Skip to content

U
ubains-module-test
提交 35387cfc authored 作者: 陈泽健's avatar 陈泽健
浏览文件
操作

Revert "feat(server): 增强服务器自检脚本功能" 

This reverts commit f4c2be9d.
上级 f4c2be9d
显示空白字符变更
内嵌 并排
正在显示 包含 1 行增加 和 749 行删除
AuxiliaryTool/ScriptTool/新服务自检/check_server_health.ps1
浏览文件 @ 35387cfc
......@@ -384,42 +384,6 @@ function Test-SystemBasicInfo {
# 系统负载
$info.LoadAverage = Invoke-SSHCommand "cat /proc/loadavg | awk '{print `$1, `$2, `$3}'" -Timeout 10
# 内核启动参数
$cmdline = Invoke-SSHCommand "cat /proc/cmdline 2>/dev/null" -Timeout 10
if ($cmdline) {
$info.KernelCmdline = $cmdline
}
# 系统资源限制
$ulimitInfo = Invoke-SSHCommand "ulimit -a 2>/dev/null | grep -E 'open files|max user processes'" -Timeout 10
if ($ulimitInfo) {
$info.UlimitInfo = $ulimitInfo
}
# 内核关键参数
$kernelParams = Invoke-SSHCommand "sysctl fs.file-max fs.inotify.max_user_watches net.core.somaxconn 2>/dev/null" -Timeout 10
if ($kernelParams) {
$info.KernelParams = $kernelParams
}
# 内存压力检测
$memPressure = Invoke-SSHCommand "cat /proc/pressure/memory 2>/dev/null | head -3" -Timeout 10
if ($memPressure) {
$info.MemoryPressure = $memPressure
}
# 虚拟内存统计
$vmstat = Invoke-SSHCommand "cat /proc/vmstat 2>/dev/null | grep -E 'nr_dirty|nr_writeback|nr_slab' | head -10" -Timeout 10
if ($vmstat) {
$info.VMStat = $vmstat
}
# 大页内存使用
$hugePages = Invoke-SSHCommand "cat /proc/meminfo 2>/dev/null | grep -E 'HugePages|Hugepagesize'" -Timeout 10
if ($hugePages) {
$info.HugePages = $hugePages
}
$script:systemInfo = $info
$script:检测结果["系统基础信息"] = $info
......@@ -494,58 +458,6 @@ function Test-CPUResource {
}
}
# ========== 新增：中断统计检测 ==========
try {
$interruptInfo = Invoke-SSHCommand "cat /proc/interrupts | grep -E '^[0-9]+:' | head -1" -Timeout 10
if ($interruptInfo) {
# 解析中断总数（格式：CPU0: 123456 789 0 ...）
$totalInterrupts = 0
$fields = $interruptInfo -split "\s+"
foreach ($field in $fields[1..($fields.Length-1)]) {
if ($field -match "^\d+$") {
$totalInterrupts += [int]$field
}
}
$results += [PSCustomObject]@{
Name = "CPU中断统计"
Value = "{0:N0}" -f $totalInterrupts
Threshold = "-"
Status = "正常"
Message = "CPU中断总次数"
}
}
}
catch {
Write-Log "中断统计检测失败: $($_.Exception.Message)" "WARN"
}
# ========== 新增：软中断统计检测 ==========
try {
$softirqInfo = Invoke-SSHCommand "cat /proc/softirqs | grep -E '^[0-9]+:' | head -1" -Timeout 10
if ($softirqInfo) {
# 解析软中断总数
$totalSoftirqs = 0
$fields = $softirqInfo -split "\s+"
foreach ($field in $fields[1..($fields.Length-1)]) {
if ($field -match "^\d+$") {
$totalSoftirqs += [int]$field
}
}
$results += [PSCustomObject]@{
Name = "CPU软中断统计"
Value = "{0:N0}" -f $totalSoftirqs
Threshold = "-"
Status = "正常"
Message = "CPU软中断总次数"
}
}
}
catch {
Write-Log "软中断统计检测失败: $($_.Exception.Message)" "WARN"
}
# 保存结果
foreach ($result in $results) {
Save-TestResult "CPU资源" $result
......@@ -838,79 +750,6 @@ function Test-DiskResource {
}
}
# 磁盘挂载情况检测
Write-Log "检测磁盘挂载情况..."
