feat: add GlobalPlanner component for centralized scaling (#5702)

Signed-off-by: Daiyaan <darfeen@nvidia.com>
Signed-off-by: daiyaanarfeen <darfeen@nvidia.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>

.github/filters.yaml

@@ -109,7 +109,9 @@ deploy:
 
 planner:
   - 'components/src/dynamo/planner/**'
+  - 'components/src/dynamo/global_planner/**'
   - 'tests/planner/**'
+  - 'tests/global_planner/**'
   - 'components/src/dynamo/profiler/**'
   - 'components/src/dynamo/global_router/**'
 

CODEOWNERS

@@ -27,6 +27,7 @@ CODEOWNERS @ai-dynamo/Devops
 # Planner
 /components/src/dynamo/planner/ @ai-dynamo/python-codeowners @ai-dynamo/Devops
 /components/src/dynamo/global_router/ @ai-dynamo/python-codeowners @ai-dynamo/Devops
+/components/src/dynamo/global_planner/ @ai-dynamo/python-codeowners @ai-dynamo/Devops
 /examples/hierarchical_planner/ @ai-dynamo/python-codeowners @ai-dynamo/Devops
 /components/src/dynamo/profiler/ @ai-dynamo/python-codeowners @ai-dynamo/Devops
 /tests/planner/ @ai-dynamo/python-codeowners @ai-dynamo/Devops

components/src/dynamo/global_planner/__init__.py

+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""
+GlobalPlanner - Centralized scaling execution service.
+
+The GlobalPlanner is a standalone component that receives scale requests from
+Planners and executes them via the Kubernetes API. It provides centralized
+scaling management across multiple deployments and namespaces.
+
+Architecture:
+- Planners make scaling decisions (observe, predict, decide)
+- Planners in delegating mode send requests to GlobalPlanner
+- GlobalPlanner executes scaling via Kubernetes API
+- GlobalPlanner is stateless and can scale horizontally
+
+Usage:
+    DYN_NAMESPACE=global-infra python -m dynamo.global_planner \
+        --managed-namespaces app-ns-1 app-ns-2
+"""
+
+__all__ = [
+    "ScaleRequestHandler",
+]
+
+from dynamo.global_planner.scale_handler import ScaleRequestHandler

components/src/dynamo/global_planner/__main__.py

+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""
+GlobalPlanner - Centralized Scaling Execution Service
+
+Entry point for the GlobalPlanner component.
+
+Usage:
+    DYN_NAMESPACE=global-infra python -m dynamo.global_planner
+
+With authorization:
+    DYN_NAMESPACE=global-infra python -m dynamo.global_planner \\
+        --managed-namespaces app-ns-1 app-ns-2
+"""
+
+import asyncio
+import logging
+import os
+
+from pydantic import BaseModel
+
+from dynamo.global_planner.argparse_config import create_global_planner_parser
+from dynamo.global_planner.scale_handler import ScaleRequestHandler
+from dynamo.runtime import DistributedRuntime, dynamo_worker
+from dynamo.runtime.logging import configure_dynamo_logging
+
+configure_dynamo_logging()
+logger = logging.getLogger(__name__)
+
+
+class HealthCheckRequest(BaseModel):
+    """Request type for health check endpoint"""
+
+    text: str = "ping"
+
+
+@dynamo_worker()
+async def main(runtime: DistributedRuntime, args):
+    """Initialize and run GlobalPlanner.
+
+    The GlobalPlanner is a centralized scaling service that:
+    1. Listens for scale requests from Planners
+    2. Validates caller authorization (optional)
+    3. Executes scaling via Kubernetes API
+    4. Returns success/failure status
+
+    Args:
+        runtime: Dynamo runtime instance
+        args: Parsed command-line arguments
+    """
+    # Get Dynamo namespace from environment variable
+    namespace = os.environ.get("DYN_NAMESPACE")
+    if not namespace:
+        raise ValueError(
+            "DYN_NAMESPACE environment variable is required but not set. "
+            "Please set DYN_NAMESPACE to specify the Dynamo namespace for GlobalPlanner."
+        )
+
+    logger.info("=" * 60)
+    logger.info("Starting GlobalPlanner")
+    logger.info("=" * 60)
+    logger.info(f"Namespace: {namespace}")
+    logger.info(f"Environment: {args.environment}")
+
+    if args.managed_namespaces:
+        logger.info("Authorization: ENABLED")
+        logger.info(f"Authorized namespaces: {args.managed_namespaces}")
+    else:
+        logger.info("Authorization: DISABLED (accepting all namespaces)")
+
+    logger.info("=" * 60)
+
+    # Create the GlobalPlanner component
+    component = runtime.namespace(namespace).component("GlobalPlanner")
+
+    # Get K8s namespace (where GlobalPlanner pod is running)
+    k8s_namespace = os.environ.get("POD_NAMESPACE", "default")
+    logger.info(f"Running in Kubernetes namespace: {k8s_namespace}")
+
+    # Create scale request handler
+    handler = ScaleRequestHandler(
+        runtime=runtime,
+        managed_namespaces=args.managed_namespaces,
+        k8s_namespace=k8s_namespace,
+    )
+
+    # Serve scale_request endpoint
+    logger.info("Serving endpoints...")
+    scale_endpoint = component.endpoint("scale_request")
+    await scale_endpoint.serve_endpoint(handler.scale_request)
+    logger.info("  ✓ scale_request - Receives scaling requests from Planners")
+
+    # Serve health check endpoint
+    async def health_check(request: HealthCheckRequest):
+        """Health check endpoint for monitoring"""
+        yield {
+            "status": "healthy",
+            "component": "GlobalPlanner",
+            "namespace": namespace,
+            "managed_namespaces": args.managed_namespaces or "all",
+        }
+
+    health_endpoint = component.endpoint("health")
+    await health_endpoint.serve_endpoint(health_check)
+    logger.info("  ✓ health - Health check endpoint")
+
+    logger.info("=" * 60)
+    logger.info("GlobalPlanner is ready and waiting for scale requests")
+    logger.info("=" * 60)
+
+    # Keep running forever (process scale requests as they come)
+    await asyncio.Event().wait()
+
+
+if __name__ == "__main__":
+    parser = create_global_planner_parser()
+    args = parser.parse_args()
+    asyncio.run(main(args))

