Fix memory allocation bug in FsExporter

run-ai · Apr 2, 2024 · 7211148 · 7211148
1 parent 064cf86
commit 7211148
Showing 1 changed file with 6 additions and 2 deletions.
diff --git a/internal/status-exporter/export/fs/exporter.go b/internal/status-exporter/export/fs/exporter.go
@@ -51,15 +51,15 @@ func (e *FsExporter) export(nodeTopology *topology.NodeTopology) {
 			log.Printf("Exporting pod %s gpu stats to filesystem", podUuid)
 
 			path := fmt.Sprintf("runai/proc/pod/%s/metrics/gpu/%d", podUuid, gpuIdx)
-			if err := os.MkdirAll(filepath.Dir(path), 0644); err != nil {
+			if err := os.MkdirAll(path, 0644); err != nil {
 				log.Printf("Failed creating directory for pod %s: %s", podUuid, err.Error())
 			}
 
 			if err := writeFile(filepath.Join(path, "utilization.sm"), []byte(strconv.Itoa(gpuUsageStatus.Utilization.Random()))); err != nil {
 				log.Printf("Failed exporting utilization for pod %s: %s", podUuid, err.Error())
 			}
 
-			if err := writeFile(filepath.Join(path, "memory.allocated"), []byte(strconv.Itoa(gpuUsageStatus.FbUsed))); err != nil {
+			if err := writeFile(filepath.Join(path, "memory.allocated"), []byte(strconv.Itoa(mbToBytes(gpuUsageStatus.FbUsed)))); err != nil {
 				log.Printf("Failed exporting memory for pod %s: %s", podUuid, err.Error())
 			}
 		}
@@ -72,3 +72,7 @@ func writeFile(path string, content []byte) error {
 	}
 	return nil
 }
+
+func mbToBytes(mb int) int {
+	return mb * 1024 * 1024
+}