分享一个很实用的K8S巡检脚本

我科绝伦（Huanhuan Zhou）2026-04-08 12:18
bash 复制代码 
#!/bin/bash

# ============================================================================
# K8s 集群全面巡检脚本
# 功能：检查节点、Pod、Deployment、Service、Ingress、PVC、事件等资源状态
# 输出：生成炫酷的HTML巡检报告
# 依赖：kubectl, jq
# ============================================================================

set -e

# ----------------------------- 配置区域 -------------------------------------
REPORT_FILE="k8s_inspection_report_$(date +%Y%m%d_%H%M%S).html"
KUBECTL="kubectl"
JQ="jq"
NAMESPACES=""  # 留空表示所有命名空间，可设置为 "default,kube-system" 等
TIMEOUT=10      # kubectl 命令超时时间（秒）
LOG_LEVEL="INFO" # 日志级别 INFO/WARN/ERROR

# 颜色输出（仅用于脚本日志）
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[0;33m'
BLUE='\033[0;34m'
NC='\033[0m' # No Color

# ----------------------------- 函数定义 -------------------------------------

# 增强的日志函数，支持日志级别
log() {
    local level=$1
    shift
    local message="$*"
    local timestamp=$(date '+%Y-%m-%d %H:%M:%S')
    
    case $level in
        ERROR)
            echo -e "${RED}[ERROR] [$timestamp]${NC} $message" >&2
            ;;
        WARN)
            echo -e "${YELLOW}[WARN] [$timestamp]${NC} $message"
            ;;
        INFO)
            echo -e "${GREEN}[INFO] [$timestamp]${NC} $message"
            ;;
        DEBUG)
            if [[ "$LOG_LEVEL" == "DEBUG" ]]; then
                echo -e "${BLUE}[DEBUG] [$timestamp]${NC} $message"
            fi
            ;;
        *)
            echo -e "[UNKNOWN] [$timestamp] $message"
            ;;
    esac
}

log_info() { log "INFO" "$@"; }
log_warn() { log "WARN" "$@"; }
log_error() { log "ERROR" "$@"; }

# 带超时的 kubectl 命令执行函数
run_kubectl_with_timeout() {
    local cmd="$*"
    timeout $TIMEOUT $cmd 2>/dev/null || {
        log_error "kubectl command failed or timed out: $cmd"
        return 1
    }
}

# 检查依赖并验证集群连通性
check_dependencies() {
    log_info "检查依赖..."
    if ! command -v $KUBECTL &> /dev/null; then
        log_error "kubectl 未安装或不在PATH中"
        exit 1
    fi
    if ! command -v $JQ &> /dev/null; then
        log_error "jq 未安装或不在PATH中，请先安装 jq"
        exit 1
    fi
    
    log_info "验证集群连通性..."
    if ! run_kubectl_with_timeout $KUBECTL cluster-info &> /dev/null; then
        log_error "无法连接到Kubernetes集群，请检查kubeconfig配置"
        exit 1
    fi
    
    # 检查 jq 是否能正确处理 JSON
    if ! echo '{"test": "ok"}' | $JQ '.' &> /dev/null; then
        log_error "jq 安装有问题，无法处理 JSON"
        exit 1
    fi
    
    log_info "依赖检查通过，开始巡检集群..."
}

# 获取所有命名空间（逗号分隔），处理错误情况
get_namespaces() {
    if [ -z "$NAMESPACES" ]; then
        NAMESPACES=$(run_kubectl_with_timeout $KUBECTL get namespaces -o jsonpath='{.items[*].metadata.name}' 2>/dev/null | tr ' ' ',' || echo "")
        if [ -z "$NAMESPACES" ]; then
            log_warn "无法获取命名空间列表，将使用 --all-namespaces 选项"
            NAMESPACES="--all-namespaces"
        fi
    fi
    echo "$NAMESPACES"
}

# 收集节点信息，优化性能和错误处理
collect_nodes() {
    log_info "收集节点信息..."
    local NODES_JSON
    NODES_JSON=$(run_kubectl_with_timeout $KUBECTL get nodes -o json)
    if [ $? -ne 0 ] || [ -z "$NODES_JSON" ]; then
        log_error "无法获取节点信息"
        NODES_COUNT=0
        NODES_READY=0
        NODES_NOT_READY=0
        NODES_TABLE="<tr><td colspan='4'>无法获取节点信息</td></tr>"
        return
    fi

    NODES_COUNT=$(echo "$NODES_JSON" | $JQ '.items | length')
    NODES_READY=$(echo "$NODES_JSON" | $JQ '[.items[] | select(.status.conditions[] | select(.type=="Ready" and .status=="True"))] | length')
    NODES_NOT_READY=$((NODES_COUNT - NODES_READY))

    # 优化节点信息收集，提取更多有用指标
    NODES_TABLE=""
    while IFS=$'\t' read -r name status cpu mem allocatable_cpu allocatable_mem taints addr_internal addr_external os arch kernel; do
        local status_class="status-bad"
        if [ "$status" == "True" ]; then 
            status_class="status-good"
        fi
        
        # 格式化显示资源信息
        local cpu_display="$cpu/$allocatable_cpu"
        local mem_display="$mem/$allocatable_mem"
        
        # 处理 Taints 为空的情况
        taints=${taints:-"None"}
        addr_internal=${addr_internal:-"N/A"}
        addr_external=${addr_external:-"N/A"}
        
