cgroups: v2: always init parent dirs

By skipping parents we were hitting an issue with the way k8s cgroups are managed since it is creating those cgroups. This change ensures we apply subtree_control to all parents as needed. Before this change pods would fail to start due to missing `cpu.weight` since we were not delegating cgroup control. Signed-off-by: Brian Goff <[email protected]>
containerd · Jan 4, 2023 · 0af69a6 · 0af69a6
1 parent b7e3ec2
commit 0af69a6
Showing 1 changed file with 0 additions and 6 deletions.
diff --git a/crates/containerd-shim-wasm/src/sandbox/cgroups/cgroupv2.rs b/crates/containerd-shim-wasm/src/sandbox/cgroups/cgroupv2.rs
@@ -38,12 +38,6 @@ impl CgroupV2 {
         for d in p.iter() {
             full = safe_join(full.clone(), PathBuf::from(d))?;
 
-            if full.exists() {
-                debug!("skipping creation of existing cgroup: {}", full.display());
-                // This is a pre-existing cgroup, so we won't touch subtree control.
-                continue;
-            }
-
             if full.to_str() == self.full_path()?.to_str() {
                 // This is the last element in the path, so we don't need to create it.
                 // The cgroup will get created later when we try to write to a file in it.