Description:  I updated to 6.3.1 last night, and the server seemed to restart fine but when I got up this morning to start work, I noticed that all of my VMs were down.  I tried to start one up manually and get a popup execution error with:

 

Invalid value '1' for 'cpuset.cpus': Invalid argument

 

I went to the dashboard to see if there was some sort of error, and this is what my CPUs/cores look like (http://i.imgur.com/WZGp4iy.png).  The number of CPUs is still correct (16), but I guess it just isn't using odd numbers for some reason?  When I go to edit the VMs, it still shows the normal count and numbers, CPU 0-CPU 15.

 

I'm not sure if the diagnostics have this information, but my system is a HP Proliant DL380 G6 w/ 2 E5540s (quad-core)

 

How to reproduce:  Update to 6.3.1 from 6.2.4

 

Expected results:  CPUs should be numbered 0-15 as they were before the update

 

Actual results:  CPUs are numbered with only even numbers (0, 2, 4,... 30), but VMs still show CPU 0-15 in the create/edit interface.

 

Other information:  VMs do start and seem to function as normal, as long as there are no odd-numbered CPUs, even if I edit the XML to use CPUs not listed on the VM edit interface.

tower-diagnostics-20170209-0744.zip

Can you reboot your system into safe mode and verify that this issue persists?  We're not seeing this on our own test systems.

I to am experiencing the exact same issue after the update. I get the message:

 

Execution error

Invalid value '9' for 'cpuset.cpus': Invalid argument

 

I have two CPUs each with 8 cores each. This ran just fine before the update.

I am also having a similar numbering issue.

My CPU's were numbered from 0-23 previously but now skip 12-15

cpu 0 / 16

cpu 1 / 17

cpu 2 / 18

cpu 3 / 19

cpu 4 / 20

cpu 5 / 21

cpu 6 / 22

cpu 7 / 23

cpu 8 / 24

cpu 9 / 25

cpu 10 / 26

cpu 11 / 27

 

The System Status does not show CPU % for CPUs 24-27

 

I get the same Invalid Value error when trying to use CPUs 12-15

I can manually edit the XML to change to use CPU's 24-27 and the VMs will run without error.

I upgraded yesterday from 6.2.4 to 6.3.1

 

Thank you

 

Edit to add my Specs in case it helps

Server: HP Proliant DL360 G7

CPU:  2 x X5675 3.07Ghz

Memory: 144GB DDR3

BIOS info: (I do not have access to upgrade as HP requires an active service contract for bios updates.

        Version: P68

        Release Date: 12/01/2010

        Firmware Revision: 1.15

HDD: 3x 1TB WD Blue

          1x 256GB corsiar SSD

 

 

Can you post the result of this command?

 

cat /sys/devices/system/cpu/*/topology/thread_siblings_list

 

Can you post the result of this command?

 

cat /sys/devices/system/cpu/*/topology/thread_siblings_list

 

 

0,16

1,17

10,26

11,27

0,16

1,17

2,18

3,19

2,18

4,20

5,21

6,22

7,23

8,24

9,25

10,26

11,27

3,19

4,20

5,21

6,22

7,23

8,24

9,25

 

I am also having a similar numbering issue.

My CPU's were numbered from 0-23 previously but now skip 12-15

cpu 0 / 16

cpu 1 / 17

cpu 2 / 18

cpu 3 / 19

cpu 4 / 20

cpu 5 / 21

cpu 6 / 22

cpu 7 / 23

cpu 8 / 24

cpu 9 / 25

cpu 10 / 26

cpu 11 / 27

 

The System Status does not show CPU % for CPUs 24-27

 

I get the same Invalid Value error when trying to use CPUs 12-15

I can manually edit the XML to change to use CPU's 24-27 and the VMs will run without error.

I upgraded yesterday from 6.2.4 to 6.3.1

 

Thank you

 

Edit to add my Specs in case it helps

Server: HP Proliant DL360 G7

CPU:  2 x X5675 3.07Ghz

Memory: 144GB DDR3

BIOS info: (I do not have access to upgrade as HP requires an active service contract for bios updates.