        NODES_TABLE+="<tr><td>$name</td><td class=\"$status_class\">$status</td><td>$cpu_display</td><td>$mem_display</td><td>$taints</td><td>$addr_internal</td><td>$addr_external</td><td>$os/$arch</td><td>$kernel</td></tr>"
    done < <(echo "$NODES_JSON" | $JQ -r '.items[] | [
        .metadata.name,
        (.status.conditions[] | select(.type=="Ready") | .status),
        (.status.capacity.cpu // "?"),
        (.status.capacity.memory // "?"),
        (.status.allocatable.cpu // "?"),
        (.status.allocatable.memory // "?"),
        ([.spec.taints[]?.key + "=" + (.spec.taints[]?.value // "") + ":" + (.spec.taints[]?.effect // "")] | join(", ") // ""),
        (.status.addresses[] | select(.type=="InternalIP") | .address // ""),
        (.status.addresses[] | select(.type=="ExternalIP") | .address // ""),
        (.status.nodeInfo.osImage // "?"),
        (.status.nodeInfo.architecture // "?"),
        (.status.nodeInfo.kernelVersion // "?")
    ] | @tsv')
}

# 收集 Pod 信息（所有命名空间，重点异常状态），优化性能
collect_pods() {
    log_info "收集Pod信息..."
    local PODS_JSON
    PODS_JSON=$(run_kubectl_with_timeout $KUBECTL get pods --all-namespaces -o json)
    if [ $? -ne 0 ] || [ -z "$PODS_JSON" ]; then
        log_error "无法获取Pod信息"
        TOTAL_PODS=0
        RUNNING_PODS=0
        PENDING_PODS=0
        FAILED_PODS=0
        UNKNOWN_PODS=0
        ABNORMAL_CONTAINERS=0
        OOMKILLED=0
        ABNORMAL_PODS_TABLE="<tr><td colspan='5'>无法获取Pod信息</td></tr>"
        return
    fi

    TOTAL_PODS=$(echo "$PODS_JSON" | $JQ '.items | length')

    # 统计各种状态
    RUNNING_PODS=$(echo "$PODS_JSON" | $JQ '[.items[] | select(.status.phase=="Running")] | length')
    PENDING_PODS=$(echo "$PODS_JSON" | $JQ '[.items[] | select(.status.phase=="Pending")] | length')
    FAILED_PODS=$(echo "$PODS_JSON" | $JQ '[.items[] | select(.status.phase=="Failed")] | length')
    UNKNOWN_PODS=$(echo "$PODS_JSON" | $JQ '[.items[] | select(.status.phase=="Unknown")] | length')

    # 异常容器状态 (CrashLoopBackOff, ImagePullBackOff, ErrImagePull, OOMKilled等)
    ABNORMAL_CONTAINERS=$(echo "$PODS_JSON" | $JQ '[.items[].status.containerStatuses[]? | select(.state.waiting != null and (.state.waiting.reason == "CrashLoopBackOff" or .state.waiting.reason == "ImagePullBackOff" or .state.waiting.reason == "ErrImagePull"))] | length')
    OOMKILLED=$(echo "$PODS_JSON" | $JQ '[.items[].status.containerStatuses[]? | select(.state.terminated != null and .state.terminated.reason == "OOMKilled")] | length')

    # 获取所有异常 Pod 列表 (状态非 Running 或 Ready 不为 True)，优化查询
    ABNORMAL_PODS_TABLE=""
    while IFS=$'\t' read -r namespace name phase ready_status last_state_reason restart_count age_str qos_class; do
        local status_class="status-bad"
        local reason_display="$last_state_reason"
        
        # 如果没有具体的容器状态原因，则显示Phase
        if [ -z "$reason_display" ] || [ "$reason_display" == "null" ]; then
            reason_display="$phase"
        fi
        
        # 计算年龄（如果可用）
        age_str=${age_str:-"N/A"}
        qos_class=${qos_class:-"N/A"}
        
        ABNORMAL_PODS_TABLE+="<tr><td>$namespace</td><td>$name</td><td class=\"$status_class\">$reason_display</td><td>$restart_count</td><td>$age_str</td><td>$qos_class</td></tr>"
    done < <(echo "$PODS_JSON" | $JQ -r '
        .items[] | select(
            .status.phase != "Running" or 
            ([.status.conditions[]? | select(.type=="Ready") | .status] | first // "False") == "False"
        ) | [
            .metadata.namespace,
            .metadata.name,
            .status.phase,
            ([.status.conditions[]? | select(.type=="Ready") | .status] | first // "Unknown"),
            (
                (.status.containerStatuses[]?.state.waiting?.reason // .status.containerStatuses[]?.state.terminated?.reason // "") as $cont_reason |
                if $cont_reason != "" then $cont_reason else empty end
            ),
            (.status.containerStatuses[]?.restartCount // 0 | tostring),
            (.metadata.creationTimestamp | sub("\\.[0-9]+Z$"; "Z") | .[0:19] | . + "Z" | .[0:10] + " " + .[11:]) as $created_time |
            (now - ($created_time | fromdateiso8601)) as $age_seconds |
            if $age_seconds < 60 then ($age_seconds | tostring) + "s"
            elif $age_seconds < 3600 then (($age_seconds / 60) | floor | tostring) + "m"
            elif $age_seconds < 86400 then (($age_seconds / 3600) | floor | tostring) + "h"
            else (($age_seconds / 86400) | floor | tostring) + "d" end
            ),
            (.status.qosClass // "N/A")
        ] | @tsv' | sort -k1,1 -k2,2)
}

# 收集 Deployment 信息，优化性能
collect_deployments() {
    log_info "收集Deployment信息..."
    local DEPLOY_JSON
    DEPLOY_JSON=$(run_kubectl_with_timeout $KUBECTL get deployments --all-namespaces -o json)
    if [ $? -ne 0 ] || [ -z "$DEPLOY_JSON" ]; then
        log_error "无法获取Deployment信息"
        TOTAL_DEPLOY=0
        UNHEALTHY_DEPLOY=0
        DEPLOY_TABLE="<tr><td colspan='6'>无法获取Deployment信息</td></tr>"
        return
    fi

