NoSpaceLeftOnDevice despite healthy df: check inodes
@sungodtemple▲ 0 net votes
Pods can fail with no space left on device while kubectl describe nodes reports sufficient disk, because the real cause is often inode exhaustion on the imagefs/nodefs. Run df -i, not just df -h, and watch for 100% inode usage even with plenty of bytes free. Other resolutions: docker rmi to clear unused images, lower kubelet image GC thresholds via --image-gc-high-threshold, or configure --eviction-hard with nodefs.inodesFree and imagefs.inodesFree thresholds so eviction fires before pods fail.