RHOAIENG-30720: Remove GCS FT for Lifecycled RayClusters

Author: kryanbeane

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
-name = "codeflare-sdk"
-version = "0.0.0-dev"
+name = "codeflare-sdk-dev"
+version = "0.0.1.dev1"
 description = "Python SDK for codeflare client"
 
 license = "Apache-2.0"
@@ -19,6 +19,8 @@ homepage = "https://github.com/project-codeflare/codeflare-sdk"
 
 keywords = ['codeflare', 'python', 'sdk', 'client', 'batch', 'scale']
 
+packages = [{include = "codeflare_sdk", from = "src"}]
+
 [tool.poetry.dependencies]
 python = "^3.11"
 openshift-client = "1.0.18"
@@ -29,7 +31,8 @@ cryptography = "43.0.3"
 executing = "1.2.0"
 pydantic = "< 2"
 ipywidgets = "8.1.2"
-python-client = { git = "https://github.com/ray-project/kuberay.git", subdirectory = "clients/python-client", rev = "d1e750d9beac612ad455b951c1a789f971409ab3" }
+# Temporarily commented out for test.pypi publishing (git dependencies not allowed)
+# python-client = { git = "https://github.com/ray-project/kuberay.git", subdirectory = "clients/python-client", rev = "4ff766f695cb66c5538072a07723b5e1df3deaa9" }
 
 [[tool.poetry.source]]
 name = "pypi"

src/codeflare_sdk/ray/rayjobs/config.py

@@ -139,20 +139,12 @@ class ManagedClusterConfig:
             A list of V1Volume objects to add to the Cluster
         volume_mounts:
             A list of V1VolumeMount objects to add to the Cluster
-        enable_gcs_ft:
-            A boolean indicating whether to enable GCS fault tolerance.
-        redis_address:
-            The address of the Redis server to use for GCS fault tolerance, required when enable_gcs_ft is True.
-        redis_password_secret:
-            Kubernetes secret reference containing Redis password. ex: {"name": "secret-name", "key": "password-key"}
-        external_storage_namespace:
-            The storage namespace to use for GCS fault tolerance. By default, KubeRay sets it to the UID of RayCluster.
     """
 
     head_cpu_requests: Union[int, str] = 2
     head_cpu_limits: Union[int, str] = 2
-    head_memory_requests: Union[int, str] = 8
-    head_memory_limits: Union[int, str] = 8
+    head_memory_requests: Union[int, str] = 2
+    head_memory_limits: Union[int, str] = 2
     head_accelerators: Dict[str, Union[str, int]] = field(default_factory=dict)
     head_tolerations: Optional[List[V1Toleration]] = None
     worker_cpu_requests: Union[int, str] = 1
@@ -173,35 +165,10 @@ class ManagedClusterConfig:
     annotations: Dict[str, str] = field(default_factory=dict)
     volumes: list[V1Volume] = field(default_factory=list)
     volume_mounts: list[V1VolumeMount] = field(default_factory=list)
-    enable_gcs_ft: bool = False
-    redis_address: Optional[str] = None
-    redis_password_secret: Optional[Dict[str, str]] = None
-    external_storage_namespace: Optional[str] = None
 
     def __post_init__(self):
         self.envs["RAY_USAGE_STATS_ENABLED"] = "0"
 
-        if self.enable_gcs_ft:
-            if not self.redis_address:
-                raise ValueError(
-                    "redis_address must be provided when enable_gcs_ft is True"
-                )
-
-            if self.redis_password_secret and not isinstance(
-                self.redis_password_secret, dict
-            ):
-                raise ValueError(
-                    "redis_password_secret must be a dictionary with 'name' and 'key' fields"
-                )
-
-            if self.redis_password_secret and (
-                "name" not in self.redis_password_secret
-                or "key" not in self.redis_password_secret
-            ):
-                raise ValueError(
-                    "redis_password_secret must contain both 'name' and 'key' fields"
-                )
-
         self._validate_types()
         self._memory_to_string()
         self._validate_gpu_config(self.head_accelerators)
@@ -286,11 +253,6 @@ def build_ray_cluster_spec(self, cluster_name: str) -> Dict[str, Any]:
             "workerGroupSpecs": [self._build_worker_group_spec(cluster_name)],
         }
 
-        # Add GCS fault tolerance if enabled
-        if self.enable_gcs_ft:
-            gcs_ft_options = self._build_gcs_ft_options()
-            ray_cluster_spec["gcsFaultToleranceOptions"] = gcs_ft_options
-
         return ray_cluster_spec
 
     def _build_head_group_spec(self) -> Dict[str, Any]:
@@ -494,28 +456,6 @@ def _build_env_vars(self) -> list:
         """Build environment variables list."""
         return [V1EnvVar(name=key, value=value) for key, value in self.envs.items()]
 
-    def _build_gcs_ft_options(self) -> Dict[str, Any]:
-        """Build GCS fault tolerance options."""
-        gcs_ft_options = {"redisAddress": self.redis_address}
-
-        if (
-            hasattr(self, "external_storage_namespace")
-            and self.external_storage_namespace
-        ):
-            gcs_ft_options["externalStorageNamespace"] = self.external_storage_namespace
-
-        if hasattr(self, "redis_password_secret") and self.redis_password_secret:
-            gcs_ft_options["redisPassword"] = {
-                "valueFrom": {
-                    "secretKeyRef": {
-                        "name": self.redis_password_secret["name"],
-                        "key": self.redis_password_secret["key"],
-                    }
-                }
-            }
-
-        return gcs_ft_options
-
     def add_script_volumes(
         self, configmap_name: str, mount_path: str = "/home/ray/scripts"
     ):

src/codeflare_sdk/ray/rayjobs/rayjob.py

@@ -541,8 +541,6 @@ def _update_existing_cluster_for_scripts(
             config_builder: ManagedClusterConfig instance for building specs
         """
 
-        # Get existing RayCluster
-        api_instance = client.CustomObjectsApi(get_api_client())
         try:
             ray_cluster = self._cluster_api.get_ray_cluster(
                 name=self.cluster_name,

src/codeflare_sdk/ray/rayjobs/test_rayjob.py

@@ -1007,27 +1007,6 @@ def test_rayjob_user_override_shutdown_behavior(mocker):
     assert rayjob_override_priority.shutdown_after_job_finishes is True
 
 
-def test_build_ray_cluster_spec_with_gcs_ft(mocker):
-    """Test build_ray_cluster_spec with GCS fault tolerance enabled."""
-    from codeflare_sdk.ray.rayjobs.config import ManagedClusterConfig
-
-    # Create a test cluster config with GCS FT enabled
-    cluster_config = ManagedClusterConfig(
-        enable_gcs_ft=True,
-        redis_address="redis://redis-service:6379",
-        external_storage_namespace="storage-ns",
-    )
-
-    # Build the spec using the method on the cluster config
-    spec = cluster_config.build_ray_cluster_spec("test-cluster")
-
-    # Verify GCS fault tolerance options
-    assert "gcsFaultToleranceOptions" in spec
-    gcs_ft = spec["gcsFaultToleranceOptions"]
-    assert gcs_ft["redisAddress"] == "redis://redis-service:6379"
-    assert gcs_ft["externalStorageNamespace"] == "storage-ns"
-
-
 class TestRayVersionValidation:
     """Test Ray version validation in RayJob."""