feat(RHOAIENG-25241): disable ray usage metrics by default

Author: kryanbeane

docs/sphinx/user-docs/cluster-configuration.rst

@@ -44,6 +44,21 @@ requirements for creating the Ray Cluster.
   documentation on building a custom image
   `here <https://github.com/opendatahub-io/distributed-workloads/tree/main/images/runtime/examples>`__.
 
+Ray Usage Statistics
+-------------------
+
+By default, Ray usage statistics collection is disabled in CodeFlare SDK clusters. This stops statistics from being sent to AnyScale. If you want to enable usage statistics collection, you can set the ``RAY_USAGE_STATS_ENABLED`` environment variable to ``1`` in your cluster configuration:
+
+.. code:: python
+
+   from codeflare_sdk import Cluster, ClusterConfiguration
+
+   cluster = Cluster(ClusterConfiguration(
+       name='ray-example',
+       namespace='default',
+       envs={'RAY_USAGE_STATS_ENABLED': '1'}  # Enable usage statistics
+   ))
+
 The ``labels={"exampleLabel": "example"}`` parameter can be used to
 apply additional labels to the RayCluster resource.
 

src/codeflare_sdk/ray/cluster/config.py

@@ -161,6 +161,10 @@ def __post_init__(self):
                 "Warning: TLS verification has been disabled - Endpoint checks will be bypassed"
             )
 
+        # Set default environment variable to disable Ray usage stats if not already set
+        if "RAY_USAGE_STATS_ENABLED" not in self.envs:
+            self.envs["RAY_USAGE_STATS_ENABLED"] = "0"
+
         if self.enable_gcs_ft:
             if not self.redis_address:
                 raise ValueError(

src/codeflare_sdk/ray/cluster/test_config.py

@@ -24,6 +24,7 @@
 import filecmp
 import pytest
 import os
+import yaml
 
 parent = Path(__file__).resolve().parents[4]  # project directory
 expected_clusters_dir = f"{parent}/tests/test_cluster_yamls"
@@ -85,7 +86,11 @@ def test_config_creation_all_parameters(mocker):
     assert cluster.config.worker_memory_requests == "12G"
     assert cluster.config.worker_memory_limits == "16G"
     assert cluster.config.appwrapper == False
-    assert cluster.config.envs == {"key1": "value1", "key2": "value2"}
+    assert cluster.config.envs == {
+        "key1": "value1",
+        "key2": "value2",
+        "RAY_USAGE_STATS_ENABLED": "0",
+    }
     assert cluster.config.image == "example/ray:tag"
     assert cluster.config.image_pull_secrets == ["secret1", "secret2"]
     assert cluster.config.write_to_file == True
@@ -206,6 +211,46 @@ def test_gcs_fault_tolerance_config_validation():
         )
 
 
+def test_ray_usage_stats_default(mocker):
+    mocker.patch("kubernetes.client.ApisApi.get_api_versions")
+    mocker.patch("kubernetes.client.CustomObjectsApi.list_namespaced_custom_object")
+
+    cluster = Cluster(
+        ClusterConfiguration(name="default-usage-stats-cluster", namespace="ns")
+    )
+
+    # Verify that usage stats are disabled by default
+    assert cluster.config.envs["RAY_USAGE_STATS_ENABLED"] == "0"
+
+    # Check that the environment variable is set in the YAML
+    head_container = cluster.resource_yaml["spec"]["headGroupSpec"]["template"]["spec"][
+        "containers"
+    ][0]
+    env_vars = {env["name"]: env["value"] for env in head_container["env"]}
+    assert env_vars["RAY_USAGE_STATS_ENABLED"] == "0"
+
+
+def test_ray_usage_stats_enabled(mocker):
+    mocker.patch("kubernetes.client.ApisApi.get_api_versions")
+    mocker.patch("kubernetes.client.CustomObjectsApi.list_namespaced_custom_object")
+
+    cluster = Cluster(
+        ClusterConfiguration(
+            name="usage-stats-enabled-cluster",
+            namespace="ns",
+            envs={"RAY_USAGE_STATS_ENABLED": "1"},
+        )
+    )
+
+    assert cluster.config.envs["RAY_USAGE_STATS_ENABLED"] == "1"
+
+    head_container = cluster.resource_yaml["spec"]["headGroupSpec"]["template"]["spec"][
+        "containers"
+    ][0]
+    env_vars = {env["name"]: env["value"] for env in head_container["env"]}
+    assert env_vars["RAY_USAGE_STATS_ENABLED"] == "1"
+
+
 # Make sure to always keep this function last
 def test_cleanup():
     os.remove(f"{aw_dir}test-all-params.yaml")

