Remove device name and ip address from required parameters of configure_nw_interfaces.sh

[hgreebe]

Description of changes

• Remove device name and ip address from required parameters of configure_nw_interfaces.sh
• Device name and ip address do not exist for efa-only nw interfaces.
• This adds a check that is device name and ip address are not included, then it is an efa-only interface so the rest of the script is skipped.
• We use ec2 instance metadata to get the parameters like ip address, network card index etc. This instance metadata does not have interface_type so we can not use it to see if it is efa-only. This limitation is why we are validating that it is efa-only using device_name and ip addess.

Tests

• Ran 'test_multiple_nics` integration test

References

• CLI change: Change network interface setup logic to account for gb200 aws-parallelcluster#6930

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[gmarciani]

Where it is documented that efa-only interfaces do not have a device name?

[hgreebe]

I couldn't find official documentation but when you try to run this script with an efa-only interface type it has the following input:

{"DEVICE_NAME"=>"", "DEVICE_NUMBER"=>"1", "NETWORK_CARD_INDEX"=>"1", "GW_IP_ADDRESS"=>"192.168.112.1", "DEVICE_IP_ADDRESS"=>nil, "CIDR_PREFIX_LENGTH"=>"20", "NETMASK"=>"255.255.240.0", "CIDR_BLOCK"=>"192.168.112.0/20", "MAC"=>"02:6e:6f:a7:24:8d"}

So it doesn't input a device name or ip address.

[gmarciani]

This logic is ok for EFA only, but removing the check on DEVICE_NAME and DEVICE_IP_ADDRESS (lines above) we are accepting invalid values for the other types of interface.

Before your change we required all network interfaces to have both DEVICE_NAME and DEVICE_IP_ADDRESS. Now we don't check anymore. Removing the check makes sense for EFA only interfaces, but it does not for the other types.

A more robust logic should be: require all variables are set, unless both DEVICE_NAME and DEVICE_IP_ADDRESS are unset

[gmarciani]

this change contributes to supporting GB200 instance types. Let' make sure to capture an entry in the changelog

[gmarciani]

Please specify the name of variables that may be missing.
This script does not have -x, so if we see from the logs the message 'One or more environment variables missing' we do not know if it is the first or second check that is failing.

[gmarciani]

thank you for aligning all Os specific scripts.
Since we are doing it, please use the same execution options: set -ex everywhere to facilitate troubleshooting in case of issues