chore: add limits and requests to pods

Signed-off-by: SdgJlbl <[email protected]>

backend/builder/image_builder/image_builder.py

@@ -22,6 +22,7 @@
 from substrapp.compute_tasks.volumes import get_worker_subtuple_pvc_name
 from substrapp.docker_registry import USER_IMAGE_REPOSITORY
 from substrapp.kubernetes_utils import delete_pod
+from substrapp.kubernetes_utils import get_resources_requirements
 from substrapp.kubernetes_utils import get_security_context
 from substrapp.lock_local import lock_resource
 from substrapp.utils import timeit
@@ -291,6 +292,7 @@ def _build_container(dockerfile_mount_path: str, image_tag: str) -> kubernetes.c
         args=args,
         volume_mounts=volume_mounts,
         security_context=container_security_context,
+        resources=get_resources_requirements(cpu_request="1000m", memory_request="2Gi", memory_limit="128Gi"),
     )
 
 

backend/substrapp/compute_tasks/compute_pod.py

@@ -6,6 +6,7 @@
 
 from substrapp.kubernetes_utils import delete_pod
 from substrapp.kubernetes_utils import get_pod_security_context
+from substrapp.kubernetes_utils import get_resources_requirements
 from substrapp.kubernetes_utils import get_security_context
 
 NAMESPACE = settings.NAMESPACE
@@ -112,6 +113,7 @@ def create_pod(
         args=None,
         volume_mounts=volume_mounts + gpu_volume_mounts,
         security_context=get_security_context(),
+        resources=get_resources_requirements(cpu_request="1000m", memory_request="1Gi", memory_limit="256Gi"),
         env=[kubernetes.client.V1EnvVar(name=env_name, value=env_value) for env_name, env_value in environment.items()],
         **container_optional_kwargs,
     )

backend/substrapp/kubernetes_utils.py

@@ -47,6 +47,14 @@ def get_security_context(root: bool = False, capabilities: list[str] = None) ->
     return security_context
 
 
+def get_resources_requirements(
+    *, cpu_request: str = "1000m", memory_request: str = "100M", memory_limit: str = "1G"
+) -> kubernetes.client.V1ResourceRequirements:
+    return kubernetes.client.V1ResourceRequirements(
+        requests={"cpu": cpu_request, "memory": memory_request}, limits={"memory": memory_limit}
+    )
+
+
 def pod_exists_by_label_selector(k8s_client: kubernetes.client.CoreV1Api, label_selector: str) -> bool:
     """Return True if the pod exists, else False.
 

charts/substra-backend/templates/deployment-api-events.yaml

@@ -41,6 +41,12 @@ spec:
         - name: api-event-app
           image: {{ include "substra-backend.images.name" (dict "img" .Values.api.events.image "defaultTag" $.Chart.AppVersion) }}
           imagePullPolicy: {{ .Values.api.events.image.pullPolicy }}
+          resources:
+            requests:
+              memory: "200Mi"
+              cpu: "500m"
+            limits:
+              memory: "400Mi"
           command: ["/bin/bash"]
           {{- if eq .Values.settings "prod" }}
           args: ["-c", "python manage.py consume"]

charts/substra-backend/templates/deployment-registry-prepopulate.yaml

@@ -23,9 +23,21 @@ spec:
       initContainers:
       - name: wait-registry
         image: jwilder/dockerize:0.6.1
+        resources:
+          requests:
+            memory: "200Mi"
+            cpu: "500m"
+          limits:
+            memory: "400Mi"
         command: ['dockerize', '-wait', 'tcp://{{ $.Release.Name }}-docker-registry:5000']
       - name: kaniko
         image: {{ include "common.images.name" $.Values.kaniko.image }}
+        resources:
+          requests:
+            memory: "2Gi"
+            cpu: "1000m"
+          limits:
+            memory: "4Gi"
         args:
           - "--context=/docker-context"
           {{- if .dstImage }}
@@ -48,6 +60,13 @@ spec:
       containers:
       - image: gcr.io/google-containers/pause:3.2
         name: pause
+        resources:
+          requests:
+            memory: "64Mi"
+            cpu: "250m"
+          limits:
+            memory: "128Mi"
+            cpu: "500m"
       volumes:
       - name: kaniko-dir
         emptyDir: {}

charts/substra-backend/templates/deployment-worker-events.yaml

@@ -41,6 +41,12 @@ spec:
         - name: worker-event-app
           image: {{ include "substra-backend.images.name" (dict "img" .Values.worker.events.image "defaultTag" $.Chart.AppVersion) }}
           imagePullPolicy: {{ .Values.worker.events.image.pullPolicy }}
+          resources:
+            requests:
+              memory: "200Mi"
+              cpu: "500m"
+            limits:
+              memory: "400Mi"
           command: ["/bin/bash"]
           {{- if eq .Values.settings "prod" }}
           args: ["-c", "python manage.py consume"]

charts/substra-backend/values.yaml

@@ -181,7 +181,15 @@ server:
   ##     cpu: 100m
   ##     memory: 128Mi
   ##
-  resources: {}
+  resources: {
+    requests: {
+      cpu: "1000m",
+      memory: "4Gi"
+    },
+    limits: {
+      memory: "32Gi"
+    }
+  }
 
   persistence:
     ## @param server.persistence.storageClass Specify the _StorageClass_ used to provision the volume. Or the default _StorageClass_ will be used. Set it to `-` to disable dynamic provisioning
@@ -828,6 +836,15 @@ postgresql:
       capabilities:
         drop:
         - ALL
+  resources: {
+    requests: {
+      cpu: "1000m",
+      memory: "200Mi"
+    },
+    limits: {
+      memory: "64Gi"
+    }
+  }
 
 ## @skip redis
 ##
@@ -849,6 +866,15 @@ redis:
     appendonly yes
     # Disable RDB persistence since AOF persistence is enabled
     save ""
+  resources: {
+    requests: {
+      cpu: "1000m",
+      memory: "512Mi"
+    },
+    limits: {
+      memory: "1024Mi"
+    }
+  }
 
 ## @skip docker-registry
 ##
@@ -861,6 +887,15 @@ docker-registry:
     deleteEnabled: true
   service:
     type: NodePort
+  resources: {
+    requests: {
+      cpu: "1000m",
+      memory: "16Gi"
+    },
+    limits: {
+      memory: "64Gi"
+    }
+  }
 
 ## @skip minio
 ##
@@ -881,6 +916,15 @@ minio:
     capabilities:
       drop:
         - ALL
+  resources: {
+    requests: {
+      cpu: "1000m",
+      memory: "16Gi"
+    },
+    limits: {
+      memory: "64Gi"
+    }
+}
 
 ## @skip localstack
 ##