Merge pull request #2239 from ConnorJC3/nvme-fix

Restrict NVMe Metrics to EBS CSI Managed Volumes
kubernetes-sigs · Nov 25, 2024 · 50f6b2e · 50f6b2e
2 parents 0301ff3 + 1f45d80
commit 50f6b2e
Show file tree

Hide file tree

Showing 4 changed files with 10 additions and 6 deletions.
diff --git a/charts/aws-ebs-csi-driver/templates/_node.tpl b/charts/aws-ebs-csi-driver/templates/_node.tpl
@@ -78,7 +78,7 @@ spec:
             - --http-endpoint=0.0.0.0:3302
             {{- end}}
             {{- with .Values.node.kubeletPath }}
-            - --csi-mount-point-prefix={{ . }}
+            - --csi-mount-point-prefix={{ . }}/plugins/kubernetes.io/csi/ebs.csi.aws.com/
             {{- end}}
             {{- with .Values.node.volumeAttachLimit }}
             - --volume-attach-limit={{ . }}

diff --git a/deploy/kubernetes/base/node.yaml b/deploy/kubernetes/base/node.yaml
@@ -61,7 +61,7 @@ spec:
           args:
             - node
             - --endpoint=$(CSI_ENDPOINT)
-            - --csi-mount-point-prefix=/var/lib/kubelet
+            - --csi-mount-point-prefix=/var/lib/kubelet/plugins/kubernetes.io/csi/ebs.csi.aws.com/
             - --logging-format=text
             - --v=2
           env:

diff --git a/pkg/metrics/nvme.go b/pkg/metrics/nvme.go
@@ -26,6 +26,7 @@ import (
 	"math"
 	"os"
 	"os/exec"
+	"path/filepath"
 	"strings"
 	"time"
 	"unsafe"
@@ -122,7 +123,9 @@ func NewNVMECollector(path, instanceID string) *NVMECollector {
 			"read_io_latency_histogram":                  prometheus.NewDesc("read_io_latency_histogram", "Histogram of read I/O latencies (in microseconds)", variableLabels, constLabels),
 			"write_io_latency_histogram":                 prometheus.NewDesc("write_io_latency_histogram", "Histogram of write I/O latencies (in microseconds)", variableLabels, constLabels),
 		},
-		csiMountPointPath: path,
+		// Clean CSI mount point path to normalize path
+		// Add trailing slash back that Clean prunes
+		csiMountPointPath: filepath.Clean(path) + "/",
 		instanceID:        instanceID,
 		collectionDuration: prometheus.NewHistogram(prometheus.HistogramOpts{
 			Name:        "nvme_collector_duration_seconds",

diff --git a/pkg/metrics/nvme_test.go b/pkg/metrics/nvme_test.go
@@ -27,7 +27,8 @@ import (
 )
 
 func TestNewNVMECollector(t *testing.T) {
-	testPath := "/test/path"
+	testPath := "/test//unclean/../path"
+	expectedPath := "/test/path/"
 	testInstanceID := "test-instance-1"
 
 	collector := NewNVMECollector(testPath, testInstanceID)
@@ -36,8 +37,8 @@ func TestNewNVMECollector(t *testing.T) {
 		t.Fatal("NewNVMECollector returned nil")
 	}
 
-	if collector.csiMountPointPath != testPath {
-		t.Errorf("csiMountPointPath = %v, want %v", collector.csiMountPointPath, testPath)
+	if collector.csiMountPointPath != expectedPath {
+		t.Errorf("csiMountPointPath = %v, want %v", collector.csiMountPointPath, expectedPath)
 	}
 
 	if collector.instanceID != testInstanceID {