not make a difference in stability.

Best regards,
 
 Roalt Zijlstra
  Teamleader Infra & Deliverability
   
 roalt.zijlstra@xxxxxxxxxxx
 +31 342 423 262
 roalt.zijlstra
 https://www.webpower-group.com
 
 
Facebook Twitter Linkedin
Barcelona | Barneveld | Beijing | Chengdu | Guangzhou
Hamburg | Shanghai | Shenzhen | Stockholm
 


Op ma 5 nov. 2018 om 18:24 schreef John Naggets <hostingnuggets@xxxxxxxxx>:
It could be as you mention... your domU are they PV? I am using paravirtualization exclusively and on this specific server have the following CPU:

Intel(R) Xeon(R) CPU           E5645  @ 2.40GHz

Do you have the intel-microcode Debian package from the non-free repo installed on your servers? I currently don't...

J.


On Mon, Nov 5, 2018 at 3:04 PM Roalt Zijlstra | webpower <roalt.zijlstra@xxxxxxxxxxx> wrote:
Hi John,

It could very well be that it is also restricted to some CPUs, but I am inclinded to believe that the used DomU kernels can influence stability.  We did have a pretty busy SSL offloader running on a 3.16 kernel, which might have caused the crashes. 

Just for reference, we have the following two CPUs causing us trouble, but I am not sure if it matters.
Intel(R) Xeon(R) CPU E5-2640 0 @ 2.50GHz
Intel(R) Xeon(R) CPU E5-2670 v3 @ 2.30GHz

Roalt


Op ma 5 nov. 2018 om 10:45 schreef John Naggets <hostingnuggets@xxxxxxxxx>:
Hi,

Thanks for your feedback. I was wondering because I have just upgraded a Debian 9 server to the latest kernel with the latest Xen packages from the official Debian repo. The only difference is that I have an older IBM server which is already ~7 years old patched with the latest BIOS/UEFI and so far so good no crash. The uptime is 6 days for now. Here are the details about my kernel and xen packages.

ii  xen-hypervisor-4.8-amd64       4.8.4+xsa273+shim4.10.1+xsa273-1+deb9u10 amd64        Xen Hypervisor on AMD64
ii  linux-image-4.9.0-8-amd64      4.9.110-3+deb9u6                         amd64        Linux 4.9 for 64-bit PCs

Regards,
J.


On Fri, Nov 2, 2018 at 7:57 PM Volker Janzen <volker@xxxxxxxxxx> wrote:
Hi John,

the problem is that I cannot provide any metrics or logfiles showing an error. I can only tell that dom0 is rebooting for a reason that is not logged. I have no physical access to the server. I got one other report about this kind of issue.

My assumption the cause are the backported patches is based on the current 16 day uptime. 16 days ago the server rebooted every 3-5 days. It won’t be a useful bug report from my point of view.

The other thing is that my two servers are now running upstream Xen and kernel and I might not go back to both old versions in Debian stretch. The other server had always running upstream versions and had never a problem, that’s why I updated the other, too.


Best regards
    Volker


Am 02.11.2018 um 17:23 schrieb John Naggets <hostingnuggets@xxxxxxxxx>:

I was wondering if any of you guys reported this bug/issue/problem back to the Debian community? For example on their bugs.debian org web site?

On Thu, Nov 1, 2018 at 1:47 PM Volker Janzen <volker@xxxxxxxxxx> wrote:
Hi,

I had these crash problems with the Xen version in Debian stretch, too. After 3 to 7 days the Xen server rebooted without log entry or something else to observe. The problems started when the first patches were applied by Debian. Some updates made it better, the last worse again. I checked hard drives, RAM and closely monitored metrics what might be the cause.

My solution after no longer suspecting a hardware fault: build upstream Xen 4.11 for Debian stretch. I am currently running this setup with my own build of kernel 4.19. The machines are now working stable again.


    Volker


Am 29.10.2018 um 13:13 schrieb Roalt Zijlstra | webpower <roalt.zijlstra@xxxxxxxxxxx>:

Hi there,

Ever since all the Meltdown and Spectre kernel updates and possibly also Xen 4.8 updates, we experience crashes of the Dom0 just out of the blue. Sometimes after 1 day, sometimes after a few days or even 14 days, completely random.

We have two Dell P730 servers and two Dell P720 servers with this behaviour. One thing is that we updated these machine to the latest available firmware, because that is the most secure way. Then we installed Debian Stretch with Xen 4.8 support

We have done serveral installs and 4 servers seem to crash pretty fast and other don't. In the end we think that we can lead it back to the xen-4.8.4-pre version being stable and the xen-4.8.5-pre being unstable. This was kinda independent of the kernel that we were using 4.14 or 4.9.0-8-amd64. This is off course all Debian package numbering.

As last resort  we updated on one server all DomU kernels of our Jessie servers on this Dom0 to 4.9.0 from backports instead of the 3.16 kernel. For now that seems to work, but the crashes are random so it could happen any time again. The idea is that these kernels are completely spectre& meltdown unaware and might cause trouble in Xen kernel support. I am not sure if this is true at all, but we are pretty lost what the actual cause is.

We also tested with CentOS and we also had these crashes there with certain combinations of kernel/Xen. The most recent updates seem to be more stable tough. The most frustrating part is the there is absolutely no logs to be found. No kernel oops or what.. the server just resets and boots again.

Are there others experiencing problems like this? Do you see more frequent server/kernel crashes on production servers?  

Best regards,
 
Roalt Zijlstra

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-users

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.