DigitalOcean deployments break on upgrade to unstable channel

[dhess]

On all of my DigitalOcean deployments, upon upgrading to the unstable channel (and possibly earlier releases), the kernel's primary NIC name changes from enp0s3 to ens3.

Because DO does not support DHCP, this means that upon the first reboot after upgrading to the unstable channel, the host loses network connectivity, because the deployed configuration tries to statically configure the no-longer-existing enp0s3 interface.

The only fix I've been able to come up with is to login to the droplet via the DO web console and bring the interface up by hand. Then, from the deployment host, add the new networking config to the machine definition, deploy (which breaks as soon as the new config is activated and the connection is terminated), reboot the droplet, and deploy again to finish the job.

[pbogdan]

For reference this is mentioned in release notes of 17.09 though I'm still not sure how to easily deal with it.

For brand new deployments I bumped NIX_CHANNEL in nixops/data/nixos-infect to 17.09 and ran sed -i "s/enp0/en/g" on nixops/data/nixos-infect & nixops/backends/digital_ocean.py.

[dhess]

Thanks. It would be useful if this were mentioned in the DigitalOcean NixOps manual.

Also, I'm not sure if NixOps releases track NixOS releases in any way, but it might be a good idea to change NixOps's bundled nixos-infect to use 17.09 going forward. It seems reasonable to assume that anyone using a recent NixOps version and making new DigitalOcean deployments will also be deploying 17.09 or later.

[cleverca22]

if you switch over to this module, then nixos can configure the right IP on bootup every time
#605

[corpix]

Hello. NixOS newcomer is here! :)
Provisioned a DigitalOcean machine with NixOps 1.5.2, no network after reboot.
So I think the title could be changed from

DigitalOcean deployments break on upgrade to unstable channel

to

DigitalOcean deployments break after reboot