Unable to start VE - Invalid ID

[ramnet]

We currently are using Proxmox with the 2.6.35 kernel.

Upon becoming aware that 2.6.35 is no longer supported, we attempted to downgrade a server from 2.6.35 to 2.6.32. Everything seems fine, except for one thing: large VE ID's don't seem to work anymore.

Given a node with a VE numbered 123 and a VE numbered 5678912, the larger numbered one refuses to start:

# qm start 5003165
Invalid ID
start failed: command '/usr/bin/kvm -monitor unix:/var/run/qemu-server/5678912.mon,server,nowait -vnc unix:/var/run/qemu-server/5678912.vnc,password -pidfile /var/run/qemu-server/5678912.pid -daemonize -usbdevice tablet -name coland2 -smp sockets=1,cores=1 -nodefaults -boot menu=on -vga cirrus -tdf -drive file=/var/lib/vz/images/5678912/vm-5678912-disk-1.raw,if=scsi,index=0,cache=writeback,boot=on -m 256 -netdev type=tap,id=vlan1d0,ifname=tap5678912i1d0,script=/var/lib/qemu-server/bridge-vlan -device e1000,romfile=,mac=56:49:0E:60:C2:9F,netdev=vlan1d0 -id 5678912 -cpuunits 1000' failed with exit code 1

Whereas the VE numbered 123 starts fine.

Reverting to 2.6.35 resolves the problem.

This issue is currently preventing us from switching from the 2.6.35 kernel to the 2.6.32 kernel. While we could renumber the affected VE's, we have over a thousand of such VE's that would need renumbering, and thus it would be a great inconvenience to change all the VE IDs.

From my research I haven't come across anything about the max size ID you can use, so I'm hoping support for larger VE ID's will be restored, or any guidance on how this can be fixed without VE ID renumbering.

 

[dietmar]

Reverting to 2.6.35 resolves the problem.

This issue is currently preventing us from switching from the 2.6.35 kernel to the 2.6.32 kernel.

Something is wrong with that claim. That check is done inside package 'pve-qemu-kvm', so this is not related to the kernel versions. So what version of the pve-qemu-kvm package do you use (both on the working and the 'not working' system)?

# dpkg -l pve-qemu-kvm

 

[ramnet]

dpkg -l pve-qemu-kvm
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Inst/Cfg-files/Unpacked/Failed-cfg/Half-inst/trig-aWait/Trig-pend
|/ Err?=(none)/Hold/Reinst-required/X=both-problems (Status,Err: uppercase=bad)
||/ Name Version Description
+++-==============-==============-============================================
ii pve-qemu-kvm 0.15.0-2 Full virtualization on x86 hardware


I can confirm that this package does not change when I switch kernels. The only packages that are changed are proxmox-ve-2.6.35 to proxmox-ve-2.6.32 and pve-kernel-2.6.35-2-pve to pve-kernel-2.6.32-7-pve

 

[dietmar]

I can confirm that this package does not change when I switch kernels. The only packages that are changed are proxmox-ve-2.6.35 to proxmox-ve-2.6.32 and pve-kernel-2.6.35-2-pve to pve-kernel-2.6.32-7-pve[/QUOTE]

Sorry, but what you claim simply can't be true - please test again. I get the same error message when using kernel 2.6.35.

Note: the restriction is only for KVM VMs, not for containers.

 

[ramnet]

I can reproduce the issue on multiple systems.

How to reproduce:

1) install minimal Debian 5 x64 with default partition layout.
2) install Proxmox with 2.6.35 kernel following http://pve.proxmox.com/wiki/Install_Proxmox_VE_on_Debian_Lenny
3) create guest vm with proxmox web interface, specify a high VMID such as 10383279
4) use vm as normal - all works well
5) using aptitude, remove proxmox-ve-2.6.35 and pve-kernel-2.6.35-2-pve, install pve-kernel-2.6.32-7-pve and proxmox-ve-2.6.32
6) reboot into new kernel
7) guest vm with high VMID will not start.

That check is done inside package 'pve-qemu-kvm', so this is not related to the kernel versions. So what version of the pve-qemu-kvm package do you use (both on the working and the 'not working' system)?

Can you clarify specifically in which file this check is done?

And for that matter, why a check for maximum VMID exists at all?

This is a sample VM config:

vz1:/etc/qemu-server# cat 25253442.conf
ostype: other
vlan1: e1000=B6:E9:25:0A:97:3E
memory: 512
sockets: 1
onboot: 1
name: myserver
bootdisk: scsi0
scsi0: local:25253442/vm-25253442-disk-1.raw,cache=writeback

 

[dietmar]

1) install minimal Debian 5 x64 with default partition layout.
2) install Proxmox with 2.6.35 kernel following http://pve.proxmox.com/wiki/Install_Proxmox_VE_on_Debian_Lenny
3) create guest vm with proxmox web interface, specify a high VMID such as 10383279
4) use vm as normal - all works well

The check is done in the 'kvm' binary. So I can't see why it work with kernel 2.6.35 - That is simply not possible! What is the output of

# md5sum /usr/bin/kvm

Should be the same on both systems.

I already removed the restriction in PVE 2.0 code (see source repository).

 

[ramnet]

f089e52c10192ec211350c417dd58586 /usr/bin/kvm

md5sum is same in both cases.

My best guess is that the check simply isn't being done on 2.6.35 but is being done on 2.6.32. Since the kvm binary is dependent on the kernel for a lot of what it does, it's not unreasonable to expect a difference even with the same kvm binary.

I already removed the restriction in PVE 2.0 code (see source repository).

Great Thank you.

Would it be possible for you to release an updated package to do the same thing for the current stable 1.9 branch please? The change looks fairly trivial.

 

[dietmar]

Would it be possible for you to release an updated package to do the same thing for the current stable 1.9 branch please? The change looks fairly trivial.

PVE 1.9 is frozen - we will not add new features there.