Explore options for graceful shutdown of Linux-based VMs

[cmainas]

When we deploy Linux-based containers over Linux VMs, we strongly encourage the use of urunit as an init process for the VM. When the container's application exits inside the VM, then urunit will perform a graceful shutdown of the VM. However, in all other cases we do not perform a graceful shutdown, which in some cases might cause issues (e.g. corruption of block devices without unmounting) . A very typical such scenario is in a deployment where a container needs to restart or stop. In that scenario urunc will simply kill the VM and therefore no graceful shutdown takes place.

In that context, it would be helpful to explore ways we can notify urunit to perform the graceful shutdown for both Qemu and Firecrakcer in all supported architectures (x86 and arm64).

For example, in Firecracker, we can use SendCtrlAltDel bu this only works for x86 architectures.

[codesmith25103]

Hey @cmainas
I propose implementing this using a VSock-based control channel between the host and the guest.

Guest Side (urunit) We can update urunit to start a listener on a specific VSock port. If it receives a "SHUTDOWN" command, urunit will handle the graceful teardown by:

1. Sending SIGTERM to the container process.
2. Waiting for the process to exit.
3. Finally invoking the reboot(LINUX_REBOOT_CMD_POWER_OFF) syscall.

Host Side (urunc) In the urunc stop logic, instead of immediately killing the VMM process, urunc will first attempt to dial the VM's VSock CID. It will send the shutdown command and wait for a specified timeout. If the timeout expires or the connection fails, only then will it proceed to a hard kill.

[cmainas]

Hello @codesmith25103 , thank you for the suggestion.

However, we can discuss this idea in more details in the next community meeting on Wednesday, January 21st 2026. Feel free to add it in the agenda.