components/src/dynamo/global_planner/argparse_config.py

+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Argument parsing for GlobalPlanner."""
+
+import argparse
+
+
+def create_global_planner_parser() -> argparse.ArgumentParser:
+    """Create and configure the argument parser for GlobalPlanner.
+
+    Returns:
+        argparse.ArgumentParser: Configured argument parser for GlobalPlanner
+    """
+    parser = argparse.ArgumentParser(
+        description="GlobalPlanner - Centralized Scaling Execution Service",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+        epilog="""
+Examples:
+  # Simple deployment (accept all namespaces)
+  DYN_NAMESPACE=global-infra python -m dynamo.global_planner
+
+  # With authorization
+  DYN_NAMESPACE=global-infra python -m dynamo.global_planner \\
+    --managed-namespaces app-ns-1 app-ns-2 app-ns-3
+
+  # Custom environment
+  DYN_NAMESPACE=global-infra python -m dynamo.global_planner \\
+    --environment=kubernetes
+        """,
+    )
+
+    parser.add_argument(
+        "--managed-namespaces",
+        type=str,
+        nargs="+",
+        default=None,
+        help="Optional: List of namespaces authorized to use this GlobalPlanner (default: accept all)",
+    )
+
+    parser.add_argument(
+        "--environment",
+        default="kubernetes",
+        choices=["kubernetes"],
+        help="Environment type (currently only kubernetes supported)",
+    )
+
+    return parser

components/src/dynamo/global_planner/scale_handler.py

+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Handler for scale_request endpoint in GlobalPlanner."""
+
+import logging
+
+from dynamo.planner import KubernetesConnector
+from dynamo.planner.scale_protocol import ScaleRequest, ScaleResponse, ScaleStatus
+from dynamo.runtime import DistributedRuntime, dynamo_endpoint
+
+logger = logging.getLogger(__name__)
+
+# Model name used for KubernetesConnector in remote execution mode
+MANAGED_MODEL_NAME = "managed"
+
+
+class ScaleRequestHandler:
+    """Handles incoming scale requests in GlobalPlanner.
+
+    This handler:
+    1. Receives scale requests from Planners
+    2. Validates caller authorization (optional)
+    3. Caches KubernetesConnector per DGD for efficiency
+    4. Executes scaling via Kubernetes API
+    5. Returns current replica counts
+    """
+
+    def __init__(
+        self, runtime: DistributedRuntime, managed_namespaces: list, k8s_namespace: str
+    ):
+        """Initialize the scale request handler.
+
+        Args:
+            runtime: Dynamo runtime instance
+            managed_namespaces: List of authorized namespaces (None = accept all)
+            k8s_namespace: Kubernetes namespace where GlobalPlanner is running
+        """
+        self.runtime = runtime
+        # If managed_namespaces is None, accept all namespaces
+        self.managed_namespaces = (
+            set(managed_namespaces) if managed_namespaces else None
+        )
+        self.k8s_namespace = k8s_namespace
+        self.connectors = {}  # Cache of KubernetesConnector per DGD
+
+        if self.managed_namespaces:
+            logger.info(
+                f"ScaleRequestHandler initialized for namespaces: {managed_namespaces}"
+            )
+        else:
+            logger.info("ScaleRequestHandler initialized (accepting all namespaces)")
+
+    @dynamo_endpoint(ScaleRequest, ScaleResponse)
+    async def scale_request(self, request: ScaleRequest):
+        """Process scaling request from a Planner.
+
+        Args:
+            request: ScaleRequest with target replicas and DGD info
+
+        Yields:
+            ScaleResponse with status and current replica counts
+        """
+        try:
+            # Validate caller namespace (if authorization is enabled)
+            if (
+                self.managed_namespaces is not None
+                and request.caller_namespace not in self.managed_namespaces
+            ):
+                yield {
+                    "status": ScaleStatus.ERROR.value,
+                    "message": f"Namespace {request.caller_namespace} not authorized",
+                    "current_replicas": {},
+                }
+                return
+
+            logger.info(
+                f"Processing scale request from {request.caller_namespace} "
+                f"for DGD {request.graph_deployment_name} "
+                f"in K8s namespace {request.k8s_namespace}"
+            )
+
+            # Get or create connector for this DGD
+            connector_key = f"{request.k8s_namespace}/{request.graph_deployment_name}"
+            if connector_key not in self.connectors:
+                connector = KubernetesConnector(
+                    dynamo_namespace=request.caller_namespace,
+                    model_name=MANAGED_MODEL_NAME,  # Not used for remote execution
+                    k8s_namespace=request.k8s_namespace,
+                    parent_dgd_name=request.graph_deployment_name,
+                )
+                await connector._async_init()
+                self.connectors[connector_key] = connector
+                logger.debug(f"Created new connector for {connector_key}")
+            else:
+                connector = self.connectors[connector_key]
+                logger.debug(f"Reusing cached connector for {connector_key}")
+
+            # Execute scaling (request.target_replicas is already List[TargetReplica])
+            await connector.set_component_replicas(
+                request.target_replicas, blocking=request.blocking
+            )
+
+            # Get current replica counts
+            current_replicas = {}
+            deployment = connector.kube_api.get_graph_deployment(
+                connector.parent_dgd_name
+            )
+            for service_name, service_spec in deployment["spec"]["services"].items():
+                sub_type = service_spec.get("subComponentType", "")
+                if sub_type:
+                    current_replicas[sub_type] = service_spec.get("replicas", 0)
+
+            logger.info(
+                f"Successfully scaled {request.graph_deployment_name}: {current_replicas}"
+            )
+            yield {
+                "status": ScaleStatus.SUCCESS.value,
+                "message": f"Scaled {request.graph_deployment_name} successfully",
+                "current_replicas": current_replicas,
+            }
+
+        except Exception as e:
+            logger.exception(f"Error processing scale request: {e}")
+            yield {
+                "status": ScaleStatus.ERROR.value,
+                "message": str(e),
+                "current_replicas": {},
+            }

