stela ops usage fixes roles propagation and timoeut, one account to support multi tenants, migrations consolidation, search to support documentation, doctor and open api vector db search

2026-02-22 19:27:54 +02:00
parent a29f438f53
commit bd8fee6ed8
373 changed files with 832097 additions and 3369 deletions
--- a/devops/compose/scripts/router-mode-redeploy.ps1
+++ b/devops/compose/scripts/router-mode-redeploy.ps1
@@ -0,0 +1,129 @@
+param(
+    [ValidateSet("microservice", "reverseproxy")]
+    [string]$Mode = "microservice",
+    [string]$ComposeFile = "docker-compose.stella-ops.yml",
+    [int]$WaitTimeoutSeconds = 1200,
+    [int]$RecoveryAttempts = 2,
+    [int]$RecoveryWaitSeconds = 180
+)
+
+Set-StrictMode -Version Latest
+$ErrorActionPreference = "Stop"
+$ProgressPreference = "SilentlyContinue"
+
+$configPath = switch ($Mode) {
+    "microservice" { "./router-gateway-local.json" }
+    "reverseproxy" { "./router-gateway-local.reverseproxy.json" }
+    default { throw "Unsupported mode: $Mode" }
+}
+
+Write-Host "Redeploy mode: $Mode"
+Write-Host "Gateway config: $configPath"
+Write-Host "Compose file: $ComposeFile"
+
+$env:ROUTER_GATEWAY_CONFIG = $configPath
+
+function Invoke-Compose {
+    param(
+        [Parameter(Mandatory = $true)]
+        [string[]]$Args,
+        [switch]$IgnoreExitCode
+    )
+
+    & docker compose -f $ComposeFile @Args
+    $exitCode = $LASTEXITCODE
+    if (-not $IgnoreExitCode -and $exitCode -ne 0) {
+        throw "docker compose $($Args -join ' ') failed with exit code $exitCode."
+    }
+
+    return $exitCode
+}
+
+function Get-UnhealthyContainers {
+    $containers = & docker ps --filter "health=unhealthy" --format "{{.Names}}"
+    if ($LASTEXITCODE -ne 0) {
+        throw "Failed to query unhealthy containers."
+    }
+
+    $filtered = @($containers | Where-Object { -not [string]::IsNullOrWhiteSpace($_) -and $_ -like "stellaops-*" })
+    return [string[]]$filtered
+}
+
+function Get-ComposeServiceName {
+    param(
+        [Parameter(Mandatory = $true)]
+        [string]$ContainerName
+    )
+
+    $service = & docker inspect --format "{{ index .Config.Labels \"com.docker.compose.service\" }}" $ContainerName 2>$null
+    if ($LASTEXITCODE -ne 0 -or [string]::IsNullOrWhiteSpace($service)) {
+        return $null
+    }
+
+    return $service.Trim()
+}
+
+function Wait-ForContainerHealth {
+    param(
+        [Parameter(Mandatory = $true)]
+        [string]$ContainerName,
+        [Parameter(Mandatory = $true)]
+        [int]$TimeoutSeconds
+    )
+
+    $deadline = (Get-Date).AddSeconds($TimeoutSeconds)
+    while ((Get-Date) -lt $deadline) {
+        $status = (& docker inspect --format "{{if .State.Health}}{{.State.Health.Status}}{{else}}none{{end}}" $ContainerName 2>$null).Trim()
+        if ($LASTEXITCODE -ne 0) {
+            return $false
+        }
+
+        if ($status -eq "healthy" -or $status -eq "none") {
+            return $true
+        }
+
+        Start-Sleep -Seconds 5
+    }
+
+    return $false
+}
+
+Invoke-Compose -Args @("down", "-v", "--remove-orphans") | Out-Null
+
+$upExitCode = Invoke-Compose -Args @("up", "-d", "--wait", "--wait-timeout", $WaitTimeoutSeconds.ToString()) -IgnoreExitCode
+if ($upExitCode -ne 0) {
+    Write-Warning "docker compose up returned exit code $upExitCode. Running unhealthy-service recovery."
+}
+
+for ($attempt = 1; $attempt -le $RecoveryAttempts; $attempt++) {
+    $unhealthyContainers = @(Get-UnhealthyContainers)
+    if ($unhealthyContainers.Count -eq 0) {
+        break
+    }
+
+    Write-Warning "Recovery attempt ${attempt}: unhealthy containers detected: $($unhealthyContainers -join ', ')"
+    $services = New-Object System.Collections.Generic.HashSet[string]([System.StringComparer]::OrdinalIgnoreCase)
+
+    foreach ($containerName in $unhealthyContainers) {
+        $serviceName = Get-ComposeServiceName -ContainerName $containerName
+        if (-not [string]::IsNullOrWhiteSpace($serviceName)) {
+            [void]$services.Add($serviceName)
+        }
+    }
+
+    foreach ($serviceName in $services) {
+        Write-Host "Recreating service: $serviceName"
+        Invoke-Compose -Args @("up", "-d", "--force-recreate", "--no-deps", $serviceName) | Out-Null
+    }
+
+    foreach ($containerName in $unhealthyContainers) {
+        [void](Wait-ForContainerHealth -ContainerName $containerName -TimeoutSeconds $RecoveryWaitSeconds)
+    }
+}
+
+$remainingUnhealthy = @(Get-UnhealthyContainers)
+if ($remainingUnhealthy.Count -gt 0) {
+    throw "Redeploy completed with unresolved unhealthy containers: $($remainingUnhealthy -join ', ')"
+}
+
+Write-Host "Redeploy complete for mode '$Mode'."