Sane-behavior cannot set when enabling hierarchical cgroup v1 blkio throttling

[liuyingshanx]

Question

OS: Ubuntu20.04 5.15.0-46-generic docker: 20.10.12 containerd: 20.10.12-0 runc: 1.1.0-0 cgroup version: v1 kubernetes: 1.22.0

Hello,

I implemeted OpenLocal with kubernetes 1.22.0, and trying to throttle LVM block IO following https://github.com/alibaba/open-local/blob/main/docs/user-guide/type-lvm_zh_CN.md.

The cgroup block io bps and iops limit is all in pod level. Throttling implements hierarchy support, it is reasonable to limit all containers' total block iops and bps under the value that set to the pod's blkio.throttle.read_bps_device, blkio.throttle.write_bps_device, blkio.throttle.read_iops_device and blkio.throttle.write_iops_device.

However, throttling's hierarchy support is enabled only if "sane_behavior" is enabled from cgroup side, and cgroup.sane_behavior(ReadOnly) is set to 0 by default. Ref: https://www.kernel.org/doc/Documentation/cgroup-v1/blkio-controller.txt To limit container's blkio, I must enable "sane_behavior" by remounting /sys/fs/cgroup/blkio with the flag -o __DEVEL__sane_behavior. The problem is that it seems the kernel desn't recognize the flag.

$ umount -l /sys/fs/cgroup/blkio $ mount -t cgroup -o blkio -o __DEVEL__sane_behavior none /sys/fs/cgroup/blkio mount: /sys/fs/cgroup: wrong fs type, bad option, bad superblock on none, missing codepage or helper program, or other error

Is any one tried the io throttling function with success? Could you please give me some info about your kernel version or any problem with my configuration? Thanks!

[Clara12062]

Do you have any new developments? I'm also having the same issue on version 0.7.1.

• kernel: 3.10.0-1160.6.1.el7.x86_64.
• k8s: 1.23.17
• docker: 20.10.24 This is environmentally effective.

I deployed it as https://github.com/alibaba/open-local/blob/main/docs/user-guide/type-lvm_zh_CN.md. When I use kernel version 4.18.0-372.9.1.15.po1.x86_64, pod-level limits are successfully written. But the current limit didn't take effect. Then I wrote the limit to the container hierarchy of the inner layer, the IO limit takes effect.

I also suspect something to do with sane_behavior.