chore: all resources in values

Signed-off-by: ThibaultFy <thibault.fouqueray@gmail.com>

backend/backend/settings/common.py

@@ -197,6 +197,7 @@
     "KANIKO_MIRROR": to_bool(os.environ.get("KANIKO_MIRROR", False)),
     "KANIKO_IMAGE": os.environ.get("KANIKO_IMAGE"),
     "KANIKO_DOCKER_CONFIG_SECRET_NAME": os.environ.get("KANIKO_DOCKER_CONFIG_SECRET_NAME"),
+    "KANIKO_RESOURCES": os.environ.get("KANIKO_RESOURCES"),
     "COMPUTE_POD_STARTUP_TIMEOUT_SECONDS": int(os.environ.get("COMPUTE_POD_STARTUP_TIMEOUT_SECONDS", 300)),
     "PRIVATE_CA_ENABLED": to_bool(os.environ.get("PRIVATE_CA_ENABLED")),
 }
@@ -223,6 +224,7 @@
 COMPUTE_POD_RUN_AS_GROUP = os.environ.get("COMPUTE_POD_RUN_AS_GROUP")
 COMPUTE_POD_FS_GROUP = os.environ.get("COMPUTE_POD_FS_GROUP")
 COMPUTE_POD_GKE_GPUS_LIMITS = int(os.environ.get("COMPUTE_POD_GKE_GPUS_LIMITS", 0))
+COMPUTE_POD_RESOURCES = os.environ.get("COMPUTE_POD_RESOURCES")
 
 # Prometheus configuration
 ENABLE_METRICS = to_bool(os.environ.get("ENABLE_METRICS", False))

backend/builder/image_builder/image_builder.py

@@ -23,7 +23,7 @@
 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_resources_requirements_from_yaml
 from substrapp.kubernetes_utils import get_security_context
 from substrapp.lock_local import lock_resource
 from substrapp.utils import timeit
@@ -43,6 +43,7 @@
 IMAGE_BUILD_TIMEOUT = settings.IMAGE_BUILD_TIMEOUT
 KANIKO_CONTAINER_NAME = "kaniko"
 HOSTNAME = settings.HOSTNAME
+KANIKO_RESOURCES = settings.KANIKO_RESOURCES
 
 
 def container_image_tag_from_function(function: orchestrator.Function) -> str:
@@ -307,7 +308,7 @@ 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="4Gi", memory_limit="32Gi"),
+        resources=get_resources_requirements_from_yaml(yaml_resources=KANIKO_RESOURCES),
     )
 
 

backend/substrapp/compute_tasks/compute_pod.py

@@ -6,10 +6,11 @@
 
 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_resources_requirements_from_yaml
 from substrapp.kubernetes_utils import get_security_context
 
 NAMESPACE = settings.NAMESPACE
+COMPUTE_POD_RESOURCES = settings.COMPUTE_POD_RESOURCES
 logger = structlog.get_logger(__name__)
 
 
@@ -113,7 +114,7 @@ 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="64Gi"),
+        resources=get_resources_requirements_from_yaml(yaml_resources=COMPUTE_POD_RESOURCES),
         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

@@ -1,4 +1,5 @@
 import kubernetes
+import yaml
 import structlog
 from django.conf import settings
 
@@ -47,11 +48,13 @@ 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 = "200M", memory_limit: str = "2G"
+def get_resources_requirements_from_yaml(
+    *,
+    yaml_resources: str,
 ) -> kubernetes.client.V1ResourceRequirements:
+    resources_dict = yaml.load(yaml_resources, Loader=yaml.FullLoader)
     return kubernetes.client.V1ResourceRequirements(
-        requests={"cpu": cpu_request, "memory": memory_request}, limits={"memory": memory_limit}
+        requests=resources_dict["requests"], limits=resources_dict["limits"]
     )
 
 

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

@@ -29,7 +29,7 @@ spec:
       - name: kaniko
         image: {{ include "common.images.name" $.Values.kaniko.image }}
         resources:
-          {{- toYaml $.Values.registryPrepopulate.kaniko.resources | nindent 12 }}
+          {{- toYaml $.Values.kaniko.resources | nindent 12 }}
         args:
           - "--context=/docker-context"
           {{- if .dstImage }}

charts/substra-backend/templates/statefulset-builder.yaml

@@ -160,6 +160,8 @@ spec:
             value: {{ .Values.kaniko.dockerConfigSecretName | quote }}
           - name: OBJECTSTORE_URL
             value: {{ include "substra-backend.objectStore.url" . | quote }}
+          - name: KANIKO_RESOURCES
+            value: {{ toYaml .Values.kaniko.resources | quote }}
         ports:
           - name: http
             containerPort: 8000

charts/substra-backend/templates/statefulset-worker.yaml

@@ -109,6 +109,8 @@ spec:
               valueFrom:
                 fieldRef:
                   fieldPath: spec.nodeName
+            - name: COMPUTE_POD_RESOURCES
+              value: {{ toYaml .Values.worker.computePod.resources | quote }}
             - name: COMPUTE_POD_MAX_STARTUP_WAIT_SECONDS
               value: {{ .Values.worker.computePod.maxStartupWaitSeconds | quote }}
             - name: OBJECTSTORE_URL

charts/substra-backend/values.yaml

@@ -355,6 +355,16 @@ worker:
       fsGroup: 1001
       runAsUser: 1001
       runAsGroup: 1001
+    ## @param worker.computePod.resources.requests.cpu Worker compute pod container cpu request
+    ## @param worker.computePod.resources.requests.memory Worker compute pod container memory request
+    ## @param worker.computePod.resources.limits.memory Worker compute pod container memory limit
+    ##
+    resources:
+      requests:
+        cpu: "500m"
+        memory: "512Mi"
+      limits:
+        memory: "64Gi"
   events:
     ## @param worker.events.enabled Enable event service
     ##
@@ -577,7 +587,6 @@ builder:
     limits:
       cpu: "2000m"
       memory: "8Gi"
-
   ## @param builder.nodeSelector Node labels for pod assignment
   ##
   nodeSelector: { }
@@ -749,6 +758,16 @@ kaniko:
     registry: gcr.io
     repository: kaniko-project/executor
     tag: v1.8.1
+  ## @param kaniko.resources.requests.cpu Kaniko container cpu request
+  ## @param kaniko.resources.requests.memory Kaniko container memory request
+  ## @param kaniko.resources.limits.memory Kaniko container memory limit
+  ##
+  resources:
+    requests:
+      cpu: "500m"
+      memory: "256Mi"
+    limits:
+      memory: "32Gi"
   ## @param kaniko.mirror If set to `true` pull base images from the local registry.
   ##
   mirror: false
@@ -814,17 +833,6 @@ registryPrepopulate:
         cpu: "100m"
       limits:
         memory: "400Mi"
-  ## @param registryPrepopulate.kaniko.resources.requests.cpu Kaniko container cpu request
-  ## @param registryPrepopulate.kaniko.resources.requests.memory Kaniko container memory request
-  ## @param registryPrepopulate.kaniko.resources.limits.memory Kaniko container memory limit
-  ##
-  kaniko:
-    resources:
-      requests:
-        memory: "256Mi"
-        cpu: "500m"
-      limits:
-        memory: "8Gi"
   ## @param registryPrepopulate.pause.resources.requests.cpu Pause container cpu request
   ## @param registryPrepopulate.pause.resources.requests.memory Pause container memory request
   ## @param registryPrepopulate.pause.resources.limits.memory Pause container memory limit