    TOTAL_DEPLOY=$(echo "$DEPLOY_JSON" | $JQ '.items | length')
    UNHEALTHY_DEPLOY=0
    DEPLOY_TABLE=""
    
    while IFS=$'\t' read -r ns name desired available ready updated ready_replicas unavailable; do
        local status="健康"
        local status_class="status-good"
        if [ "$desired" != "$available" ] || [ "$desired" != "$ready" ] || [ "$unavailable" != "0" ]; then
            status="异常"
            status_class="status-bad"
            ((UNHEALTHY_DEPLOY++))
        fi
        
        # 格式化显示，处理空值
        desired=${desired:-0}
        available=${available:-0}
        ready=${ready:-0}
        updated=${updated:-0}
        ready_replicas=${ready_replicas:-0}
        unavailable=${unavailable:-0}
        
        DEPLOY_TABLE+="<tr><td>$ns</td><td>$name</td><td>$desired</td><td>$available</td><td>$ready</td><td class=\"$status_class\">$status</td></tr>"
    done < <(echo "$DEPLOY_JSON" | $JQ -r '.items[] | [
        .metadata.namespace,
        .metadata.name,
        (.spec.replicas // 0),
        (.status.availableReplicas // 0),
        (.status.readyReplicas // 0),
        (.status.updatedReplicas // 0),
        (.status.replicas // 0),
        (.status.unavailableReplicas // 0)
    ] | @tsv')
}

# 收集 Service 信息，优化性能
collect_services() {
    log_info "收集Service信息..."
    local SVC_JSON
    SVC_JSON=$(run_kubectl_with_timeout $KUBECTL get services --all-namespaces -o json)
    if [ $? -ne 0 ] || [ -z "$SVC_JSON" ]; then
        log_error "无法获取Service信息"
        TOTAL_SVC=0
        LB_PENDING=0
        SVC_TABLE="<tr><td colspan='4'>无法获取Service信息</td></tr>"
        return
    fi

    TOTAL_SVC=$(echo "$SVC_JSON" | $JQ '.items | length')
    LB_PENDING=0
    SVC_TABLE=""
    
    while IFS=$'\t' read -r ns name svcType clusterIP externalIP ports selector session_affinity type; do
        SVC_TABLE+="<tr><td>$ns</td><td>$name</td><td>$svcType</td><td>$clusterIP</td><td>$externalIP</td><td>$ports</td><td>$selector</td><td>$session_affinity</td></tr>"
        
        # 检查 LoadBalancer pending 状态
        if [ "$svcType" == "LoadBalancer" ] && [ -z "$externalIP" ] && [ "$externalIP" != "<pending>" ]; then
            ((LB_PENDING++))
        fi
    done < <(echo "$SVC_JSON" | $JQ -r '.items[] | [
        .metadata.namespace,
        .metadata.name,
        .spec.type,
        (.spec.clusterIP // "-"),
        (.status.loadBalancer.ingress[0].ip // .status.loadBalancer.ingress[0].hostname // "<pending>"),
        ([.spec.ports[] | "\(.port):\(.targetPort)/\(.protocol)"] | join(", ")),
        ([.spec.selector | to_entries[] | "\(.key)=\(.value)"] | join(",")),
        (.spec.sessionAffinity // "None"),
        (.spec.type // "ClusterIP")
    ] | @tsv')
    
    # 重新计算 LB pending 数量（更精确）
    LB_PENDING=$(echo "$SVC_JSON" | $JQ '[.items[] | select(.spec.type=="LoadBalancer" and (.status.loadBalancer.ingress == null or .status.loadBalancer.ingress | length == 0 or (.status.loadBalancer.ingress[0].ip == null and .status.loadBalancer.ingress[0].hostname == null)))] | length')
}

# 收集 Ingress 信息，优化性能
collect_ingress() {
    log_info "收集Ingress信息..."
    local INGRESS_JSON
    if run_kubectl_with_timeout $KUBECTL get ingress --all-namespaces &>/dev/null; then
        INGRESS_JSON=$(run_kubectl_with_timeout $KUBECTL get ingress --all-namespaces -o json)
        if [ $? -ne 0 ] || [ -z "$INGRESS_JSON" ]; then
            log_error "无法获取Ingress信息"
            TOTAL_INGRESS=0
            INGRESS_TABLE="<tr><td colspan='4'>无法获取Ingress信息</td></tr>"
            return
        fi
        
        TOTAL_INGRESS=$(echo "$INGRESS_JSON" | $JQ '.items | length')
        INGRESS_TABLE=""
        
        while IFS=$'\t' read -r ns name hosts tls_enabled backend_service paths age_str; do
            # 处理空值
            hosts=${hosts:-"N/A"}
            tls_enabled=${tls_enabled:-"No"}
            backend_service=${backend_service:-"N/A"}
            paths=${paths:-"N/A"}
            age_str=${age_str:-"N/A"}
            
            INGRESS_TABLE+="<tr><td>$ns</td><td>$name</td><td>$hosts</td><td>$tls_enabled</td><td>$backend_service</td><td>$paths</td><td>$age_str</td></tr>"
        done < <(echo "$INGRESS_JSON" | $JQ -r '
            .items[] | [
                .metadata.namespace,
                .metadata.name,
                ([.spec.rules[].host] | unique | join(",")),
                (if .spec.tls and (.spec.tls | length > 0) then "Yes" else "No" end),
                ([.spec.rules[].http.paths[].backend.service.name] | unique | join(",")),
                ([.spec.rules[].http.paths[].path // "/*"] | unique | join("<br>")),
                (.metadata.creationTimestamp | sub("\\.[0-9]+Z$"; "Z") | .[0:19] | . + "Z" | .[0:10] + " " + .[11:]) as $created_time |
                (now - ($created_time | fromdateiso8601)) as $age_seconds |
                if $age_seconds < 60 then ($age_seconds | tostring) + "s"
                elif $age_seconds < 3600 then (($age_seconds / 60) | floor | tostring) + "m"
                elif $age_seconds < 86400 then (($age_seconds / 3600) | floor | tostring) + "h"
                else (($age_seconds / 86400) | floor | tostring) + "d" end
            ] | @tsv')
    else
        TOTAL_INGRESS=0
        INGRESS_TABLE="<tr><td colspan='4'>未发现 Ingress 资源或 Ingress 资源不可用</td></tr>"
    fi
}

