[00:03] <minimal> sbraz: looking at the manpage for systemd-udev-settle.service the suggest instead subscribing to udev events and react to those
[00:04] <minimal> that's basically what the hotplug functionality, recently added to cloud-init, does
[00:05] <minimal> however it's intended to react to *additional* network cards appearing later once cloud-init has run, rather than the 1st network appearing later (as it is needed for metadata server access)
[00:05] <sbraz> ah :/
[00:06] <minimal> hotplug is supported for Ec2 and Openstack datasources (from memory)
[04:04] <holmanb> sbraz - I agree the right fix would probably be to wait for nics to be ready, and preferably via the kernel directly independant of init system, but I'm not sure how to do that. I wonder if operstate is available before the module is completely loaded - might be a way to instrument that.
[04:17] <holmanb> one possible option: https://github.com/systemd/systemd/blob/4e3a50169bef2952544cd6418d9ffd5075202ca2/docs/NETWORK_ONLINE.md?plain=1#L202
[04:20] <holmanb> I doubt that ^ approach is something that would be done quickly
[04:20] <holmanb> sbraz: any luck with the initramfs workaround?
[04:48] <holmanb> sbraz: I think an upstream bug would be appropriate. Other distros might see this in the future.
[10:26] <sbraz> holmanb: i'll look into the initramfs thing today; re. that doc you linked, how would cloud-init decide what interface to wait for? would it wait for the one with the right mac address to appear?
[12:22] <sbraz> holmanb: the initrams trick seems to work very nicely, the nic gets initialised 6 seconds before cloud init local, that's very comfortable :)
[12:27] <sbraz> and that's probably also why debian doesn't have an issue: its initramfs also contains mlx5_core.ko
[13:03] <holmanb> Good question, I would have to read more code to answer that.
[13:04] <holmanb> Glad you got a workaround \o/
[22:15] <holmanb> Is there anywhere (distros, etc) that regularly define a ds-id config? I haven't seen it actually used anywhere to my knowledge, and I'm curious whether it's more common than I think it is.
[22:18] <meena> hmm?
[22:18] <meena> where else do we have that pattern?
[22:19] <holmanb> I've never seen /etc/cloud/ds-identify.cfg used anywhere, even though we have code to parse it, just whether it actually gets used
[22:20] <minimal> holmanb: ds-id? you mean ds-identify?
[22:20] <meena> I have never seen that file
[22:20] <holmanb> yes
[22:22] <minimal> I've never seen that file either
[22:23] <holmanb> its only documentation so far as I can tell is the source of ds-identify