High inode usage on /run due to persistent libpod directories prevents pod scheduling on RHCOS nodes (OCP 4.17)
Issue
- New pods fail to start
- Node becomes unschedulable
- Core system functionality (container runtime) is affected
- On OpenShift Container Platform 4.17, nodes become unschedulable due to
/run
filling up with podman-related directories, leading to inode exhaustion. This prevents new pods from being scheduled and results in container creation failures. - The
/run
directory, backed bytmpfs
, fills up due to thousands of leftover files under/run/libpod/persist/
, resulting in 100% inode usage. This leads to failure in creating containers:
Error: creating container storage: open /run/containers/storage/overlay-layers/.tmp-mountpoints.jsonXXXXXX: no space left on device
Environment
- OpenShift Container Platform: 4.17
- RHCOS version:
417.94.202502111408-0
- Podman version:
podman-5.2.2-3.rhaos4.17.el9.x86_64
- Rootless containers
- Architecture:
x86_64
- Platform: RHEL CoreOS 9.4
Subscriber exclusive content
A Red Hat subscription provides unlimited access to our knowledgebase, tools, and much more.