# 收集 PVC 信息 (检查绑定状态)，优化性能
collect_pvcs() {
    log_info "收集PVC信息..."
    local PVC_JSON
    PVC_JSON=$(run_kubectl_with_timeout $KUBECTL get pvc --all-namespaces -o json)
    if [ $? -ne 0 ] || [ -z "$PVC_JSON" ]; then
        log_error "无法获取PVC信息"
        TOTAL_PVC=0
        PVC_PENDING=0
        PVC_TABLE="<tr><td colspan='6'>无法获取PVC信息</td></tr>"
        return
    fi

    TOTAL_PVC=$(echo "$PVC_JSON" | $JQ '.items | length')
    PVC_PENDING=0
    PVC_TABLE=""
    
    while IFS=$'\t' read -r ns name status volume access_modes storage_class requested capacity age_str; do
        local status_class="status-good"
        if [ "$status" == "Pending" ]; then
            status_class="status-bad"
            ((PVC_PENDING++))
        fi
        
        # 处理空值
        volume=${volume:-"-"}
        access_modes=${access_modes:-"N/A"}
        storage_class=${storage_class:-"default"}
        requested=${requested:-"N/A"}
        capacity=${capacity:-"N/A"}
        age_str=${age_str:-"N/A"}
        
        PVC_TABLE+="<tr><td>$ns</td><td>$name</td><td class=\"$status_class\">$status</td><td>$volume</td><td>$access_modes</td><td>$storage_class</td><td>$requested</td><td>$capacity</td><td>$age_str</td></tr>"
    done < <(echo "$PVC_JSON" | $JQ -r '
        .items[] | [
            .metadata.namespace,
            .metadata.name,
            .status.phase,
            (.spec.volumeName // "-"),
            ([.spec.accessModes[]] | join(",")),
            (.spec.storageClassName // "default"),
            (.spec.resources.requests.storage // "N/A"),
            (.status.capacity.storage // "N/A"),
            (.metadata.creationTimestamp | sub("\\.[0-9]+Z$"; "Z") | .[0:19] | . + "Z" | .[0:10] + " " + .[11:]) as $created_time |
            (now - ($created_time | fromdateiso8601)) as $age_seconds |
            if $age_seconds < 60 then ($age_seconds | tostring) + "s"
            elif $age_seconds < 3600 then (($age_seconds / 60) | floor | tostring) + "m"
            elif $age_seconds < 86400 then (($age_seconds / 3600) | floor | tostring) + "h"
            else (($age_seconds / 86400) | floor | tostring) + "d" end
        ] | @tsv')
}

# 收集事件 (Warning 级别)，优化性能，限制数量
collect_events() {
    log_info "收集Warning事件..."
    local EVENTS_JSON
    EVENTS_JSON=$(run_kubectl_with_timeout $KUBECTL get events --all-namespaces --field-selector type=Warning -o json)
    if [ $? -ne 0 ] || [ -z "$EVENTS_JSON" ]; then
        log_error "无法获取事件信息"
        TOTAL_WARNINGS=0
        EVENTS_TABLE="<tr><td colspan='6'>无法获取事件信息</td></tr>"
        return
    fi

    TOTAL_WARNINGS=$(echo "$EVENTS_JSON" | $JQ '.items | length')
    EVENTS_TABLE=""
    
    # 只取最近的20个事件，按时间倒序排列
    while IFS=$'\t' read -r ns kind name reason message timestamp count first_timestamp; do
        # 截取过长消息，保留重要信息
        local message_display=$(echo "$message" | cut -c1-100)
        local tooltip_msg=$(echo "$message" | sed 's|"|&quot;|g')
        
        # 格式化时间
        timestamp=${timestamp:-"N/A"}
        first_timestamp=${first_timestamp:-"N/A"}
        count=${count:-"1"}
        
        EVENTS_TABLE+="<tr><td>$ns</td><td>$kind</td><td>$name</td><td>$reason</td><td title=\"$tooltip_msg\">${message_display}...</td><td>$timestamp</td><td>$count</td><td>$first_timestamp</td></tr>"
    done < <(echo "$EVENTS_JSON" | $JQ -r '
        .items | 
        map(. | .firstTimestamp = (.firstTimestamp // .eventTime)) |
        sort_by(.lastTimestamp // .eventTime) | 
        reverse | 
        .[:20] | 
        .[] | [
            .metadata.namespace,
            .involvedObject.kind,
            .involvedObject.name,
            .reason,
            .message,
            (.lastTimestamp // .eventTime // "N/A"),
            (.count // 1 | tostring),
            (.firstTimestamp // .eventTime // "N/A")
        ] | @tsv')
}

# 收集集群总体信息，增加更多指标
collect_cluster_info() {
    log_info "收集集群基本信息..."
    CLUSTER_INFO=$(run_kubectl_with_timeout $KUBECTL cluster-info 2>/dev/null | head -n1 || echo "无法获取集群信息")
    K8S_VERSION=$(run_kubectl_with_timeout $KUBECTL version -o json 2>/dev/null | $JQ -r '.serverVersion.gitVersion' 2>/dev/null || echo "unknown")
    CURRENT_CONTEXT=$(run_kubectl_with_timeout $KUBECTL config current-context 2>/dev/null || echo "unknown")
    SERVER_ADDR=$(run_kubectl_with_timeout $KUBECTL config view -o jsonpath="{.clusters[?(@.name==\"$CURRENT_CONTEXT\")].cluster.server}" 2>/dev/null || echo "unknown")
    
