helm uninstall runs a hook first. It stops koperator, applies uninstall.policy to every agent
that koperator applied, reports the result to the Komodor Agentic Operation Platform (KAOP), and
releases koperator’s namespace claims. Set the policy in your values before you uninstall.
Choose the uninstall policy
keep is the default because an accidental uninstall cannot turn it into an outage. delete never
deletes an object koperator did not apply, such as a Secret you labelled agent-usable. In every
case the agent stays in KAOP with its worker token. To revoke the token, delete the agent in KAOP:
that is the kill switch that works even when the hook does not run.
With keep or scale_to_zero, each agent’s Deployment panel shows Unmanaged on the deploy
target, and its open changes are closed. With keep, KAOP also discards each open revision
on the deploy target: its objects stay but get no traffic, and its worker token is revoked. Remove
them with:
When the uninstall fails
If the policy fails for an agent, the hook starts koperator again and fails, sohelm uninstall
stops and the release stays. Read the hook log, fix the cause and run helm uninstall again:
kubectl delete lease command. If you deleted a managed
namespace first, the hook cannot start: uninstall without it, as below.
Uninstall without the hook
To uninstall without the hook, usehelm uninstall --no-hooks. Koperator then applies no policy, the
agents keep running, and the claims stay. Delete the claims before you install again:
Next steps
Argo CD and Flux
A GitOps tool can skip the uninstall hook.
Install koperator
Install koperator again on a deploy target.