        Version: P68

        Release Date: 12/01/2010

        Firmware Revision: 1.15

HDD: 3x 1TB WD Blue

          1x 256GB corsiar SSD

 

Hi Liemander,

 

I'm going to send you a PM in a minute here.  What I'd like to do is give you a coupon code for a free services session that you can schedule with us so I can take a closer look at your system using a remote connection.  We haven't experienced the issue you're having and so far, it appears you're the only one experiencing this, but I'd like to get hands on to experiment a bit further.

I'm having a similar issue.

I'm on a HP ProLiant DL370 G6.
root@Test-Server:~# cat /sys/devices/system/cpu/*/topology/thread_siblings_list
0,16
10,26
12,28
14,30
0,16
2,18
2,18
4,20
6,22
8,24
10,26
12,28
14,30
4,20
6,22
8,24

 

I'm running on unRAID 6.3.2, sorry for poking the thread.

Edited March 6, 20179 yr by AnnabellaRenee87
Added version I am running and a screenshot.

On 3/6/2017 at 6:58 AM, AnnabellaRenee87 said:

I'm on a HP ProLiant DL370 G6.

 

Is there anyone having this issue who is not running an HP ProLiant motherboard?

I upgraded one of my dl380 g6 server to 6.3.2 (rest are still on 6.2.4) and can confirm the same issue. But I come bearing gifts: diagnostics on 6.2.4 from just before updating, and diagnostic from after. Not sure if they will be that helpful, but they can't hurt.

 

modified vm xml to reflect new cpu layout and it booted fine but is only showing usage on 2 threads of the 4 assigned... interesting stuff....

 

-editing to update as I experiment-

If I assign 8,10,12,14,  the dashboard shows usage on 4,6 (not reflected in images uploaded.)

If I assign 24,26,28,30, the dashboard shows usage on 12, 14. 

 

So, it is taking 2 assigned threads, halving the number, and slamming them on 1 host thread? The vm reports 4 cores and shows 4 cores of activity in both circumstances.

 

 

b3-diagnostics-20170314-0610 6.2.4.zip

b3-diagnostics-20170314-0624 6.3.2.zip

Edited March 14, 20179 yr by 1812
playing with cores

We bisected the kernel between 4.8 (working) and 4.9 (broken) and determined this problem was probably introduced by this linux kernel patch:

https://patchwork.kernel.org/patch/9346709/

 

We don't know how to make a workaround or why it has a issue with these HP servers in particular.  Seems to also not work with 4.10 kernel.  We're trying to contact the patch author.  Obviously no one in kernel development is using HP Proliants.  Usually with this kind of thing, eventually enough people complain and either a correction is made in the kernel, or more often, the manufacturer issues a bios update.

 

 

9 minutes ago, limetech said:

We bisected the kernel between 4.8 (working) and 4.9 (broken) and determined this problem was probably introduced by this linux kernel patch:

https://patchwork.kernel.org/patch/9346709/

 

We don't know how to make a workaround or why it has a issue with these HP servers in particular.  Seems to also not work with 4.10 kernel.  We're trying to contact the patch author.  Obviously no one in kernel development is using HP Proliants.  Usually with this kind of thing, eventually enough people complain and either a correction is made in the kernel, or more often, the manufacturer issues a bios update.

 

 

 

Thank you for the followup! 

Hi, 

 

yesterday i just wanted to try out Unraid, means I'm totally new too this hole thing.

I run into the same issue trying to start up my first VM. Like Liemander mention it works if i select a special number of CPU Core but in my case only one of 6 Cores. Also on the Dashboard i only could see 4 of 6 Cores working.

 

my Hardware specs:

Asus M5A99X EVO AMD 990X

AMD FX Series FX-6100 6x 3.30GHz

16GB Corsair Vengeance Red DDR3-1866 DIMM 

 

Is there any way to work around it? Or must i install a previous Version to the Stick?

 

Best regards

Edited April 6, 20179 yr by Tharosh

1 hour ago, Tharosh said:

Is there any way to work around it? Or must i install a previous Version to the Stick?

I found a work around is to to the main dashboard and look at what cpu numbers are listed under system status and only use those numbers.
If you want to use a cpu number that is listed under system status but not in the gui for the VM then you need to edit the XML to do so.

