[epic] Operator Controller supports safe CRD upgrades

[everettraven]

Since operator-controller is focused on lifecycling Kubernetes extensions, an important part of this is ensuring that when upgrades occur that CRDs are upgraded safely. For more information see Brief: CRD Upgrade Safety

Epic project view: https://github.com/orgs/operator-framework/projects/8/views/32

At a high-level this looks like:

• Updating carvel-dev/kapp to add a preflight check for CRD Upgrade Safety (Note: While we are no longer continuing to build on the carvel tool suite, we will continue to contribute the more generic CRD validations to the kapp project and re-use/re-purpose that code in operator-controller)

  • Carvel Proposal: https://github.com/carvel-dev/carvel/pull/729/files
    • While the proposal is still being ironed out, changes will be made to https://github.com/everettraven/kapp/tree/feature/crd-upgrade-safety-preflight
  • Generic implementation of CRD upgrade preflight check #661
  • Add a check for existing version that is a stored version being removed #662
  • Add a check for new required field #663
  • Add a check for removing an existing field #664
  • Add a check for scope change (Namespace <--> Cluster) #665
  • Add an extensible generic check method for evaluating updates to an existing field #666
  • Update general preflight package to allow for preflight checks to specify configuration options #667
  • Add configuration options to CRD Upgrade Safety preflight check for error/warn modes and failing open/closed. #668
  • Spike: Investigate valid changes to existing fields (should result in more checks to be written) #669
  • Add a check for handling updates to existing fields to decrease the minimums #713
  • Add a check for handling updates to existing fields to increase the maximums #714
  • Add a check for ensuring that existing required fields can be marked as optional #715
  • Add a check for ensuring that default values are not changed #797
  • Add a check for ensuring that more enum values can be added for an existing field #716

• Updating carvel-dev/kapp-controller to allow configuration of preflight checks (Note: We are no longer doing this as we are no longer building on top of the carvel tool suite)

  • Based on Optional ability to check if client has permissions to perform CRUD actions before trying them carvel-dev/kapp-controller#1381 (comment) - this should be a straightforward change once kapp-controller is updated to use a version of kapp that has the preflight check functionality (i believe this bump is done automatically when a new release of kapp is created. see: carvel-dev/kapp-controller@6911c80)
  • Update kapp-controller to allow the use of the --preflight flag when using kapp as the deployment method #688

• Updating operator-controller to run CRD upgrade safety validation

  • Add configuration to the Extension API for disabling CRD upgrade safety validations (should be enabled by default) #658
  • Update kapp-controller version to one that uses a version of kapp with the CRD upgrade safety preflight check functionality #660
  • Update logic to create App resources to include enabling/disabling the kapp CRD Upgrade Safety preflight check #659
  • Implement generic, extensible pre-flight check functionality based on the new Helm direction #741 (merge in rukpak first)
  • Create CRDUpgradeSafety pre-flight check that re-uses the CRD change validation logic contributed to carvel-dev/kapp #742 (merge in rukpak first)
  • Add configuration to the ClusterExtension API for disabling CRD Upgrade Safety validations #745 (replicate in rukpak first)
  • Transplant pre-flight logic introduced in rukpak to operator-controller #877

• Documentation: Docs: CRD Upgrade Safety #746

[everettraven]

All items in this task list are completed. Closing this issue.