components/src/dynamo/planner/__init__.py

@@ -5,12 +5,14 @@ __all__ = [
     "PlannerConnector",
     "KubernetesConnector",
     "VirtualConnector",
+    "GlobalPlannerConnector",
     "SLAPlannerDefaults",
     "TargetReplica",
     "SubComponentType",
 ]
 # Import the classes
 from dynamo.planner.defaults import SLAPlannerDefaults, SubComponentType
+from dynamo.planner.global_planner_connector import GlobalPlannerConnector
 from dynamo.planner.kubernetes_connector import KubernetesConnector, TargetReplica
 from dynamo.planner.planner_connector import PlannerConnector
 from dynamo.planner.virtual_connector import VirtualConnector

components/src/dynamo/planner/global_planner_connector.py

+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Connector for delegating scaling decisions to a centralized GlobalPlanner."""
+
+import logging
+import os
+import time
+from typing import Optional
+
+from dynamo.planner.defaults import SubComponentType
+from dynamo.planner.kubernetes_connector import TargetReplica
+from dynamo.planner.planner_connector import PlannerConnector
+from dynamo.planner.remote_planner_client import RemotePlannerClient
+from dynamo.planner.scale_protocol import ScaleRequest, ScaleStatus
+from dynamo.planner.utils.exceptions import EmptyTargetReplicasError
+from dynamo.runtime import DistributedRuntime
+from dynamo.runtime.logging import configure_dynamo_logging
+
+configure_dynamo_logging()
+logger = logging.getLogger(__name__)
+
+
+class GlobalPlannerConnector(PlannerConnector):
+    """
+    Connector that delegates scaling decisions to a centralized GlobalPlanner.
+
+    This connector wraps RemotePlannerClient and implements the PlannerConnector
+    interface, allowing planner_core.py to treat global-planner environment mode
+    consistently with kubernetes and virtual modes.
+    """
+
+    def __init__(
+        self,
+        runtime: DistributedRuntime,
+        dynamo_namespace: str,
+        global_planner_namespace: str,
+        global_planner_component: str = "GlobalPlanner",
+        model_name: Optional[str] = None,
+    ):
+        """
+        Initialize GlobalPlannerConnector.
+
+        Args:
+            runtime: Distributed runtime for communication
+            dynamo_namespace: Local dynamo namespace (caller identification)
+            global_planner_namespace: Namespace where GlobalPlanner is deployed
+            global_planner_component: Component name of GlobalPlanner (default: "GlobalPlanner")
+            model_name: Optional model name (will be managed remotely if not provided)
+        """
+        self.runtime = runtime
+        self.dynamo_namespace = dynamo_namespace
+        self.global_planner_namespace = global_planner_namespace
+        self.global_planner_component = global_planner_component
+        self.model_name = model_name
+        self.remote_client: Optional[RemotePlannerClient] = None
+
+        # Cache for predicted load (will be set by planner before scaling)
+        self.last_predicted_load: Optional[dict] = None
+
+    async def _async_init(self):
+        """Async initialization - creates RemotePlannerClient"""
+        self.remote_client = RemotePlannerClient(
+            self.runtime,
+            self.global_planner_namespace,
+            self.global_planner_component,
+        )
+        logger.info(
+            f"GlobalPlannerConnector initialized: will delegate to {self.global_planner_namespace}.{self.global_planner_component}"
+        )
+
+    def set_predicted_load(
+        self, num_requests: Optional[float], isl: Optional[float], osl: Optional[float]
+    ):
+        """
+        Set predicted load for inclusion in next scale request.
+
+        This is called by planner_core.py before calling set_component_replicas.
+        """
+        self.last_predicted_load = {
+            "num_requests": num_requests,
+            "isl": isl,
+            "osl": osl,
+        }
+
+    async def set_component_replicas(
+        self, target_replicas: list[TargetReplica], blocking: bool = True
+    ):
+        """
+        Set component replicas by delegating to GlobalPlanner.
+
+        Sends a ScaleRequest to the GlobalPlanner with the target replica configurations.
+
+        Args:
+            target_replicas: List of target replica configurations
+            blocking: Whether to wait for scaling completion (passed to GlobalPlanner)
+
+        Raises:
+            EmptyTargetReplicasError: If target_replicas is empty
+            RuntimeError: If remote_client is not initialized or response indicates error
+        """
+        if not target_replicas:
+            raise EmptyTargetReplicasError()
+
+        if self.remote_client is None:
+            raise RuntimeError(
+                "GlobalPlannerConnector not initialized. Call _async_init() first."
+            )
+
+        # Get DGD info from environment variables
+        graph_deployment_name = os.environ.get("DYN_PARENT_DGD_K8S_NAME")
+        if not graph_deployment_name:
+            raise ValueError(
+                "DYN_PARENT_DGD_K8S_NAME environment variable is required but not set. "
+                "Please set DYN_PARENT_DGD_K8S_NAME to specify the parent DGD name."
+            )
+        k8s_namespace = os.environ.get("POD_NAMESPACE")
+        if not k8s_namespace:
+            raise ValueError(
+                "POD_NAMESPACE environment variable is required but not set. "
+                "Please set POD_NAMESPACE to specify the Kubernetes namespace."
+            )
+
+        # Create scale request
+        request = ScaleRequest(
+            caller_namespace=self.dynamo_namespace,
+            graph_deployment_name=graph_deployment_name,
+            k8s_namespace=k8s_namespace,
+            target_replicas=target_replicas,
+            blocking=blocking,
+            timestamp=time.time(),
+            predicted_load=self.last_predicted_load,
+        )
+
+        logger.info(
+            f"Delegating scale request to GlobalPlanner: "
+            f"DGD={graph_deployment_name}, "
+            f"prefill={[r.desired_replicas for r in target_replicas if r.sub_component_type == SubComponentType.PREFILL]}, "
+            f"decode={[r.desired_replicas for r in target_replicas if r.sub_component_type == SubComponentType.DECODE]}"
+        )
+
+        # Send request to GlobalPlanner
+        response = await self.remote_client.send_scale_request(request)
+
+        # Check response status
+        if response.status == ScaleStatus.SUCCESS:
+            logger.info(f"GlobalPlanner scaling successful: {response.message}")
+        elif response.status == ScaleStatus.ERROR:
+            logger.error(f"GlobalPlanner scaling failed: {response.message}")
+            raise RuntimeError(f"GlobalPlanner scaling failed: {response.message}")
+        else:
+            logger.warning(
+                f"GlobalPlanner returned status '{response.status.value}': {response.message}"
+            )
+
+    async def add_component(
+        self, sub_component_type: SubComponentType, blocking: bool = True
+    ):
+        """
+        Add a component (not supported for GlobalPlanner).
+
+        GlobalPlanner only supports batch operations via set_component_replicas.
+        """
+        raise NotImplementedError(
+            "GlobalPlannerConnector only supports batch operations via set_component_replicas(). "
+            "Use set_component_replicas() to scale components."
+        )
+
+    async def remove_component(
+        self, sub_component_type: SubComponentType, blocking: bool = True
+    ):
+        """
+        Remove a component (not supported for GlobalPlanner).
+
+        GlobalPlanner only supports batch operations via set_component_replicas.
+        """
+        raise NotImplementedError(
+            "GlobalPlannerConnector only supports batch operations via set_component_replicas(). "
+            "Use set_component_replicas() to scale components."
+        )
+
+    async def validate_deployment(
+        self,
+        prefill_component_name: Optional[str] = None,
+        decode_component_name: Optional[str] = None,
+    ):
+        """
+        Validate deployment (no-op for GlobalPlanner).
+
+        The GlobalPlanner validates the deployment on its side, so local
+        validation is not needed in delegating mode.
+        """
+        logger.info(
+            "GlobalPlannerConnector: Skipping local deployment validation "
+            "(GlobalPlanner will validate on its side)"
+        )
+
+    async def wait_for_deployment_ready(self):
+        """
+        Wait for deployment to be ready (no-op for GlobalPlanner).
+
+        The GlobalPlanner manages deployment state, so we don't need to
+        wait locally in delegating mode.
+        """
+        logger.info(
+            "GlobalPlannerConnector: Skipping deployment ready check "
+            "(GlobalPlanner manages deployment state)"
+        )
+
+    def get_model_name(self) -> str:
+        """
+        Get model name.
+
+        Returns the model name if provided during initialization, otherwise
+        returns a placeholder indicating the model is managed remotely.
+        """
+        if self.model_name:
+            return self.model_name
+        return "managed-remotely"

