[WIP] Install gather support for baremetal platform

[stbenjam]

fixes #79

This adds support to the 'gather' command for baremetal. The bootstrap
node's IP address is retrieved via libvirt's DHCP records. The master's
IP's are retrieved using Ironic introspection data, that is stored in
the terraform state's introspection data source (added in
terraform-provider-ironic v0.1.4)

How to test this:

• Kick off an install with this PR checked out, ensuring you set KNI_INSTALL_FROM_GIT=true env variable
• When the installer reaches the point where it's waiting for the bootstrap API to come up, hit control + C to abort
• Run the gather command:

[root@dell-r730-021 dev-scripts]# ~/go/src/github.com/openshift-metalkube/kni-installer/bin/kni-install gather bootstrap --dir ocp
INFO Use the following commands to gather logs from the cluster 
INFO ssh -A core@192.168.111.38 '/usr/local/bin/installer-gather.sh 172.22.0.97 172.22.0.10 172.22.0.14' 
INFO scp core@192.168.111.38:~/log-bundle.tar.gz . 

[stbenjam]

This is all bit complex, since we have to retrieve the lease information from libvirt. The libvirt platform generates the IP's: https://github.com/openshift/installer/blob/master/pkg/tfvars/libvirt/libvirt.go, maybe we could use that approach instead.

We're also affected by openshift-metal3/dev-scripts#512 here

[hardys]

What do we need the BootstrapIP for? I'm wondering if we can use the static IP configured for the provisioning network in #100 instead?

[stbenjam]

Debug gathering gets logs from bootstrap if install fails while bootstrap is still up, I think it also uses it as a bastion host to get to the masters in that case as well.

But we can definitely use the static IP instead of this.

[hardys]

I've run into issues getting ironic to work reliably on the boostrap VM, but I may try breaking #100 into two or three PRs, then we can enable the provisioning nic so that this PR can use it, then potentially host the images to allow fixing #58, and finally switch the ironic endpoint over to the bootstrap VM.

@stbenjam let me know if the above decoupling of PRs would be helpful, if so I'll rebase and open some additional PRs.

[stbenjam]

I think that makes sense. I don't think we need install gather to open the first PR to openshift/installer but we do need it eventually.

Could we move the masters to static IP's as well? Or do you think the approach of getting them out of ironic introspection data is reliable?

[hardys]

Yes I think we'll need to switch the masters to static IPs to enable self-hosted ironic where we'll need to run dnsmasq and the other ironic bits on a master to deploy the workers - ref discussion on metal3-io/metal3-docs#39

I'm not yet sure what the best way to inject that configuration is though, e.g we could use ignition, something run via the BMO or perhaps the MCO, and there's the nmstate operator to investigate too.

[derekhiggins]

Build SUCCESS, see build http://10.8.144.11:8080/job/dev-tools/746/

[stbenjam]

Marking this WIP again, as it sounds like we can just use static IP's

[stbenjam]

I'm going to close this, we can open a PR to enable gather when we're using static ip's.

[hardys]

I'm going to close this, we can open a PR to enable gather when we're using static ip's.

Note that openshift/installer#1941 landed yesterday, I'm testing locally with the new RHCOS version and plan to update #100 soon, so that should hopefully provide the static provisioning IP needed for the bootstrap VM.

[stbenjam]

I'm going to close this, we can open a PR to enable gather when we're using static ip's.

Note that openshift/installer#1941 landed yesterday, I'm testing locally with the new RHCOS version and plan to update #100 soon, so that should hopefully provide the static provisioning IP needed for the bootstrap VM.

We're going to move to static IP's for masters on the provisioning network as well, aren't we? Or do we let DHCP handle it and just get the data from introspection for install-gather?