This is currently broken and causing a bunch of spam in our logs that looks like "Unable to fetch pod log stats" -- leading to some nodes (like jaws, with a 64G disk) to become NotReady due to disk pressure.
This commit works around the above issue by subverting the string check using the fact that /run is symlinked to /var/run.
As explained in https://github.com/kubernetes/kubernetes/issues/106957#issuecomment-1167388634 -- the CRI-O team found performance issues in the CRI stats provider, so there's a hack in kubelet which falls back to cadvisor when crio is being used.
This is currently broken and causing a bunch of spam in our logs that looks like "Unable to fetch pod log stats" -- leading to some nodes (like jaws, with a 64G disk) to become NotReady due to disk pressure.
This commit works around the above issue by subverting the string check using the fact that
/run
is symlinked to/var/run
.