portal-api/app/Services/VirtualMachine/Rebuild/VmRebuildCloneService.php

<?php

declare(strict_types=1);

namespace App\Services\VirtualMachine\Rebuild;

use App\DTOs\VirtualMachine\Rebuild\RebuildContext;
use App\Http\Services\vCenter as VCenterService;
use App\Models\ModuleDatastore;
use App\Models\vCenter as VCenterModel;
use App\Support\TlsHttpClientFactory;
use Illuminate\Support\Facades\Log;
use RuntimeException;

/**
 * VmRebuildCloneService
 *
 * Clones a fresh VM from the chosen OS template and powers it on.
 * Uses the vCenter REST API directly — never dispatches CreateVirtualMachineJob.
 */
final class VmRebuildCloneService
{
    private const CLONE_TIMEOUT      = 300;
    private const POWER_ON_TIMEOUT   = 30;
    private const TOOLS_POLL_DELAY   = 10;
    private const TOOLS_MAX_ATTEMPTS = 30;
    /**
     * Extra grace after VMware Tools reports an IP. VMware Tools advertises
     * the guest IP as soon as the kernel finishes networking, but
     * Invoke-VMScript / guest ops daemon needs another ~20–40 s before it
     * actually accepts script invocations on a fresh clone. Without this
     * delay assignIP/verifyIP scripts often return "FAIL" on the first run.
     */
    private const TOOLS_READY_GRACE_SECONDS = 45;

    public function __construct(private readonly VCenterService $vCenter)
    {
    }

    /**
     * Clone a new VM from the OS template specified in $ctx.
     *
     * @return string vSphere vm_id of the newly cloned VM
     */
    public function clone(RebuildContext $ctx): string
    {
        $vcenter = VCenterModel::find($ctx->vcenterId);
        if (!$vcenter || !$this->vCenter->loginServer($vcenter)) {
            throw new RuntimeException("Rebuild clone: cannot login to vCenter #{$ctx->vcenterId}");
        }

        $datastore = ModuleDatastore::where('module_host_id', $ctx->hostId)->first();
        $datastoreRef = $datastore?->datastore ?: 'datastore-3004';

        $body = [
            'spec' => [
                'name'    => $ctx->transientName,
                'source'  => $ctx->template,
                'placement' => [
                    'datacenter'    => $ctx->datacenterRef,
                    'folder'        => 'group-v1008',
                    'host'          => $ctx->hostRef,
                    'resource_pool' => 'resgroup-3002',
                    'datastore'     => $datastoreRef,
                ],
                'power_on' => false,
                'fallback_to_datastore_default_policy' => true,
            ],
        ];

        Log::info('VmRebuildCloneService: clone start', [
            'module_vmware_id' => $ctx->moduleVmwareId,
            'transient_name'   => $ctx->transientName,
            'template'         => $ctx->template,
        ]);

        $response = TlsHttpClientFactory::vCenter()
            ->timeout(self::CLONE_TIMEOUT)
            ->connectTimeout(30)
            ->withHeaders([
                'vmware-api-session-id' => $this->vCenter->token,
                'Content-Type'          => 'application/json',
            ])
            ->post(
                "https://{$this->vCenter->apiUri}/rest/vcenter/vm?action=clone",
                $body,
            );

        $data = $response->json();
        if (!$response->successful() || isset($data['type'])) {
            throw new RuntimeException(
                'Rebuild clone failed: ' . json_encode($data ?: $response->body())
            );
        }

        $newVmId = (string) ($data['value'] ?? '');
        if ($newVmId === '' || in_array($newVmId, ['unknown', 'UNKNOWN'], true)) {
            throw new RuntimeException('Rebuild clone: empty vm_id returned by vCenter');
        }

