fix(app): use 3 scenarios for session cpu limits

helm-chart/renku-notebooks/templates/deployment.yaml

@@ -119,6 +119,8 @@ spec:
                 fieldRef:
                   apiVersion: v1
                   fieldPath: metadata.namespace
+            - name: ENFORCE_CPU_LIMITS
+              value: {{ .Values.enforceCPULimits | quote }}
           ports:
             - name: http
               containerPort: 8000

helm-chart/renku-notebooks/templates/test.yaml

@@ -76,6 +76,8 @@ spec:
           value: |
             {{- toYaml . | nindent 12 }}
         {{- end }}
+        - name: ENFORCE_CPU_LIMITS
+          value: {{ $.Values.enforceCPULimits | quote }}
       command:
         - pipenv
         - run

helm-chart/renku-notebooks/values.yaml

@@ -152,6 +152,13 @@ serverDefaults:
   gpu_request: 0
   lfs_auto_fetch: false
 
+## How to enforce CPU limits for sessions, options are "lax", "off" or "strict"
+## - "strict" = CPU limit equals cpu request
+## - "lax" = CPU limit equals maximum from server_options, if CPU is not in server options then
+##   CPU limit is set to the CPU request
+## - "off" = no CPU limits at all
+enforceCPULimits: "off"
+
 ## Default image used when the requested image for a user session cannot be found.
 defaultSessionImage: "renku/renkulab-py:3.9-0.10.1"
 

renku_notebooks/api/classes/server.py

@@ -210,14 +210,23 @@ def _get_registry_secret(self, b64encode=True):
         return output
 
     def _get_session_k8s_resources(self):
-        cpu = float(self.server_options["cpu_request"])
+        cpu_request = float(self.server_options["cpu_request"])
         mem = self.server_options["mem_request"]
         gpu_req = self.server_options.get("gpu_request", {})
         gpu = {"nvidia.com/gpu": str(gpu_req)} if gpu_req else None
         resources = {
-            "requests": {"memory": mem, "cpu": cpu},
-            "limits": {"memory": mem, "cpu": cpu},
+            "requests": {"memory": mem, "cpu": cpu_request},
+            "limits": {"memory": mem},
         }
+        if current_app.config["ENFORCE_CPU_LIMITS"] == "lax":
+            if "cpu_request" in current_app.config["SERVER_OPTIONS_UI"]:
+                resources["limits"]["cpu"] = max(
+                    current_app.config["SERVER_OPTIONS_UI"]["cpu_request"]["options"]
+                )
+            else:
+                resources["limits"]["cpu"] = cpu_request
+        elif current_app.config["ENFORCE_CPU_LIMITS"] == "strict":
+            resources["limits"]["cpu"] = cpu_request
         if gpu:
             resources["requests"] = {**resources["requests"], **gpu}
             resources["limits"] = {**resources["limits"], **gpu}

renku_notebooks/config.py

@@ -118,3 +118,4 @@
 SESSION_NODE_SELECTOR = safe_load(os.environ.get("SESSION_NODE_SELECTOR", "{}"))
 SESSION_AFFINITY = safe_load(os.environ.get("SESSION_AFFINITY", "{}"))
 SESSION_TOLERATIONS = safe_load(os.environ.get("SESSION_TOLERATIONS", "[]"))
+ENFORCE_CPU_LIMITS = os.getenv("ENFORCE_CPU_LIMITS", "off")