Is this page helpful?

Part 2: Importing the OS Image into BCM#

The steps in this part are the same for all distributions. The only difference is the --distro flag passed to cm-image create. Supported values are determined by the XML selection files provided by cluster-tools and present on your head node. To see which distributions your BCM installation supports, check:

# Command must be executed at the OS/shell level on the BCM head node, not within cmsh.

ls /cm/local/apps/cluster-tools/config/

Each XML file in that directory corresponds to a supported distro. The filename prefix is the value to pass to --distro. If a distro XML is not present, it is not supported by your current cluster-tools version.

Verify the tarball integrity.

Run an integrity check before transferring to confirm the archive is not corrupted:

# Run command at the OS/shell level on the DGX host used for Phase 1-4

tar -tzf /srv/rhel9-dgx-image.tar.gz | tail -20

# Expected: the last 20 entries in the archive show familiar Linux paths.
# If the command completes without errors, the tarball is good.

Transfer the tarball to the head node.

For B200 deployment context, refer to the NVIDIA DGX SuperPOD and BasePOD with DGX B200 Deployment Guide. From the DGX bare-metal node:
```
# Run command at the OS/shell level on the DGX host used for Phase 1-4

scp /srv/rhel9-dgx-image.tar.gz root@<head-node-ip>:/root/
```

Run cm-image create on the head node.

Using the tarball transferred in Step 2 as the source, run the following on the head node:

Important

Ensure /cm/shared is not being used by any services to allow BCM successfully migrate it to the new per-OS path (/cm/shared-rhel9-x86_64).

Note

When adding a RHEL image to a BCM cluster that was originally Ubuntu-only, cm-image create all first rebuilds the Ubuntu ramdisk and all existing image ramdisks before creating the RHEL ramdisk. This is expected because BCM is migrating the cluster from a single shared directory (/cm/shared) to a per-OS model (/cm/shared-ubuntu2404-x86_64, /cm/shared-rhel9-x86_64). All existing image paths are updated automatically. Do not interrupt this process.

# Command must be executed at the OS/shell level on the BCM head node, not within cmsh

cm-image --verbose \
   --logfile cm-image-rhel9u6-$(date +%Y%m%d-%H%M%S).log \
   create all \
   --arch x86_64 \
   --distro rhel9u6 \
   --dgx \
   --dgx-type dgx_b200 \
   --source /root/rhel9-dgx-image.tar.gz

# cm-image create will proceed through all phases automatically:

Note

For the NAS mount phase, indicate a mount path that has enough available space if the default path does not have adequate space.

Creating software image dgx-rhel9u6-x86_64........................[  OK  ]
Locking software image dgx-rhel9u6-x86_64...
Creating node installer image.....................................[  OK  ]
NAS mount from 10.96.2.34 detected
Please create/export a directory on the NAS to store /cm/shared-rhel9-x86_64 contents

   Enter NAS directory mount path (default: 10.96.2.34:/cm/shared-rhel9-x86_64, press Enter to use it): 10.96.2.34:/cm_shared/shared-rhel9-x86_64

Enter mount options, press Enter for defaults:
Enter filesystem type: (default: nfs):
Running mount -t nfs -o defaults 10.96.2.34:/cm/shared-rhel9-x86_64 /cm/shared-rhel9-x86_64
Creating cm-shared image at /cm/shared-rhel9-x86_64...
Unmounting /cm/shared-rhel9-x86_64
Adding fsmount for /cm/shared-rhel9-x86_64 from 10.96.2.34:/cm/shared-rhel9-x86_64
Adding cmshared entry /cm/shared-rhel9-x86_64
Adding cm-shared image entities...................................[  OK  ]
Updating cmd entities
Updating head node fsmounts
Changing fspart /cm/shared to /cm/shared-ubuntu2404-x86_64
Unmounting network mount at /cm/shared
Successfully unmounted /cm/shared
Changing fspart /cm/node-installer to /cm/node-installer-ubuntu2404-x86_64
First run NAS setup on secondary

Running while [ ! -d /cm/shared-ubuntu2404-x86_64 ]; do sleep 1; done;

Running while [ -d /cm/shared ]; do mountpoint -q /cm/shared && umount -l /cm/shared || rm -rf /cm/shared; sleep 1; done

Running ln -sf /cm/shared-ubuntu2404-x86_64 /cm/shared

Running while [ ! -d /cm/node-installer-ubuntu2404-x86_64 ]; do sleep 1; done; rm -rf /cm/node-installer && ln -sf /cm/node-installer-ubuntu2404-x86_64 /cm/node-installer

Creating ramdisk for ubuntu2404/x86_64 (this may take a few minutes)...
Creating ramdisk for ubuntu2404/x86_64............................[  OK  ]
Recreating other ramdisks (this may take a few minutes)...

Creating ramdisk for dgx-image...
Creating ramdisk for dgx-b200-slurm-image...
Creating ramdisk for slogin-image...

Recreating other ramdisks.........................................[  OK  ]
Adding fsmount for /cm/shared from 10.96.2.34:/cm_shared/shared-rhel9-x86_64
Added new category: dgx-rhel9-x86_64
    Use this category for adding nodes

Unlocking software image dgx-rhel9u6-x86_64...
Creating ramdisk for rhel9u6/x86_64 (this may take a few minutes)...
Creating ramdisk for rhel9u6/x86_64...............................[  OK  ]

The final line “Creating ramdisk for rhel9u6/x86_64……[ OK ]” confirms cm-image create all completed successfully.

Remove entitlement certificates from the BCM-managed image.

Now that cm-create image has completed, the Red Hat entitlement certificates are no longer needed inside the image. Remove them from the BCM-managed image path to ensure they are not distributed to provisioned nodes.

# Command must be executed at the OS/shell level on the BCM head node, not within cmsh

rm -f /cm/images/dgx-rhel9u6-x86_64/etc/pki/entitlement/<ID>.pem
rm -f /cm/images/dgx-rhel9u6-x86_64/etc/pki/entitlement/<ID>-key.pem
rm -f /cm/images/dgx-rhel9u6-x86_64/etc/rhsm/ca/redhat-uep.pem

Note

Each node should be registered with Red Hat independently using its own credentials after provisioning.