diff --git a/argocd/apps/plane/values-plane.yaml b/argocd/apps/plane/values-plane.yaml index 4dc5bf8..5510a34 100644 --- a/argocd/apps/plane/values-plane.yaml +++ b/argocd/apps/plane/values-plane.yaml @@ -6,14 +6,12 @@ metadata: - resources-finalizer.argocd.argoproj.io spec: ignoreDifferences: - # PVCs - creationTimestamp und Status ignorieren - group: "" kind: PersistentVolumeClaim jsonPointers: - /metadata/creationTimestamp - /status - # StatefulSets - alle volumeClaimTemplates komplett ignorieren - group: apps kind: StatefulSet jsonPointers: @@ -21,7 +19,6 @@ spec: - /status - /spec/replicas - # Worker Deployment - Replica Status ignorieren - group: apps kind: Deployment name: plane-worker-wl @@ -59,47 +56,53 @@ spec: aws_secret_access_key: "0d54ee2f943f2a56b8cafc3afe9cb1e2f9fecac2" aws_region: "eu-central-1" aws_s3_endpoint_url: "https://sws3.innovation-hub-niedersachsen.de" - # Celery Worker Konfiguration - Optimiert für Stabilität - CELERY_WORKER_CONCURRENCY: "2" - CELERY_WORKER_MAX_TASKS_PER_CHILD: "100" - CELERY_WORKER_MAX_MEMORY_PER_CHILD: "200000" - # Task Timeouts zur Vermeidung von Memory-Leaks - CELERY_TASK_SOFT_TIME_LIMIT: "300" - CELERY_TASK_TIME_LIMIT: "600" + + # Celery Worker - Aggressive Memory-Begrenzung + CELERY_WORKER_CONCURRENCY: "1" + CELERY_WORKER_MAX_TASKS_PER_CHILD: "10" + CELERY_WORKER_MAX_MEMORY_PER_CHILD: "500000" + CELERY_WORKER_PREFETCH_MULTIPLIER: "1" + + # Task Limits - Sehr restriktiv + CELERY_TASK_SOFT_TIME_LIMIT: "120" + CELERY_TASK_TIME_LIMIT: "180" + CELERY_TASK_ACKS_LATE: "true" + CELERY_TASK_REJECT_ON_WORKER_LOST: "true" + + # Python Memory Management + PYTHONMALLOC: "malloc" + MALLOC_TRIM_THRESHOLD_: "65536" + + # Logging reduzieren + CELERY_WORKER_LOG_LEVEL: "WARNING" - # Worker-spezifische Einstellungen worker: replicas: 1 - # Reduzierte Concurrency für bessere Memory-Nutzung concurrency: 1 + resources: requests: - memory: "4Gi" + memory: "2Gi" cpu: "500m" limits: - memory: "8Gi" - cpu: "2000m" - # Vereinfachte Health Checks + memory: "6Gi" + cpu: "1500m" + + # Health Checks komplett deaktiviert für Debugging readinessProbe: exec: - command: - - /bin/sh - - -c - - celery -A plane inspect ping -d celery@$HOSTNAME -t 5 - initialDelaySeconds: 60 - periodSeconds: 60 - timeoutSeconds: 15 - failureThreshold: 5 + command: ["/bin/true"] + initialDelaySeconds: 10 + periodSeconds: 300 + livenessProbe: exec: - command: - - /bin/sh - - -c - - celery -A plane inspect ping -d celery@$HOSTNAME -t 5 - initialDelaySeconds: 120 - periodSeconds: 120 - timeoutSeconds: 15 - failureThreshold: 5 + command: ["/bin/true"] + initialDelaySeconds: 30 + periodSeconds: 300 + + # Graceful Shutdown + terminationGracePeriodSeconds: 60 destination: server: 'https://kubernetes.default.svc'