From 3fad3f19d7d181a67e6bba561df1a188521474d8 Mon Sep 17 00:00:00 2001
From: Kristi Nikolla <kristi@nikolla.me>
Date: Tue, 8 Oct 2024 12:42:41 -0400
Subject: [PATCH] Updated job resources and added mount on /dev/shm

/dev/shm being too small was causing a crash in the DataLoader.
---
 k8s/base/job.yaml | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/k8s/base/job.yaml b/k8s/base/job.yaml
index 4e9a49a..fb6e446 100644
--- 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