nokmem

Bug details

The node has sufficient disks, but it keeps reporting that the disk is insufficient to create a Pod.

Scope

When the node is used for a long time, it prompts an error that the remaining space is insufficient. The error message is as follows:

mkdir: cannot create directory '/sys/fs/cgroup/memory/8': No space left on device

The node disk is sufficient but reports this error, and the creation of Pod always fails. This is a potential “time bomb”.

All environments that use early-version kernels and Kubernetes versions before 1.21 will be affected. In runc 1.0.0-rc94 (opencontainers/runc#2840) it has been fixed (removed directly).

Prevention

  • Upgrade the system kernel
  • Kubernetes 1.14 to 1.20
    • Rebuild Kubelet with -tags=nokmem
  • Kubernetes 1.14 or earlier
  • Kubernetes 1.21 or higher
    • not affected

Fixed by KLTS


Last modified March 18, 2022 : Fix release (068dbc21)