components/src/dynamo/planner/kubernetes_connector.py

@@ -51,6 +51,7 @@ class KubernetesConnector(PlannerConnector):
         dynamo_namespace: str,
         model_name: Optional[str] = None,
         k8s_namespace: Optional[str] = None,
+        parent_dgd_name: Optional[str] = None,
     ):
         self.kube_api = KubernetesAPI(k8s_namespace)
 
@@ -60,13 +61,19 @@ class KubernetesConnector(PlannerConnector):
                 model_name.lower()
             )  # normalize model name to lowercase (MDC)
 
+        # Allow overriding parent DGD name for centralized planner
+        if parent_dgd_name:
+            self.parent_dgd_name = parent_dgd_name
+        else:
             graph_deployment_name = os.getenv("DYN_PARENT_DGD_K8S_NAME")
             if not graph_deployment_name:
                 raise DeploymentValidationError(
                     ["DYN_PARENT_DGD_K8S_NAME environment variable is not set"]
                 )
+            self.parent_dgd_name = graph_deployment_name
 
-        self.graph_deployment_name = graph_deployment_name
+        # For backwards compatibility
+        self.graph_deployment_name = self.parent_dgd_name
 
     async def add_component(
         self, sub_component_type: SubComponentType, blocking: bool = True
@@ -411,7 +418,9 @@ if __name__ == "__main__":
     )
     parser.add_argument("--blocking", action="store_true")
     args = parser.parse_args()
-    connector = KubernetesConnector(args.dynamo_namespace, args.k8s_namespace)
+    connector = KubernetesConnector(
+        args.dynamo_namespace, k8s_namespace=args.k8s_namespace
+    )
 
     if args.action == "add":
         task = connector.add_component(SubComponentType(args.component), args.blocking)

components/src/dynamo/planner/planner_sla.py

@@ -77,4 +77,5 @@ async def init_planner(runtime: DistributedRuntime, args):
 if __name__ == "__main__":
     parser = create_sla_planner_parser()
     args = parser.parse_args()
+    validate_sla_planner_args(args)
     asyncio.run(init_planner(args))

