Updated job resources and added mount on /dev/shm

/dev/shm being too small was causing a crash in the DataLoader.
CCI-MOC · Oct 8, 2024 · 3fad3f1 · 3fad3f1
1 parent a6c81c2
commit 3fad3f1
Showing 1 changed file with 12 additions and 0 deletions.
diff --git a/k8s/base/job.yaml b/k8s/base/job.yaml
@@ -12,9 +12,21 @@ spec:
           volumeMounts:
             - mountPath: /storage/unet3d_data
               name: unet3d-data
+            - mountPath: /dev/shm
+              name: shm
+          resources:
+            requests:
+              memory: "16Gi"
+              cpu: "500m"
+            limits:
+              memory: "64Gi"
+              cpu: "4"
       restartPolicy: Never
       volumes:
         - name: unet3d-data
           persistentVolumeClaim:
             claimName: mlperf-storage-data
+        - name: shm
+          emptyDir:
+            medium: Memory
   backoffLimit: 4