[FEATURE] Memory Limits (#494, @konradmalik)

cmd/cluster/clusterCreate.go

@@ -24,12 +24,13 @@ package cluster
 
 import (
 	"fmt"
-	"github.com/docker/go-connections/nat"
 	"os"
 	"runtime"
 	"strings"
 	"time"
 
+	"github.com/docker/go-connections/nat"
+
 	"github.com/spf13/cobra"
 	"github.com/spf13/viper"
 	"gopkg.in/yaml.v2"
@@ -316,6 +317,12 @@ func NewCmdClusterCreate() *cobra.Command {
 	cmd.Flags().String("gpus", "", "GPU devices to add to the cluster node containers ('all' to pass all GPUs) [From docker]")
 	_ = cfgViper.BindPFlag("options.runtime.gpurequest", cmd.Flags().Lookup("gpus"))
 
+	cmd.Flags().String("servers-memory", "", "Memory limit imposed on the server nodes [From docker]")
+	_ = cfgViper.BindPFlag("options.runtime.serversmemory", cmd.Flags().Lookup("servers-memory"))
+
+	cmd.Flags().String("agents-memory", "", "Memory limit imposed on the agents nodes [From docker]")
+	_ = cfgViper.BindPFlag("options.runtime.agentsmemory", cmd.Flags().Lookup("agents-memory"))
+
 	/* Image Importing */
 	cmd.Flags().Bool("no-image-volume", false, "Disable the creation of a volume for importing images")
 	_ = cfgViper.BindPFlag("options.k3d.disableimagevolume", cmd.Flags().Lookup("no-image-volume"))

cmd/node/nodeCreate.go

@@ -27,6 +27,7 @@ import (
 
 	"github.com/spf13/cobra"
 
+	dockerunits "github.com/docker/go-units"
 	"github.com/rancher/k3d/v4/cmd/util"
 	k3dc "github.com/rancher/k3d/v4/pkg/client"
 	"github.com/rancher/k3d/v4/pkg/runtimes"
@@ -67,6 +68,7 @@ func NewCmdNodeCreate() *cobra.Command {
 	}
 
 	cmd.Flags().StringP("image", "i", fmt.Sprintf("%s:%s", k3d.DefaultK3sImageRepo, version.GetK3sVersion(false)), "Specify k3s image used for the node(s)")
+	cmd.Flags().String("memory", "", "Memory limit imposed on the node [From docker]")
 
 	cmd.Flags().BoolVar(&createNodeOpts.Wait, "wait", false, "Wait for the node(s) to be ready before returning.")
 	cmd.Flags().DurationVar(&createNodeOpts.Timeout, "timeout", 0*time.Second, "Maximum waiting time for '--wait' before canceling/returning.")
@@ -112,6 +114,16 @@ func parseCreateNodeCmd(cmd *cobra.Command, args []string) ([]*k3d.Node, *k3d.Cl
 		Name: clusterName,
 	}
 
+	// --memory
+	memory, err := cmd.Flags().GetString("memory")
+	if err != nil {
+		log.Errorln("No memory specified")
+		log.Fatalln(err)
+	}
+	if _, err := dockerunits.RAMInBytes(memory); memory != "" && err != nil {
+		log.Errorf("Provided memory limit value is invalid")
+	}
+
 	// generate list of nodes
 	nodes := []*k3d.Node{}
 	for i := 0; i < replicas; i++ {
@@ -123,6 +135,7 @@ func parseCreateNodeCmd(cmd *cobra.Command, args []string) ([]*k3d.Node, *k3d.Cl
 				k3d.LabelRole: roleStr,
 			},
 			Restart: true,
+			Memory:  memory,
 		}
 		nodes = append(nodes, node)
 	}

go.mod

@@ -12,6 +12,7 @@ require (
 	github.com/docker/distribution v2.7.1+incompatible // indirect
 	github.com/docker/docker v20.10.5+incompatible
 	github.com/docker/go-connections v0.4.0
+	github.com/docker/go-units v0.4.0
 	github.com/go-test/deep v1.0.4
 	github.com/gogo/protobuf v1.3.2 // indirect
 	github.com/heroku/docker-registry-client v0.0.0-20190909225348-afc9e1acc3d5

pkg/client/node.go

@@ -26,13 +26,17 @@ import (
 	"bytes"
 	"context"
 	"fmt"
+	"os"
 	"reflect"
 	"strings"
 	"time"
 
