clear pod cpuset

When the kubelet configures cpuManagerPolicy as static, the Pod will specify CPU IDs for CPU affinity. Due to the different cpusets of the guest OS and host OS, this can lead to configuration failures. There is no need for CPU affinity in the guest OS, so the Pod's cpuset should be cleared.

Signed-off-by: MorningTZH <[email protected]>

vmm/sandbox/src/container/handler/spec.rs

@@ -22,6 +22,7 @@ use containerd_sandbox::{
     error::Error,
     spec::{JsonSpec, Mount},
 };
+use log::debug;
 use path_clean::clean;
 use vmm_common::{
     ETC_HOSTNAME, ETC_HOSTS, ETC_RESOLV, HOSTNAME_FILENAME, HOSTS_FILENAME, KUASAR_STATE_DIR,
@@ -69,6 +70,21 @@ where
         if let Some(p) = spec.process.as_mut() {
             p.apparmor_profile = "".to_string();
         }
+
+        // When the kubelet configures cpuManagerPolicy as static, the Pod will specify CPU IDs
+        // for CPU affinity. Due to the different cpusets of the guest OS and host OS, this can
+        // lead to configuration failures. There is no need for CPU affinity in the guest OS,
+        // so the Pod's cpuset should be cleared.
+        spec.linux
+            .as_mut()
+            .and_then(|l| l.resources.as_mut())
+            .and_then(|r| {
+                r.cpu.as_mut().map(|cpu| {
+                    debug!("Container cpuset will be clear: {:?}", &cpu.cpus);
+                    cpu.cpus = "".to_string();
+                })
+            });
+
         // Update sandbox files mounts for container
         container_mounts(&shared_path, spec);
         let spec_str = serde_json::to_string(spec)