Notes That Apply to Major Upgrades

The notes in this section apply to major upgrades.

Required Task: Add Missing MAC Addresses

Before you begin to upgrade the HP XC Version 2.1 system to this release, you must add missing MAC addresses to the database for external devices on client nodes

Follow this procedure to populate the external device MAC address field in the database for those client nodes with external connections. These MAC addresses are required by the upgrade process to ensure that the external device information is populated correctly in the database and that all external interfaces work properly after the upgrade.

Before turning off power to the client nodes in preparation for the upgrade, invoke the ifconfig command on all nodes with external connections to capture the external device MAC address.
# ifconfig | more
Write down the node name and its associated MAC address.
After you have upgraded the system, but just before invoking the cluster_config --migrate | --init command, invoke the following device_config command to set the external device MAC address in the database for a specific node. Run the device_config command for every node with an external connection.
# cd /opt/hptc/config/sbin # ./device_config --type E --host host_name MAC_address
Verify that the MAC addresses have been populated in the database:
# shownode config | grep -A 3 External:

Required Task: Determine MAC Addresses For Client Nodes With External Devices

Before you upgrade the HP XC Version 2.1 system to this release, determine the MAC address for client nodes with external devices. Later, you will add the MAC addresses to the database.

The first step is to determine the MAC address of the affected client nodes:

Before turning off power to the client nodes in preparation for the upgrade, invoke the ifconfig command on all nodes with external connections to capture the external device MAC address.
# ifconfig | more
Write down the node name and the MAC address associated with the external device. You will use this address in “Required Task: Set External MAC Addresses Before Running the cluster_config Utility”.

Required Task: Set External MAC Addresses Before Running the cluster_config Utility

Perform this task after you have completed upgrade tasks up to including invoking the upgradesys command, but just before invoking the cluster_config utility.

Follow this procedure to populate the external device MAC address field in the database for client nodes with external connections. These MAC addresses are required by the upgrade process to ensure that the external device information is populated correctly in the database and that all external interfaces work properly after the upgrade.

Run the following device_config command to set the external device MAC address in the database for a specific node. Use the MAC addresses you obtained in “Required Task: Determine MAC Addresses For Client Nodes With External Devices”.
# cd /opt/hptc/config/sbin # ./device_config --type E MAC_address --host host_name
Run a similar device_config command for every node that has an external connection.
Verify that the MAC addresses have been populated in the database:
# shownode config | grep -A 3 External:

Required Task: Remove SystemImager Override Symbolic Links Before Running the cluster_config Utility

Because the master autoinstallation scripts (named *.master.0) have changed in this release, you must remove the symbolic links in the /var/lib/systemimager/scripts directory before invoking the cluster_config --init | --migrate command. After you do so, the cluster_config utility automatically creates the correct symbolic links.

If you are using file overrides to the golden image, you must also re-create the *.master.0 override file or files so that the file or files use the new version of the master autoinstallation script.

Re-create the *.master.0 override files and associated symbolic links after invoking the cluster_config utility (but before re-imaging) because the new autoinstallation script is not created until after cluster_config is invoked.

Required Task on HP Integrity Head Nodes: Edit the elilo.conf File

If the head node is an HP Integrity model server, perform this task on the head node after you upgrade and reboot the head node:

Use the text editor of your choice to edit the following file:
/boot/efi/efi/lnxhpc/elilo.conf
Make the following changes to the line that begins with the word kernel:
- Remove rhgb
- Remove quiet
Save your changes and exit the text editor.
Reboot the head node.

Required Task for HP Integrity Hardware: Set the IPMI Password

After you upgrade the system, follow the instructions in the HP XC System Software Administration Guide to set the IPMI password on HP Integrity hardware models.

InfiniBand Directory is Missing from the /etc/profile File

The following note applies only to systems using an InfiniBand interconnect.

The ibhost-biz RPM adds InfiniBand directories to the PATH variable in the /etc/profile file. When you remove or upgrade the previous ibhost-biz RPM, the RPM does not undo the /etc/profile changes, nor does it modify the file since it changed from HP XC Version 2.1 to Version 3.1.

This is not critical because HP does not recommend using Voltaire MPI, which is located in /usr/voltaire/mpi/bin/mpicc. However, these tools are available for diagnostic purposes, and you might want to use them.

For proper HP-MPI behavior, HP recommends that you module load hp/mpi/default, which sets up mpicc and mpirun as the HP-MPI versions.

If you want to use the Voltaire MPI tools, you can work around this issue by adding /usr/voltaire/mpi/bin to the PATH in the /etc/profile file, similar to the following:

# Voltaire InfiniBand Stack Services - Start
PATH="${PATH}:/usr/voltaire/bin:/usr/voltaire/scripts:/usr/mellanox/bin:\
/usr/voltaire/mpi/bin"
export MTHOME="/usr/mellanox"
export NOGAMLA=1
export MTCONF=release
# Voltaire InfiniBand Stack Services - End

Invoke the cluster_config Utility From the Console

It is possible for an external Ethernet connection to occasionally fail to come up after invoking the cluster_config --init | --migrate command. You may see messages similar to the following in the /var/log/nconfig.log file:

Bringing up interface eth2: SIOCSIFFLAGS: Cannot allocate memory Failed to bring up eth2.

If the output from the ifconfig command does not contain the external interface, invoke the ifup command. In the following example, external_device_name is the name of an external device such as eth2.

# ifup external_device_name

Because this issue is intermittent, HP recommends that you invoke the cluster_config utility directly from the console (and not remotely) in case you lose the external connection.

HP Integrity System Does Not Boot After an Upgrade

If an HP Integrity system does not boot after a software upgrade, select the option to boot HPC Linux, and at the boot prompt, boot the system to single-user mode as follows:

> linux single

Then, edit the elilo.conf file as described in “Required Task on HP Integrity Head Nodes: Edit the elilo.conf File” , and boot the system to multiuser mode.