deploy most units after 'cluster-machines-ready' unit
With this MR, most units will be deployed on a cluster after all Machines are fully ready:
- for the management cluster installation
- for the workload cluster installation
Before this MR, the following MR have introduced the same idea for upgrades: reconciling most units (the ones on which the cluster
unit does not depend) only after all Machines are ready:
- for mgmt cluster upgrades (!2032 (merged) + !2054 (merged))
- for workload clusters upgrades (!2052 (merged))
This significantly changes the cinematic of the mgmt cluster installation, will less things happening in parallel.
It is expected that avoiding having unit deployments in parallel with CAPI node creations will make deployments more predictable.
It will also help having the repartition of pods on nodes closer between two deployments (before this change, many pods would end up on the first nodes, and with a repartition that would largely depend on how fast nodes came up vs. units started reconciling). See #563 (closed) for more discussion about that.
It will also:
- help notice when #1144 (closed) occurs, by failing earlier
- close #1112 (closed)