Move validation from controllers to pkg/backup, pkg/restore

[ncdc]

Make the validation the first thing we do in Backup() and Restore(). WDYT @skriss?

[skriss]

yeah, makes sense to me to put it in the library

[rosskukulinski]

Internal development enhancement, but this is P1 because right now we have backup and restore logic split in 2 different packages. In addition, if anyone wants to vendor ark to do backups/restores, it's not pretty. We want a self-contained package for each.

[skriss]

Here's an initial mini-design for this issue:

Goals

• Move all backup/restore-processing logic into pkg/backup, pkg/restore respectively so there's a single entrypoint for running a backup/restore which can be more easily tested, vendored, etc.
• Make controllers "dumb" - only responsible for watching for new/updated custom resources, enqueuing them, and dequeueing them and passing them to the appropriate packages for work

Non-Goals

• Full refactoring of pkg/backup & pkg/restore. The goal of this initial effort is just to draw a clear boundary around the end-to-end backup/restore processes. From there, integration tests can be written against this boundary to provide guardrails for refactoring the implementations.
• Similar refactoring for other controllers. This should probably be undertaken later, but first priority is for core backup and restore flows.

Design
Entrypoints to packages should be very simple:

// <pkg/backup>
type Pipeline interface {
    Execute(namespace, name string) error
}

• Backup pipeline will:
  • Ensure API object exists
  • Ensure API object has the correct phase for processing (blank or New)
  • Default unspecified API object fields
    • .spec.ttl
    • .spec.storageLocation
    • .spec.volumeSnapshotLocations
    • .status.version
  • Validate API object spec
    • included/excluded resources
    • included/excluded namespaces
    • backup storage location
    • volume snapshot locations
  • Update API object's status to either FailedValidation (with errors) or InProgress
  • Instantiate all the things
  • Run API-scraping & related actions/hooks
  • Upload artifacts (archive, log, volumesnapshots file) to storage
  • Update API object's final status
• Restore pipeline will work similarly (to be detailed after going through backup pipeline & learning from it)

cc @carlisia @nrb @wwitzel3

[eleanor-millman]

Closing because it should naturally fall out of the Astrolabe work if we do that.