Serve preview URLs over Traefik+TLS, stabilize them, and speed up builds.

Ingress / preview URLs:
- Default the app Ingress class and ACME HTTP-01 solver to Traefik
  (k3s default) via a new INGRESS_CLASS env, instead of hardcoding nginx —
  fixes 404s on clusters without ingress-nginx.
- Only put public, real-TLD hosts (custom domain + preview) in the TLS
  block; the internal *.apps.cloudhost.local host no longer poisons the
  Let's Encrypt order, so certs actually issue.
- Make the per-app preview number stable across redeploys so URLs stop
  breaking, and let PREVIEW_BASE_DOMAIN configure the base domain.

Registry pulls:
- Point the k3s registries.yaml mirror endpoint at the registry NodePort on
  loopback so node containerd never depends on cluster DNS (image pulls
  survive node restarts).

Builds:
- Pin the Kaniko image, use IfNotPresent pull policy, drop the dead build
  queue/processor, and retry transient Kubernetes API errors while polling
  build jobs.

Logs & apps list:
- fluent-bit reads log files from head so startup output reaches
  Elasticsearch.
- Order joined deployments newest-first so the apps list shows the latest
  deployment status.

Allocation:
- Reserve in-flight (pending/building) capacity and stop globally degrading
  the cluster on a single allocation failure, so concurrent deploys don't
  starve or wrongly report "no healthy cluster".

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

backend/.env.example

@@ -43,6 +43,12 @@ BUILD_SERVICE_ACCOUNT=kaniko-builder
 
 # Platform
 PLATFORM_DOMAIN=apps.cloudhost.local
+# Base domain for per-user preview URLs (<userId>-<7-digit>.<base-domain>).
+# Falls back to the root domain derived from FRONTEND_URL when unset.
+# PREVIEW_BASE_DOMAIN=3fase.ir
+# Ingress controller class for app Ingress + ACME HTTP-01 solver.
+# k3s default is Traefik; use "nginx" only on clusters running ingress-nginx.
+# INGRESS_CLASS=traefik
 UPLOAD_DIR=./uploads
 # PVC resize: use a dynamic StorageClass with allowVolumeExpansion (k3s: rancher.io/local-path)
 # k3s: use local-path and skip creating a custom class (set CREATE=false)

backend/helm/cloudhost-app/templates/deployment.yaml

@@ -26,7 +26,9 @@ spec:
       containers:
         - name: {{ $name }}
           image: {{ .Values.app.image | quote }}
-          imagePullPolicy: Always
+          # Image tags are unique per build (name:timestamp) and immutable —
+          # IfNotPresent avoids re-pulling on every restart/scale-up.
+          imagePullPolicy: IfNotPresent
           {{- if include "cloudhost-app.loggingWrapEnabled" . }}
           command: ["sh", "-c"]
           args:

backend/helm/cloudhost-app/templates/fluent-bit-configmap.yaml

@@ -26,6 +26,10 @@ data:
         Refresh_Interval 5
         Mem_Buf_Limit 5MB
         Skip_Long_Lines On
+        # Read the whole file (incl. startup output written before fluent-bit
+        # attached) instead of only new lines — otherwise an idle app that
+        # logged only at boot would ship nothing to Elasticsearch.
+        Read_from_Head On
 
     [FILTER]
         Name          record_modifier

backend/helm/cloudhost-app/templates/ingress.yaml

@@ -3,6 +3,13 @@
 {{- $ns   := include "cloudhost-app.namespace" . -}}
 {{- $host := printf "%s.%s" (default $name .Values.ingress.subdomain) .Values.ingress.domain -}}
 {{- $previewHost := .Values.ingress.previewHost | default "" -}}
+{{- $className := .Values.ingress.className | default "traefik" -}}
+{{- /* Only public, real-TLD hosts (custom domain + preview) can get a managed cert.
+       The internal *.apps.cloudhost.local host is not a valid public suffix and would
+       make Let's Encrypt reject the whole order. */ -}}
+{{- $tlsHosts := list -}}
+{{- if .Values.ingress.customDomain }}{{- $tlsHosts = append $tlsHosts .Values.ingress.customDomain -}}{{- end -}}
+{{- if $previewHost }}{{- $tlsHosts = append $tlsHosts $previewHost -}}{{- end -}}
 apiVersion: networking.k8s.io/v1
 kind: Ingress
 metadata:
@@ -10,10 +17,12 @@ metadata:
   namespace: {{ $ns }}
   labels:
     {{- include "cloudhost-app.labels" . | nindent 4 }}
+  {{- if gt (len $tlsHosts) 0 }}
   annotations:
     cert-manager.io/cluster-issuer: {{ .Values.ingress.clusterIssuer | quote }}
+  {{- end }}
 spec:
-  ingressClassName: nginx
+  ingressClassName: {{ $className }}
   rules:
     - host: {{ $host }}
       http:
@@ -49,14 +58,12 @@ spec:
                 port:
                   number: 80
     {{- end }}
+  {{- if gt (len $tlsHosts) 0 }}
   tls:
     - hosts:
-        - {{ $host }}
-        {{- if .Values.ingress.customDomain }}
-        - {{ .Values.ingress.customDomain }}
-        {{- end }}
-        {{- if $previewHost }}
-        - {{ $previewHost }}
+        {{- range $tlsHosts }}
+        - {{ . }}
         {{- end }}
       secretName: {{ $name }}-tls
+  {{- end }}
 {{- end }}