$mountOutput = Invoke-SSHCommand "mount 2>/dev/null | grep -E 'nfs|cifs|fuse.sshfs'" -Timeout 10
if ($mountOutput) {
if ($mountOutput -is [array]) { $mountOutput = $mountOutput -join "`n" }
$remoteMounts = @()
foreach ($mountLine in $mountOutput -split "`n") {
if ($mountLine -match "on\s+(\S+)\s+type") {
$remoteMounts += $matches[1]
}
}
if ($remoteMounts.Count -gt 0) {
$results += [PSCustomObject]@{
Name = "远程磁盘挂载"
Value = "$($remoteMounts.Count)个"
Threshold = "-"
Status = "正常"
Message = "远程挂载: $($remoteMounts -join ', ')"
}
}
}
# RAID状态检测
Write-Log "检测RAID状态..."
$mdstat = Invoke-SSHCommand "cat /proc/mdstat 2>/dev/null" -Timeout 10
if ($mdstat -and $mdstat -match "md\d+ : active") {
$raidDevices = @()
foreach ($line in $mdstat -split "`n") {
if ($line -match "(md\d+)\s*:\s*active\s+(\S+)") {
$raidDevice = $matches[1]
$raidLevel = $matches[2]
$raidDevices += "${raidDevice}(${raidLevel})"
}
}
if ($raidDevices.Count -gt 0) {
# 检查RAID状态是否正常
$raidStatus = if ($mdstat -match "degraded|fail") { "警告" } else { "正常" }
$results += [PSCustomObject]@{
Name = "RAID状态"
Value = ($raidDevices -join ", ")
Threshold = "-"
Status = $raidStatus
Message = "软件RAID设备状态"
}
if ($raidStatus -ne "正常") {
Add-Issue -Message "RAID状态异常: $($raidDevices -join ', ')" -Level $raidStatus
}
}
}
# 磁盘调度算法检测
Write-Log "检测磁盘调度算法..."
$schedulerOutput = Invoke-SSHCommand "ls /sys/block/*/queue/scheduler 2>/dev/null | head -5 | xargs -I {} sh -c 'echo {}: \`cat {}\`'" -Timeout 10
if ($schedulerOutput) {
if ($schedulerOutput -is [array]) { $schedulerOutput = $schedulerOutput -join "`n" }
$schedulerInfo = @()
foreach ($schedLine in $schedulerOutput -split "`n") {
if ($schedLine -match "\[([^\]]+)\]") {
$schedulerInfo += $schedLine
}
}
if ($schedulerInfo.Count -gt 0) {
$results += [PSCustomObject]@{
Name = "磁盘调度器"
Value = "已配置"
Threshold = "-"
Status = "正常"
Message = ($schedulerInfo -join "; ")
}
}
}
# 保存结果
foreach ($result in $results) {
Save-TestResult "磁盘资源" $result
......@@ -1091,64 +930,6 @@ function Test-ProcessStatus {
}
}
# ========== 新增：按进程统计打开文件数 ==========
try {
$fdByProcess = Invoke-SSHCommand "for pid in \$(ps -eo pid | head -20); do echo -n \"\$pid \"; ls /proc/\$pid/fd 2>/dev/null | wc -l; done | sort -t: -k2 -rn | head -10" -Timeout 15
if ($fdByProcess) {
if ($fdByProcess -is [array]) { $fdByProcess = $fdByProcess -join "`n" }
$processFds = @()
$totalFds = 0
foreach ($line in $fdByProcess -split "`n") {
if ($line -match "^(\d+)\s+(\d+)") {
$processFds += "PID:$($matches[1]):$($matches[2])个"
$totalFds += [int]$matches[2]
}
}
if ($processFds.Count -gt 0) {
$results += [PSCustomObject]@{
Name = "打开文件数TOP10"
Value = "已统计"
Threshold = "-"
Status = "正常"
Message = "打开文件最多的10个进程: " + ($processFds -join ", ")
}
}
}
}
catch {
Write-Log "按进程统计打开文件数失败: $($_.Exception.Message)" "WARN"