components/src/dynamo/planner/remote_planner_client.py

+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Client for calling remote planner's scale_request endpoint."""
+
+import asyncio
+import logging
+
+from dynamo.planner.defaults import SubComponentType
+from dynamo.planner.scale_protocol import ScaleRequest, ScaleResponse
+from dynamo.runtime import DistributedRuntime
+
+logger = logging.getLogger(__name__)
+
+
+class RemotePlannerClient:
+    """Client for delegating scaling requests to centralized planner"""
+
+    def __init__(
+        self,
+        runtime: DistributedRuntime,
+        central_namespace: str,
+        central_component: str,
+        connection_timeout: float = 30.0,
+        max_retries: int = 3,
+    ):
+        self.runtime = runtime
+        self.central_namespace = central_namespace
+        self.central_component = central_component
+        self.connection_timeout = connection_timeout
+        self.max_retries = max_retries
+        self._client = None
+
+    async def _ensure_client(self):
+        """Lazy initialization of endpoint client with retry mechanism"""
+        if self._client is None:
+            endpoint = (
+                self.runtime.namespace(self.central_namespace)
+                .component(self.central_component)
+                .endpoint("scale_request")
+            )
+
+            # Retry logic with exponential backoff
+            last_error = None
+            for attempt in range(self.max_retries):
+                try:
+                    logger.info(
+                        f"Attempting to connect to GlobalPlanner at "
+                        f"{self.central_namespace}.{self.central_component} "
+                        f"(attempt {attempt + 1}/{self.max_retries})"
+                    )
+
+                    self._client = await endpoint.client()
+
+                    # Wait for instances with timeout
+                    await asyncio.wait_for(
+                        self._client.wait_for_instances(),
+                        timeout=self.connection_timeout,
+                    )
+
+                    logger.info(
+                        f"Successfully connected to centralized planner at "
+                        f"{self.central_namespace}.{self.central_component}"
+                    )
+                    return
+
+                except asyncio.TimeoutError as e:
+                    last_error = e
+                    logger.warning(
+                        f"Connection attempt {attempt + 1} timed out after "
+                        f"{self.connection_timeout}s"
+                    )
+                    self._client = None
+
+                except Exception as e:
+                    last_error = e
+                    logger.warning(f"Connection attempt {attempt + 1} failed: {e}")
+                    self._client = None
+
+                # Exponential backoff before retry (except on last attempt)
+                if attempt < self.max_retries - 1:
+                    backoff = 2**attempt  # 1s, 2s, 4s, ...
+                    logger.info(f"Retrying in {backoff}s...")
+                    await asyncio.sleep(backoff)
+
+            # All retries exhausted
+            raise RuntimeError(
+                f"Failed to connect to GlobalPlanner at "
+                f"{self.central_namespace}.{self.central_component} after "
+                f"{self.max_retries} attempts. Last error: {last_error}"
+            )
+
+    async def send_scale_request(self, request: ScaleRequest) -> ScaleResponse:
+        """Send scale request to centralized planner"""
+        await self._ensure_client()
+
+        logger.info(
+            f"Sending scale request to centralized planner: "
+            f"prefill={[r.desired_replicas for r in request.target_replicas if r.sub_component_type == SubComponentType.PREFILL]}, "
+            f"decode={[r.desired_replicas for r in request.target_replicas if r.sub_component_type == SubComponentType.DECODE]}"
+        )
+
+        # Send request to single endpoint
+        request_json = request.model_dump_json()
+
+        response_data = await self._client.scale_request(request_json)
+
+        if response_data is None:
+            raise RuntimeError("No response from centralized planner")
+
+        # Parse response
+        response = ScaleResponse(**response_data)
+        logger.info(f"Scale request response: {response.status} - {response.message}")
+
+        return response

components/src/dynamo/planner/scale_protocol.py

+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Data structures for scale request/response protocol between delegating and centralized planners."""
+
+from enum import Enum
+from typing import List, Optional
+
+from pydantic import BaseModel
+
+from dynamo.planner.kubernetes_connector import TargetReplica
+
+
+class ScaleStatus(str, Enum):
+    """Status values for scaling operations"""
+
+    SUCCESS = "success"
+    ERROR = "error"
+    SCALING = "scaling"
+
+
+class ScaleRequest(BaseModel):
+    """Request to scale a deployment"""
+
+    # Caller identification
+    caller_namespace: str
+
+    # Target deployment
+    graph_deployment_name: str  # K8s DynamoGraphDeployment name
+    k8s_namespace: str  # K8s namespace
+
+    # Scaling targets
+    target_replicas: List[TargetReplica]
+
+    # Execution options
+    blocking: bool = False
+
+    # Optional context (for debugging/logging)
+    timestamp: Optional[float] = None
+    predicted_load: Optional[dict] = None
+
+
+class ScaleResponse(BaseModel):
+    """Response from scaling operation"""
+
+    status: ScaleStatus
+    message: str
+    current_replicas: dict  # {"prefill": 3, "decode": 5}

components/src/dynamo/planner/utils/planner_argparse.py

@@ -28,8 +28,8 @@ def create_sla_planner_parser() -> argparse.ArgumentParser:
     parser.add_argument(
         "--environment",
         default=SLAPlannerDefaults.environment,
-        choices=["kubernetes", "virtual"],
-        help="Environment type",
+        choices=["kubernetes", "virtual", "global-planner"],
+        help="Environment type: kubernetes (direct K8s scaling), virtual (dynamo runtime scaling), global-planner (delegate to GlobalPlanner)",
     )
     parser.add_argument(
         "--namespace",
@@ -177,6 +177,14 @@ def create_sla_planner_parser() -> argparse.ArgumentParser:
         help="Model name of deployment (only required for virtual environment)",
     )
 
+    # For global-planner environment mode
+    parser.add_argument(
+        "--global-planner-namespace",
+        type=str,
+        default=None,
+        help="Namespace of GlobalPlanner component (required when environment=global-planner)",
+    )
+
     # Scaling mode flags
     parser.add_argument(
         "--enable-throughput-scaling",
@@ -238,6 +246,16 @@ def create_sla_planner_parser() -> argparse.ArgumentParser:
     return parser
 
 
+def validate_planner_args(args):
+    """Validate planner configuration"""
+    if args.environment == "global-planner":
+        if not args.global_planner_namespace:
+            raise ValueError(
+                "--global-planner-namespace required when environment=global-planner. "
+                "Please specify the namespace where GlobalPlanner is running."
+            )
+
+
 def validate_sla_planner_args(args: argparse.Namespace) -> None:
     """Validate and normalize SLA planner arguments.
 

