Merge pull request #1823 from Sneha-at/data-cache

Fix cachesize parsing and setup logic

Author: k8s-ci-robot

cmd/gce-pd-csi-driver/main.go

@@ -242,10 +242,10 @@ func handle() {
 
 	if *enableNodeDataCacheFlag {
 		if nodeName == nil || *nodeName == "" {
-			klog.Fatalf("Data cache enabled, but --node-name not passed")
+			klog.Errorf("Data cache enabled, but --node-name not passed")
 		}
 		if err := setupDataCache(ctx, *nodeName); err != nil {
-			klog.Fatalf("DataCache setup failed: %v", err)
+			klog.Errorf("DataCache setup failed: %v", err)
 		}
 	}
 
@@ -299,24 +299,26 @@ func urlFlag(target **url.URL, name string, usage string) {
 }
 
 func setupDataCache(ctx context.Context, nodeName string) error {
-	cfg, err := rest.InClusterConfig()
-	if err != nil {
-		return err
-	}
-	kubeClient, err := kubernetes.NewForConfig(cfg)
-	if err != nil {
-		return err
-	}
-	node, err := kubeClient.CoreV1().Nodes().Get(ctx, nodeName, metav1.GetOptions{})
-	if err != nil {
-		// We could retry, but this error will also crashloop the driver which may be as good a way to retry as any.
-		return err
-	}
-	if val, found := node.GetLabels()[dataCacheLabel]; !found || val != dataCacheLabelValue {
-		klog.V(2).Infof("Datacache not enabled for node %s; node label %s=%s and not %s", nodeName, dataCacheLabel, val, dataCacheLabelValue)
-		return nil
+	klog.V(2).Infof("Seting up data cache for node %s", nodeName)
+	if nodeName != common.TestNode {
+		cfg, err := rest.InClusterConfig()
+		if err != nil {
+			return err
+		}
+		kubeClient, err := kubernetes.NewForConfig(cfg)
+		if err != nil {
+			return err
+		}
+		node, err := kubeClient.CoreV1().Nodes().Get(ctx, nodeName, metav1.GetOptions{})
+		if err != nil {
+			// We could retry, but this error will also crashloop the driver which may be as good a way to retry as any.
+			return err
+		}
+		if val, found := node.GetLabels()[dataCacheLabel]; !found || val != dataCacheLabelValue {
+			klog.V(2).Infof("Datacache not enabled for node %s; node label %s=%s and not %s", nodeName, dataCacheLabel, val, dataCacheLabelValue)
+			return nil
+		}
 	}
-	// Setup data cache only if enabled fro nodes
 	klog.V(2).Info("Raiding local ssds to setup data cache")
 	if err := driver.RaidLocalSsds(); err != nil {
 		return fmt.Errorf("Failed to Raid local SSDs, unable to setup data caching, got error %v", err)

pkg/common/constants.go

@@ -42,4 +42,6 @@ const (
 
 	// Keys in the publish context
 	ContexLocalSsdCacheSize = "local-ssd-cache-size"
+	// Node name for E2E tests
+	TestNode = "test-node-csi-e2e"
 )

pkg/common/parameters.go

@@ -251,7 +251,7 @@ func ExtractAndDefaultParameters(parameters map[string]string, driverName string
 				return p, d, fmt.Errorf("parameters contain invalid dataCacheSize parameter: %w", err)
 			}
 			d.DataCacheSize = strconv.FormatInt(paramDataCacheSize, 10)
-			klog.V(2).Infof("====== Data cache size is %v ======", v)
+			klog.V(2).Infof("====== Data cache size is %v GiB ======", d.DataCacheSize)
 		case ParameterKeyDataCacheMode:
 			if !enableDataCache {
 				return p, d, fmt.Errorf("data caching enabled %v; parameters contains invalid option %q", enableDataCache, ParameterKeyDataCacheSize)
@@ -260,7 +260,7 @@ func ExtractAndDefaultParameters(parameters map[string]string, driverName string
 				return p, d, fmt.Errorf("parameters contains invalid option: %w", err)
 			}
 			d.DataCacheMode = v
-			klog.V(2).Infof("====== Data cache mode is %v ======", v)
+			klog.V(2).Infof("====== Data cache mode is %v ======", d.DataCacheMode)
 		case ParameterKeyResourceTags:
 			if err := extractResourceTagsParameter(v, p.ResourceTags); err != nil {
 				return p, d, err

pkg/gce-pd-csi-driver/cache.go

@@ -164,14 +164,15 @@ func setupCaching(devicePath string, req *csi.NodeStageVolumeRequest, nodeId str
 	} else {
 		fastCacheSize := req.GetPublishContext()[common.ContexLocalSsdCacheSize]
 		chunkSize := "960" // Cannot use default chunk size(64KiB) as it errors on maxChunksAllowed. Unit - KiB
-		klog.V(2).Infof("============================== fastCacheSize is %v ==============================", fastCacheSize)
+		klog.V(2).Infof("============================== fastCacheSize is %v GiB ==============================", fastCacheSize)
 		klog.V(2).Infof("============================== lvcreate fast cache layer again with the VolumeGroup %v==============================", volumeGroupName)
 		args = []string{
 			"--yes",
 			"-n",
 			cacheLvName,
 			"-L",
-			fastCacheSize,
+			// ConvertGiStringToInt64 converts the input size to GiB so default to "g" for cache size - LVM g|G is GiB.
+			fastCacheSize + "g",
 			volumeGroupName,
 			raidedLocalSsdPath,
 		}

test/e2e/utils/utils.go

@@ -62,6 +62,7 @@ func GCEClientAndDriverSetup(instance *remote.InstanceInfo, computeEndpoint stri
 		"--use-instance-api-to-list-volumes-published-nodes",
 		"--enable-controller-data-cache",
 		"--enable-node-data-cache",
+		fmt.Sprintf("--node-name=%s", utilcommon.TestNode),
 	}
 	extra_flags = append(extra_flags, fmt.Sprintf("--compute-endpoint=%s", computeEndpoint))