backend/src/applications/applications.service.ts

@@ -159,7 +159,9 @@ export class ApplicationsService {
       .createQueryBuilder('app')
       .leftJoinAndSelect('app.deployments', 'deployments')
       .where('app.userId = :userId', { userId })
-      .orderBy('app.createdAt', 'DESC');
+      .orderBy('app.createdAt', 'DESC')
+      // Ensure deployments[0] is the most recent so the UI shows the latest status.
+      .addOrderBy('deployments.createdAt', 'DESC');
 
     if (options?.productType === 'application') {
       qb.andWhere(
@@ -184,7 +186,9 @@ export class ApplicationsService {
       .createQueryBuilder('app')
       .leftJoinAndSelect('app.user', 'user')
       .leftJoinAndSelect('app.deployments', 'deployments')
-      .orderBy('app.createdAt', 'DESC');
+      .orderBy('app.createdAt', 'DESC')
+      // Ensure deployments[0] is the most recent so the UI shows the latest status.
+      .addOrderBy('deployments.createdAt', 'DESC');
 
     if (search && search.trim()) {
       const s = `%${search.trim()}%`;

backend/src/build/build.module.ts

@@ -1,17 +1,14 @@
 import { Module, forwardRef } from '@nestjs/common';
-import { BullModule } from '@nestjs/bull';
 import { BuildService } from './build.service';
-import { BuildProcessor } from './build.processor';
 import { KubernetesModule } from '../kubernetes/kubernetes.module';
 import { ClustersModule } from '../clusters/clusters.module';
 
 @Module({
   imports: [
-    BullModule.registerQueue({ name: 'build' }),
     forwardRef(() => KubernetesModule),
     ClustersModule,
   ],
-  providers: [BuildService, BuildProcessor],
+  providers: [BuildService],
   exports: [BuildService],
 })
 export class BuildModule {}

backend/src/build/build.processor.ts

@@ -1,42 +0,0 @@
-import { Process, Processor } from '@nestjs/bull';
-import { Logger } from '@nestjs/common';
-import { Job } from 'bull';
-import { BuildService } from './build.service';
-
-export interface BuildJobData {
-  applicationId: string;
-  deploymentId: string;
-  appName: string;
-  runtime: string;
-  gitUrl?: string;
-  codePath?: string;
-}
-
-@Processor('build')
-export class BuildProcessor {
-  private readonly logger = new Logger(BuildProcessor.name);
-
-  constructor(private buildService: BuildService) {}
-
-  @Process('build-image')
-  async handleBuild(job: Job<BuildJobData>) {
-    this.logger.log(`Processing build job ${job.id} for app: ${job.data.appName}`);
-
-    try {
-      await job.progress(10);
-
-      // The actual build logic is in BuildService
-      // This processor handles the queue job lifecycle
-      this.logger.log(`Build job ${job.id} started for ${job.data.appName}`);
-
-      await job.progress(50);
-      await job.progress(100);
-
-      this.logger.log(`Build job ${job.id} completed for ${job.data.appName}`);
-      return { status: 'completed', appName: job.data.appName };
-    } catch (error: any) {
-      this.logger.error(`Build job ${job.id} failed: ${error.message}`);
-      throw error;
-    }
-  }
-}

