Description

We deploy SSR on 6.1.2. Upon upgrade to 6.2 PCI address mappings are no longer the same. 

There was a kernel upgrade between 6.1.2 and 6.2.3 to resolve many OS vulnerabilities 

This is likely the trigger to cause the assignment change when upgrading to this version. It is coincidence that 6.2 causes this mapping change. As noted earlier from: Oracle Linux: Give Network Interfaces Persistent Names Without Changing The GRUB Boot Configuration (Kernel Parameter biosdevname Or net.ifnames) the interface names should be mapped.
https://support.oracle.com/knowledge/Oracle%20Linux%20and%20Virtualization/2521902_1.html 

You will note that any interfaces with the prefix ens stay on the same topology, and any of the eth interfaces moved around.

Network interface naming, i.e. eth0, eth1, isn’t guaranteed ordering. The cloud-init udev rule option should work and it appears to at least rename a handful of the interface from the original description. Basically, it is working as designed.

Symptoms

Upgrade from 6.1.2 to 6.2 results in broken PCI mapping in AWS

Solution

There are plans in the works for a new unified AMI for both Mist and conductor managed deployments. This AMI will do the mapping automatically on initial boot.  Please check with your account team when that will be available and in what version.

For the workaround:
The way we’ve handled this in virtual environments is to create udev rules that ensure a consistent interface naming by PCI address across reboots. This article from our documentation discusses that: Deployment Using QCOW2 | SSN Docs. Note: this document discusses our qcow2 image used for virtual environments. Regardless of what is being used, the concept of creating the udev rules should also work for customers in AWS.

Modification History

2024-05-31 : Article Created