We are running several POWER9 and Power10 servers, use HMC an Novalink to manage the systems and use PowerVC (in combination with Novalink) for daily tasks, deploy VM etc. To be able to properly deploy an AIX LPAR, cloud-init RPM must be installed and configured on the image that is eventually used to deploy AIX. This used to work flawlessly.

We have switched to newer AIX levels and have also started to use a newer version of cloud-init (22.1.5 from the AIX Toolbox). This exihibits a strange phenomenon. We configure an IPv4 address on the first three Ethernet interfaces (in different vlans), so on en0, en1 and en2. We automatically get an IPv6 address on the next adapter (en3) for RMC connection between the AIX LPAR and the Novalink partition.

Since cloud-init 22.1.5 we need to reboot the AIX LPAR in order to get RMC to work (we never needed to do that). That is cumbersome because to finish the configuration (possibly automated) we need to be able to dynamically add/resize disks etc. that requires an RMC connection. It makes it much more complicated if you have to reboot the LPAR and wait for it to come up again, before RMC starts to work. Eventually I found out the aix.py component of cloud-init performs and autoconf of IPv6 on en0, even when the configuration details that are handled over from PowerVC to cloud-init do not ask for one.

In case you are running into the same problem, in our environment I fixed it by changing the file /opt/freeware/lib/python3.9/site-packages/cloudinit/distros/aix.py (after making a copy of course).

The original line 138 through 141 read

if run_autoconf6:
                       command = "/usr/sbin/autoconf6 -i en0"
                       output = subprocess.check_output(command, shell=True, stderr=subprocess.STDOUT, universal_newlines=True)
                       time.sleep(2)

 I changed that to comment-out the autoconf6 command:

if run_autoconf6:
   # command = "/usr/sbin/autoconf6 -i en0"
   # output = subprocess.check_output(command, shell=True, stderr=subprocess.STDOUT, universal_newlines=True)
   # time.sleep(2)

Now we only get our "desired" IPv6 RMC interface on en3 (or whatever the 'next available' adapter will be) and RMC comes up a minute or so after deploy finishes.

Let me know if this works for you; or if you found a better (permanent) solution for this issue.

Please try using `cloud-init-22.1-7`, as the issue has been addressed in that version.

We are running several POWER9 and Power10 servers, use HMC an Novalink to manage the systems and use PowerVC (in combination with Novalink) for daily tasks, deploy VM etc. To be able to properly deploy an AIX LPAR, cloud-init RPM must be installed and configured on the image that is eventually used to deploy AIX. This used to work flawlessly.

We have switched to newer AIX levels and have also started to use a newer version of cloud-init (22.1.5 from the AIX Toolbox). This exihibits a strange phenomenon. We configure an IPv4 address on the first three Ethernet interfaces (in different vlans), so on en0, en1 and en2. We automatically get an IPv6 address on the next adapter (en3) for RMC connection between the AIX LPAR and the Novalink partition.

Since cloud-init 22.1.5 we need to reboot the AIX LPAR in order to get RMC to work (we never needed to do that). That is cumbersome because to finish the configuration (possibly automated) we need to be able to dynamically add/resize disks etc. that requires an RMC connection. It makes it much more complicated if you have to reboot the LPAR and wait for it to come up again, before RMC starts to work. Eventually I found out the aix.py component of cloud-init performs and autoconf of IPv6 on en0, even when the configuration details that are handled over from PowerVC to cloud-init do not ask for one.

In case you are running into the same problem, in our environment I fixed it by changing the file /opt/freeware/lib/python3.9/site-packages/cloudinit/distros/aix.py (after making a copy of course).

The original line 138 through 141 read

if run_autoconf6:
                       command = "/usr/sbin/autoconf6 -i en0"
                       output = subprocess.check_output(command, shell=True, stderr=subprocess.STDOUT, universal_newlines=True)
                       time.sleep(2)

 I changed that to comment-out the autoconf6 command:

if run_autoconf6:
   # command = "/usr/sbin/autoconf6 -i en0"
   # output = subprocess.check_output(command, shell=True, stderr=subprocess.STDOUT, universal_newlines=True)
   # time.sleep(2)

Now we only get our "desired" IPv6 RMC interface on en3 (or whatever the 'next available' adapter will be) and RMC comes up a minute or so after deploy finishes.

Let me know if this works for you; or if you found a better (permanent) solution for this issue.