machine configs defining *.nmconnect interfaces for nics that do not exist cause OVS-configuration to fail
Issue
-
It was previously possible to define a blanket machine configuration file that specified multiple nmconnect files for all distributed interfaces in a multi-hardware-node environment, for example, 11-worker-bond0 with definitions for ens2f1, ens2f0, (which exist on some nodes) and a bond0 that matched, AND ens3f1 and ens3f0 which only existed on OTHER nodes (and a matching bond0 interface for those).
-
In OCP 4.10, defining interfaces that do not exist will cause a fail condition during execution of ovs-configure service during boot, failing networking setup and leading to degraded node status.
-
Error messaging and problems can include: Nodes Not in READY, nodes not responding to ping/ssh, nodes failing to complete version upgrades from 4.9.
Environment
- OpenShift Container Platform (OCP) 4.10+
- Clusters with different hardware profiles across different nodes affected (e.g. some nodes do not include nics that others have).
Subscriber exclusive content
A Red Hat subscription provides unlimited access to our knowledgebase, tools, and much more.