    # 收集更多集群信息
    NODE_ARCHITECTURE=$(run_kubectl_with_timeout $KUBECTL get nodes -o jsonpath='{.items[0].status.nodeInfo.architecture}' 2>/dev/null || echo "unknown")
    NODE_OS_IMAGE=$(run_kubectl_with_timeout $KUBECTL get nodes -o jsonpath='{.items[0].status.nodeInfo.osImage}' 2>/dev/null || echo "unknown")
    CNI_PLUGIN=$(run_kubectl_with_timeout $KUBECTL get pods -n kube-system -l k8s-app -o jsonpath='{.items[0].metadata.labels.k8s-app}' 2>/dev/null || echo "unknown")
}

# ----------------------------- 生成 HTML 报告 ---------------------------------
generate_html() {
    log_info "生成HTML报告: $REPORT_FILE"

    # 计算健康分数 (满分100)，引入权重
    local node_weight=0.25
    local pod_weight=0.30
    local deploy_weight=0.25
    local pvc_weight=0.20
    
    local node_score=0
    local pod_score=0
    local deploy_score=0
    local pvc_score=0
    
    if [ $NODES_COUNT -gt 0 ]; then
        node_score=$(( NODES_READY * 100 / NODES_COUNT ))
    fi
    
    if [ $TOTAL_PODS -gt 0 ]; then
        local healthy_pods=$(( TOTAL_PODS - PENDING_PODS - FAILED_PODS - UNKNOWN_PODS ))
        pod_score=$(( healthy_pods * 100 / TOTAL_PODS ))
    fi
    
    if [ $TOTAL_DEPLOY -gt 0 ]; then
        deploy_score=$(( (TOTAL_DEPLOY - UNHEALTHY_DEPLOY) * 100 / TOTAL_DEPLOY ))
    fi
    
    if [ $TOTAL_PVC -gt 0 ]; then
        pvc_score=$(( (TOTAL_PVC - PVC_PENDING) * 100 / TOTAL_PVC ))
    fi
    
    # 加权平均健康分数
    local health_score
    health_score=$(awk "BEGIN {printf \"%.0f\", ($node_score * $node_weight + $pod_score * $pod_weight + $deploy_score * $deploy_weight + $pvc_score * $pvc_weight) * 100}")
    health_score=$((health_score / 100))
    
    # 确定分数颜色
    local score_color="#10b981"
    local score_status="健康"
    if [ $health_score -ge 80 ]; then
        score_color="#10b981"
        score_status="健康"
    elif [ $health_score -ge 60 ]; then
        score_color="#f59e0b"
        score_status="警告"
    else
        score_color="#ef4444"
        score_status="危险"
    fi

    # 定义图表数据
    local chart_data_node="[$node_score, $((100 - node_score))]"
    local chart_data_pod="[$pod_score, $((100 - pod_score))]"
    local chart_data_deploy="[$deploy_score, $((100 - deploy_score))]"
    local chart_data_pvc="[$pvc_score, $((100 - pvc_score))]"
    local chart_data_total="[$health_score, $((100 - health_score))]"

