Cgroups mount fix #130
Cgroups mount fix #130
Conversation
Signed-off-by: Alexander Morozov <[email protected]>
Signed-off-by: Alexander Morozov <[email protected]>
| @@ -180,18 +180,22 @@ func mountToRootfs(m *configs.Mount, rootfs, mountLabel string) error { | |||
| if err != nil { | |||
| return err | |||
| } | |||
| relDir, err := filepath.Rel(mm.Root, dir) | |||
There was a problem hiding this comment.
If I am reading the code correctly, this doesn't have any effect unless we are planning to set Root to something else.
There was a problem hiding this comment.
I don't understand. Can you elaborate?
It seems like this mm.Root always prepended to cgroup path in /proc/self/cgroup
For example if I'm in parent container, I have mm.Root like
/system.slice/docker-bae3c1be58e13929aa6ae2f10ec958e5e27e9a4fb178f256e229c5bc54212d66.scope
and in new container after creation will be in /proc/self/cgroup something like:
/system.slice/docker-bae3c1be58e13929aa6ae2f10ec958e5e27e9a4fb178f256e229c5bc54212d66.scope/docker/ID
I'll retry now though
There was a problem hiding this comment.
I tried to print stuff and it looks exactly like I described. Mountpoints in container somehow became pretty weird, like:
cope/docker/8cccc84d71a08a64948c297f7117b777e69fc337b75f2df3fa78c1515f908933Maybe I did logical mistake somewhere, or maybe it is expected.
There was a problem hiding this comment.
@LK4D4 Ack, I will test out more from within a docker container.
|
LGTM |
Fix LinuxRuntime field
Because of this there was problems with mounting cgroups in children containers. It tried to mount whole path, which exists on host, but not in container.
ping @mrunalp @hqhq @avagin