fix(app): correct the worker KEDA PostgreSQL scaler (#12055)

Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
Co-authored-by: Utwo <mihai.legat@gmail.com>

contrib/k8s/helm/prowler-app/templates/worker/scaled-object.yaml

@@ -18,15 +18,12 @@ spec:
   triggers:
     - type: {{ .Values.worker.keda.triggerType }}
       metadata:
-        userName: "postgres"
+        userName: {{ .Values.worker.keda.postgresql.userName | quote }}
-        passwordFromEnv: POSTGRES_ADMIN_PASSWORD
+        passwordFromEnv: {{ .Values.worker.keda.postgresql.passwordFromEnv | quote }}
-        host: {{ .Release.Name }}-postgresql
+        host: {{ .Values.worker.keda.postgresql.host | default (printf "%s-postgresql.%s.svc.cluster.local" .Release.Name .Release.Namespace) | quote }}
-        port: {{ .Values.postgresql.port | quote }}
+        port: {{ .Values.worker.keda.postgresql.port | quote }}
-        dbName: {{ .Values.postgresql.auth.database | quote }}
+        dbName: {{ .Values.worker.keda.postgresql.database | default .Values.postgresql.auth.database | quote }}
-        sslmode: disable
+        sslmode: {{ .Values.worker.keda.postgresql.sslmode | quote }}
-        # Query for KEDA to count the number of scans that are in executing, available, or scheduled states,
+        query: {{ .Values.worker.keda.query | quote }}
-        # where the scheduled time is within the last 2 hours and is before NOW(). Used for scaling workers.
+        targetQueryValue: {{ .Values.worker.keda.targetQueryValue | quote }}
-        query: >-
-          SELECT COUNT(*) FROM scans WHERE ((state='executing' OR state='available' OR state='scheduled') and scheduled_at < NOW() and scheduled_at > NOW() - INTERVAL '2 hours')
-        targetQueryValue: "1"
 {{- end }}

contrib/k8s/helm/prowler-app/values.yaml

@@ -427,10 +427,61 @@ worker:
     pollingInterval: 30
     # -- The cooldown period in seconds for scaling
     cooldownPeriod: 120
-    # -- The trigger type for scaling (cpu or memory)
+    # -- The KEDA scaler type. Only `postgresql` is supported by the default query below.
     triggerType: "postgresql"
-    # -- The target utilization percentage for the worker pods
+    # PostgreSQL connection used by the scaler query. The KEDA operator opens this
-    value: "50"
+    # connection from its own namespace, so `host` must resolve from there. The
+    # defaults target the bundled postgresql subchart; set them explicitly when
+    # using an external database (postgresql.enabled: false).
+    postgresql:
+      # -- Scaler database host. Defaults to the bundled "<release>-postgresql.<namespace>.svc.cluster.local" service.
+      host: ""
+      # -- Scaler database port.
+      port: "5432"
+      # -- Scaler database name. Defaults to `postgresql.auth.database`.
+      database: ""
+      # -- User the scaler authenticates as.
+      userName: "postgres"
+      # -- Name of an env var on the worker container holding the password.
+      passwordFromEnv: "POSTGRES_ADMIN_PASSWORD"
+      # -- sslmode for the scaler connection.
+      sslmode: "disable"
+    # -- The scaler divides the query result by this value to get the desired replica count.
+    targetQueryValue: "1"
+    # -- Query the scaler runs to measure pending work. It replaces the previous
+    # 2-hour scheduled-only window, which missed manual scans, older backlogs and
+    # in-progress scans. Override to tune scaling for your workload.
+    #
+    # The default sums three signals:
+    #   1. Scans executing or available, bounded to rows updated in the last 24h so
+    #      orphaned rows do not pin the worker up, plus scheduled scans that are due
+    #      (no lower bound, so an overdue backlog still scales up).
+    #   2. Scan tasks published in the last 48h that no worker has finished. A PENDING
+    #      TaskResult is written at publish time (before_task_publish in api/signals.py),
+    #      so Beat's daily publishes are visible even with zero workers. Signal 1 alone
+    #      deadlocks with minReplicas 0: every scan row after the first is created by
+    #      the worker, so once the initial row ages out of the 24h bound there is
+    #      nothing to count and nothing to create more.
+    #   3. Non-scan tasks pending in the last hour. Provider connection checks,
+    #      deletions, reports and backfills never touch the scans table, so without
+    #      this they are never picked up while the worker is scaled to zero.
+    #      This includes reconcile-orphan-tasks, a Beat watchdog that runs every two
+    #      minutes, so with minReplicas 0 the worker is woken about that often. Add
+    #      it to the excluded task names below, or raise cooldownPeriod, if you would
+    #      rather trade watchdog latency for longer idle periods.
+    query: >-
+      SELECT
+      (SELECT COUNT(*) FROM scans
+      WHERE (state IN ('executing', 'available') AND updated_at > NOW() - INTERVAL '24 hours')
+      OR (state = 'scheduled' AND scheduled_at < NOW()))
+      + (SELECT COUNT(*) FROM django_celery_results_taskresult
+      WHERE task_name IN ('scan-perform', 'scan-perform-scheduled')
+      AND status IN ('PENDING', 'RECEIVED', 'STARTED')
+      AND date_created > NOW() - INTERVAL '48 hours')
+      + (SELECT COUNT(*) FROM django_celery_results_taskresult
+      WHERE task_name NOT IN ('scan-perform', 'scan-perform-scheduled')
+      AND status IN ('PENDING', 'RECEIVED', 'STARTED')
+      AND date_created > NOW() - INTERVAL '1 hour')
 
 worker_beat:
   # This will set the replicaset count more information can be found here: https://kubernetes.io/docs/concepts/workloads/controllers/replicaset/