backend/src/build/build.service.ts

@@ -45,6 +45,12 @@ export class BuildService {
   private readonly logger = new Logger(BuildService.name);
   private readonly progressMap = new Map<string, BuildProgress>();
   private readonly activeBuilds = new Map<string, ActiveBuildSession>();
+  /**
+   * Kaniko executor image. Pinned (not `:latest`) so it can be cached on the node
+   * with imagePullPolicy=IfNotPresent — avoids re-pulling the ~250MB image on every build.
+   */
+  private readonly kanikoImage =
+    process.env.KANIKO_IMAGE || 'gcr.io/kaniko-project/executor:v1.23.2';
 
   constructor(
     private configService: ConfigService,
@@ -332,6 +338,7 @@ export class BuildService {
       initContainers.push({
         name: 'unzip-source',
         image: 'alpine:3.19',
+        imagePullPolicy: 'IfNotPresent',
         command: ['sh', '-c', `
           apk add --no-cache unzip tar gzip &&
           cp /workspace/Dockerfile /workspace-out/Dockerfile &&
@@ -390,6 +397,7 @@ export class BuildService {
       initContainers.push({
         name: 'git-clone',
         image: 'alpine/git:2.43.0',
+        imagePullPolicy: 'IfNotPresent',
         command: ['sh', '-c', `
           echo ">>> Cloning branch '${branch}' from ${app.gitUrl}" &&
           git clone --depth 1 --branch ${branch} ${cloneUrl} /workspace-out/source &&
@@ -417,6 +425,7 @@ export class BuildService {
       initContainers.push({
         name: 'prepare-workspace',
         image: 'alpine:3.19',
+        imagePullPolicy: 'IfNotPresent',
         command: ['sh', '-c', `
           mkdir -p /workspace-out/source &&
           cp /dockerfile/Dockerfile /workspace-out/Dockerfile &&
@@ -447,7 +456,8 @@ export class BuildService {
             containers: [
               {
                 name: 'kaniko',
-                image: 'gcr.io/kaniko-project/executor:latest',
+                image: this.kanikoImage,
+                imagePullPolicy: 'IfNotPresent',
                 args: kanikoArgs,
                 volumeMounts: kanikoVolumeMounts,
                 resources: {
@@ -519,17 +529,6 @@ export class BuildService {
     }
   }
 
-  /**
-   * Poll the helper pod until netcat finishes writing and reports the expected byte count.
-   */
-  private async waitForRemoteUploadDone(
-    kubeconfig: string,
-    namespace: string,
-    podName: string,
-    expectedSize: number,
-    deploymentId?: string,
-  ): Promise<void> { /* unused — kept for API compat */ }
-
   /**
    * Upload a local file to the helper pod using kubectl cp with progress tracking.
    * kubectl cp uses tar over the k8s exec API — reliable for any file size.
@@ -668,6 +667,7 @@ export class BuildService {
         containers: [{
           name: 'helper',
           image: 'alpine:3.19',
+          imagePullPolicy: 'IfNotPresent',
           command: ['sh', '-c', 'sleep 3600'],
           volumeMounts: [{ name: 'source', mountPath: '/data' }],
           resources: {
@@ -1467,6 +1467,30 @@ CMD ["sh", "-c", "DLL=$(find . -maxdepth 1 -name '*.dll' ! -name '*.deps.dll' !
 `;
   }
 
+  /**
+   * Whether a K8s API error is a transient connectivity/availability blip that
+   * should be retried rather than failing the operation. Covers socket-level
+   * errors, request timeouts (incl. the apiserver "request did not complete
+   * within the allotted timeout" message), DNS hiccups and 5xx/429 responses.
+   */
+  private isTransientK8sError(err: any): boolean {
+    const statusCode = err?.statusCode ?? err?.response?.statusCode ?? err?.body?.code;
+    if (typeof statusCode === 'number' && (statusCode >= 500 || statusCode === 429)) {
+      return true;
+    }
+    const haystack = [
+      err?.code,
+      err?.message,
+      err?.body?.message,
+      err?.cause?.code,
+    ]
+      .filter(Boolean)
+      .join(' ');
+    return /ECONNRESET|ECONNREFUSED|ETIMEDOUT|ESOCKETTIMEDOUT|EPIPE|EAI_AGAIN|ENOTFOUND|ENETUNREACH|socket hang up|timed? ?out|allotted timeout|did not complete|Client network socket disconnected/i.test(
+      haystack,
+    );
+  }
+
   private async waitForJobCompletion(
     batchApi: k8s.BatchV1Api,
     coreApi: k8s.CoreV1Api,
@@ -1493,9 +1517,12 @@ CMD ["sh", "-c", "DLL=$(find . -maxdepth 1 -name '*.dll' ! -name '*.deps.dll' !
       try {
         job = await batchApi.readNamespacedJob(jobName, namespace);
       } catch (pollErr: any) {
-        const code = pollErr?.code || pollErr?.message || '';
-        if (/ECONNRESET|ECONNREFUSED|ETIMEDOUT|socket hang up/i.test(String(code))) {
-          this.logger.warn(`Transient K8s API error polling job ${jobName}: ${code} — retrying in 5s`);
+        // The Kaniko job keeps running independently of these status polls.
+        // A single API blip (timeout, reset, 5xx, DNS) must NOT abort a build
+        // that is still progressing — just retry on the next poll tick.
+        if (this.isTransientK8sError(pollErr)) {
+          const detail = pollErr?.code || pollErr?.message || pollErr?.statusCode || 'unknown';
+          this.logger.warn(`Transient K8s API error polling job ${jobName}: ${detail} — retrying in 5s`);
           await new Promise(r => setTimeout(r, 5000));
           continue;
         }

backend/src/clusters/cluster-tools.service.ts

@@ -25,7 +25,11 @@ const CERT_MANAGER_CHART = 'jetstack/cert-manager';
 
 const CLUSTER_ISSUER_NAME = 'letsencrypt-prod';
 const ACME_PROD_SERVER = 'https://acme-v02.api.letsencrypt.org/directory';
-const DEFAULT_INGRESS_CLASS = 'nginx';
+// k3s ships Traefik by default; match the app Ingress class (INGRESS_CLASS)
+// so the ACME HTTP-01 solver Ingress is actually served by the controller.
+const DEFAULT_INGRESS_CLASS = (process.env.INGRESS_CLASS || 'traefik')
+  .trim()
+  .toLowerCase();
 
 const ISSUER_GROUP = 'cert-manager.io';
 const ISSUER_VERSION = 'v1';

backend/src/clusters/clusters.service.ts

@@ -300,12 +300,15 @@ export class ClustersService implements OnModuleInit, OnModuleDestroy {
     }
 
     const candidates = await this.getCachedHealthyClusters(pool || undefined, options.excludeClusterIds || []);
-    const appCounts = await this.getAppCounts(candidates.map((cluster) => cluster.id));
+    const clusterIds = candidates.map((cluster) => cluster.id);
+    const appCounts = await this.getAppCounts(clusterIds);
+    const reservations = await this.getInFlightReservations(clusterIds, options.applicationId);
     const candidateScores: Record<string, any>[] = [];
     const rejectionReasons: Record<string, any>[] = [];
 
     for (const cluster of candidates) {
-      const rejection = this.getClusterRejectionReason(cluster, estimatedRequest);
+      const reserved = reservations.get(cluster.id) || { cpuMillicores: 0, memoryMi: 0, pods: 0 };
+      const rejection = this.getClusterRejectionReason(cluster, estimatedRequest, reserved);
       if (rejection) {
         rejectionReasons.push({ clusterId: cluster.id, clusterName: cluster.name, reason: rejection });
         continue;
@@ -313,8 +316,8 @@ export class ClustersService implements OnModuleInit, OnModuleDestroy {
 
       const strategy = this.resolveStrategy(pool?.strategy);
       const appCount = appCounts.get(cluster.id) || 0;
-      const score = this.scoreCluster(cluster, estimatedRequest, appCount, strategy, (dto as any).region);
-      const resourceMetrics = this.getResourceMetrics(cluster, estimatedRequest, appCount);
+      const score = this.scoreCluster(cluster, estimatedRequest, appCount, strategy, (dto as any).region, reserved);
+      const resourceMetrics = this.getResourceMetrics(cluster, estimatedRequest, appCount, reserved);
       candidateScores.push({
         clusterId: cluster.id,
         clusterName: cluster.name,
@@ -382,11 +385,13 @@ export class ClustersService implements OnModuleInit, OnModuleDestroy {
     message: string,
   ): Promise<void> {
     if (clusterId) {
-      await this.clustersRepository.update(clusterId, {
-        healthStatus: 'degraded',
-        healthMessage: message,
-        lastHealthCheckedAt: new Date(),
-      });
+      // A single deployment failure (quota, image pull, app bug, transient
+      // scheduling pressure) does NOT mean the cluster is unhealthy — flipping
+      // its healthStatus to 'degraded' would poison it for every other app and,
+      // when there's only one cluster, cause "no healthy cluster has capacity"
+      // for the next deployment. The fallback loop already excludes this cluster
+      // for *this* app via excludeClusterIds, so here we only invalidate the
+      // cached capacity snapshot to force a fresh read on the next allocation.
       this.clusterHealthCache.delete(clusterId);
     }
 
@@ -1023,6 +1028,13 @@ export class ClustersService implements OnModuleInit, OnModuleDestroy {
 
     const { username, password } = this.registryService.getRegistryCredentials();
     const dsName = 'cloudhost-k3s-registry-mirrors';
+    // The mirror endpoint must be reachable by the node's containerd, which does
+    // NOT use cluster DNS — so we point it at the registry NodePort on loopback
+    // (http://127.0.0.1:<nodePort>) instead of the in-cluster service DNS name.
+    // Otherwise image pulls break whenever node-level resolution of
+    // *.svc.cluster.local is unavailable (e.g. right after a node restart).
+    const registryNodePort = 30500;
+    const nodePortHost = `127.0.0.1:${registryNodePort}`;
     const configureScript = [
       'set -e',
       'REG=/host/etc/rancher/k3s/registries.yaml',
@@ -1031,12 +1043,16 @@ export class ClustersService implements OnModuleInit, OnModuleDestroy {
       'mirrors:',
       `  "${registryUrl}":`,
       '    endpoint:',
-      `      - "http://${registryUrl}"`,
+      `      - "http://${nodePortHost}"`,
       'configs:',
       `  "${registryUrl}":`,
       '    auth:',
       `      username: ${JSON.stringify(username)}`,
       `      password: ${JSON.stringify(password)}`,
+      `  "${nodePortHost}":`,
+      '    auth:',
+      `      username: ${JSON.stringify(username)}`,
+      `      password: ${JSON.stringify(password)}`,
       'EOFREG',
       'if [ ! -f "$REG" ] || ! cmp -s /tmp/cloudhost-registries.yaml "$REG"; then',
       '  cp /tmp/cloudhost-registries.yaml "$REG"',
@@ -1212,7 +1228,11 @@ export class ClustersService implements OnModuleInit, OnModuleDestroy {
     }
   }
 
-  private getClusterRejectionReason(cluster: Cluster, estimatedRequest: Record<string, any>): string | null {
+  private getClusterRejectionReason(
+    cluster: Cluster,
+    estimatedRequest: Record<string, any>,
+    reserved: { cpuMillicores: number; memoryMi: number; pods: number } = { cpuMillicores: 0, memoryMi: 0, pods: 0 },
+  ): string | null {
     if (cluster.status !== ClusterStatus.ACTIVE) {
       return `status=${cluster.status}`;
     }
@@ -1221,7 +1241,7 @@ export class ClustersService implements OnModuleInit, OnModuleDestroy {
       return `health=${cluster.healthStatus}`;
     }
 
-    const metrics = this.getResourceMetrics(cluster, estimatedRequest, 0);
+    const metrics = this.getResourceMetrics(cluster, estimatedRequest, 0, reserved);
     const { available, utilization } = metrics;
 
     if (available.cpuMillicores > 0 && available.cpuMillicores < estimatedRequest.cpuMillicores) {
@@ -1252,8 +1272,9 @@ export class ClustersService implements OnModuleInit, OnModuleDestroy {
     appCount: number,
     strategy: PoolStrategy,
     desiredRegion?: string,
+    reserved: { cpuMillicores: number; memoryMi: number; pods: number } = { cpuMillicores: 0, memoryMi: 0, pods: 0 },
   ): number {
-    const metrics = this.getResourceMetrics(cluster, estimatedRequest, appCount);
+    const metrics = this.getResourceMetrics(cluster, estimatedRequest, appCount, reserved);
     const capacityScore = metrics.capacityScore;
     const appPenalty = Math.min(appCount, 100) * 0.75;
 
@@ -1281,6 +1302,7 @@ export class ClustersService implements OnModuleInit, OnModuleDestroy {
     cluster: Cluster,
     estimatedRequest: Record<string, any>,
     appCount: number,
+    reserved: { cpuMillicores: number; memoryMi: number; pods: number } = { cpuMillicores: 0, memoryMi: 0, pods: 0 },
   ): {
     available: { cpuMillicores: number; memoryMi: number; storageMi: number; pods: number };
     utilization: { cpu: number; memory: number; storage: number; pods: number; appPressure: number; average: number };
@@ -1297,17 +1319,23 @@ export class ClustersService implements OnModuleInit, OnModuleDestroy {
     const nodeCount = Number(resources.nodeCount || 0);
     const podCapacity = Number(resources.podCapacity || nodeCount * 110 || 0);
 
+    // Include capacity reserved by in-flight deployments whose pods don't exist
+    // on the cluster yet, so concurrent allocations don't over-commit the node.
+    const cpuUsed = cpuRequested + (reserved.cpuMillicores || 0);
+    const memoryUsed = memoryRequested + (reserved.memoryMi || 0);
+    const podsUsed = podCount + (reserved.pods || 0);
+
     const available = {
-      cpuMillicores: Math.max(cpuCapacity - cpuRequested, 0),
-      memoryMi: Math.max(memoryCapacity - memoryRequested, 0),
+      cpuMillicores: Math.max(cpuCapacity - cpuUsed, 0),
+      memoryMi: Math.max(memoryCapacity - memoryUsed, 0),
       storageMi: storageCapacity > 0 ? Math.max(storageCapacity - storageUsed, 0) : 0,
-      pods: podCapacity > 0 ? Math.max(podCapacity - podCount, 0) : 0,
+      pods: podCapacity > 0 ? Math.max(podCapacity - podsUsed, 0) : 0,
     };
     const utilization = {
-      cpu: this.utilizationRatio(cpuRequested, cpuCapacity),
-      memory: this.utilizationRatio(memoryRequested, memoryCapacity),
+      cpu: this.utilizationRatio(cpuUsed, cpuCapacity),
+      memory: this.utilizationRatio(memoryUsed, memoryCapacity),
       storage: this.utilizationRatio(storageUsed, storageCapacity),
-      pods: this.utilizationRatio(podCount, podCapacity),
+      pods: this.utilizationRatio(podsUsed, podCapacity),
       appPressure: Math.min(appCount / 100, 1),
       average: 0,
     };
@@ -1377,6 +1405,60 @@ export class ClustersService implements OnModuleInit, OnModuleDestroy {
     return new Map(rows.map((row) => [row.clusterId, parseInt(row.count, 10)]));
   }
 
+  /**
+   * Capacity that has been allocated to deployments which are still building but
+   * whose pods don't exist on the cluster yet (status pending/building). The
+   * cluster's `availableResources` snapshot only reflects pods that already exist
+   * (and is cached up to cacheTtlMs), so without this, two deployments started
+   * within the same window both read the same stale capacity and can both be
+   * placed on a cluster that only has room for one — the second then can't be
+   * scheduled and hangs. We subtract these in-flight requests during allocation.
+   *
+   * @param excludeApplicationId skip an app's own in-flight deployment (used on
+   *        fallback re-allocation so the app doesn't reserve capacity against itself).
+   */
+  private async getInFlightReservations(
+    clusterIds: string[],
+    excludeApplicationId?: string,
+  ): Promise<Map<string, { cpuMillicores: number; memoryMi: number; pods: number }>> {
+    const result = new Map<string, { cpuMillicores: number; memoryMi: number; pods: number }>();
+    if (clusterIds.length === 0) {
+      return result;
+    }
+
+    const rows: any[] = await this.dataSource.query(`
+      SELECT a.id as "applicationId",
+             a."clusterId" as "clusterId",
+             a."cpuRequest" as "cpuRequest",
+             a."memoryRequest" as "memoryRequest",
+             a.replicas as "replicas",
+             a."databaseType" as "databaseType",
+             a."enableRedis" as "enableRedis",
+             a."enableRabbitmq" as "enableRabbitmq",
+             a."appStorageSize" as "appStorageSize",
+             a."dbStorageSize" as "dbStorageSize",
+             a."optionalServiceResources" as "optionalServiceResources"
+      FROM deployments d
+      JOIN applications a ON a.id = d."applicationId"
+      WHERE a."clusterId" = ANY($1)
+        AND d.status IN ('pending', 'building')
+    `, [clusterIds]);
+
+    for (const row of rows) {
+      if (excludeApplicationId && row.applicationId === excludeApplicationId) {
+        continue;
+      }
+      const est = this.estimateApplicationRequest(row as CreateApplicationDto);
+      const current = result.get(row.clusterId) || { cpuMillicores: 0, memoryMi: 0, pods: 0 };
+      current.cpuMillicores += est.cpuMillicores;
+      current.memoryMi += est.memoryMi;
+      current.pods += est.podEstimate;
+      result.set(row.clusterId, current);
+    }
+
+    return result;
+  }
+
   private parseStorageToMi(storage: string): number {
     if (!storage) return 0;
     if (storage.endsWith('Ki')) return parseFloat(storage) / 1024;

