feat: support vclusters

To make vclusters work properly, we need to get the vcluster pod
name/namespace in some places but keep using the host name/namespace in
others. We solve this by passing the vcluster annotations that contain
the name/namespace to the shim and then use the host/vcluster names
where appropriate.

cmd/installer/main.go

@@ -72,7 +72,9 @@ network-lock skip
     "zeropod.ctrox.dev/scaledown-duration",
     "zeropod.ctrox.dev/disable-checkpointing",
     "zeropod.ctrox.dev/pre-dump",
-    "io.containerd.runc.v2.group"
+    "io.containerd.runc.v2.group",
+    "vcluster.loft.sh/name",
+    "vcluster.loft.sh/namespace"
   ]
 `
 )

zeropod/config.go

@@ -23,6 +23,8 @@ const (
 	PreDumpAnnotationKey             = "zeropod.ctrox.dev/pre-dump"
 	CRIContainerNameAnnotation       = "io.kubernetes.cri.container-name"
 	CRIContainerTypeAnnotation       = "io.kubernetes.cri.container-type"
+	VClusterNameAnnotationKey        = "vcluster.loft.sh/name"
+	VClusterNamespaceAnnotationKey   = "vcluster.loft.sh/namespace"
 
 	defaultScaleDownDuration = time.Minute
 	containersDelim          = ","
@@ -43,6 +45,8 @@ type annotationConfig struct {
 	PodName               string `mapstructure:"io.kubernetes.cri.sandbox-name"`
 	PodNamespace          string `mapstructure:"io.kubernetes.cri.sandbox-namespace"`
 	PodUID                string `mapstructure:"io.kubernetes.cri.sandbox-uid"`
+	VClusterPodName       string `mapstructure:"vcluster.loft.sh/name"`
+	VClusterPodNamespace  string `mapstructure:"vcluster.loft.sh/namespace"`
 }
 
 type Config struct {
@@ -53,11 +57,13 @@ type Config struct {
 	PreDump               bool
 	ContainerName         string
 	ContainerType         string
-	PodName               string
-	PodNamespace          string
-	PodUID                string
+	podName               string
+	podNamespace          string
+	podUID                string
 	ContainerdNamespace   string
 	spec                  *specs.Spec
+	vclusterPodName       string
+	vclusterPodNamespace  string
 }
 
 // NewConfig uses the annotations from the container spec to create a new
@@ -141,11 +147,13 @@ func NewConfig(ctx context.Context, spec *specs.Spec) (*Config, error) {
 		ZeropodContainerNames: containerNames,
 		ContainerName:         cfg.ContainerName,
 		ContainerType:         cfg.ContainerType,
-		PodName:               cfg.PodName,
-		PodNamespace:          cfg.PodNamespace,
-		PodUID:                cfg.PodUID,
 		ContainerdNamespace:   ns,
+		podName:               cfg.PodName,
+		podNamespace:          cfg.PodNamespace,
+		podUID:                cfg.PodUID,
 		spec:                  spec,
+		vclusterPodName:       cfg.VClusterPodName,
+		vclusterPodNamespace:  cfg.VClusterPodNamespace,
 	}, nil
 }
 
@@ -159,3 +167,29 @@ func (cfg Config) IsZeropodContainer() bool {
 	// if there is none specified, every one of them is considered.
 	return len(cfg.ZeropodContainerNames) == 0
 }
+
+func (cfg Config) HostPodName() string {
+	return cfg.podName
+}
+
+func (cfg Config) HostPodNamespace() string {
+	return cfg.podNamespace
+}
+
+func (cfg Config) HostPodUID() string {
+	return cfg.podUID
+}
+
+func (cfg Config) PodName() string {
+	if cfg.vclusterPodName != "" {
+		return cfg.vclusterPodName
+	}
+	return cfg.podName
+}
+
+func (cfg Config) PodNamespace() string {
+	if cfg.vclusterPodNamespace != "" {
+		return cfg.vclusterPodNamespace
+	}
+	return cfg.podNamespace
+}

zeropod/container.go

@@ -64,7 +64,7 @@ func New(ctx context.Context, cfg *Config, cr *sync.Mutex, container *runc.Conta
 		return nil, err
 	}
 
-	logPath, err := getLogPath(ctx, cfg)
+	logPath, err := getLogPath(cfg)
 	if err != nil {
 		return nil, fmt.Errorf("unable to get log path: %w", err)
 	}
@@ -174,8 +174,8 @@ func (c *Container) Status() *v1.ContainerStatus {
 	return &v1.ContainerStatus{
 		Id:           c.ID(),
 		Name:         c.cfg.ContainerName,
-		PodName:      c.cfg.PodName,
-		PodNamespace: c.cfg.PodNamespace,
+		PodName:      c.cfg.PodName(),
+		PodNamespace: c.cfg.PodNamespace(),
 		Phase:        phase,
 	}
 }

zeropod/log.go

@@ -1,7 +1,6 @@
 package zeropod
 
 import (
-	"context"
 	"fmt"
 	"os"
 	"path/filepath"
@@ -14,8 +13,8 @@ import (
 // containerd only passes that to the sandbox container (pause). One possible
 // solution would be to implement log restoring in the sandbox container
 // instead of the zeropod.
-func getLogPath(ctx context.Context, cfg *Config) (string, error) {
-	logDir := fmt.Sprintf("/var/log/pods/%s_%s_%s/%s", cfg.PodNamespace, cfg.PodName, cfg.PodUID, cfg.ContainerName)
+func getLogPath(cfg *Config) (string, error) {
+	logDir := fmt.Sprintf("/var/log/pods/%s_%s_%s/%s", cfg.HostPodNamespace(), cfg.HostPodName(), cfg.HostPodUID(), cfg.ContainerName)
 
 	dir, err := os.Open(logDir)
 	if err != nil {

zeropod/metrics.go

@@ -74,8 +74,8 @@ func NewRegistry() *prometheus.Registry {
 func (c *Container) labels() map[string]string {
 	return map[string]string{
 		labelContainerName: c.cfg.ContainerName,
-		LabelPodName:       c.cfg.PodName,
-		LabelPodNamespace:  c.cfg.PodNamespace,
+		LabelPodName:       c.cfg.PodName(),
+		LabelPodNamespace:  c.cfg.PodNamespace(),
 	}
 }