=== Juesto is now known as Juest [02:45] lxd-agent did not have any reported errors / functions correctly when started. It seems lxd-agent was started by a udev trigger, after cloud-init started. I have set lxd-agent to enabled but it still does not start before cloud-init. Strangely I have seen other Ubuntu systems where lxd-agent starts before cloud-init-local [02:53] Is it an official ubuntu lxd image? If so, which image (release/arch/serial/vm) is this? [12:19] simpoir: the offical lxd images from ubuntu: work just fine, lxd-agent starts well before cloud-init-local. I am using a ubuntu cloud image (from ubuntu) via packer applying configuration via ansible. certainly something in the ansible could affect things but I do not see anything directly affecting lxd-agent… and I would have guessed that [12:19] enabling the systemd lxd-agent service (instead of being activated by a trigger) would have fixed this, an enabled service with Before= set would then be lauched before.. Any suggestions as to what could cause an enabled service to have Before= ignored (but eventually start)? [13:01] smcavoy: which ubuntu release is this? There are some differences between releases, like 20.04 having a ConditionPathExists=/dev/virtio-ports/org.linuxcontainers.lxd on the unit. if there is any delay in driver loading, then the agent would initially not start, but would be triggered later. [13:14] If that's your case, I suspect you might have to tweak your image so the drivers (virtiofs?) are loaded earlier. [15:15] simpoir: 24.04, I have not tested our 22.04 and 20.04 is not used.. [15:16] we do use the generic kernel instead of the kvm.. [15:17] ah! that might be it then. The kvm kernel has a couple of builtin modules [15:24] okay, so maybe specifying the vsock modules in initramfs.conf / updating should make the kernel module install at boot… [15:45] simpoir: no change, setting vsock in either /etc/initramfs-tools/modules or /etc/modules-load.d/modules.conf does not change the lxd start time/order [15:54] the default for vsock is to be loaded about the time of raid modules, so no changes needed there it seems === at_work_ is now known as at_work