add enroot config and ibvervs-utils to image build

Author: RachaelSTamakloe

examples/machine-learning/build-service-images/a4x/blueprint.yaml

@@ -65,7 +65,9 @@ deployment_groups:
                 - hold
                 - "{{ item }}"
               loop: "{{ nvidia_packages_to_hold }}"
-
+      # An unlimited stack size resulted in PMIx errors during
+      # NCCL testing. Setting the stack size to 8192 resolves these failures
+      # and is an intentional change.
       - type: data
         destination: /etc/security/limits.d/99-unlimited.conf
         content: |
@@ -75,6 +77,14 @@ deployment_groups:
           * - nofile 1048576
           * - cpu unlimited
           * - rtprio unlimited
+      - type: data
+        destination: /etc/enroot/enroot.conf
+        content: |
+          ENROOT_CONFIG_PATH     ${HOME}/.enroot
+          ENROOT_RUNTIME_PATH    /mnt/localssd/${UID}/enroot/runtime
+          ENROOT_CACHE_PATH      /mnt/localssd/${UID}/enroot/cache
+          ENROOT_DATA_PATH       /mnt/localssd/${UID}/enroot/data
+          ENROOT_TEMP_PATH       /mnt/localssd/${UID}/enroot
       - type: ansible-local
         destination: update_settings.yml
         content: |
@@ -185,6 +195,19 @@ deployment_groups:
                 name: nvidia-persistenced.service
                 state: stopped
                 enabled: false
+      - type: ansible-local
+        destination: install_ibverbs_utils.yml
+        content: |
+          ---
+          - name: Install ibverbs-utils
+            hosts: all
+            become: true
+            tasks:
+            - name: Install Linux Modules Extra
+              ansible.builtin.package:
+                name:
+                - ibverbs-utils
+                state: present
       # The script below is intended to bypass the packer script that fails on Ubuntu 24.04 images
       # Once Ubuntu 24.04 no longer sets ${HOSTNAME} to the FQDN, this can be removed
       - type: shell