Remove early exit from installer

[stbenjam]

Currently we exit early from the installer, and do a bunch of hacks in 06_create_cluster.sh.

• Remove the NoSchedule taint from masters

  • PR: Add support for baremetal platform openshift/machine-config-operator#846 Merged
  • Can remove on next rebase

• Remove patch_ep_host_etcd

  • PR: Remove etcd patch and enforce hostname requirements dev-scripts#620

• Remove csr signing Certs not getting auto approved, stuck in pending state dev-scripts#260

  • Alternative merged: moved csr signing before running installer Move CSR signing to before starting cluster install dev-scripts#621

• Remove add-machine-ips.sh script https://github.com/openshift-metal3/dev-scripts/blob/master/06_create_cluster.sh#L85-L94

[russellb]

One complication here is that we don't want this done in every case of the baremetal platform.

We don't have a declarative way to specify this right now. Once it's configurable somewhere, this might be a customization that happens at the "create manifests" stage, or remain as a day 2 operation.

If we want to re-phrase this issue as "remove early exit", there's another path to doing that.

• First we need to register BareMeralHosts for masters: installer should register hosts for masters and workers #46
• We also need to register 1 or more BareMetalHosts that can be provisioned as workers

With that in place, the installer can provision a cluster, including a worker, that should run to completion.

Making workloads run on masters is still required if we only have 3 hosts to deploy to, though.

[stbenjam]

I re-opened the MCO PR, openshift/machine-config-operator#846, to remove the no schedule taints. Until workers are being deployed with the installer, this is required because ingress, monitoring, and registry operators otherwise won't come up. We can remove the kubelet override once that happens, but I don't want to block opening the openshift/installer PR on waiting for that.

[markmc]

It looks like we're mostly just waiting on a replacement for add-machine-ips.sh ?

I see progress on e.g. metal3-io/cluster-api-provider-baremetal#49 but it's a struggle to piece it all together into a coherent summary of status on this ...

[russellb]

I don’t think we need early exit for that. It can be moved to post install. The purpose of that is to enable auto approval of CSRs, which the cron job handles in the meantime.

For a more detailed history of the CSR issue, see openshift-metal3/dev-scripts#260

I need to check back into this. I think we may have enough finished that everything works normally for workers. We lack introspection data on the BareMetalHost objects for masters. That data is needed so the actuator can copy it over to the Machine.