+	dockerunits "github.com/docker/go-units"
 	"github.com/imdario/mergo"
 	"github.com/rancher/k3d/v4/pkg/runtimes"
+	"github.com/rancher/k3d/v4/pkg/runtimes/docker"
 	k3d "github.com/rancher/k3d/v4/pkg/types"
+	"github.com/rancher/k3d/v4/pkg/util"
 	log "github.com/sirupsen/logrus"
 	"golang.org/x/sync/errgroup"
 )
@@ -329,6 +333,37 @@ func NodeCreate(ctx context.Context, runtime runtimes.Runtime, node *k3d.Node, c
 		}
 	}
 
+	// memory limits
+	if node.Memory != "" {
+		if runtime != runtimes.Docker {
+			log.Warn("ignoring specified memory limits as runtime is not Docker")
+		} else {
+			memory, err := dockerunits.RAMInBytes(node.Memory)
+			if err != nil {
+				return fmt.Errorf("Invalid memory limit format: %+v", err)
+			}
+			// mount fake meminfo as readonly
+			fakemempath, err := util.MakeFakeMeminfo(memory, node.Name)
+			if err != nil {
+				return fmt.Errorf("Failed to create fake meminfo: %+v", err)
+			}
+			node.Volumes = append(node.Volumes, fmt.Sprintf("%s:%s:ro", util.MemInfoPath, fakemempath))
+			// mount empty edac folder, but only if it exists
+			exists, err := docker.CheckIfDirectoryExists(ctx, node.Image, util.EdacFolderPath)
+			if err != nil {
+				return fmt.Errorf("Failed to check for the existence of edac folder: %+v", err)
+			}
+			if exists {
+				log.Debugln("Found edac folder")
+				fakeedacpath, err := util.MakeFakeEdac(node.Name)
+				if err != nil {
+					return fmt.Errorf("Failed to create fake edac: %+v", err)
+				}
+				node.Volumes = append(node.Volumes, fmt.Sprintf("%s:%s:ro", util.EdacFolderPath, fakeedacpath))
+			}
+		}
+	}
+
 	/*
 	 * CREATION
 	 */
@@ -346,6 +381,17 @@ func NodeDelete(ctx context.Context, runtime runtimes.Runtime, node *k3d.Node, o
 		log.Error(err)
 	}
 
+	// delete fake folder created for limits
+	if node.Memory != "" {
+		log.Debug("Cleaning fake files folder from k3d config dir for this node...")
+		filepath, err := util.GetNodeFakerDirOrCreate(node.Name)
+		err = os.RemoveAll(filepath)
+		if err != nil {
+			// this err prob should not be fatal, just log it
+			log.Errorf("Could not remove fake files folder for node %s: %+v", node.Name, err)
+		}
+	}
+
 	// update the server loadbalancer
 	if !opts.SkipLBUpdate && (node.Role == k3d.ServerRole || node.Role == k3d.AgentRole) {
 		cluster, err := ClusterGet(ctx, runtime, &k3d.Cluster{Name: node.Labels[k3d.LabelClusterName]})

pkg/config/transform.go

@@ -107,6 +107,7 @@ func TransformSimpleToClusterConfig(ctx context.Context, runtime runtimes.Runtim
 			Image:      simpleConfig.Image,
 			Args:       simpleConfig.Options.K3sOptions.ExtraServerArgs,
 			ServerOpts: k3d.ServerOpts{},
+			Memory:     simpleConfig.Options.Runtime.ServersMemory,
 		}
 
 		// first server node will be init node if we have more than one server specified but no external datastore
@@ -120,9 +121,10 @@ func TransformSimpleToClusterConfig(ctx context.Context, runtime runtimes.Runtim
 
