Kubernetes 容器内存使用巡检脚本

最终版的 Kubernetes 容器内存使用巡检脚本，特点如下：

• 基于 crictl inspect 和 jq，适配 containerd 运行时
• 排除 sandbox 容器（pause 容器）
• 安全提取容器名、Pod 名、Namespace（防止 null）
• 显示容器的 Page Cache（cache）、RSS 内存占用
• 计算 Page Cache 占节点总内存百分比
• 额外显示容器镜像名和内存限制
• 支持参数指定 Top N 容器，默认 10

---

脚本内容：top_container_memory_final.sh

#!/bin/bash

TOP_N=${1:-10}

echo -e "Collecting container memory usage on node...\n"

# 检查依赖
for cmd in crictl jq; do
  command -v $cmd >/dev/null 2>&1 || { echo >&2 "$cmd is required but not installed. Aborting."; exit 1; }
done

TOTAL_MEM_KB=$(grep MemTotal /proc/meminfo | awk '{print $2}')
[[ -z "$TOTAL_MEM_KB" ]] && { echo "Cannot detect total memory"; exit 1; }

TMPFILE=$(mktemp)

for cid in $(crictl ps -q); do
  INFO=$(crictl inspect $cid 2>/dev/null)
  [[ -z "$INFO" ]] && continue

  # 排除 sandbox (pause) 容器
  IS_SANDBOX=$(echo "$INFO" | jq -r '.info.config.labels."io.kubernetes.container.name" // empty')
  if [[ "$IS_SANDBOX" == "POD" ]]; then
    continue
  fi

  # 安全提取字段，避免 null
  CONTAINER_NAME=$(echo "$INFO" | jq -r '.info.name // .info.config.metadata.name // "unknown-container"')
  NAMESPACE=$(echo "$INFO" | jq -r '.info.config.metadata.namespace // "unknown-namespace"')
  POD_NAME=$(echo "$INFO" | jq -r '.info.config.metadata.name // "unknown-pod"')
  IMAGE=$(echo "$INFO" | jq -r '.info.imageSpec.image // "unknown-image"')
  MEM_LIMIT=$(echo "$INFO" | jq -r '.info.config.linux.resources.memoryLimitInBytes // "unlimited"')

  # 找 cgroup memory.stat 路径
  CGROUP_PATH=$(find /sys/fs/cgroup -type f -name memory.stat -path "*$cid*" 2>/dev/null | head -n 1)
  [[ ! -f "$CGROUP_PATH" ]] && continue

  CACHE_KB=$(grep '^cache ' "$CGROUP_PATH" | awk '{print $2}')
  RSS_KB=$(grep '^rss ' "$CGROUP_PATH" | awk '{print $2}')
  [[ -z "$CACHE_KB" ]] && CACHE_KB=0
  [[ -z "$RSS_KB" ]] && RSS_KB=0

  # 计算占比
  CACHE_PERCENT=$(awk "BEGIN {printf \"%.2f\", ($CACHE_KB/$TOTAL_MEM_KB)*100}")

  # 转换限制内存为 MB，方便阅读
  if [[ "$MEM_LIMIT" =~ ^[0-9]+$ ]]; then
    MEM_LIMIT_MB=$(awk "BEGIN {printf \"%.2f\", $MEM_LIMIT/1024/1024}")
  else
    MEM_LIMIT_MB="$MEM_LIMIT"
  fi

  echo -e "${CACHE_KB}\t${RSS_KB}\t${CACHE_PERCENT}\t${cid}\t${NAMESPACE}\t${POD_NAME}\t${CONTAINER_NAME}\t${IMAGE}\t${MEM_LIMIT_MB}MB" >> "$TMPFILE"
done

# 输出表头
echo -e "CACHE(KB)\tRSS(KB)\tCACHE%\tCONTAINER_ID\tNAMESPACE\tPOD\tCONTAINER\tIMAGE\tMEM_LIMIT"
echo "-----------------------------------------------------------------------------------------------------------------------"

sort -nr "$TMPFILE" | head -n "$TOP_N"

rm -f "$TMPFILE"

---

使用示例

bash top_container_memory_final.sh 10

---

输出示例

CACHE(KB)       RSS(KB)     CACHE%  CONTAINER_ID            NAMESPACE       POD                     CONTAINER       IMAGE                           MEM_LIMIT
-----------------------------------------------------------------------------------------------------------------------
10381238        2833723     65.88   4222c3ab...            prometheus      prometheus              prometheus      prom/prometheus:v2.42.0        2048.00MB
1359884         828960      14.50   b9b8b7b4...            database        postgres-db             postgres        postgres:13.3-alpine           unlimited
...

---