components/src/dynamo/planner/utils/planner_core.py

@@ -18,6 +18,7 @@ from dynamo.planner import (
     VirtualConnector,
 )
 from dynamo.planner.defaults import WORKER_COMPONENT_NAMES
+from dynamo.planner.global_planner_connector import GlobalPlannerConnector
 from dynamo.planner.utils.exceptions import DeploymentValidationError
 from dynamo.planner.utils.load_predictor import LOAD_PREDICTORS
 from dynamo.planner.utils.perf_interpolation import (
@@ -267,8 +268,16 @@ class BasePlanner:
             self.namespace = args.namespace
 
             if not args.no_operation:
-                if connector is not None:
-                    self.connector = connector
+                # Initialize connector based on environment
+                if args.environment == "global-planner":
+                    # Use GlobalPlannerConnector to delegate to GlobalPlanner
+                    self.connector = GlobalPlannerConnector(
+                        runtime,
+                        self.namespace,
+                        args.global_planner_namespace,
+                        "GlobalPlanner",
+                        getattr(args, "model_name", None),
+                    )
                 elif args.environment == "kubernetes":
                     self.connector = KubernetesConnector(
                         self.namespace, self.model_name

tests/global_planner/unit/test_scale_request_handler.py

+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Unit tests for ScaleRequestHandler."""
+
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from dynamo.global_planner.scale_handler import ScaleRequestHandler
+from dynamo.planner import SubComponentType, TargetReplica
+from dynamo.planner.scale_protocol import ScaleRequest
+
+pytestmark = [
+    pytest.mark.gpu_0,
+    pytest.mark.pre_merge,
+    pytest.mark.unit,
+    pytest.mark.planner,
+    pytest.mark.filterwarnings("ignore::pydantic.warnings.PydanticDeprecatedSince20"),
+]
+
+
+@pytest.fixture
+def mock_runtime():
+    """Create a mock DistributedRuntime."""
+    return MagicMock()
+
+
+@pytest.mark.asyncio
+async def test_handler_authorization_success(mock_runtime):
+    """Test handler authorizes requests from managed namespaces."""
+    handler = ScaleRequestHandler(
+        runtime=mock_runtime, managed_namespaces=["app-ns"], k8s_namespace="default"
+    )
+
+    request = ScaleRequest(
+        caller_namespace="app-ns",
+        graph_deployment_name="my-dgd",
+        k8s_namespace="default",
+        target_replicas=[
+            TargetReplica(
+                sub_component_type=SubComponentType.PREFILL, desired_replicas=3
+            )
+        ],
+    )
+
+    # Mock KubernetesConnector
+    with patch(
+        "dynamo.global_planner.scale_handler.KubernetesConnector"
+    ) as mock_connector_cls:
+        mock_connector = AsyncMock()
+        mock_connector_cls.return_value = mock_connector
+        mock_connector._async_init = AsyncMock()
+        mock_connector.set_component_replicas = AsyncMock()
+        mock_connector.kube_api = MagicMock()
+        mock_connector.kube_api.get_graph_deployment = MagicMock(
+            return_value={
+                "spec": {
+                    "services": {
+                        "prefill-svc": {"subComponentType": "prefill", "replicas": 3},
+                        "decode-svc": {"subComponentType": "decode", "replicas": 5},
+                    }
+                }
+            }
+        )
+
+        # Process request (pass as dict to match endpoint behavior)
+        results = []
+        async for response in handler.scale_request(request.model_dump()):
+            results.append(response)
+
+        assert len(results) == 1
+        response = results[0]
+        assert response["status"] == "success"
+        assert "Scaled" in response["message"]
+        assert response["current_replicas"]["prefill"] == 3
+        assert response["current_replicas"]["decode"] == 5
+
+
+@pytest.mark.asyncio
+async def test_handler_authorization_failure(mock_runtime):
+    """Test handler rejects requests from unauthorized namespaces."""
+    handler = ScaleRequestHandler(
+        runtime=mock_runtime,
+        managed_namespaces=["authorized-ns"],
+        k8s_namespace="default",
+    )
+
+    request = ScaleRequest(
+        caller_namespace="unauthorized-ns",
+        graph_deployment_name="my-dgd",
+        k8s_namespace="default",
+        target_replicas=[
+            TargetReplica(
+                sub_component_type=SubComponentType.PREFILL, desired_replicas=3
+            )
+        ],
+    )
+
+    # Process request
+    results = []
+    async for response in handler.scale_request(request.model_dump()):
+        results.append(response)
+
+    assert len(results) == 1
+    response = results[0]
+    assert response["status"] == "error"
+    assert "not authorized" in response["message"]
+    assert response["current_replicas"] == {}
+
+
+@pytest.mark.asyncio
+async def test_handler_multiple_dgds(mock_runtime):
+    """Test handler creates separate connectors for different DGDs (and caches them)."""
+    handler = ScaleRequestHandler(
+        runtime=mock_runtime, managed_namespaces=["app-ns"], k8s_namespace="default"
+    )
+
+    request1 = ScaleRequest(
+        caller_namespace="app-ns",
+        graph_deployment_name="dgd-1",
+        k8s_namespace="default",
+        target_replicas=[
+            TargetReplica(
+                sub_component_type=SubComponentType.PREFILL, desired_replicas=2
+            )
+        ],
+    )
+
+    request2 = ScaleRequest(
+        caller_namespace="app-ns",
+        graph_deployment_name="dgd-2",  # Different DGD
+        k8s_namespace="default",
+        target_replicas=[
+            TargetReplica(
+                sub_component_type=SubComponentType.PREFILL, desired_replicas=4
+            )
+        ],
+    )
+
+    with patch(
+        "dynamo.global_planner.scale_handler.KubernetesConnector"
+    ) as mock_connector_cls:
+        mock_connector = AsyncMock()
+        mock_connector_cls.return_value = mock_connector
+        mock_connector._async_init = AsyncMock()
+        mock_connector.set_component_replicas = AsyncMock()
+        mock_connector.kube_api = MagicMock()
+        mock_connector.kube_api.get_graph_deployment = MagicMock(
+            return_value={"spec": {"services": {}}}
+        )
+
+        # Process both requests
+        async for _ in handler.scale_request(request1.model_dump()):
+            pass
+        async for _ in handler.scale_request(request2.model_dump()):
+            pass
+
+        # Verify two connectors were created
+        assert "default/dgd-1" in handler.connectors
+        assert "default/dgd-2" in handler.connectors
+        assert mock_connector_cls.call_count == 2
+
+
+@pytest.mark.asyncio
+async def test_handler_error_handling(mock_runtime):
+    """Test handler error handling during scaling."""
+    handler = ScaleRequestHandler(
+        runtime=mock_runtime, managed_namespaces=["app-ns"], k8s_namespace="default"
+    )
+
+    request = ScaleRequest(
+        caller_namespace="app-ns",
+        graph_deployment_name="my-dgd",
+        k8s_namespace="default",
+        target_replicas=[
+            TargetReplica(
+                sub_component_type=SubComponentType.PREFILL, desired_replicas=3
+            )
+        ],
+    )
+
+    with patch(
+        "dynamo.global_planner.scale_handler.KubernetesConnector"
+    ) as mock_connector_cls:
+        mock_connector = AsyncMock()
+        mock_connector_cls.return_value = mock_connector
+        mock_connector._async_init = AsyncMock()
+        # Simulate error during scaling
+        mock_connector.set_component_replicas = AsyncMock(
+            side_effect=Exception("Scaling failed")
+        )
+
+        # Process request (pass as dict to match endpoint behavior)
+        results = []
+        async for response in handler.scale_request(request.model_dump()):
+            results.append(response)
+
+        assert len(results) == 1
+        response = results[0]
+        assert response["status"] == "error"
+        assert "Scaling failed" in response["message"]
+
+
+@pytest.mark.asyncio
+async def test_handler_blocking_mode(mock_runtime):
+    """Test handler respects blocking mode."""
+    handler = ScaleRequestHandler(
+        runtime=mock_runtime, managed_namespaces=["app-ns"], k8s_namespace="default"
+    )
+
+    request = ScaleRequest(
+        caller_namespace="app-ns",
+        graph_deployment_name="my-dgd",
+        k8s_namespace="default",
+        target_replicas=[
+            TargetReplica(
+                sub_component_type=SubComponentType.PREFILL, desired_replicas=3
+            )
+        ],
+        blocking=True,  # Request blocking mode
+    )
+
+    with patch(
+        "dynamo.global_planner.scale_handler.KubernetesConnector"
+    ) as mock_connector_cls:
+        mock_connector = AsyncMock()
+        mock_connector_cls.return_value = mock_connector
+        mock_connector._async_init = AsyncMock()
+        mock_connector.set_component_replicas = AsyncMock()
+        mock_connector.kube_api = MagicMock()
+        mock_connector.kube_api.get_graph_deployment = MagicMock(
+            return_value={"spec": {"services": {}}}
+        )
+
+        # Process request (pass as dict to match endpoint behavior)
+        async for _ in handler.scale_request(request.model_dump()):
+            pass
+
+        # Verify blocking=True was passed to connector
+        mock_connector.set_component_replicas.assert_called_once()
+        call_args = mock_connector.set_component_replicas.call_args
+        assert call_args[1]["blocking"] is True

