What causes "deregistering cloud instance %s of group %q: %w" in kubernetes/kops?

Check the wrapped error status — if 404, the instance is already gone and deregistration is moot.

deregistering cloud instance %s of group %q: %w

Error message

deregistering cloud instance %s of group %q: %w

What it means

DeregisterInstance removes a node from Scaleway load-balancers before it leaves the cluster. The first step fetches the server with instanceAPI.GetServer; any non-nil error is wrapped as "deregistering cloud instance %s of group %q: %w". Unlike DeleteInstance there is no 404 special-casing here, so an already-deleted instance also fails deregistration.

Source

Thrown at upup/pkg/fi/cloudup/scaleway/cloud.go:260

		}
		return fmt.Errorf("deleting cloud instance %s of group %s: %w", i.ID, i.CloudInstanceGroup.HumanName, err)
	}

	err = s.DeleteServer(server.Server)
	if err != nil {
		return fmt.Errorf("deleting cloud instance %s of group %s: %w", i.ID, i.CloudInstanceGroup.HumanName, err)
	}

	return nil
}

func (s *scwCloudImplementation) DeregisterInstance(i *cloudinstances.CloudInstance) error {
	server, err := s.instanceAPI.GetServer(&instance.GetServerRequest{
		Zone:     s.zone,
		ServerID: i.ID,
	})
	if err != nil {
		return fmt.Errorf("deregistering cloud instance %s of group %q: %w", i.ID, i.CloudInstanceGroup.HumanName, err)
	}
	serverIP, err := s.GetServerIP(server.Server.ID, server.Server.Zone)
	if err != nil {
		return fmt.Errorf("deregistering cloud instance %s of group %q: %w", i.ID, i.CloudInstanceGroup.HumanName, err)
	}

	// We remove the instance's IP from load-balancers
	lbs, err := s.GetClusterLoadBalancers(s.ClusterName(server.Server.Tags))
	if err != nil {
		return fmt.Errorf("deregistering cloud instance %s of group %q: %w", i.ID, i.CloudInstanceGroup.HumanName, err)
	}
	for _, loadBalancer := range lbs {
		backEnds, err := s.lbAPI.ListBackends(&lb.ZonedAPIListBackendsRequest{
			Zone: s.zone,
			LBID: loadBalancer.ID,
		}, scw.WithAllPages())
		if err != nil {
			return fmt.Errorf("deregistering cloud instance %s of group %q: listing load-balancer's back-ends for instance creation: %w", i.ID, i.CloudInstanceGroup.HumanName, err)

View on GitHub (pinned to 4c8573c808)

Solutions

Check the wrapped error status — if 404, the instance is already gone and deregistration is moot.
Verify s.zone matches the server's zone (cluster tags vs Scaleway profile defaults).
Ensure credentials grant servers:read in the project.
Retry after transient API errors.

Example fix

// before: deleting a node then deregistering it
kops delete instance   // server gone -> GetServer 404 -> error
// after: deregister BEFORE deleting the instance

Defensive patterns

Strategy: type-guard

Type guard

func isAlreadyDeleted(err error) bool {
  var respErr *scw.ResponseError
  return errors.As(err, &respErr) && respErr.StatusCode == http.StatusNotFound
}

Try / catch

if err := cloud.DeregisterInstance(inst); err != nil {
  if isAlreadyDeleted(err) {
    klog.Infof("instance %s already removed; skipping deregister", inst.ID)
    return nil
  }
  return err
}

Prevention

Deregister instances before deleting them, not after.
Handle 404 as success for deregistration flows, like DeleteInstance does.
Keep server IDs in kops state in sync with the Scaleway console.

When it happens

Trigger: instanceAPI.GetServer(&instance.GetServerRequest{Zone: s.zone, ServerID: i.ID}) returns an error: server already deleted (404), wrong zone, bad server ID, insufficient permissions, or API/network failure.

Common situations: Deregistering a node that was already terminated out-of-band (404 surfaces as this error); zone mismatch between the kOps config and the server's actual zone; IAM key lacking servers:read.

Related errors

AI-assisted analysis of kubernetes/kops@4c8573c808 (2026-09-05). Data as JSON: /api/errors/6661d8edd2ac7a82. Report an issue: GitHub.