RHOAIENG-27828: Template runtime image SHA

Author: kryanbeane

demo-notebooks/guided-demos/notebook-ex-outputs/0_basic_ray.ipynb

@@ -49,7 +49,7 @@
     "\n",
     "NOTE: The default images used by the CodeFlare SDK for creating a RayCluster resource depend on the installed Python version:\n",
     "\n",
-    "- For Python 3.11: 'quay.io/modh/ray:2.46.0-py311-cu121'\n",
+    "- For Python 3.11: 'quay.io/modh/ray:2.47.1-py311-cu121'\n",
     "\n",
     "If you prefer to use a custom Ray image that better suits your needs, you can specify it in the image field to override the default."
    ]

src/codeflare_sdk/common/utils/constants.py

@@ -1 +1,3 @@
 RAY_VERSION = "2.47.1"
+# Below references ray:2.47.1-py311-cu121
+CUDA_RUNTIME_IMAGE = "quay.io/modh/ray@sha256:6d076aeb38ab3c34a6a2ef0f58dc667089aa15826fa08a73273c629333e12f1e"

src/codeflare_sdk/common/utils/unit_test_support.py

@@ -14,6 +14,7 @@
 
 import string
 import sys
+from codeflare_sdk.common.utils import constants
 from codeflare_sdk.ray.cluster.cluster import (
     Cluster,
     ClusterConfiguration,
@@ -68,7 +69,7 @@ def create_cluster_wrong_type():
         worker_extended_resource_requests={"nvidia.com/gpu": 7},
         appwrapper=True,
         image_pull_secrets=["unit-test-pull-secret"],
-        image="quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707",
+        image=constants.CUDA_RUNTIME_IMAGE,
         write_to_file=True,
         labels={1: 1},
     )
@@ -148,19 +149,29 @@ def get_cluster_object(file_a, file_b):
 
 def get_ray_obj(group, version, namespace, plural):
     # To be used for mocking list_namespaced_custom_object for Ray Clusters
-    rc_a_path = f"{parent}/tests/test_cluster_yamls/support_clusters/test-rc-a.yaml"
-    rc_b_path = f"{parent}/tests/test_cluster_yamls/support_clusters/test-rc-b.yaml"
-    rc_a, rc_b = get_cluster_object(rc_a_path, rc_b_path)
+    rc_a = apply_template(
+        f"{parent}/tests/test_cluster_yamls/support_clusters/test-rc-a.yaml",
+        get_template_variables(),
+    )
+    rc_b = apply_template(
+        f"{parent}/tests/test_cluster_yamls/support_clusters/test-rc-b.yaml",
+        get_template_variables(),
+    )
 
     rc_list = {"items": [rc_a, rc_b]}
     return rc_list
 
 
 def get_ray_obj_with_status(group, version, namespace, plural):
     # To be used for mocking list_namespaced_custom_object for Ray Clusters with statuses
-    rc_a_path = f"{parent}/tests/test_cluster_yamls/support_clusters/test-rc-a.yaml"
-    rc_b_path = f"{parent}/tests/test_cluster_yamls/support_clusters/test-rc-b.yaml"
-    rc_a, rc_b = get_cluster_object(rc_a_path, rc_b_path)
+    rc_a = apply_template(
+        f"{parent}/tests/test_cluster_yamls/support_clusters/test-rc-a.yaml",
+        get_template_variables(),
+    )
+    rc_b = apply_template(
+        f"{parent}/tests/test_cluster_yamls/support_clusters/test-rc-b.yaml",
+        get_template_variables(),
+    )
 
     rc_a.update(
         {
@@ -205,19 +216,29 @@ def get_ray_obj_with_status(group, version, namespace, plural):
 
 def get_aw_obj(group, version, namespace, plural):
     # To be used for mocking list_namespaced_custom_object for AppWrappers
-    aw_a_path = f"{parent}/tests/test_cluster_yamls/support_clusters/test-aw-a.yaml"
-    aw_b_path = f"{parent}/tests/test_cluster_yamls/support_clusters/test-aw-b.yaml"
-    aw_a, aw_b = get_cluster_object(aw_a_path, aw_b_path)
+    aw_a = apply_template(
+        f"{parent}/tests/test_cluster_yamls/support_clusters/test-aw-a.yaml",
+        get_template_variables(),
+    )
+    aw_b = apply_template(
+        f"{parent}/tests/test_cluster_yamls/support_clusters/test-aw-b.yaml",
+        get_template_variables(),
+    )
 
     aw_list = {"items": [aw_a, aw_b]}
     return aw_list
 
 
 def get_aw_obj_with_status(group, version, namespace, plural):
     # To be used for mocking list_namespaced_custom_object for AppWrappers with statuses
-    aw_a_path = f"{parent}/tests/test_cluster_yamls/support_clusters/test-aw-a.yaml"
-    aw_b_path = f"{parent}/tests/test_cluster_yamls/support_clusters/test-aw-b.yaml"
-    aw_a, aw_b = get_cluster_object(aw_a_path, aw_b_path)
+    aw_a = apply_template(
+        f"{parent}/tests/test_cluster_yamls/support_clusters/test-aw-a.yaml",
+        get_template_variables(),
+    )
+    aw_b = apply_template(
+        f"{parent}/tests/test_cluster_yamls/support_clusters/test-aw-b.yaml",
+        get_template_variables(),
+    )
 