tests/planner/unit/test_planner_argparse.py

+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Unit tests for planner argument parsing and validation."""
+
+import pytest
+
+from dynamo.planner.utils.planner_argparse import (
+    create_sla_planner_parser,
+    validate_planner_args,
+)
+
+pytestmark = [
+    pytest.mark.gpu_0,
+    pytest.mark.pre_merge,
+    pytest.mark.unit,
+    pytest.mark.planner,
+]
+
+
+def test_parser_global_planner_mode():
+    """Test parser accepts global-planner environment mode arguments."""
+    parser = create_sla_planner_parser()
+    args = parser.parse_args(
+        [
+            "--namespace",
+            "test-ns",
+            "--environment",
+            "global-planner",
+            "--global-planner-namespace",
+            "global-ns",
+        ]
+    )
+
+    assert args.environment == "global-planner"
+    assert args.global_planner_namespace == "global-ns"
+
+
+def test_validate_global_planner_mode_without_namespace():
+    """Test validation fails for global-planner environment without GlobalPlanner namespace."""
+    parser = create_sla_planner_parser()
+    args = parser.parse_args(
+        ["--namespace", "test-ns", "--environment", "global-planner"]
+    )
+
+    with pytest.raises(ValueError, match="global-planner-namespace required"):
+        validate_planner_args(args)
+
+
+def test_parser_invalid_environment():
+    """Test parser rejects invalid environment."""
+    parser = create_sla_planner_parser()
+
+    with pytest.raises(SystemExit):
+        parser.parse_args(
+            ["--namespace", "test-ns", "--environment", "invalid-environment"]
+        )
+
+
+def test_parser_all_existing_args_still_work():
+    """Test that existing planner arguments still work."""
+    parser = create_sla_planner_parser()
+    args = parser.parse_args(
+        [
+            "--namespace",
+            "test-ns",
+            "--backend",
+            "vllm",
+            "--environment",
+            "kubernetes",
+            "--ttft",
+            "200",
+            "--itl",
+            "50",
+            "--max-gpu-budget",
+            "16",
+            "--adjustment-interval",
+            "60",
+        ]
+    )
+
+    assert args.namespace == "test-ns"
+    assert args.backend == "vllm"
+    assert args.environment == "kubernetes"
+    assert args.ttft == 200
+    assert args.itl == 50
+    assert args.max_gpu_budget == 16
+    assert args.adjustment_interval == 60

