Use the bundled red launcher against a non-secret colors.yml, with
KUBECONFIG pointing at the target cluster and kube-context naming its
context.
./red build # render manifests and the RedisDeployment under .colors/
./red create --dry-run # walk the create graph without contacting the cluster
./red create # install the controller, apply the resource, wait for Ready
./red check # Ready at the current generation and an authenticated PING
./red rehearse # suspend, run the backup rehearsal in the controller, resume
./red restart # restart the controller and prove the Droplet survives
COLORS_PAR_DRILL_DELETE_OWNED_DROPLET=true ./red drill # delete the Droplet, prove recovery
COLORS_PAR_COMPUTE_PREVENT_DESTROY=false ./red delete # Destroy through the finalizer
Read references/configuration.md before editing desired state. Put the five
credentials only in the ignored .envrc.private as COLORS_PAR_*; create
copies them into the redis-credentials Secret. Never export
COLORS_PAR_PROFILE, edit .colors/, weaken compute-prevent-destroy, reset
spec.suspend on a resource by hand, or run a real create, drill, or
delete without authorization. drill deletes a live Droplet; rehearse
leaves the resource suspended when it cannot prove the remote workflow
stopped, and says so.