}
# 进程可执行文件路径检测
try {
$exePaths = Invoke-SSHCommand "ps -eo pid,comm --no-headers | head -10 | while read pid comm; do echo -n \"PID:\$pid \"; readlink /proc/\$pid/exe 2>/dev/null || echo 'N/A'; done" -Timeout 15
if ($exePaths) {
if ($exePaths -is [array]) { $exePaths = $exePaths -join "`n" }
$exeDetails = @()
foreach ($exeLine in $exePaths -split "`n") {
if ($exeLine -match "PID:\d+\s+(.+)") {
$exeDetails += $matches[1].Trim()
}
}
if ($exeDetails.Count -gt 0) {
$results += [PSCustomObject]@{
Name = "进程可执行路径"
Value = "已获取"
Threshold = "-"
Status = "正常"
Message = "TOP10进程可执行文件路径"
}
}
}
}
catch {
Write-Log "进程可执行路径检测失败: $($_.Exception.Message)" "WARN"
}
# 保存结果
foreach ($result in $results) {
Save-TestResult "进程状态" $result
......@@ -1275,106 +1056,6 @@ function Test-NetworkStatus {
}
}
# ========== 新增：TCP参数配置检测 ==========
try {
$tcpParams = Invoke-SSHCommand "sysctl net.ipv4.tcp_fin_timeout net.ipv4.tcp_keepalive_time net.ipv4.tcp_max_syn_backlog net.ipv4.tcp_tw_reuse 2>&1" -Timeout 10
if ($tcpParams -and $tcpParams -notmatch "error|Error") {
if ($tcpParams -is [array]) { $tcpParams = $tcpParams -join "`n" }
$paramDetails = @()
foreach ($line in $tcpParams -split "`n") {
if ($line -match "net\.ipv4\.tcp_\w+\s*=\s*(.+)") {
$paramName = $line -replace "net\.ipv4\.tcp_", "" -replace "\s*=", ""
$paramValue = $matches[1].Trim()
$paramDetails += "${paramName}:${paramValue}"
}
}
if ($paramDetails.Count -gt 0) {
$results += [PSCustomObject]@{
Name = "TCP参数配置"
Value = "已获取"
Threshold = "-"
Status = "正常"
Message = ($paramDetails -join ", ")
}
}
}
}
catch {
Write-Log "TCP参数配置检测失败: $($_.Exception.Message)" "WARN"
}
# ========== 新增：TCP扩展统计检测 ==========
try {
$tcpStats = Invoke-SSHCommand "cat /proc/net/snmp | grep Tcp: | head -1" -Timeout 10
if ($tcpStats) {
# 解析TCP统计：Tcp: RtoAlgorithm RtoMin RtoMax MaxConn ActiveOpens PassiveOpens AttemptFails ...
$tcpFields = $tcpStats -split "\s+"
if ($tcpFields.Count -gt 10) {
$activeOpens = if ($tcpFields[5] -match "\d+") { $tcpFields[5] } else { "0" }
$passiveOpens = if ($tcpFields[6] -match "\d+") { $tcpFields[6] } else { "0" }
$attemptFails = if ($tcpFields[7] -match "\d+") { $tcpFields[7] } else { "0" }
$estabResets = if ($tcpFields[8] -match "\d+") { $tcpFields[8] } else { "0" }
$results += [PSCustomObject]@{
Name = "TCP扩展统计"
Value = "已统计"
Threshold = "-"
Status = "正常"
Message = "主动打开:$activeOpens, 被动打开:$passiveOpens, 尝试失败:$attemptFails, 重置:$estabResets"
}
}
}
}
catch {
Write-Log "TCP扩展统计检测失败: $($_.Exception.Message)" "WARN"
}
# ========== 新增：ARP表检测 ==========
try {
$arpTable = Invoke-SSHCommand "ip neigh show | wc -l" -Timeout 10
if ($arpTable -match "\d+") {
$arpCount = [int]$matches[0]
$arpStale = Invoke-SSHCommand "ip neigh show | grep -c STALE" -Timeout 10
$staleCount = if ($arpStale -match "\d+") { [int]$matches[0] } else { 0 }
