k8s自动清理pod脚本分享

检查会遇到集群节点内存消耗超过90%，我们可以筛选一些可以进行重启的pods，如脚本中涉及svc-开头的，进行触发即重启的shell编写。此项会涉及metrics组件需要安装。

#!/bin/bash

# 设置内存使用率阈值为90%
MEMORY_THRESHOLD=90

# 初始化一个数组来存储需要处理的节点名
EXCEED_NODES=()

# 获取所有节点的内存使用信息
NODES=$(sudo kubectl top node)

# 逐行处理节点信息
while IFS= read -r line; do
    # 跳过表头行
    if [[ $line =~ ^NAME ]]; then
        continue
    fi
    
    # 提取节点名称和内存使用百分比
    NODE=$(echo "$line" | awk '{print $1}')
    MEMORY_PERCENT=$(echo "$line" | awk '{print $4}' | tr -d '%')
    
    # 检查是否超过阈值
    if [[ $MEMORY_PERCENT -gt $MEMORY_THRESHOLD ]]; then
        EXCEED_NODES+=("$NODE")
        echo "Node $NODE memory usage is over $MEMORY_THRESHOLD%, checking for svc- prefixed Pods..."
    else
        echo "Node $NODE is within acceptable memory usage."
    fi
done <<< "$(echo "$NODES" | awk '/^[^[:space:]]/ {print}')"

# 特定命名空间
NAMESPACE="test"

# 对于每一个超过阈值的节点，查找并重启svc-开头的Pods
for NODE in "${EXCEED_NODES[@]}"; do
    # 获取该节点上所有名称以svc-开头的Pods
    PODS=$(sudo kubectl get pods -n $NAMESPACE -o wide | grep "$NODE" | grep "^.*\ssvc-" | awk '{print $1}')
    
    # 如果有符合条件的Pods，则尝试删除它们
    if [ -n "$PODS" ]; then
        for POD in $PODS; do
            sudo kubectl delete pod $POD -n $NAMESPACE
            echo "Deleted Pod $NAMESPACE/$POD"
        done
    else
        echo "No svc- prefixed Pods found on Node $NODE."
    fi
done

# 输出最终状态
if [ ${#EXCEED_NODES[@]} -eq 0 ]; then
    echo "No Pods deleted, all nodes are below the threshold."
else
    echo "Some Pods were deleted due to high memory usage."
fi