 	for i := 0; i < simpleConfig.Agents; i++ {
 		agentNode := k3d.Node{
-			Role:  k3d.AgentRole,
-			Image: simpleConfig.Image,
-			Args:  simpleConfig.Options.K3sOptions.ExtraAgentArgs,
+			Role:   k3d.AgentRole,
+			Image:  simpleConfig.Image,
+			Args:   simpleConfig.Options.K3sOptions.ExtraAgentArgs,
+			Memory: simpleConfig.Options.Runtime.AgentsMemory,
 		}
 		newCluster.Nodes = append(newCluster.Nodes, &agentNode)
 	}
@@ -226,6 +228,9 @@ func TransformSimpleToClusterConfig(ctx context.Context, runtime runtimes.Runtim
 		DisableLoadBalancer:        simpleConfig.Options.K3dOptions.DisableLoadbalancer,
 		K3sServerArgs:              simpleConfig.Options.K3sOptions.ExtraServerArgs,
 		K3sAgentArgs:               simpleConfig.Options.K3sOptions.ExtraAgentArgs,
+		GPURequest:                 simpleConfig.Options.Runtime.GPURequest,
+		ServersMemory:              simpleConfig.Options.Runtime.ServersMemory,
+		AgentsMemory:               simpleConfig.Options.Runtime.AgentsMemory,
 		GlobalLabels:               map[string]string{}, // empty init
 		GlobalEnv:                  []string{},          // empty init
 	}

pkg/config/v1alpha2/schema.json

@@ -172,6 +172,12 @@
           "properties": {
             "gpuRequest": {
               "type": "string"
+            },
+            "serversMemory": {
+              "type": "string"
+            },
+            "agentsMemory": {
+              "type": "string"
             }
           }
         }

pkg/config/v1alpha2/types.go

@@ -97,7 +97,9 @@ type SimpleConfigOptions struct {
 }
 
 type SimpleConfigOptionsRuntime struct {
-	GPURequest string `mapstructure:"gpuRequest" yaml:"gpuRequest"`
+	GPURequest    string `mapstructure:"gpuRequest" yaml:"gpuRequest"`
+	ServersMemory string `mapstructure:"serversMemory" yaml:"serversMemory"`
+	AgentsMemory  string `mapstructure:"agentsMemory" yaml:"agentsMemory"`
 }
 
 type SimpleConfigOptionsK3d struct {

pkg/config/validate.go

@@ -34,6 +34,7 @@ import (
 
 	"fmt"
 
+	dockerunits "github.com/docker/go-units"
 	log "github.com/sirupsen/logrus"
 )
 
@@ -63,6 +64,21 @@ func ValidateClusterConfig(ctx context.Context, runtime runtimes.Runtime, config
 		return fmt.Errorf("The API Port can not be changed when using 'host' network")
 	}
 
+	// memory limits must have proper format
+	// if empty we don't care about errors in parsing
+	if config.ClusterCreateOpts.ServersMemory != "" {
+		if _, err := dockerunits.RAMInBytes(config.ClusterCreateOpts.ServersMemory); err != nil {
+			return fmt.Errorf("Provided servers memory limit value is invalid")
+		}
+
+	}
+
+	if config.ClusterCreateOpts.AgentsMemory != "" {
+		if _, err := dockerunits.RAMInBytes(config.ClusterCreateOpts.AgentsMemory); err != nil {
+			return fmt.Errorf("Provided agents memory limit value is invalid")
+		}
+	}
+
 	// validate nodes one by one
 	for _, node := range config.Cluster.Nodes {
 

pkg/runtimes/docker/container.go

@@ -180,3 +180,62 @@ func getNodeContainer(ctx context.Context, node *k3d.Node) (*types.Container, er
 	return &containers[0], nil
 
 }
+
+// executes an arbitrary command in a container while returning its exit code.
+// useful to check something in docker env
+func executeCheckInContainer(ctx context.Context, image string, cmd []string) (int64, error) {
+	docker, err := client.NewClientWithOpts(client.FromEnv, client.WithAPIVersionNegotiation())
+	if err != nil {
+		log.Errorln("Failed to create docker client")
+		return -1, err
+	}
+	defer docker.Close()
+
+	if err = pullImage(ctx, docker, image); err != nil {
+		return -1, err
+	}
+
+	resp, err := docker.ContainerCreate(ctx, &container.Config{
+		Image:      image,
+		Cmd:        cmd,
+		Tty:        false,
+		Entrypoint: []string{},
+	}, nil, nil, nil, "")
+	if err != nil {
+		log.Errorf("Failed to create container from image %s with cmd %s", image, cmd)
+		return -1, err
+	}
+
+	if err = startContainer(ctx, resp.ID); err != nil {
+		return -1, err
+	}
+
+	exitCode := -1
+	statusCh, errCh := docker.ContainerWait(ctx, resp.ID, container.WaitConditionNotRunning)
+	select {
+	case err := <-errCh:
+		if err != nil {
+			log.Errorf("Error while waiting for container %s to exit", resp.ID)
+			return -1, err
+		}
+	case status := <-statusCh:
+		exitCode = int(status.StatusCode)
+	}
+
+	if err = removeContainer(ctx, resp.ID); err != nil {
+		return -1, err
+	}
+
+	return int64(exitCode), nil
+}
+
+// CheckIfDirectoryExists checks for the existence of a given path inside the docker environment
+func CheckIfDirectoryExists(ctx context.Context, image string, dir string) (bool, error) {
+	log.Tracef("checking if dir %s exists in docker environment...", dir)
+	shellCmd := fmt.Sprintf("[ -d \"%s\" ] && exit 0 || exit 1", dir)
+	cmd := []string{"sh", "-c", shellCmd}
+	exitCode, err := executeCheckInContainer(ctx, image, cmd)
+	log.Tracef("check dir container returned %d exist code", exitCode)
+	return exitCode == 0, err
+
+}

