createNode: do not copy '--cluster-init' when creating a new master node, as this will break the node

pkg/cluster/node.go

@@ -107,6 +107,25 @@ func AddNodeToCluster(ctx context.Context, runtime runtimes.Runtime, node *k3d.N
 		}
 	}
 
+	if node.Role == k3d.MasterRole {
+		for _, forbiddenCmd := range k3d.DoNotCopyMasterFlags {
+			for i, cmd := range node.Cmd {
+				// cut out the '--cluster-init' flag as this should only be done by the initializing master node
+				if cmd == forbiddenCmd {
+					log.Debugf("Dropping '%s' from node's cmd", forbiddenCmd)
+					node.Cmd = append(node.Cmd[:i], node.Cmd[i+1:]...)
+				}
+			}
+			for i, arg := range node.Args {
+				// cut out the '--cluster-init' flag as this should only be done by the initializing master node
+				if arg == forbiddenCmd {
+					log.Debugf("Dropping '%s' from node's args", forbiddenCmd)
+					node.Args = append(node.Args[:i], node.Args[i+1:]...)
+				}
+			}
+		}
+	}
+
 	if err := CreateNode(ctx, runtime, node); err != nil {
 		return err
 	}

pkg/types/types.go

@@ -102,6 +102,11 @@ const DefaultAPIPort = "6443"
 // DefaultAPIHost defines the default host (IP) for the Kubernetes API
 const DefaultAPIHost = "0.0.0.0"
 
+// DoNotCopyMasterFlags defines a list of commands/args that shouldn't be copied from an existing node when adding a similar node to a cluster
+var DoNotCopyMasterFlags = []string{
+	"--cluster-init",
+}
+
 // CreateClusterOpts describe a set of options one can set when creating a cluster
 type CreateClusterOpts struct {
 	DisableImageVolume  bool

tests/common.sh

@@ -83,7 +83,7 @@ check_multi_node() {
     passed "cluster $cluster has $expectedNodeCount nodes, as expected"
   else
     warn "cluster $cluster has incorrect number of nodes: $nodeCount != $expectedNodeCount"
-    kubectl get nodes -o=custom-columns=NAME:.metadata.name --no-headers
+    kubectl get nodes
     docker ps -a
     return 1
   fi

tests/test_full_lifecycle.sh

@@ -38,6 +38,16 @@ check_clusters "$clustername" || failed "error checking cluster"
 info "Checking that we have 2 nodes online..."
 check_multi_node "$clustername" 2 || failed "failed to verify number of nodes"
 
+# 4. adding another worker node
+# info "Adding one worker node..."
+# LOG_LEVEL=debug $EXE create node "extra-worker" --cluster "$clustername" --role "worker" || failed "failed to add worker node"
+#
+# info "Waiting for a bit to give the new node enough time to boot and register..."
+# sleep 10
+#
+# info "Checking that we have 3 nodes available now..."
+# check_multi_node "$clustername" 3 || failed "failed to verify number of nodes"
+
 # 4. load an image into the cluster
 info "Loading an image into the cluster..."
 docker pull nginx:latest > /dev/null