FIX: #412 set Restart flag to true before creation (#413, thanks @fabricev)

cmd/node/nodeCreate.go

@@ -122,6 +122,7 @@ func parseCreateNodeCmd(cmd *cobra.Command, args []string) ([]*k3d.Node, *k3d.Cl
 			Labels: map[string]string{
 				k3d.LabelRole: roleStr,
 			},
+			Restart: true,
 		}
 		nodes = append(nodes, node)
 	}

pkg/cluster/cluster.go

@@ -180,7 +180,7 @@ func ClusterCreate(ctx context.Context, runtime k3drt.Runtime, cluster *k3d.Clus
 
 		node.Name = generateNodeName(cluster.Name, node.Role, suffix)
 		node.Network = cluster.Network.Name
-
+		node.Restart = true
 		node.GPURequest = cluster.CreateClusterOpts.GPURequest
 
 		// create node
@@ -343,6 +343,7 @@ func ClusterCreate(ctx context.Context, runtime k3drt.Runtime, cluster *k3d.Clus
 				Role:    k3d.LoadBalancerRole,
 				Labels:  k3d.DefaultObjectLabels, // TODO: createLoadBalancer: add more expressive labels
 				Network: cluster.Network.Name,
+				Restart: true,
 			}
 			cluster.Nodes = append(cluster.Nodes, lbNode) // append lbNode to list of cluster nodes, so it will be considered during rollback
 			log.Infof("Creating LoadBalancer '%s'", lbNode.Name)

pkg/cluster/node.go

@@ -324,7 +324,7 @@ func NodeGet(ctx context.Context, runtime runtimes.Runtime, node *k3d.Node) (*k3
 	return node, nil
 }
 
-// NodeWaitForLogMessage follows the logs of a node container and returns if it finds a specific line in there (or timeout is reached)
+//NodeWaitForLogMessage follows the logs of a node container and returns if it finds a specific line in there (or timeout is reached)
 func NodeWaitForLogMessage(ctx context.Context, runtime runtimes.Runtime, node *k3d.Node, message string, since time.Time) error {
 	for {
 		select {
@@ -353,8 +353,15 @@ func NodeWaitForLogMessage(ctx context.Context, runtime runtimes.Runtime, node *
 		if nRead > 0 && strings.Contains(output, message) {
 			break
 		}
+
+		// check if the container is restarting
+		running, status, _ := runtime.GetNodeStatus(ctx, node)
+		if running && status == k3d.NodeStatusRestarting {
+			return fmt.Errorf("Node %s is restarting, early exit to avoid crash loop", node.Name)
+		}
+
+		time.Sleep(500 * time.Millisecond) // wait for half a second to avoid overloading docker (error `socket: too many open files`)
 	}
-	time.Sleep(500 * time.Millisecond) // wait for half a second to avoid overloading docker (error `socket: too many open files`)
 	log.Debugf("Finished waiting for log message '%s' from node '%s'", message, node.Name)
 	return nil
 }

pkg/runtimes/containerd/node.go

@@ -118,6 +118,11 @@ func (d Containerd) GetNode(ctx context.Context, node *k3d.Node) (*k3d.Node, err
 	return nil, nil
 }
 
+// GetNodeStatus returns the status of a node (Running, Started, etc.)
+func (d Containerd) GetNodeStatus(ctx context.Context, node *k3d.Node) (bool, string, error) {
+	return true, "", nil
+}
+
 // GetNodeLogs returns the logs from a given node
 func (d Containerd) GetNodeLogs(ctx context.Context, node *k3d.Node, since time.Time) (io.ReadCloser, error) {
 	return nil, nil

pkg/runtimes/runtime.go

@@ -55,6 +55,7 @@ type Runtime interface {
 	DeleteNode(context.Context, *k3d.Node) error
 	GetNodesByLabel(context.Context, map[string]string) ([]*k3d.Node, error)
 	GetNode(context.Context, *k3d.Node) (*k3d.Node, error)
+	GetNodeStatus(context.Context, *k3d.Node) (bool, string, error)
 	CreateNetworkIfNotPresent(context.Context, string) (string, bool, error) // @return NETWORK_NAME, EXISTS, ERROR
 	GetKubeconfig(context.Context, *k3d.Node) (io.ReadCloser, error)
 	DeleteNetwork(context.Context, string) error

pkg/types/types.go

@@ -54,6 +54,9 @@ var ReadyLogMessageByRole = map[Role]string{
 	LoadBalancerRole: "start worker processes",
 }
 
+// NodeStatusRestarting defines the status string that signals the node container is restarting
+const NodeStatusRestarting = "restarting"
+
 // Role defines a k3d node role
 type Role string