    cat > "$REPORT_FILE" <<EOF
<!DOCTYPE html>
<html lang="zh-CN">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>K8s 集群巡检报告 - ${CURRENT_CONTEXT}</title>
    <!-- Font Awesome & Google Fonts -->
    <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.0.0-beta3/css/all.min.css">
    <link href="https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700;800&display=swap" rel="stylesheet">
    <!-- Chart.js CDN -->
    <script src="https://cdn.jsdelivr.net/npm/chart.js@4.4.0/dist/chart.umd.min.js"></script>
    <style>
        * {
            margin: 0;
            padding: 0;
            box-sizing: border-box;
        }
        body {
            background: linear-gradient(135deg, #0f172a 0%, #1e293b 100%);
            font-family: 'Inter', sans-serif;
            color: #e2e8f0;
            padding: 2rem;
            line-height: 1.6;
        }
        .container {
            max-width: 1600px;
            margin: 0 auto;
        }
        /* 头部区域 */
        .header {
            text-align: center;
            margin-bottom: 3rem;
            animation: fadeInDown 0.8s ease;
        }
        .header h1 {
            font-size: 2.5rem;
            font-weight: 800;
            background: linear-gradient(135deg, #60a5fa, #a78bfa);
            -webkit-background-clip: text;
            background-clip: text;
            color: transparent;
            margin-bottom: 0.5rem;
        }
        .header .sub {
            color: #94a3b8;
            font-size: 0.95rem;
        }
        .badge {
            display: inline-block;
            background: #1e293b;
            border-radius: 40px;
            padding: 0.25rem 1rem;
            font-size: 0.8rem;
            margin-top: 0.5rem;
            margin-right: 0.5rem;
        }
        /* 卡片网格 */
        .grid {
            display: grid;
            grid-template-columns: repeat(auto-fit, minmax(280px, 1fr));
            gap: 1.5rem;
            margin-bottom: 2rem;
        }
        .card {
            background: rgba(30, 41, 59, 0.7);
            backdrop-filter: blur(10px);
            border-radius: 24px;
            padding: 1.5rem;
            border: 1px solid rgba(255,255,255,0.1);
            transition: transform 0.2s, box-shadow 0.2s;
            box-shadow: 0 8px 20px rgba(0,0,0,0.2);
            position: relative;
            overflow: hidden;
        }
        .card::before {
            content: '';
            position: absolute;
            top: 0;
            left: 0;
            right: 0;
            height: 3px;
            background: linear-gradient(90deg, #60a5fa, #a78bfa);
        }
        .card:hover {
            transform: translateY(-4px);
            box-shadow: 0 20px 30px -12px rgba(0,0,0,0.4);
            border-color: rgba(96,165,250,0.4);
        }
        .card-title {
            font-size: 1.1rem;
            font-weight: 600;
            text-transform: uppercase;
            letter-spacing: 1px;
            color: #94a3b8;
            margin-bottom: 1rem;
            display: flex;
            align-items: center;
            gap: 0.5rem;
        }
        .card-value {
            font-size: 2.5rem;
            font-weight: 800;
            margin-bottom: 0.5rem;
        }
        .card-desc {
            font-size: 0.8rem;
            color: #cbd5e1;
            white-space: nowrap;
            overflow: hidden;
            text-overflow: ellipsis;
        }
        .status-good {
            color: #10b981;
            font-weight: 600;
        }
        .status-bad {
            color: #ef4444;
            font-weight: 600;
        }
        .status-warning {
            color: #f59e0b;
        }
        /* 健康仪表盘圆形 */
        .score-container {
            display: flex;
            justify-content: center;
            align-items: center;
            flex-direction: column;
        }
        canvas.healthChart {
            width: 120px !important;
            height: 120px !important;
            margin-bottom: 1rem;
        }
        .score-text {
            font-size: 1.2rem;
            font-weight: 700;
        }
        /* 表格样式 */
        .table-wrapper {
            overflow-x: auto;
            border-radius: 20px;
            background: rgba(15, 23, 42, 0.6);
            padding: 0.5rem;
            margin-bottom: 2rem;
            box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1);
        }
        table {
            width: 100%;
            border-collapse: collapse;
            font-size: 0.85rem;
        }
        th {
            text-align: left;
            padding: 0.8rem 0.8rem;
            background: #0f172a;
            color: #94a3b8;
            font-weight: 600;
            border-bottom: 1px solid #334155;
            position: sticky;
            top: 0;
            z-index: 10;
        }
        td {
            padding: 0.7rem 0.8rem;
            border-bottom: 1px solid #1e293b;
        }
        tr:hover {
            background: rgba(51, 65, 85, 0.4);
        }
        .section-title {
            font-size: 1.5rem;
            font-weight: 700;
            margin: 2rem 0 1rem 0;
            display: flex;
            align-items: center;
            gap: 0.8rem;
            color: #e2e8f0;
        }
        .footer {
            text-align: center;
            margin-top: 3rem;
            padding: 1rem;
            font-size: 0.75rem;
            color: #475569;
            border-top: 1px solid #334155;
            padding-top: 1rem;
        }
        @keyframes fadeInDown {
            from {
                opacity: 0;
                transform: translateY(-20px);
            }
            to {
                opacity: 1;
                transform: translateY(0);
            }
        }
        i.icon {
            width: 28px;
            color: #60a5fa;
        }
        .cluster-info {
            background: rgba(0,0,0,0.3);
            border-radius: 16px;
            padding: 1rem;
            margin-bottom: 2rem;
            font-size: 0.9rem;
            display: flex;
            flex-wrap: wrap;
            justify-content: space-between;
            gap: 1rem;
        }
        .chart-container {
            display: flex;
            justify-content: space-around;
            flex-wrap: wrap;
            gap: 1rem;
            margin: 2rem 0;
        }
        .chart-item {
            text-align: center;
            flex: 1;
            min-width: 150px;
        }
        .chart-title {
            font-size: 0.9rem;
            color: #94a3b8;
            margin-bottom: 0.5rem;
        }
        .collapsible-table {
            cursor: pointer;
        }
        .collapsible-table .table-header {
            background: #1e293b;
            padding: 0.5rem;
            border-radius: 10px;
            margin-bottom: 0.5rem;
            display: flex;
            justify-content: space-between;
            align-items: center;
        }
        .collapsible-table .table-content {
            max-height: 400px;
            overflow-y: auto;
        }
        .toggle-icon {
            transition: transform 0.3s;
        }
        .collapsed .toggle-icon {
            transform: rotate(-90deg);
        }
        .collapsed .table-content {
            display: none;
        }
        @media (max-width: 768px) {
            body {
                padding: 1rem;
            }
            .cluster-info {
                flex-direction: column;
                align-items: flex-start;
            }
            .grid {
                grid-template-columns: 1fr;
            }
            .chart-container {
                flex-direction: column;
            }
        }
    </style>
</head>
<body>
<div class="container">
    <div class="header">
        <h1><i class="fas fa-kubernetes"></i> K8s 集群巡检报告</h1>
        <div class="sub">全面健康检查 & 异常诊断 | 生成时间: $(date '+%Y-%m-%d %H:%M:%S')</div>
        <div>
            <div class="badge"><i class="fas fa-check-circle"></i> 集群: ${CURRENT_CONTEXT}</div>
            <div class="badge"><i class="fas fa-server"></i> 版本: ${K8S_VERSION}</div>
            <div class="badge"><i class="fas fa-microchip"></i> 架构: ${NODE_ARCHITECTURE}</div>
            <div class="badge"><i class="fas fa-desktop"></i> 系统: ${NODE_OS_IMAGE}</div>
        </div>
    </div>

    <div class="cluster-info">
        <span><i class="fas fa-server"></i> <strong>API Server:</strong> ${SERVER_ADDR}</span>
        <span><i class="fab fa-kubernetes"></i> <strong>版本:</strong> ${K8S_VERSION}</span>
        <span><i class="fas fa-cogs"></i> <strong>CNI:</strong> ${CNI_PLUGIN}</span>
        <span><i class="fas fa-chart-line"></i> <strong>巡检范围:</strong> 全集群资源</span>
    </div>

