diff --git a/crates/runc-shim/src/runc.rs b/crates/runc-shim/src/runc.rs
index 1e6146a..97903e8 100644
--- a/crates/runc-shim/src/runc.rs
+++ b/crates/runc-shim/src/runc.rs
@@ -14,6 +14,8 @@
    limitations under the License.
 */
 
+#[cfg(target_os = "linux")]
+use std::sync::RwLock;
 use std::{
     convert::TryFrom,
     os::{
@@ -29,6 +31,8 @@ use std::{
 };
 
 use async_trait::async_trait;
+#[cfg(target_os = "linux")]
+use cgroups_rs::fs::Cgroup;
 use containerd_shim::{
     api::{CreateTaskRequest, ExecProcessRequest, Options, Status},
     asynchronous::monitor::{monitor_subscribe, monitor_unsubscribe, Subscription},
@@ -66,19 +70,6 @@ use crate::{
     io::Stdio,
 };
 
-/// check the process is zombie
-#[cfg(target_os = "linux")]
-fn is_zombie_process(pid: i32) -> bool {
-    if let Ok(status) = std::fs::read_to_string(format!("/proc/{}/status", pid)) {
-        for line in status.lines() {
-            if line.starts_with("State:") && line.contains('Z') {
-                return true;
-            }
-        }
-    }
-    false
-}
-
 pub type ExecProcess = ProcessTemplate<RuncExecLifecycle>;
 pub type InitProcess = ProcessTemplate<RuncInitLifecycle>;
 
@@ -139,6 +130,12 @@ impl ContainerFactory<RuncContainer> for RuncFactory {
 
         let config = CreateConfig::default();
         self.do_create(&mut init, config).await?;
+        #[cfg(target_os = "linux")]
+        {
+            *init.lifecycle.cgroup_cache.write().unwrap() =
+                containerd_shim::cgroup::get_cgroup(init.pid as u32).ok();
+        }
+
         let container = RuncContainer {
             id: id.to_string(),
             bundle: bundle.to_string(),
@@ -151,6 +148,7 @@ impl ContainerFactory<RuncContainer> for RuncFactory {
             },
             processes: Default::default(),
         };
+
         Ok(container)
     }
 
@@ -270,6 +268,9 @@ pub struct RuncInitLifecycle {
     opts: Options,
     bundle: String,
     exit_signal: Arc<ExitSignal>,
+    /// Cache for cgroup paths to avoid repeated /proc/<pid>/cgroup parsing
+    #[cfg(target_os = "linux")]
+    cgroup_cache: RwLock<Option<Cgroup>>,
 }
 
 #[async_trait]
@@ -327,15 +328,18 @@ impl ProcessLifecycle<InitProcess> for RuncInitLifecycle {
             ));
         }
 
-        // check the process is zombie
-        if is_zombie_process(p.pid) {
+        // Check if cgroup still exists before attempting update
+        if !self.ensure_init_cgroup_exists().await {
             return Err(other!(
-                "failed to update resources because process {} is a zombie",
+                "failed to update resources because cgroup for process {} has been released",
                 p.pid
             ));
         }
-
-        containerd_shim::cgroup::update_resources(p.pid as u32, resources)
+        let cgroup_guard = p.lifecycle.cgroup_cache.read().unwrap();
+        let cgroup = cgroup_guard
+            .as_ref()
+            .ok_or_else(|| other!("cgroup cache is empty for process {}", p.pid))?;
+        containerd_shim::cgroup::update_resources(cgroup, resources)
     }
 
     #[cfg(not(target_os = "linux"))]
@@ -352,15 +356,18 @@ impl ProcessLifecycle<InitProcess> for RuncInitLifecycle {
             ));
         }
 
-        // check the process is zombie
-        if is_zombie_process(p.pid) {
+        // Check if cgroup still exists before attempting to collect stats
+        if !self.ensure_init_cgroup_exists().await {
             return Err(other!(
-                "failed to collect metrics because process {} is a zombie",
+                "failed to collect metrics because cgroup for process {} has been released",
                 p.pid
             ));
         }
-
-        containerd_shim::cgroup::collect_metrics(p.pid as u32)
+        let cgroup_guard = p.lifecycle.cgroup_cache.read().unwrap();
+        let cgroup = cgroup_guard
+            .as_ref()
+            .ok_or_else(|| other!("cgroup cache is empty for process {}", p.pid))?;
+        containerd_shim::cgroup::collect_metrics(cgroup)
     }
 
     #[cfg(not(target_os = "linux"))]
@@ -436,6 +443,20 @@ impl RuncInitLifecycle {
             opts,
             bundle: bundle.to_string(),
             exit_signal: Default::default(),
+            #[cfg(target_os = "linux")]
+            cgroup_cache: RwLock::new(None),
+        }
+    }
+
+    /// Ensure cgroup exists and cache the path information
+    /// Returns true if cgroup exists, false if released
+    #[cfg(target_os = "linux")]
+    async fn ensure_init_cgroup_exists(&self) -> bool {
+        let cache = self.cgroup_cache.read().unwrap();
+        if let Some(ref cached) = *cache {
+            cached.exists()
+        } else {
+            false
         }
     }
 }
diff --git a/crates/shim/src/cgroup.rs b/crates/shim/src/cgroup.rs
index 1c8b912..c901ee8 100644
--- a/crates/shim/src/cgroup.rs
+++ b/crates/shim/src/cgroup.rs
@@ -109,13 +109,11 @@ fn write_process_oom_score(pid: u32, score: i64) -> Result<()> {
 
 /// Collect process cgroup stats, return only necessary parts of it
 #[cfg_attr(feature = "tracing", tracing::instrument(level = "info"))]
-pub fn collect_metrics(pid: u32) -> Result<Metrics> {
+pub fn collect_metrics(cgroup: &Cgroup) -> Result<Metrics> {
     let mut metrics = Metrics::new();
 
-    let cgroup = get_cgroup(pid)?;
-
     // to make it easy, fill the necessary metrics only.
-    for sub_system in Cgroup::subsystems(&cgroup) {
+    for sub_system in Cgroup::subsystems(cgroup) {
         match sub_system {
             Subsystem::Cpu(cpu_ctr) => {
                 let mut cpu_usage = CPUUsage::new();
@@ -204,7 +202,7 @@ pub fn collect_metrics(pid: u32) -> Result<Metrics> {
 
 // get_cgroup will return either cgroup v1 or v2 depending on system configuration
 #[cfg_attr(feature = "tracing", tracing::instrument(level = "info"))]
-fn get_cgroup(pid: u32) -> Result<Cgroup> {
+pub fn get_cgroup(pid: u32) -> Result<Cgroup> {
     let hierarchies = hierarchies::auto();
     let cgroup = if hierarchies.v2() {
         let path = get_cgroups_v2_path_by_pid(pid)?;
@@ -250,11 +248,8 @@ fn parse_cgroups_v2_path(content: &str) -> Result<PathBuf> {
 
 /// Update process cgroup limits
 #[cfg_attr(feature = "tracing", tracing::instrument(level = "info"))]
-pub fn update_resources(pid: u32, resources: &LinuxResources) -> Result<()> {
-    // get container main process cgroup
-    let cgroup = get_cgroup(pid)?;
-
-    for sub_system in Cgroup::subsystems(&cgroup) {
+pub fn update_resources(cgroup: &Cgroup, resources: &LinuxResources) -> Result<()> {
+    for sub_system in Cgroup::subsystems(cgroup) {
         match sub_system {
             Subsystem::Pid(pid_ctr) => {
                 // set maximum number of PIDs