$arpStatus = if ($staleCount -gt $arpCount * 0.1) { "警告" } else { "正常" }
$results += [PSCustomObject]@{
Name = "ARP表"
Value = "$arpCount 项"
Threshold = "-"
Status = $arpStatus
Message = "ARP邻居表条目数，过期:$staleCount"
}
}
}
catch {
Write-Log "ARP表检测失败: $($_.Exception.Message)" "WARN"
}
# ========== 新增：路由表检测 ==========
try {
$routeCount = Invoke-SSHCommand "ip route show | wc -l" -Timeout 10
$defaultRoute = Invoke-SSHCommand "ip route show | grep default" -Timeout 10
if ($routeCount -match "\d+") {
$routeCount = [int]$matches[0]
$defaultInfo = if ($defaultRoute) { "有默认网关" } else { "无默认网关" }
$results += [PSCustomObject]@{
Name = "路由表"
Value = "$routeCount 条路由"
Threshold = "-"
Status = "正常"
Message = $defaultInfo
}
}
}
catch {
Write-Log "路由表检测失败: $($_.Exception.Message)" "WARN"
}
# 保存结果
foreach ($result in $results) {
Save-TestResult "网络连接" $result
......@@ -1553,128 +1234,6 @@ function Test-SecurityStatus {
}
}
# ========== 新增：iptables规则检测 ==========
try {
$iptablesRules = Invoke-SSHCommand "iptables -L -n --line-numbers 2>/dev/null | head -20" -Timeout 10
if ($iptablesRules -and $iptablesRules -notmatch "error") {
if ($iptablesRules -is [array]) { $iptablesRules = $iptablesRules -join "`n" }
# 统计规则数量
$ruleCount = ($iptablesRules -split "`n" | Where-Object { $_ -match "^\d+" } | Measure-Object).Count
# 检查是否有INPUT策略
$inputPolicy = Invoke-SSHCommand "iptables -L INPUT -n | grep '^Chain' | awk '{print `$3}'" -Timeout 10
$policyInfo = if ($inputPolicy) { $inputPolicy.Trim() } else { "未知" }
$results += [PSCustomObject]@{
Name = "iptables规则"
Value = "$ruleCount 条"
Threshold = "-"
Status = "正常"
Message = "防火墙规则数量，INPUT策略:$policyInfo"
}
}
}
catch {
Write-Log "iptables规则检测失败: $($_.Exception.Message)" "WARN"
}
# ========== 新增：异常cron任务检测 ==========
try {
# 检查最近修改的cron任务
$cronFiles = Invoke-SSHCommand "find /etc/cron.* /var/spool/cron -type f -mtime -7 2>/dev/null | head -10" -Timeout 10
if ($cronFiles) {
if ($cronFiles -is [array]) { $cronFiles = $cronFiles -join "`n" }
$cronList = $cronFiles -split "`n" | Where-Object { $_ -match "\S+" }
if ($cronList.Count -gt 0) {
$results += [PSCustomObject]@{
Name = "最近修改的cron任务"
Value = "$($cronList.Count) 个"
Threshold = "-"
Status = "正常"
Message = "7天内修改的定时任务文件"
}
}
}
# 检查可疑的cron任务（包含wget、curl等远程命令）
$suspiciousCron = Invoke-SSHCommand "grep -r 'wget\|curl\|bash.*http' /etc/cron.* /var/spool/cron 2>/dev/null | head -5" -Timeout 10
if ($suspiciousCron) {
if ($suspiciousCron -is [array]) { $suspiciousCron = $suspiciousCron -join "`n" }
$suspiciousCount = ($suspiciousCron -split "`n" | Where-Object { $_ -match "\S+" } | Measure-Object).Count
if ($suspiciousCount -gt 0) {
$results += [PSCustomObject]@{
Name = "可疑cron任务"
Value = "$suspiciousCount 个"
Threshold = ">0"
Status = "警告"
Message = "发现包含远程命令的定时任务"
}
Add-Issue -Message "发现可疑cron任务: $suspiciousCount 个" -Level "警告"
}
}
}
catch {
Write-Log "异常cron任务检测失败: $($_.Exception.Message)" "WARN"
}
# ========== 新增：空密码账户检测 ==========
try {
# 检测空密码账户，排除系统服务账户
$emptyPasswordAccounts = Invoke-SSHCommand 'awk -F: '"'"'($2 == "" || $2 == "!") {print $1}'"'"' /etc/shadow 2>/dev/null | grep -v -E "^((systemd|dbus|polkitd|saslauth|dhcpd|sshd|tss|unbound|chrony|nginx|apache|www|mysql|redis|emqx|nobody|mail|ftp|games|operator|sync|shutdown|halt|news|uucp|proxy|list|irc)|$)' | head -10' -Timeout 10
if ($emptyPasswordAccounts) {
if ($emptyPasswordAccounts -is [array]) { $emptyPasswordAccounts = $emptyPasswordAccounts -join "`n" }
$emptyAccounts = $emptyPasswordAccounts -split "`n" | Where-Object { $_ -match "\S+" }
if ($emptyAccounts.Count -gt 0) {
$results += [PSCustomObject]@{
Name = "空密码账户"
Value = "$($emptyAccounts.Count) 个"
Threshold = ">0"
Status = "严重"
Message = "发现空密码账户: $($emptyAccounts -join ', ')"
}
Add-Issue -Message "发现空密码账户: $($emptyAccounts -join ', ')" -Level "严重"
}
}
else {
$results += [PSCustomObject]@{
Name = "空密码账户"
Value = "0 个"
Threshold = ">0"
Status = "正常"
Message = "无空密码账户（已排除系统服务）"
}
}
}
catch {
Write-Log "空密码账户检测失败: $($_.Exception.Message)" "WARN"
}
# ========== 新增：最近修改的关键系统文件 ==========
try {
$recentSystemFiles = Invoke-SSHCommand "find /etc -name '*.conf' -mtime -1 -type f 2>/dev/null | head -10" -Timeout 15
if ($recentSystemFiles) {
if ($recentSystemFiles -is [array]) { $recentSystemFiles = $recentSystemFiles -join "`n" }
$modifiedFiles = $recentSystemFiles -split "`n" | Where-Object { $_ -match "\S+" }
if ($modifiedFiles.Count -gt 0) {
$results += [PSCustomObject]@{
Name = "最近修改的系统配置"
Value = "$($modifiedFiles.Count) 个"
Threshold = "-"
Status = "正常"
Message = "24小时内修改的配置文件: " + (($modifiedFiles | ForEach-Object { $_ -replace '/etc/', '' }) -join ', ')
}
}
}
}
catch {
Write-Log "最近修改的系统文件检测失败: $($_.Exception.Message)" "WARN"
}
# 保存结果
foreach ($result in $results) {
Save-TestResult "安全合规" $result
......@@ -1774,69 +1333,6 @@ function Test-SystemLogs {
}
}
# ========== 新增：内核panic/oops检测 ==========
try {
$panicLog = Invoke-SSHCommand "dmesg | grep -iE 'panic|oops|bug' | tail -20" -Timeout 10
if ($panicLog) {
$panicCount = ($panicLog -split "`n" | Where-Object { $_ -match "panic|oops|bug" } | Measure-Object).Count
if ($panicCount -gt 0) {
$results += [PSCustomObject]@{
Name = "内核异常日志"
Value = "$panicCount 条"
Threshold = ">0"
Status = "严重"
Message = "检测到内核panic/oops/bug记录"
}
Add-Issue -Message "检测到内核异常: $panicCount 条" -Level "严重"
}
}
}
catch {
Write-Log "内核panic/oops检测失败: $($_.Exception.Message)" "WARN"
}
# ========== 新增：服务崩溃重启记录 ==========
try {
$crashLog = Invoke-SSHCommand "journalctl --since '24 hours ago' --no-pager 2>/dev/null | grep -iE 'crashed|segfault|core dumped' | tail -20" -Timeout 30
if ($crashLog) {
$crashCount = ($crashLog -split "`n" | Where-Object { $_ -match "crashed|segfault|core dumped" } | Measure-Object).Count
if ($crashCount -gt 0) {
$results += [PSCustomObject]@{
Name = "服务崩溃记录"
Value = "$crashCount 条"
Threshold = ">0"
Status = "严重"
Message = "24小时内检测到服务崩溃"
}
Add-Issue -Message "检测到服务崩溃: $crashCount 条" -Level "严重"
}
}
}
catch {
Write-Log "服务崩溃重启记录检测失败: $($_.Exception.Message)" "WARN"
}
# ========== 新增：系统资源耗尽事件 ==========
try {
$resourceEvents = Invoke-SSHCommand "journalctl --since '24 hours ago' --no-pager 2>/dev/null | grep -iE 'resource temporarily unavailable|cannot allocate memory|too many open files' | tail -20" -Timeout 30
if ($resourceEvents) {
$eventCount = ($resourceEvents -split "`n" | Where-Object { $_ -match "resource|allocate|open files" } | Measure-Object).Count
if ($eventCount -gt 0) {
$results += [PSCustomObject]@{
Name = "资源耗尽事件"
Value = "$eventCount 条"
Threshold = ">0"
Status = "严重"
Message = "24小时内检测到资源耗尽事件"
}
Add-Issue -Message "检测到资源耗尽事件: $eventCount 条" -Level "严重"
}
}
}
catch {
Write-Log "系统资源耗尽事件检测失败: $($_.Exception.Message)" "WARN"
}
# 保存结果
foreach ($result in $results) {
Save-TestResult "系统日志" $result
......@@ -2371,95 +1867,6 @@ function Test-DockerStatus {
}
}
# 容器资源限制检测
Write-Log "检测容器资源限制..."
$resourceLimits = @("umysql", "uredis", "uemqx", "ujava2")
foreach ($container in $resourceLimits) {
$limitInfo = Invoke-SSHCommand "docker inspect $container 2>/dev/null --format '{{.HostConfig.Memory}}:{{.HostConfig.NanoCpus}}'" -Timeout 10
if ($limitInfo -and $limitInfo -match "^(\d+):(\d+)$") {
$memLimit = [long]$matches[1]
$cpuLimit = [long]$matches[2]
$memLimitMB = if ($memLimit -gt 0) { "{0:N0} MB" -f ($memLimit / 1MB) } else { "无限制" }
$cpuQuota = if ($cpuLimit -gt 0) { [math]::Round($cpuLimit / 1e9, 2) } else { "无限制" }
$results += [PSCustomObject]@{
Name = "容器资源限制$container"
Value = "CPU:$cpuQuota核, 内存:$memLimitMB"
Threshold = "-"
Status = "正常"
Message = "容器资源配置限制"
}
}
}
# 容器日志错误扫描
Write-Log "扫描容器日志中的错误..."
foreach ($container in $resourceLimits) {
# 获取最近100行日志中的错误
$errorLogs = Invoke-SSHCommand "docker logs --tail 100 $container 2>&1 | grep -iE 'error|exception|fail|fatal|critical'" -Timeout 15
if ($errorLogs) {
if ($errorLogs -is [array]) { $errorLogs = $errorLogs -join "`n" }
$errorCount = ($errorLogs -split "`n" | Where-Object { $_ -match "\S+" }).Count
if ($errorCount -gt 0) {
$status = if ($errorCount -gt 10) { "警告" } else { "注意" }
$results += [PSCustomObject]@{
Name = "容器日志错误$container"
Value = "${errorCount}条"
Threshold = ">10条警告"
Status = $status
Message = "最近100行日志中发现${errorCount}个错误"
}
if ($status -eq "警告") {
Add-Issue -Message "容器${container}日志中发现${errorCount}个错误" -Level $status
}
}
}
}
# Docker事件检测
Write-Log "获取最近的Docker事件..."
