In a Linux Kubernetes cluster that has frequent pod creations and deletions along with large amounts of pod network traffic, the following error may be logged to the host system logs:
"unregister_netdevice: waiting for lo to become free. Usage count = 1"
The kernel will typically be in a semi-hung state after this, causing major system instability.
This issue is fixed upstream in the Linux Kernel by this commit which was released in version 4.4.0.
We recommend upgrading to the latest linux kernel available in your distribution.
We have seen cases where certain kernel modules can cause this issue while loaded, even on a kernel that includes the fix above. If you are running a kernel higher than 4.4.0 and still seeing this issue, try disabling any third-party kernel modules to test.
Project/OS Specific bugs:
Docker - https://github.com/moby/moby/issues/5618
RedHat - https://access.redhat.com/solutions/3105941 - https://access.redhat.com/solutions/3659011
Centos - https://bugs.centos.org/view.php?id=12711