 

only use CPUs listed here

 

If you want to use one that isn't listed here

 

 

you can choose to edit xml instead of edit and change the vcpupin vcpu=... lines to what you want and it works.

 

Edited April 6, 20179 yr by Liemander
screenshots

1 hour ago, Liemander said:

I found a work around is to to the main dashboard and look at what cpu numbers are listed under system status and only use those numbers.
If you want to use a cpu number that is listed under system status but not in the gui for the VM then you need to edit the XML to do so.

 

only use CPUs listed here

 

If you want to use one that isn't listed here

 

 

you can choose to edit xml instead of edit and change the vcpupin vcpu=... lines to what you want and it works.

 

 

Thanks for the advise and example. I will go with that! At least i can use 4 Cores that's way for hopefully a short period of time. 
 

 

You should be able to use all cores thru this method. A better example would of been a VM I have set up to use all my available cores.

 

In my case my CPU list skips cpu cores 12-15 but adds additional cores 24-27. in reality I should only have cores up to 23.

 

In the GUI I can't use cores 12-15 because they don't exist according to unRaid.

 

What I did was assign all the cores anyway and then edited the XML after the fact.

I changed the broken cores 12-15 into cores 24-27.

 

What I found would occur for my dl380 servers is that in 6.3.x unRaid would put group 2 cores from a vm together on to one thread of a lower numbered core. Weird stuff. And I had no way of accessing the higher core number since they were always being combined. Workaround don't work for mine. Had to roll back to 6.2.4 to regain full usability of all cores.

 

Might play around with this again in a few days, but my numbering issue was different than yours, as it renumbered them on even numbers only.

 

cpu 0 <===> cpu 16
cpu 2 <===> cpu 18
cpu 4 <===> cpu 20
cpu 6 <===> cpu 22
cpu 8 <===> cpu 24
cpu 10 <===> cpu 26
cpu 12 <===> cpu 28
cpu 14 <===> cpu 30

 

So if I use a older BIOS Revision (one that doesn't make the server sound like it's about to float off the ground and I can't hear myself think), the cores show up correctly.

On the newest version, the system fans run at normal speeds and it's a lot more quiet, but I am getting this now.

root@Server:~# cat /sys/devices/system/cpu/*/topology/thread_siblings_list
0,16
10,26
12,28
14,30
0,16
2,18
2,18
4,20
6,22
8,24
10,26
12,28
14,30
4,20
6,22
8,24

And if I go to the newest BIOS they have for it QEMU/KVM quits working to where I can not pass through hardware to VMs

I'm having performance issues with VMs right now >_<

I'd like to report some positive news on the horizon:

 

I had a spare usb and booted the 6.4rc on a dl380 G6 running dual E5520 processors and the numbering was correct. I did not attempt it on one of my x5670 processor servers as they can't be taken off-line at the moment. Looking forward to skipping past 6.3 when 6.4 is ready to go!

 

 

cpu 0 <===> cpu 8
cpu 1 <===> cpu 9
cpu 2 <===> cpu 10
cpu 3 <===> cpu 11
cpu 4 <===> cpu 12
cpu 5 <===> cpu 13
cpu 6 <===> cpu 14
cpu 7 <===> cpu 15

 

1 hour ago, 1812 said:

I'd like to report some positive news on the horizon:

 

I had a spare usb and booted the 6.4rc on a dl380 G6 running dual E5520 processors and the numbering was correct. I did not attempt it on one of my x5670 processor servers as they can't be taken off-line at the moment. Looking forward to skipping past 6.3 when 6.4 is ready to go!

 

 

cpu 0 <===> cpu 8
cpu 1 <===> cpu 9
cpu 2 <===> cpu 10
cpu 3 <===> cpu 11
cpu 4 <===> cpu 12
cpu 5 <===> cpu 13
cpu 6 <===> cpu 14
cpu 7 <===> cpu 15

 

 

Same here.  Upgraded to the RC and my core numbers are back to normal.

 
Same here.  Upgraded to the RC and my core numbers are back to normal.

Ditto, fixed here too.

Sent from my HTC 10 using Tapatalk

 
Is there anyone having this issue who is not running an HP ProLiant motherboard?

I have a dl160 g6 dual xeon, and I don't have an issue.

Sent from my Pixel 2 using Tapatalk