tests/test_cluster_yamls/appwrapper/unit-test-all-params.yaml

@@ -54,6 +54,8 @@ spec:
                   value: value1
                 - name: key2
                   value: value2
+                - name: RAY_USAGE_STATS_ENABLED
+                  value: "0"
                 image: example/ray:tag
                 imagePullPolicy: Always
                 lifecycle:
@@ -159,6 +161,8 @@ spec:
                   value: value1
                 - name: key2
                   value: value2
+                - name: RAY_USAGE_STATS_ENABLED
+                  value: "0"
                 image: example/ray:tag
                 imagePullPolicy: Always
                 lifecycle:

tests/test_cluster_yamls/kueue/aw_kueue.yaml

@@ -75,6 +75,9 @@ spec:
                 - mountPath: /etc/ssl/certs/odh-ca-bundle.crt
                   name: odh-ca-cert
                   subPath: odh-ca-bundle.crt
+                env:
+                - name: RAY_USAGE_STATS_ENABLED
+                  value: "0"
               volumes:
               - configMap:
                   items:
@@ -133,6 +136,9 @@ spec:
                 - mountPath: /etc/ssl/certs/odh-ca-bundle.crt
                   name: odh-ca-cert
                   subPath: odh-ca-bundle.crt
+                env:
+                - name: RAY_USAGE_STATS_ENABLED
+                  value: "0"
               volumes:
               - configMap:
                   items:

tests/test_cluster_yamls/kueue/ray_cluster_kueue.yaml

@@ -75,6 +75,9 @@ spec:
                 - mountPath: /etc/ssl/certs/odh-ca-bundle.crt
                   name: odh-ca-cert
                   subPath: odh-ca-bundle.crt
+                env:
+                - name: RAY_USAGE_STATS_ENABLED
+                  value: "0"
               volumes:
               - configMap:
                   items:
@@ -133,6 +136,9 @@ spec:
                 - mountPath: /etc/ssl/certs/odh-ca-bundle.crt
                   name: odh-ca-cert
                   subPath: odh-ca-bundle.crt
+                env:
+                - name: RAY_USAGE_STATS_ENABLED
+                  value: "0"
               volumes:
               - configMap:
                   items:

tests/test_cluster_yamls/ray/default-appwrapper.yaml

@@ -53,6 +53,9 @@ spec:
                   name: dashboard
                 - containerPort: 10001
                   name: client
+                env:
+                - name: RAY_USAGE_STATS_ENABLED
+                  value: "0"
                 resources:
                   limits:
                     cpu: 2
@@ -111,6 +114,9 @@ spec:
                       - -c
                       - ray stop
                 name: machine-learning
+                env:
+                - name: RAY_USAGE_STATS_ENABLED
+                  value: "0"
                 resources:
                   limits:
                     cpu: 1

tests/test_cluster_yamls/ray/default-ray-cluster.yaml

@@ -45,6 +45,9 @@ spec:
             name: dashboard
           - containerPort: 10001
             name: client
+          env:
+          - name: RAY_USAGE_STATS_ENABLED
+            value: "0"
           resources:
             limits:
               cpu: 2
@@ -110,6 +113,9 @@ spec:
             requests:
               cpu: 1
               memory: 2G
+          env:
+          - name: RAY_USAGE_STATS_ENABLED
+            value: "0"
           volumeMounts:
           - mountPath: /etc/pki/tls/certs/odh-trusted-ca-bundle.crt
             name: odh-trusted-ca-cert

tests/test_cluster_yamls/ray/unit-test-all-params.yaml

@@ -45,6 +45,8 @@ spec:
             value: value1
           - name: key2
             value: value2
+          - name: RAY_USAGE_STATS_ENABLED
+            value: "0"
           image: example/ray:tag
           imagePullPolicy: Always
           lifecycle:
@@ -150,6 +152,8 @@ spec:
             value: value1
           - name: key2
             value: value2
+          - name: RAY_USAGE_STATS_ENABLED
+            value: "0"
           image: example/ray:tag
           imagePullPolicy: Always
           lifecycle: