feat(deploy): 为更新组件增加定时维护与双保险

2026-10-08 09:23:25 +08:00
parent e3d760c5ab
commit 89a4ccd7e7
30 changed files with 796 additions and 7 deletions
@@ -0,0 +1,74 @@
+param([string]$ProjectRoot = (Split-Path $PSScriptRoot -Parent))
+$ErrorActionPreference = 'Stop'
+Set-Location $ProjectRoot
+. (Join-Path $ProjectRoot 'scripts/Initialize-SecurityEnv.ps1')
+$content = [System.IO.File]::ReadAllText((Join-Path $ProjectRoot '.env'))
+if ((Get-EnvValue $content 'INFRASTRUCTURE_AUTO_UPDATE_ENABLED') -eq 'false') { return }
+$state = Join-Path $ProjectRoot 'data/infrastructure-update'
+[System.IO.Directory]::CreateDirectory($state) | Out-Null
+try { $lock = [System.IO.File]::Open((Join-Path $state 'task.lock'), 'OpenOrCreate', 'ReadWrite', 'None') }
+catch [System.IO.IOException] { return }
+$log = Join-Path $state 'last-run.log'
+$previousConfig = $env:DOCKER_CONFIG
+try {
+    [System.IO.File]::WriteAllText($log, "$(Get-Date -Format 'yyyy-MM-dd HH:mm:ss') 检查 Watchtower 和 updater 更新`n")
+    $configFile = Get-EnvValue $content 'DOCKER_CONFIG_FILE'
+    if (-not [string]::IsNullOrWhiteSpace($configFile) -and (Test-Path $configFile -PathType Leaf)) {
+        $env:DOCKER_CONFIG = Split-Path $configFile -Parent
+    }
+    & docker compose version *> $null
+    $v2 = $LASTEXITCODE -eq 0
+    function Invoke-InfrastructureCompose {
+        param([string[]]$ComposeArguments)
+        if ($v2) { $output = & docker compose @ComposeArguments 2>&1 }
+        else { $output = & docker-compose @ComposeArguments 2>&1 }
+        $code = $LASTEXITCODE
+        $output | Out-File $log -Append -Encoding utf8
+        if ($code -ne 0) { throw '组件更新命令未成功，现有容器状态需查看日志' }
+    }
+    if ($v2) { $available = @(& docker compose config --services) }
+    else { $available = @(& docker-compose config --services) }
+    if ($LASTEXITCODE -ne 0 -or $available -notcontains 'watchtower') { throw '无法读取 Watchtower 部署配置' }
+    $failed = @()
+    foreach ($service in @('watchtower', 'easyai-updater')) {
+        # Do not enable updater in profile-disabled legacy deployments.
+        if ($available -notcontains $service) {
+            "$service 未启用，跳过维护" | Out-File $log -Append -Encoding utf8
+            continue
+        }
+        try {
+            "检查 $service" | Out-File $log -Append -Encoding utf8
+            Invoke-InfrastructureCompose -ComposeArguments @('pull', $service)
+            Invoke-InfrastructureCompose -ComposeArguments @('up', '-d', '--no-deps', $service)
+            $healthy = $false
+            for ($attempt=0; $attempt -lt 12; $attempt++) {
+                if ($v2) { $containerId = & docker compose ps -q $service }
+                else { $containerId = & docker-compose ps -q $service }
+                if ($LASTEXITCODE -ne 0) { throw "无法读取 $service 状态" }
+                if ($containerId) {
+                    $status = & docker inspect -f '{{.State.Running}} {{if .State.Health}}{{.State.Health.Status}}{{else}}none{{end}}' $containerId
+                    if ($LASTEXITCODE -ne 0) { throw "无法读取 $service 状态" }
+                    if ($status -eq 'true healthy') { $healthy = $true; break }
+                    if ($status -eq 'true none') {
+                        if ($service -eq 'watchtower') {
+                            & docker exec $containerId /watchtower --health-check *> $null
+                        } else {
+                            & docker exec $containerId node -e "require('http').get('http://127.0.0.1:3015/health',r=>{r.resume();process.exit(r.statusCode===200?0:1)}).on('error',()=>process.exit(1))" *> $null
+                        }
+                        if ($LASTEXITCODE -eq 0) { $healthy = $true; break }
+                    }
+                }
+                Start-Sleep -Seconds 5
+            }
+            if (-not $healthy) { throw "$service 暂未恢复可用" }
+            "$service 已恢复可用" | Out-File $log -Append -Encoding utf8
+        } catch {
+            "$service 本次维护未成功，下次定时任务将继续检查：$($_.Exception.Message)" | Out-File $log -Append -Encoding utf8
+            $failed += $service
+        }
+    }
+    if ($failed.Count -gt 0) { throw "更新组件维护未全部成功：$($failed -join ', ')" }
+} finally {
+    $env:DOCKER_CONFIG = $previousConfig
+    $lock.Dispose()
+}