    <!-- 健康分数卡片及统计 -->
    <div class="grid">
        <div class="card">
            <div class="card-title"><i class="fas fa-heartbeat icon"></i> 集群健康指数</div>
            <div class="score-container">
                <canvas id="healthChart" class="healthChart" width="120" height="120"></canvas>
                <div class="score-text" style="color: ${score_color};">${health_score}分 - ${score_status}</div>
            </div>
        </div>
        <div class="card">
            <div class="card-title"><i class="fas fa-server icon"></i> 节点</div>
            <div class="card-value">${NODES_READY} / ${NODES_COUNT}</div>
            <div class="card-desc">就绪节点 / 总数 &nbsp; <span class="${NODES_NOT_READY>0?'status-bad':'status-good'}">异常节点: ${NODES_NOT_READY}</span></div>
        </div>
        <div class="card">
            <div class="card-title"><i class="fas fa-cubes icon"></i> Pods</div>
            <div class="card-value">${RUNNING_PODS} / ${TOTAL_PODS}</div>
            <div class="card-desc">运行中 | 等待: ${PENDING_PODS} 失败: ${FAILED_PODS} <br> 异常容器: ${ABNORMAL_CONTAINERS} OOM: ${OOMKILLED}</div>
        </div>
        <div class="card">
            <div class="card-title"><i class="fas fa-rocket icon"></i> Deployments</div>
            <div class="card-value">$((TOTAL_DEPLOY - UNHEALTHY_DEPLOY)) / ${TOTAL_DEPLOY}</div>
            <div class="card-desc">健康/总数 &nbsp; 不健康副本: ${UNHEALTHY_DEPLOY}</div>
        </div>
        <div class="card">
            <div class="card-title"><i class="fas fa-database icon"></i> PersistentVolumeClaims</div>
            <div class="card-value">$((TOTAL_PVC - PVC_PENDING)) / ${TOTAL_PVC}</div>
            <div class="card-desc">已绑定 | Pending: ${PVC_PENDING}</div>
        </div>
        <div class="card">
            <div class="card-title"><i class="fas fa-exclamation-triangle icon"></i> Warning事件</div>
            <div class="card-value">${TOTAL_WARNINGS}</div>
            <div class="card-desc">最近Warning级别事件</div>
        </div>
    </div>

    <!-- 图表展示 -->
    <div class="section-title"><i class="fas fa-chart-bar"></i> 健康度分布</div>
    <div class="chart-container">
        <div class="chart-item">
            <div class="chart-title">节点健康度</div>
            <canvas id="nodeChart" class="healthChart" width="100" height="100"></canvas>
            <div>${node_score}%</div>
        </div>
        <div class="chart-item">
            <div class="chart-title">Pod健康度</div>
            <canvas id="podChart" class="healthChart" width="100" height="100"></canvas>
            <div>${pod_score}%</div>
        </div>
        <div class="chart-item">
            <div class="chart-title">Deployment健康度</div>
            <canvas id="deployChart" class="healthChart" width="100" height="100"></canvas>
            <div>${deploy_score}%</div>
        </div>
        <div class="chart-item">
            <div class="chart-title">PVC健康度</div>
            <canvas id="pvcChart" class="healthChart" width="100" height="100"></canvas>
            <div>${pvc_score}%</div>
        </div>
    </div>

    <!-- 节点详情 -->
    <div class="section-title"><i class="fas fa-server"></i> 节点状态详情</div>
    <div class="table-wrapper">
        <table>
            <thead><tr><th>节点名称</th><th>状态 (Ready)</th><th>CPU (Capacity/Allocatable)</th><th>内存 (Capacity/Allocatable)</th><th>Taints</th><th>内部IP</th><th>外部IP</th><th>系统/架构</th><th>内核版本</th></tr></thead>
            <tbody>
                ${NODES_TABLE}
            </tbody>
        </table>
    </div>

    <!-- 异常Pod列表 -->
    <div class="section-title"><i class="fas fa-exclamation-circle"></i> 异常Pod / 容器问题</div>
    <div class="table-wrapper">
        <table>
            <thead><tr><th>命名空间</th><th>Pod名称</th><th>异常原因</th><th>重启次数</th><th>运行时长</th><th>QoS等级</th></tr></thead>
            <tbody>
                ${ABNORMAL_PODS_TABLE:-<tr><td colspan="6" style="text-align:center">✅ 没有发现异常Pod</td></tr>}
            </tbody>
        </table>
    </div>

    <!-- Deployment 状态 -->
    <div class="section-title"><i class="fas fa-chart-simple"></i> Deployment 副本健康度</div>
    <div class="table-wrapper">
        <table>
            <thead><tr><th>命名空间</th><th>名称</th><th>期望副本</th><th>可用副本</th><th>就绪副本</th><th>状态</th></tr></thead>
            <tbody>
                ${DEPLOY_TABLE}
            </tbody>
        </table>
    </div>

    <!-- Service 列表 -->
    <div class="section-title"><i class="fas fa-link"></i> 服务 (Service)</div>
    <div class="table-wrapper">
        <table>
            <thead><tr><th>命名空间</th><th>服务名</th><th>类型</th><th>ClusterIP</th><th>ExternalIP</th><th>端口</th><th>Selector</th><th>会话亲和性</th></tr></thead>
            <tbody>
                ${SVC_TABLE}
            </tbody>
        </table>
        <div style="margin-top: 0.5rem; font-size:0.8rem; color:#f59e0b;">⚠️ LoadBalancer Pending数量: ${LB_PENDING}</div>
    </div>

    <!-- Ingress 列表 -->
    <div class="section-title"><i class="fas fa-globe"></i> Ingress 路由</div>
    <div class="table-wrapper">
        <table>
            <thead><tr><th>命名空间</th><th>Ingress名称</th><th>Hosts</th><th>TLS</th><th>后端服务</th><th>路径</th><th>运行时长</th></tr></thead>
            <tbody>
                ${INGRESS_TABLE}
            </tbody>
        </table>
    </div>

    <!-- PVC 列表 -->
    <div class="section-title"><i class="fas fa-hdd"></i> 持久卷声明 (PVC)</div>
    <div class="table-wrapper">
        <table>
            <thead><tr><th>命名空间</th><th>PVC名称</th><th>状态</th><th>绑定PV</th><th>访问模式</th><th>存储类</th><th>请求容量</th><th>实际容量</th><th>运行时长</th></tr></thead>
            <tbody>
                ${PVC_TABLE}
            </tbody>
        </table>
    </div>