backend/src/config/configuration.ts

@@ -28,6 +28,13 @@ function stripLeadingSubdomain(hostname: string): string {
 }
 
 function resolvePreviewRootDomainFromEnv(): string {
+  // Explicit override takes priority — the base domain used for preview URLs
+  // (e.g. PREVIEW_BASE_DOMAIN=3fase.ir) must be configurable, never hardcoded.
+  const explicit = process.env.PREVIEW_BASE_DOMAIN;
+  if (explicit && explicit.trim()) {
+    return explicit.trim().toLowerCase();
+  }
+
   const frontendUrl = process.env.FRONTEND_URL;
   if (frontendUrl) {
     try {
@@ -108,6 +115,12 @@ export default () => ({
   platform: {
     domain: resolvePlatformDomainFromEnv(),
     previewRootDomain: resolvePreviewRootDomainFromEnv(),
+    /**
+     * Ingress controller class used for app Ingress objects and the ACME
+     * HTTP-01 solver. k3s ships Traefik by default, so we default to "traefik".
+     * Set INGRESS_CLASS=nginx for clusters running ingress-nginx instead.
+     */
+    ingressClass: (process.env.INGRESS_CLASS || 'traefik').trim().toLowerCase(),
     uploadDir: process.env.UPLOAD_DIR || './uploads',
     /** StorageClass for new PVCs; must support allowVolumeExpansion for disk resize */
     storageClass: process.env.PLATFORM_STORAGE_CLASS || 'cloudhost-expandable',

backend/src/deployments/deployments.service.ts

@@ -30,13 +30,29 @@ export class DeploymentsService {
   ) {}
 
   /**
-   * Stable "number" part for preview host, derived from deployment id.
-   * We keep it numeric to match the "<number>" requirement.
+   * Random 7-digit suffix for the preview host: <userId>-<7-digit>.<baseDomain>.
+   * Generated once per application (see resolvePreviewNumber) and persisted.
    */
-  private computePreviewNumberFromDeploymentId(deploymentId: string): string {
-    const hashHex = crypto.createHash('sha256').update(deploymentId).digest('hex');
-    const num = parseInt(hashHex.slice(0, 8), 16) % 1_000_000; // 0..999999
-    return String(num).padStart(6, '0');
+  private generatePreviewNumber(): string {
+    return String(crypto.randomInt(1_000_000, 10_000_000));
+  }
+
+  /**
+   * Returns a STABLE preview number for an application: reuse the one already
+   * assigned to a previous deployment so the public preview URL never changes
+   * across redeploys (otherwise old links 404). Only generates a new number the
+   * first time the app is deployed. Apps with a custom domain get no preview host.
+   */
+  private async resolvePreviewNumber(applicationId: string): Promise<string> {
+    const existing = await this.deploymentsRepository
+      .createQueryBuilder('d')
+      .select('d.previewSubdomain', 'previewSubdomain')
+      .where('d.applicationId = :applicationId', { applicationId })
+      .andWhere('d.previewSubdomain IS NOT NULL')
+      .orderBy('d.createdAt', 'DESC')
+      .limit(1)
+      .getRawOne<{ previewSubdomain: string }>();
+    return existing?.previewSubdomain || this.generatePreviewNumber();
   }
 
   async triggerDeployment(applicationId: string, userId: string): Promise<Deployment> {
@@ -56,7 +72,7 @@ export class DeploymentsService {
     // Fill deterministic preview number after we have the deployment id.
     let previewSubdomain: string | null = null;
     if (!app.customDomain) {
-      previewSubdomain = this.computePreviewNumberFromDeploymentId(saved.id);
+      previewSubdomain = await this.resolvePreviewNumber(app.id);
       await this.deploymentsRepository.update(saved.id, { previewSubdomain });
       saved.previewSubdomain = previewSubdomain;
     }
@@ -657,7 +673,7 @@ export class DeploymentsService {
 
     let previewSubdomain: string | null = null;
     if (!app.customDomain) {
-      previewSubdomain = this.computePreviewNumberFromDeploymentId(saved.id);
+      previewSubdomain = await this.resolvePreviewNumber(app.id);
       await this.deploymentsRepository.update(saved.id, { previewSubdomain });
       saved.previewSubdomain = previewSubdomain;
     }

