Default cpu-shares are 0, but documentation says it is 1024

[marusak]

Is this a BUG REPORT or FEATURE REQUEST? (leave only one on its own line)

/kind bug

Description
From doc:

To modify the proportion from the default of 1024, use the --cpu-shares flag to set the weighting to 2 or higher.

But when I run container without specifying --cpu-shares it has cpu-shares set to 0.

Steps to reproduce the issue:

 $ sudo podman run --name is_it_1024 -dit fedora bash
c9cfa73320f239f1aa07acf94f9ee926df06b4f5bc308368fcde791b8256bea4

$ sudo podman inspect --format '{{.HostConfig.CpuShares}}' is_it_1024
0

Describe the results you expected:

1024

Output of podman version:

Version:            1.6.2
RemoteAPI Version:  1
Go Version:         go1.13.1
OS/Arch:            linux/amd64

Output of podman info --debug:

debug:
  compiler: gc
  git commit: ""
  go version: go1.13.1
  podman version: 1.6.2
host:
  BuildahVersion: 1.11.3
  CgroupVersion: v2
  Conmon:
    package: conmon-2.0.2-1.fc31.x86_64
    path: /usr/bin/conmon
    version: 'conmon version 2.0.2, commit: 186a550ba0866ce799d74006dab97969a2107979'
  Distribution:
    distribution: fedora
    version: "31"
  IDMappings:
    gidmap:
    - container_id: 0
      host_id: 1000
      size: 1
    - container_id: 1
      host_id: 1935008
      size: 65536
    uidmap:
    - container_id: 0
      host_id: 1000
      size: 1
    - container_id: 1
      host_id: 1935008
      size: 65536
  MemFree: 1709568000
  MemTotal: 16426700800
  OCIRuntime:
    name: crun
    package: crun-0.10.6-1.fc31.x86_64
    path: /usr/bin/crun
    version: |-
      crun version 0.10.6
      spec: 1.0.0
      +SYSTEMD +SELINUX +APPARMOR +CAP +SECCOMP +EBPF +YAJL
  SwapFree: 0
  SwapTotal: 0
  arch: amd64
  cpus: 4
  eventlogger: journald
  hostname: zahlmaschine
  kernel: 5.3.15-300.fc31.x86_64
  os: linux
  rootless: true
  slirp4netns:
    Executable: /usr/bin/slirp4netns
    Package: slirp4netns-0.4.0-20.1.dev.gitbbd6f25.fc31.x86_64
    Version: |-
      slirp4netns version 0.4.0-beta.3+dev
      commit: bbd6f25c70d5db2a1cd3bfb0416a8db99a75ed7e
  uptime: 44h 16m 49.74s (Approximately 1.83 days)
registries:
  blocked: null
  insecure: null
  search:
  - docker.io
  - registry.fedoraproject.org
  - registry.access.redhat.com
  - registry.centos.org
  - quay.io
store:
  ConfigFile: /home/mmarusak/.config/containers/storage.conf
  ContainerStore:
    number: 2
  GraphDriverName: overlay
  GraphOptions:
    overlay.mount_program:
      Executable: /usr/bin/fuse-overlayfs
      Package: fuse-overlayfs-0.7.2-2.fc31.x86_64
      Version: |-
        fusermount3 version: 3.6.2
        fuse-overlayfs: version 0.7.2
        FUSE library version 3.6.2
        using FUSE kernel interface version 7.29
  GraphRoot: /home/mmarusak/.local/share/containers/storage
  GraphStatus:
    Backing Filesystem: extfs
    Native Overlay Diff: "false"
    Supports d_type: "true"
    Using metacopy: "false"
  ImageStore:
    number: 3
  RunRoot: /run/user/1000

Package info (e.g. output of rpm -q podman or apt list podman):

podman-1.6.2-2.fc31.x86_64

[vrothberg]

Hi @marusak, thanks for reaching out!

@giuseppe, using --cpu-shares yields the following error:

Error: invalid configuration, cannot specify resource limits without cgroups v2 and --cgroup-manager=systemd

Is that correct or a regressions? It has worked with v1 at some point, didn't it?

[giuseppe]

Is that correct or a regressions? It has worked with v1 at some point, didn't it?

are you running as rootless?

[vrothberg]

are you running as rootless?

🙈 yes ... which totally explains the difference. So we should make the default value conditional and update the docs?

[marusak]

In my example I was running as root, as it is not supported as rootless (to my knowledge).

$ podman run --name is_it_1024 --cpu-shares=1024 -dit fedora bash
Error: writing file '/sys/fs/cgroup/user.slice/user-1000.slice/[email protected]/cgroup.subtree_control': No such file or directory: OCI runtime command not found error

(this is CGroupsV2 of Fedora 31)

[giuseppe]

see_no_evil yes ... which totally explains the difference. So we should make the default value conditional and update the docs?

1024 is the default also when running as rootless. The difference is that it cannot be modified.

I think the issue here can be that 0 means "unspecified", given that the minimum value for cpu shares on cgroup v1 is 2. I think we should just convert 0 to 1024.

[vrothberg]

@giuseppe the default is set to 0 (see https://github.com/containers/libpod/blob/master/cmd/podman/common.go#L192).

[giuseppe]

@giuseppe the default is set to 0 (see https://github.com/containers/libpod/blob/master/cmd/podman/common.go#L192).

which is fine. When it is set to 0 then its value is not set in the OCI spec file as libpod treats it as unspecified by the user: https://github.com/containers/libpod/blob/master/pkg/spec/spec.go#L187-L189

So I think the issue is only in inspect.

In the inspect output, should we convert 0 to 1024 which is the default "most of the time" value used for the cgroup? We can't really assume it though as we could re-use an existing cgroup or configure systemd to use a different value.

I don't see how we can make the inspect output better and make it clear that 0 means "it uses whatever value the cgroup is configured with". The same problem exists with other cgroup settings as well.

[mheon]

I would think that we set it to 1024 by default in the output, given that we should only be overriding it if the resource limits struct is actually present and populated in the spec?