Ran into this well-known race condition involving NRI plugins, containerd/nri#142 where when DRANET pod is undergoing some lifecycle event (e.g. recreated due to an update) and around the same time a workload-pod gets created that expects device injection, the device injection can be skipped for the duration when DRANET pods are not-started.
We should follow up on that issue to determine if there's a way to avoid this race condition.
Ran into this well-known race condition involving NRI plugins, containerd/nri#142 where when DRANET pod is undergoing some lifecycle event (e.g. recreated due to an update) and around the same time a workload-pod gets created that expects device injection, the device injection can be skipped for the duration when DRANET pods are not-started.
We should follow up on that issue to determine if there's a way to avoid this race condition.