Our current understanding of this is:
* It only triggers at cluster installation.
* It appears to be a libreswan issue: https://github.com/libreswan/libreswan/issues/390
* A liveness probe has been added at https://github.com/openshift/cluster-network-operator/pull/952. This will restart the container which fixes the problem.
As a result, we have reduced the severity to medium and it is no longer a blocker. We will continue to work with the libreswan team to resolve the underlying issue.
This issue has been resolved in Libreswan.