        Log::info('VmRebuildCloneService: clone success', [
            'module_vmware_id' => $ctx->moduleVmwareId,
            'new_vm_id'        => $newVmId,
        ]);

        return $newVmId;
    }

    /**
     * Power on the freshly cloned VM and wait for VMware Tools to report
     * a guest IP — without this, IP assignment scripts cannot run.
     */
    public function powerOnAndWaitTools(RebuildContext $ctx): void
    {
        if ($ctx->newVmId === null) {
            throw new RuntimeException('powerOnAndWaitTools: newVmId is null');
        }

        $vcenter = VCenterModel::find($ctx->vcenterId);
        if (!$vcenter || !$this->vCenter->loginServer($vcenter)) {
            throw new RuntimeException("Rebuild powerOn: cannot login to vCenter #{$ctx->vcenterId}");
        }

        $powerResp = TlsHttpClientFactory::vCenter()
            ->timeout(self::POWER_ON_TIMEOUT)
            ->withHeaders(['vmware-api-session-id' => $this->vCenter->token])
            ->post("https://{$this->vCenter->apiUri}/rest/vcenter/vm/{$ctx->newVmId}/power/start");

        if (!$powerResp->successful() && $powerResp->status() !== 400) {
            // 400 = already powered on; treat as OK.
            throw new RuntimeException(
                "Rebuild powerOn failed: status={$powerResp->status()} body={$powerResp->body()}"
            );
        }

        Log::info('VmRebuildCloneService: powered on, waiting for tools', [
            'new_vm_id' => $ctx->newVmId,
        ]);

        for ($i = 1; $i <= self::TOOLS_MAX_ATTEMPTS; $i++) {
            $info = TlsHttpClientFactory::vCenter()
                ->timeout(10)
                ->withHeaders(['vmware-api-session-id' => $this->vCenter->token])
                ->get("https://{$this->vCenter->apiUri}/rest/vcenter/vm/{$ctx->newVmId}/guest/identity")
                ->json();

            $guestIp = trim((string) ($info['value']['ip_address'] ?? ''));
            if ($guestIp !== '' && $guestIp !== '0.0.0.0') {
                Log::info('VmRebuildCloneService: tools ready — applying grace', [
                    'new_vm_id'        => $ctx->newVmId,
                    'guest_ip'         => $guestIp,
                    'attempt'          => $i,
                    'grace_seconds'    => self::TOOLS_READY_GRACE_SECONDS,
                ]);
                // Give the guest ops daemon time to accept Invoke-VMScript.
                sleep(self::TOOLS_READY_GRACE_SECONDS);
                return;
            }

            sleep(self::TOOLS_POLL_DELAY);
        }

        throw new RuntimeException(
            "Rebuild powerOn: VMware Tools never became ready on new VM {$ctx->newVmId}"
        );
    }

    /**
     * Best-effort: power off the OLD VM so it stops responding on the shared IP
     * before we reassign it inside the guest of the new VM. We do NOT delete it
     * here — deletion is the very last step, after full validation.
     */
    public function powerOffOldVm(RebuildContext $ctx): void
    {
        if ($ctx->oldVmId === '') {
            return;
        }

        $vcenter = VCenterModel::find($ctx->vcenterId);
        if (!$vcenter || !$this->vCenter->loginServer($vcenter)) {
            Log::warning('VmRebuildCloneService: powerOff old VM skipped — vCenter login failed', [
                'old_vm_id' => $ctx->oldVmId,
            ]);
            return;
        }

        try {
            TlsHttpClientFactory::vCenter()
                ->timeout(30)
                ->withHeaders(['vmware-api-session-id' => $this->vCenter->token])
                ->post("https://{$this->vCenter->apiUri}/rest/vcenter/vm/{$ctx->oldVmId}/power/stop");

            Log::info('VmRebuildCloneService: old VM powered off', [
                'old_vm_id' => $ctx->oldVmId,
            ]);
        } catch (\Throwable $e) {
            // Non-fatal — guest may already be off, or vCenter may flake.
            Log::warning('VmRebuildCloneService: powerOff old VM failed', [
                'old_vm_id' => $ctx->oldVmId,
                'error'     => $e->getMessage(),
            ]);
        }
    }
}