     aw_a.update(
         {
@@ -273,9 +294,8 @@ def apply_template(yaml_file_path, variables):
 
 
 def get_expected_image():
-    # TODO: Add Python 3.12 support
-    python_version = sys.version_info
-    return "quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707"
+    # TODO: Select image based on Python version
+    return constants.CUDA_RUNTIME_IMAGE
 
 
 def get_template_variables():

src/codeflare_sdk/ray/appwrapper/test_awload.py

@@ -12,8 +12,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from codeflare_sdk.common.utils.unit_test_support import (
+    apply_template,
     arg_check_aw_apply_effect,
     arg_check_aw_del_effect,
+    get_template_variables,
 )
 from codeflare_sdk.ray.appwrapper import AWManager
 from codeflare_sdk.ray.cluster import Cluster, ClusterConfiguration
@@ -47,8 +49,11 @@ def test_AWManager_creation(mocker):
         assert type(e) == FileNotFoundError
         assert str(e) == "[Errno 2] No such file or directory: 'fake'"
     try:
-        testaw = AWManager(
-            f"{parent}/tests/test_cluster_yamls/appwrapper/test-case-bad.yaml"
+        testaw = apply_template(
+            AWManager(
+                f"{parent}/tests/test_cluster_yamls/appwrapper/test-case-bad.yaml"
+            ),
+            get_template_variables(),
         )
     except Exception as e:
         assert type(e) == ValueError

src/codeflare_sdk/ray/cluster/build_ray_cluster.py

@@ -50,6 +50,8 @@
 import warnings
 import json
 
+from codeflare_sdk.common.utils import constants
+
 FORBIDDEN_CUSTOM_RESOURCE_TYPES = ["GPU", "CPU", "memory"]
 VOLUME_MOUNTS = [
     V1VolumeMount(
@@ -94,7 +96,7 @@
 ]
 
 SUPPORTED_PYTHON_VERSIONS = {
-    "3.11": "quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707",
+    "3.11": constants.CUDA_RUNTIME_IMAGE,
 }
 
 

tests/e2e/support.py

@@ -7,6 +7,7 @@
 from codeflare_sdk.common.kubernetes_cluster.kube_api_helpers import (
     _kube_api_error_handling,
 )
+from codeflare_sdk.common.utils import constants
 
 
 def get_ray_cluster(cluster_name, namespace):
@@ -26,8 +27,7 @@ def get_ray_cluster(cluster_name, namespace):
 
 
 def get_ray_image():
-    default_ray_image = "quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707"
-    return os.getenv("RAY_IMAGE", default_ray_image)
+    return os.getenv("RAY_IMAGE", constants.CUDA_RUNTIME_IMAGE)
 
 
 def get_setup_env_variables(**kwargs):

tests/test_cluster_yamls/appwrapper/test-case-bad.yaml

@@ -42,7 +42,7 @@ spec:
                   valueFrom:
                     fieldRef:
                       fieldPath: status.podIP
-                image: quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707
+                image: "${image}"
                 imagePullPolicy: IfNotPresent
                 lifecycle:
                   preStop:
@@ -89,7 +89,7 @@ spec:
                   valueFrom:
                     fieldRef:
                       fieldPath: status.podIP
-                image: quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707
+                image: "${image}"
                 lifecycle:
                   preStop:
                     exec:

tests/test_cluster_yamls/support_clusters/test-aw-a.yaml

@@ -38,7 +38,7 @@ spec:
           template:
             spec:
               containers:
-              - image: quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707
+              - image: "${image}"
                 imagePullPolicy: IfNotPresent
                 lifecycle:
                   preStop:
@@ -109,7 +109,7 @@ spec:
                 key: value
             spec:
               containers:
-              - image: quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707
+              - image: "${image}"
                 lifecycle:
                   preStop:
                     exec:

tests/test_cluster_yamls/support_clusters/test-aw-b.yaml

@@ -38,7 +38,7 @@ spec:
           template:
             spec:
               containers:
-              - image: quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707
+              - image: "${image}"
                 imagePullPolicy: IfNotPresent
                 lifecycle:
                   preStop:
@@ -109,7 +109,7 @@ spec:
                 key: value
             spec:
               containers:
-              - image: quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707
+              - image: "${image}"
                 lifecycle:
                   preStop:
                     exec:

tests/test_cluster_yamls/support_clusters/test-rc-a.yaml

@@ -29,7 +29,7 @@ spec:
     template:
       spec:
         containers:
-        - image: quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707
+        - image: "${image}"
           imagePullPolicy: IfNotPresent
           lifecycle:
             preStop:
@@ -100,7 +100,7 @@ spec:
           key: value
       spec:
         containers:
-        - image: quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707
+        - image: "${image}"
           lifecycle:
             preStop:
               exec:

tests/test_cluster_yamls/support_clusters/test-rc-b.yaml

@@ -29,7 +29,7 @@ spec:
     template:
       spec:
         containers:
-        - image: quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707
+        - image: "${image}"
           imagePullPolicy: IfNotPresent
           lifecycle:
             preStop:
@@ -100,7 +100,7 @@ spec:
           key: value
       spec:
         containers:
-        - image: quay.io/modh/ray@sha256:a5b7c04a14f180d7ca6d06a5697f6bb684e40a26b95a0c872cac23b552741707
+        - image: "${image}"
           lifecycle:
             preStop:
               exec: