OpenStack single-stack IPv6 support

[yasminvalim]

In #1897 issue, we are requested to create support for IPv6 so Ignition can fetch the metadata in single-stack environments.

To achieve this, we added a new URL with the IPv6 endpoint. We also created logic to first attempt the IPv4 endpoint so If this fails, it will try the IPv6 one. If both endpoints fail, it will return the appropriate error.

[travier]

Not sure that will work. What happens if the instance is IPv6 only? Do requests fail differently than if the network is not up yet?

[travier]

One of the principle behind Ignition is that it will retry indefinitely until the server explicitly returns an error.

[yasminvalim]

Not sure that will work. What happens if the instance is IPv6 only? Do requests fail differently than if the network is not up yet?

I updated my code logic. In an IPv6-only instance, IPv4 will fail because it's not found and the resources will be returned from IPv6. If both fails it will return errors. The error will be handled with a message in dispatch function above. It makes sense?

One thing I need to consider is the possibility of having both IPv4 and IPv6 simultaneously. Is that possible? If yes, I need to handle this case as well.

[yasminvalim]

One of the principle behind Ignition is that it will retry indefinitely until the server explicitly returns an error.

Nice. I guess this is being handled in func fetchConfig, right? There is a loop in there like this:

Loop:
	for {
		select {
		case <-ctx.Done():
			break Loop
		case <-errChan:
			dispatchCount--
			if dispatchCount == 0 {
				f.Logger.Info("couldn't fetch config")
				break Loop
			}
		}
	}

	return util.ParseConfig(f.Logger, data)
}

Thanks!

[travier]

One thing I need to consider is the possibility of having both IPv4 and IPv6 simultaneously. Is that possible? If yes, I need to handle this case as well.

Yes, this is very common.

[yasminvalim]

One thing I need to consider is the possibility of having both IPv4 and IPv6 simultaneously. Is that possible? If yes, I need to handle this case as well.

Yes, this is very common.

I updated the code to cover this scenarios. So, now the code looks like this:

• If both endpoints have valid data, the results are combined and returned
  I used append, not sure if is the best approach.

• If only one endpoint has valid data, that data is returned

• If both endpoints fail, the appropriate error is returned with logs

• If both endpoints return resource.ErrNotFound, the function returns nil

These should be the basic scenarios. Now, I'm working on running my code in the OpenStack environment and reading the necessary docs to test it correctly.

Thanks for reviewing and helping me with this task!

[MaysaMacedo]

Something to take into consideration:

[MaysaMacedo]

I wonder if we might need to specify the interface name like it's mentioned on the docs
https://docs.openstack.org/nova/latest/user/metadata.html#the-metadata-service
and how it's done elsewhere
https://github.com/canonical/cloud-init/blob/main/cloudinit/sources/DataSourceOpenStack.py#L76

[HuijingHei]

Maybe we need to change to [fe80::a9fe:a9fe%25%s] refer to https://docs.openstack.org/nova/latest/user/metadata.html#the-metadata-service