$dockerEvents = Invoke-SSHCommand "docker events --since '1h' --until '0s' --format '{{.Type}}:{{.Action}}:{{.Actor.Attributes.name}}' 2>&1 | head -20" -Timeout 15
if ($dockerEvents) {
if ($dockerEvents -is [array]) { $dockerEvents = $dockerEvents -join "`n" }
# 统计事件类型
$eventCount = 0
$importantEvents = @()
foreach ($eventLine in $dockerEvents -split "`n") {
if ($eventLine -match "\S+") {
$eventCount++
# 关注重要事件
if ($eventLine -match "die|oom|kill|destroy") {
$importantEvents += $eventLine
}
}
}
if ($eventCount -gt 0) {
$eventStatus = if ($importantEvents.Count -gt 0) { "警告" } else { "正常" }
$eventMessage = if ($importantEvents.Count -gt 0) {
"最近1小时${eventCount}个事件, 重要:$($importantEvents.Count)"
} else {
"最近1小时${eventCount}个事件"
}
$results += [PSCustomObject]@{
Name = "Docker事件"
Value = "${eventCount}个"
Threshold = "-"
Status = $eventStatus
Message = $eventMessage
}
if ($importantEvents.Count -gt 0) {
Add-Issue -Message "Docker重要事件: $($importantEvents -join ', ')" -Level $eventStatus
}
}
}
# 保存结果
foreach ($result in $results) {
Save-TestResult "Docker容器" $result
......@@ -4254,73 +3661,6 @@ function Test-JavaApplication {
Write-Log "JVM运行时信息检测失败: $($_.Exception.Message)" "WARN"
}
# Spring Boot Actuator端点检测
try {
Write-Log "检测Spring Boot Actuator端点..."
$actuatorCheck = Invoke-SSHCommand 'curl -s --max-time 5 http://127.0.0.1:8080/actuator/health 2>&1 | grep -oE '"'"'{"status":"[^"]+"}'"'"'' -Timeout 10
if ($actuatorCheck -match '{"status":"([^"]+)"}') {
$healthStatus = $matches[1]
$actuatorStatus = if ($healthStatus -eq "UP") { "正常" } else { "警告" }
$results += [PSCustomObject]@{
Name = "Spring Boot健康"
Value = $healthStatus
Threshold = "-"
Status = $actuatorStatus
Message = "Actuator端点状态"
}
}
}
catch {
Write-Log "Spring Boot Actuator检测失败: $($_.Exception.Message)" "WARN"
}
# Java线程dump分析
try {
Write-Log "分析Java线程状态..."
# 获取Java进程PID并分析线程状态
$threadDump = Invoke-SSHCommand 'java_pid=$(docker exec ujava2 jps 2>/dev/null | grep -v Jps | awk '"'"'NR==1{print $1}'"'"'); if [ -n "$java_pid" ]; then docker exec ujava2 jstack $java_pid 2>&1 | grep -oE '"'"'java.lang.Thread.State: [^:]+'"'"' | sort | uniq -c | sort -rn; fi' -Timeout 15
if ($threadDump) {
if ($threadDump -is [array]) { $threadDump = $threadDump -join "`n" }
$threadStates = @{}
foreach ($line in $threadDump -split "`n") {
if ($line -match "(\d+)\s+java\.lang\.Thread\.State:\s+(\S+)") {
$count = [int]$matches[1]
$state = $matches[2]
$threadStates[$state] = $count
}
}
if ($threadStates.Count -gt 0) {
# 检查阻塞和等待的线程
$blockedCount = if ($threadStates["BLOCKED"]) { $threadStates["BLOCKED"] } else { 0 }
$waitingCount = if ($threadStates["WAITING"]) { $threadStates["WAITING"] } else { 0 }
$timedWaitingCount = if ($threadStates["TIMED_WAITING"]) { $threadStates["TIMED_WAITING"] } else { 0 }
$problematicThreads = $blockedCount + $waitingCount
$threadStatus = if ($problematicThreads -gt 50) { "警告" } else { "正常" }
$stateSummary = ($threadStates.GetEnumerator() | ForEach-Object { "$($_.Key):$($_.Value)" }) -join ", "
$results += [PSCustomObject]@{
Name = "Java线程状态"
Value = "BLOCKED:$blockedCount, WAITING:$waitingCount"
Threshold = "BLOCKED+WAITING>50警告"
Status = $threadStatus