backend/src/kubernetes/kubernetes.service.ts

@@ -289,7 +289,7 @@ export class KubernetesService implements OnModuleInit {
     const namespacePrefix = app.userId.split('-')[0];
     const previewHost =
       previewNumber && !app.customDomain
-        ? `${namespacePrefix}-${previewNumber}-preview.${previewRootDomain}`
+        ? `${namespacePrefix}-${previewNumber}.${previewRootDomain}`
         : '';
     const pullRegistryUrl = this.registryService.getRegistryUrl();
     const isWordPress = app.runtime === AppRuntime.WORDPRESS;
@@ -317,6 +317,7 @@ export class KubernetesService implements OnModuleInit {
       envVars: this.resolveEnvVars(app),
       ingress: {
         enabled: true,
+        className: this.configService.get<string>('platform.ingressClass') || 'traefik',
         subdomain: app.subdomain || app.name,
         domain: domain,
         clusterIssuer: 'letsencrypt-prod',
@@ -931,8 +932,10 @@ export class KubernetesService implements OnModuleInit {
     // Main application container
     const appContainer: any = {
       name: ctx.appName,
+      // Image tags are unique per build (name:timestamp) and immutable, so
+      // IfNotPresent is correct and avoids re-pulling on every restart/scale-up.
       image: ctx.image,
-      imagePullPolicy: 'Always',
+      imagePullPolicy: 'IfNotPresent',
       ports: [{ containerPort: ctx.port }],
       envFrom,
       env: extraEnv,
@@ -1402,7 +1405,11 @@ export class KubernetesService implements OnModuleInit {
         },
       },
     ];
-    const tlsHosts = [host];
+    // Only PUBLIC, real-TLD hosts may go into the TLS block. The internal
+    // `${subdomain}.${domain}` host (e.g. *.apps.cloudhost.local) is not a valid
+    // public suffix — including it makes Let's Encrypt reject the whole order,
+    // which would also block the cert for the legitimate preview/custom domains.
+    const tlsHosts: string[] = [];
 
     if (customDomain) {
       rules.push({
@@ -1424,7 +1431,7 @@ export class KubernetesService implements OnModuleInit {
     const namespacePrefix = ctx.ownerId.split('-')[0];
     const previewHost =
       previewNumber && !customDomain
-        ? `${namespacePrefix}-${previewNumber}-preview.${previewRootDomain}`
+        ? `${namespacePrefix}-${previewNumber}.${previewRootDomain}`
         : '';
     if (previewHost) {
       rules.push({
@@ -1442,20 +1449,28 @@ export class KubernetesService implements OnModuleInit {
       tlsHosts.push(previewHost);
     }
 
+    const ingressClass =
+      this.configService.get<string>('platform.ingressClass') || 'traefik';
+    // Request a managed cert only when we actually have a public host to issue for.
+    const annotations: Record<string, string> = {};
+    const tls: k8s.V1IngressTLS[] = [];
+    if (tlsHosts.length > 0) {
+      annotations['cert-manager.io/cluster-issuer'] = 'letsencrypt-prod';
+      tls.push({ hosts: tlsHosts, secretName: `${ctx.appName}-tls` });
+    }
+
     const ingress: k8s.V1Ingress = {
       apiVersion: 'networking.k8s.io/v1',
       kind: 'Ingress',
       metadata: {
         name: ctx.appName,
         namespace: ctx.namespace,
-        annotations: {
-          'cert-manager.io/cluster-issuer': 'letsencrypt-prod',
-        },
+        annotations,
       },
       spec: {
-        ingressClassName: 'nginx',
+        ingressClassName: ingressClass,
         rules,
-        tls: [{ hosts: tlsHosts, secretName: `${ctx.appName}-tls` }],
+        ...(tls.length > 0 ? { tls } : {}),
       },
     };
 
@@ -2780,7 +2795,7 @@ export class KubernetesService implements OnModuleInit {
     if (verifiedCustomDomain) {
       ingressUrl = `https://${verifiedCustomDomain}`;
     } else if (previewNumber) {
-      ingressUrl = `https://${namespacePrefix}-${previewNumber}-preview.${previewRootDomain}`;
+      ingressUrl = `https://${namespacePrefix}-${previewNumber}.${previewRootDomain}`;
     }
 
     return {