Bug 2128413

(In reply to Jiri Kortus from comment #0)
> Description of problem:
> Installations performed by virt-install in UEFI mode with boot from network
> don't finish with a machine shutdown as expected, but reboots.
> 
> It's used in our automated tests for bare metal image installations via
> Anaconda, and the whole virt-install command looks like this (stripped from
> apparently unnecessary parts):
> virt-install --os-variant rhel9.1 --arch x86_64 --boot
> hd,network,uefi,loader=/usr/share/edk2/ovmf/OVMF_CODE.secboot.fd,
> loader_ro=yes,loader_type=pflash,nvram_template=/usr/share/edk2/ovmf/
> OVMF_VARS.fd --name rtttest --memory 4983 --disk size=10 --network
> network=netboot,mac=00:11:22:33:44:55 --serial
> file,path=/tmp/serial_output.log --noautoconsole
> 
> The '--boot network' part seems to be problematic, removing 'hd' didn't
> change anything for me.
> 
> In contrast, the same configuration, but with different --boot configuration
> and --cdrom works as expected and shuts the machine down after installation:
> 
> virt-install --os-variant rhel9.1 --arch x86_64  --cdrom /tmp/cdimage.iso
> --boot
> uefi,loader=/usr/share/edk2/ovmf/OVMF_CODE.secboot.fd,loader_ro=yes,
> loader_type=pflash,nvram_template=/usr/share/edk2/ovmf/OVMF_VARS.fd --name
> rtttest --memory 4983 --disk size=10 --network
> network=netboot,mac=00:11:22:33:44:55 --serial
> file,path=/tmp/serial_output.log
> --noautoconsole
> 
> Version-Release number of selected component (if applicable):
> virt-manager-4.0.0-1.el9
> 
> How reproducible:
> 100%
> 
> Steps to Reproduce:
> 1. Prepare an HTTP server with installation ISO (boot.iso works just fine),
> configure a libvirt network with DHCP server and HTTP boot configuration (I
> used
>   <dnsmasq:options>
>     <dnsmasq:option
> value='dhcp-vendorclass=set:efi-http,HTTPClient:Arch:00016'/>
>     <dnsmasq:option value='dhcp-option-force=tag:efi-http,60,HTTPClient'/>
>     <dnsmasq:option
> value='dhcp-boot=tag:efi-http,&quot;http://192.168.200.1/cdimage.iso&quot;'/>
>   </dnsmasq:options>
> for this purpose in network configuration).
> 2. Start virt-install RHEL-9.1 installation in UEFI mode with boot from
> network, as described above, connect to its graphical console.
> 3. Configure the necessary parts in Anaconda finish installation.
> 
> Actual results:
> The VM reboots after the installation is complete.
> 
> Expected results:
> The VM shuts down after the installation is complete.

Hello, Kortus:

After reading your description, I think add --noreboot to your cmd should do what you want.

Here is the description for --noreboot from virt-install man page

   --noreboot
       Prevent the domain from automatically rebooting after the install has completed.

Hello Hongzhou,

thank you for the suggestion, I somehow missed this option. Unfortunately, when I add '--noreboot', the VM gets created in a shut down state for some reason.

I noticed one interesting thing in the man page for virt-install in the '--autoconsole' section:
... If your command requested a multistep install, like --cdrom or --location, after the install phase is complete the VM will be shutoff, regardless of whether a reboot was requested in the VM. ...

This might mean that the shutdown really shouldn't take place when the machine is booted from network (but it's not stated explicitly and there's no definition of what 'multistep' means). Nevertheless, even if it were the case and I was supposed to use the '--noreboot' option, it doesn't work for me as expected :(.

Thank you,
Jirka

Hello Robinson and Jongsma:

Based on comment 2, is --boot network a multistep install method? 

(In reply to Jiri Kortus from comment #2)
> Hello Hongzhou,
> 
> thank you for the suggestion, I somehow missed this option. Unfortunately,
> when I add '--noreboot', the VM gets created in a shut down state for some
> reason.
>
I think this might be another problem

> I noticed one interesting thing in the man page for virt-install in the
> '--autoconsole' section:
> ... If your command requested a multistep install, like --cdrom or
> --location, after the install phase is complete the VM will be shutoff,
> regardless of whether a reboot was requested in the VM. ...
> 
> This might mean that the shutdown really shouldn't take place when the
> machine is booted from network (but it's not stated explicitly and there's
> no definition of what 'multistep' means). Nevertheless, even if it were the
> case and I was supposed to use the '--noreboot' option, it doesn't work for
> me as expected :(.
> 
Hello Robinson and Jongsma:

Based on comment 2, is --boot network a multistep install method? 

> Thank you,
> Jirka

I think the behavior is expected, but it is subtle. TLDR what you want is to remove `--boot hd,network` and use `--pxe` instead.

When you request `--boot hd,network`, this is implicitly `--import --boot hd,network`. `--import` means you are saying 'there is no install phase', like importing an existing disk image. When there's no install phase, the VM is just started and virt-install exits. If you use --noreboot here, the initial startup is not even invoked.

To request virt-install to install off PXE but then for remaining boots just boot off hd, use `--pxe` or `--install bootdev=network`. In this case, when the install completes and the VM requests reboot, virt-install will reboot the VM, then exit. If --noreboot is requested, the VM reboot request is turned into a shutdown, and the VM stays offline.

In your initial case, the VM was installing off network, and then likely requesting a reboot from inside the VM (what anaconda does after install). Since virt-install wasn't expecting a multistage install, it does not intercept the reboot, and the VM reboots as normal, and ends in booted state.

Closing the bug based on Cole's response.