pkg/runtimes/docker/translate.go

@@ -36,6 +36,7 @@ import (
 	log "github.com/sirupsen/logrus"
 
 	dockercliopts "github.com/docker/cli/opts"
+	dockerunits "github.com/docker/go-units"
 )
 
 // TranslateNodeToContainer translates a k3d node specification to a docker container representation
@@ -85,6 +86,16 @@ func TranslateNodeToContainer(node *k3d.Node) (*NodeInDocker, error) {
 		hostConfig.DeviceRequests = gpuopts.Value()
 	}
 
+	// memory limits
+	// fake meminfo is mounted to hostConfig.Binds
+	if node.Memory != "" {
+		memory, err := dockerunits.RAMInBytes(node.Memory)
+		if err != nil {
+			return nil, fmt.Errorf("Failed to set memory limit: %+v", err)
+		}
+		hostConfig.Memory = memory
+	}
+
 	/* They have to run in privileged mode */
 	// TODO: can we replace this by a reduced set of capabilities?
 	hostConfig.Privileged = true
@@ -233,6 +244,13 @@ func TranslateContainerDetailsToNode(containerDetails types.ContainerJSON) (*k3d
 		Status:  containerDetails.ContainerJSONBase.State.Status,
 	}
 
+	// memory limit
+	memoryStr := dockerunits.HumanSize(float64(containerDetails.HostConfig.Memory))
+	// no-limit is returned as 0B, filter this out
+	if memoryStr == "0B" {
+		memoryStr = ""
+	}
+
 	node := &k3d.Node{
 		Name:       strings.TrimPrefix(containerDetails.Name, "/"), // container name with leading '/' cut off
 		Role:       k3d.NodeRoles[containerDetails.Config.Labels[k3d.LabelRole]],
@@ -249,6 +267,7 @@ func TranslateContainerDetailsToNode(containerDetails types.ContainerJSON) (*k3d
 		ServerOpts: serverOpts,
 		AgentOpts:  k3d.AgentOpts{},
 		State:      nodeState,
+		Memory:     memoryStr,
 	}
 	return node, nil
 }

pkg/types/types.go

@@ -184,6 +184,8 @@ type ClusterCreateOpts struct {
 	K3sServerArgs              []string          `yaml:"k3sServerArgs" json:"k3sServerArgs,omitempty"`
 	K3sAgentArgs               []string          `yaml:"k3sAgentArgs" json:"k3sAgentArgs,omitempty"`
 	GPURequest                 string            `yaml:"gpuRequest" json:"gpuRequest,omitempty"`
+	ServersMemory              string            `yaml:"serversMemory" json:"serversMemory,omitempty"`
+	AgentsMemory               string            `yaml:"agentsMemory" json:"agentsMemory,omitempty"`
 	NodeHooks                  []NodeHook        `yaml:"nodeHooks,omitempty" json:"nodeHooks,omitempty"`
 	GlobalLabels               map[string]string `yaml:"globalLabels,omitempty" json:"globalLabels,omitempty"`
 	GlobalEnv                  []string          `yaml:"globalEnv,omitempty" json:"globalEnv,omitempty"`
@@ -328,6 +330,7 @@ type Node struct {
 	ServerOpts ServerOpts        `yaml:"serverOpts" json:"serverOpts,omitempty"`
 	AgentOpts  AgentOpts         `yaml:"agentOpts" json:"agentOpts,omitempty"`
 	GPURequest string            // filled automatically
+	Memory     string            // filled automatically
 	State      NodeState         // filled automatically
 }