    <!-- 最新Warning事件 -->
    <div class="section-title"><i class="fas fa-bell"></i> 近期Warning事件 (最多20条)</div>
    <div class="table-wrapper">
        <table>
            <thead><tr><th>命名空间</th><th>资源类型</th><th>资源名</th><th>原因</th><th>消息摘要</th><th>最后发生时间</th><th>发生次数</th><th>首次发生时间</th></tr></thead>
            <tbody>
                ${EVENTS_TABLE:-<tr><td colspan="8">无Warning事件，集群很干净~</td></tr>}
            </tbody>
        </table>
    </div>

    <div class="footer">
        <i class="fas fa-shield-alt"></i> 巡检脚本 v2.0 | 数据基于 kubectl 实时采集 | 报告生成: $(date)
    </div>
</div>

<script>
    // 通用图表创建函数
    function createDoughnutChart(ctxId, data, color) {
        const ctx = document.getElementById(ctxId).getContext('2d');
        return new Chart(ctx, {
            type: 'doughnut',
            data: {
                datasets: [{
                    data: data,
                    backgroundColor: [color, '#334155'],
                    borderWidth: 0,
                    cutout: '70%',
                    borderRadius: 10,
                    spacing: 5
                }]
            },
            options: {
                responsive: true,
                maintainAspectRatio: true,
                plugins: { 
                    tooltip: { 
                        enabled: true,
                        callbacks: {
                            label: function(context) {
                                const total = context.dataset.data.reduce((a, b) => a + b, 0);
                                const percentage = Math.round((context.raw / total) * 100);
                                return context.index === 0 ? '健康: ' + percentage + '%' : '异常: ' + percentage + '%';
                            }
                        }
                    }, 
                    legend: { display: false } 
                }
            }
        });
    }

    // 创建各种图表
    createDoughnutChart('healthChart', $chart_data_total, '${score_color}');
    createDoughnutChart('nodeChart', $chart_data_node, '${score_color}');
    createDoughnutChart('podChart', $chart_data_pod, '${score_color}');
    createDoughnutChart('deployChart', $chart_data_deploy, '${score_color}');
    createDoughnutChart('pvcChart', $chart_data_pvc, '${score_color}');

    // 表格折叠功能
    document.querySelectorAll('.collapsible-table .table-header').forEach(header => {
        header.addEventListener('click', function() {
            const tableContainer = this.parentElement;
            tableContainer.classList.toggle('collapsed');
        });
    });
</script>
</body>
</html>
EOF

    log_info "HTML报告已生成: $(pwd)/$REPORT_FILE"
}

# ----------------------------- 主函数 -------------------------------------
main() {
    log_info "开始执行K8s集群巡检脚本..."
    start_time=$(date +%s)
    
    check_dependencies
    collect_cluster_info
    collect_nodes
    collect_pods
    collect_deployments
    collect_services
    collect_ingress
    collect_pvcs
    collect_events
    generate_html

    end_time=$(date +%s)
    duration=$((end_time - start_time))
    
    echo ""
    log_info "巡检完成！耗时 ${duration} 秒"
    log_info "报告已保存至: $REPORT_FILE"
    log_info "您可以使用浏览器打开该文件查看炫酷的仪表板报告。"
    
    # 输出简要摘要
    echo ""
    log_info "=== 巡检摘要 ==="
    log_info "总节点数: $NODES_COUNT (就绪: $NODES_READY, 异常: $NODES_NOT_READY)"
    log_info "总Pod数: $TOTAL_PODS (运行中: $RUNNING_PODS, 异常: $((TOTAL_PODS - RUNNING_PODS)))"
    log_info "总Deployment数: $TOTAL_DEPLOY (健康: $((TOTAL_DEPLOY - UNHEALTHY_DEPLOY)), 异常: $UNHEALTHY_DEPLOY)"
    log_info "总PVC数: $TOTAL_PVC (已绑定: $((TOTAL_PVC - PVC_PENDING)), Pending: $PVC_PENDING)"
    log_info "Warning事件数: $TOTAL_WARNINGS"
    log_info "集群健康评分: ${health_score}/100 (${score_status})"
}

# 执行主函数
main "$@"
相关推荐
Net_Walke2 小时前
【Ubuntu】共享文件夹 /mnt/hgfs 下不显示问题解决
linux·运维·ubuntu
CHANG_THE_WORLD2 小时前
PDFIUM如何处理宽度数组
java·linux·服务器
孙同学_2 小时前
【Linux篇】应用层自定义协议与序列化
linux·服务器·网络
航Hang*2 小时前
第3章：Linux系统安全管理——第1节：Linux 防火墙部署（firewalld）
linux·服务器·网络·学习·系统安全·vmware
桦02 小时前
【Linux复习】:多线程
linux·运维
南境十里·墨染春水2 小时前
linux学习进展 基础命令 vi基础命令
linux·运维·服务器·笔记·学习
Deitymoon2 小时前
linux——读写锁
linux
迷路爸爸1802 小时前
Docker 入门学习笔记 05：卷到底是什么，为什么容器删了数据却还能保留
笔记·学习·docker
赵民勇2 小时前
locales包详解
linux
热门推荐
01GitHub 镜像站点02OpenClaw 请求超时 llm request timed out 怎么解决？3 种方案实测，附完整排查流程03AI 编程效率翻倍：Superpowers Skills 上手清单 + 完整指南04Qwen3.5-Omni与Qwen3.6模型全面解析（含测评/案例/使用教程）05VMware Workstation Pro 17 虚拟机完整安装教程（2026最新）06Claude Code 未登录 使用第三方模型07Oh My Codex 快速使用指南08UV安装并设置国内源09【技术干货】Gemma 4 上手深度指南：本地多模态大模型的新基线10最新更新版本，OpenClaw v2026.4.2 深度解读剖析：Task Flow 重磅回归与安全架构的全面硬化