Message = $stateSummary
}
if ($threadStatus -ne "正常") {
Add-Issue -Message "Java线程异常: 阻塞${blockedCount}, 等待${waitingCount}" -Level $threadStatus
}
}
}
}
catch {
Write-Log "Java线程dump分析失败: $($_.Exception.Message)" "WARN"
}
# 保存结果
foreach ($result in $results) {
Save-TestResult "Java应用" $result
......@@ -4383,41 +3723,6 @@ function Test-ApplicationLogs {
Message = "最近200行日志中无错误"
}
}
# 错误频率统计（按小时分组）
$errorFreq = Invoke-SSHCommand "grep -iE 'error|exception|fail' '$javaLogPath'/*.log 2>/dev/null | grep -oE '[0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}' | sort | uniq -c | tail -8" -Timeout 30
if ($errorFreq) {
if ($errorFreq -is [array]) { $errorFreq = $errorFreq -join "`n" }
$freqDetails = @()
$maxHourlyErrors = 0
foreach ($freqLine in $errorFreq -split "`n") {
if ($freqLine -match "(\d+)\s+(\d{4}-\d{2}-\d{2} \d{2})") {
$count = [int]$matches[1]
$hour = $matches[2]
$freqDetails += "${hour}:${count}个"
if ($count -gt $maxHourlyErrors) {
$maxHourlyErrors = $count
}
}
}
if ($freqDetails.Count -gt 0) {
$freqStatus = if ($maxHourlyErrors -gt 100) { "警告" } else { "正常" }
$results += [PSCustomObject]@{
Name = "Java错误频率"
Value = "最高${maxHourlyErrors}个/小时"
Threshold = ">100个/小时警告"
Status = $freqStatus
Message = ($freqDetails -join ", ")
}
if ($freqStatus -ne "正常") {
Add-Issue -Message "Java错误频率异常: 最高${maxHourlyErrors}个/小时" -Level $freqStatus
}
}
}
}
# 日志文件大小检测
......@@ -4539,39 +3844,6 @@ function Test-ApplicationLogs {
Message = "Nginx应用日志错误数"
}
}
# Nginx慢请求分析（访问日志）
$slowRequests = Invoke-SSHCommand "tail -1000 '$nginxLogPath'/access*.log 2>/dev/null | awk '{print \$NF}' | grep -oE '[0-9]+\.[0-9]+' | sort -rn | head -10" -Timeout 30
if ($slowRequests) {
if ($slowRequests -is [array]) { $slowRequests = $slowRequests -join "`n" }
$slowTimes = @()
$maxTime = 0
foreach ($slowLine in $slowRequests -split "`n") {
if ($slowLine -match "(\d+\.\d+)") {
$time = [double]$matches[1]
$slowTimes += $time
if ($time -gt $maxTime) { $maxTime = $time }
}
}
if ($slowTimes.Count -gt 0) {
$avgTime = [math]::Round(($slowTimes | Measure-Object -Average).Average, 2)
$slowStatus = if ($maxTime -gt 5) { "警告" } else { "正常" }
$results += [PSCustomObject]@{
Name = "Nginx慢请求"
Value = "最慢${maxTime}s, 平均${avgTime}s"
Threshold = ">5s警告"
Status = $slowStatus
Message = "最近1000条请求中最慢的10个"
}
if ($slowStatus -ne "正常") {
Add-Issue -Message "Nginx存在慢请求: 最慢${maxTime}s" -Level $slowStatus
}
}
}
}
# 保存结果
......@@ -4688,26 +3960,6 @@ $(
| CPU核心数 | $($systemInfo.CPUCores) |
| 总内存 | $($systemInfo.MemoryTotal) |
| 系统负载 | $($systemInfo.LoadAverage) |
$(
if ($systemInfo.KernelParams) {
"| 内核参数 | $($systemInfo.KernelParams -replace '\n', '<br>') |`n"
}
)
$(
if ($systemInfo.UlimitInfo) {
"| 资源限制 | $($systemInfo.UlimitInfo -replace '\n', '<br>') |`n"
}
)
$(
if ($systemInfo.MemoryPressure) {
"| 内存压力 | $($systemInfo.MemoryPressure -replace '\n', '<br>') |`n"
}
)
$(
if ($systemInfo.HugePages) {
"| 大页内存 | $($systemInfo.HugePages -replace '\n', '<br>') |`n"
}
)
---
......@@ -5004,7 +4256,7 @@ $(
*报告生成时间: $(Get-Date -Format "yyyy-MM-dd HH:mm:ss")*
*服务器健康监测脚本 v2.0*
*检测点数量: 150+*
*检测点数量: 120+*
"@
return $reportContent
......
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论