tests/planner/unit/test_remote_planner.py

+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Unit tests for remote planner components.
+
+Tests RemotePlannerClient (low-level) and GlobalPlannerConnector (high-level)
+for delegating scale requests to GlobalPlanner.
+"""
+
+import os
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from dynamo.planner import SubComponentType, TargetReplica
+from dynamo.planner.global_planner_connector import GlobalPlannerConnector
+from dynamo.planner.remote_planner_client import RemotePlannerClient
+from dynamo.planner.scale_protocol import ScaleRequest, ScaleResponse, ScaleStatus
+from dynamo.planner.utils.exceptions import EmptyTargetReplicasError
+
+pytestmark = [
+    pytest.mark.gpu_0,
+    pytest.mark.pre_merge,
+    pytest.mark.unit,
+    pytest.mark.planner,
+]
+
+
+@pytest.fixture
+def mock_runtime():
+    """Create a mock DistributedRuntime."""
+    runtime = MagicMock()
+    namespace_mock = MagicMock()
+    component_mock = MagicMock()
+    endpoint_mock = MagicMock()
+    client_mock = AsyncMock()
+
+    runtime.namespace.return_value = namespace_mock
+    namespace_mock.component.return_value = component_mock
+    component_mock.endpoint.return_value = endpoint_mock
+    endpoint_mock.client = AsyncMock(return_value=client_mock)
+    client_mock.wait_for_instances = AsyncMock()
+
+    # Mock scale_request to return a response
+    client_mock.scale_request = AsyncMock(
+        return_value={
+            "status": "success",
+            "message": "Scaled successfully",
+            "current_replicas": {"prefill": 3, "decode": 5},
+        }
+    )
+
+    return runtime, client_mock
+
+
+@pytest.mark.asyncio
+async def test_send_scale_request_success(mock_runtime):
+    """Test successful scale request (exercises protocol, client, and serialization)."""
+    runtime, mock_client = mock_runtime
+    client = RemotePlannerClient(runtime, "central-ns", "Planner")
+
+    request = ScaleRequest(
+        caller_namespace="app-ns",
+        graph_deployment_name="my-dgd",
+        k8s_namespace="default",
+        target_replicas=[
+            TargetReplica(
+                sub_component_type=SubComponentType.PREFILL, desired_replicas=3
+            ),
+            TargetReplica(
+                sub_component_type=SubComponentType.DECODE, desired_replicas=5
+            ),
+        ],
+        blocking=False,
+    )
+
+    response = await client.send_scale_request(request)
+
+    assert response.status == ScaleStatus.SUCCESS
+    assert response.message == "Scaled successfully"
+    assert response.current_replicas["prefill"] == 3
+    assert response.current_replicas["decode"] == 5
+    # Verify lazy init happened
+    assert client._client is not None
+    runtime.namespace.assert_called_once_with("central-ns")
+
+
+@pytest.mark.asyncio
+async def test_send_scale_request_error():
+    """Test scale request error handling."""
+    runtime = MagicMock()
+    namespace_mock = MagicMock()
+    component_mock = MagicMock()
+    endpoint_mock = MagicMock()
+    client_mock = AsyncMock()
+
+    runtime.namespace.return_value = namespace_mock
+    namespace_mock.component.return_value = component_mock
+    component_mock.endpoint.return_value = endpoint_mock
+    endpoint_mock.client = AsyncMock(return_value=client_mock)
+    client_mock.wait_for_instances = AsyncMock()
+
+    # Mock scale_request to return error response
+    client_mock.scale_request = AsyncMock(
+        return_value={
+            "status": "error",
+            "message": "Namespace not authorized",
+            "current_replicas": {},
+        }
+    )
+
+    client = RemotePlannerClient(runtime, "central-ns", "Planner")
+
+    request = ScaleRequest(
+        caller_namespace="unauthorized-ns",
+        graph_deployment_name="my-dgd",
+        k8s_namespace="default",
+        target_replicas=[
+            TargetReplica(
+                sub_component_type=SubComponentType.PREFILL, desired_replicas=1
+            )
+        ],
+    )
+
+    response = await client.send_scale_request(request)
+
+    assert response.status == ScaleStatus.ERROR
+    assert "not authorized" in response.message
+
+
+@pytest.mark.asyncio
+async def test_send_scale_request_no_response():
+    """Test scale request when no response is received."""
+    runtime = MagicMock()
+    namespace_mock = MagicMock()
+    component_mock = MagicMock()
+    endpoint_mock = MagicMock()
+    client_mock = AsyncMock()
+
+    runtime.namespace.return_value = namespace_mock
+    namespace_mock.component.return_value = component_mock
+    component_mock.endpoint.return_value = endpoint_mock
+    endpoint_mock.client = AsyncMock(return_value=client_mock)
+    client_mock.wait_for_instances = AsyncMock()
+
+    # Mock scale_request to return None
+    client_mock.scale_request = AsyncMock(return_value=None)
+
+    client = RemotePlannerClient(runtime, "central-ns", "Planner")
+
+    request = ScaleRequest(
+        caller_namespace="app-ns",
+        graph_deployment_name="my-dgd",
+        k8s_namespace="default",
+        target_replicas=[
+            TargetReplica(
+                sub_component_type=SubComponentType.PREFILL, desired_replicas=1
+            )
+        ],
+    )
+
+    with pytest.raises(RuntimeError, match="No response from centralized planner"):
+        await client.send_scale_request(request)
+
+
+@pytest.mark.asyncio
+async def test_multiple_requests_reuse_client(mock_runtime):
+    """Test that multiple requests reuse the same client instance."""
+    runtime, mock_client = mock_runtime
+    client = RemotePlannerClient(runtime, "central-ns", "Planner")
+
+    request1 = ScaleRequest(
+        caller_namespace="app-ns",
+        graph_deployment_name="my-dgd",
+        k8s_namespace="default",
+        target_replicas=[
+            TargetReplica(
+                sub_component_type=SubComponentType.PREFILL, desired_replicas=2
+            )
+        ],
+    )
+
+    request2 = ScaleRequest(
+        caller_namespace="app-ns",
+        graph_deployment_name="my-dgd",
+        k8s_namespace="default",
+        target_replicas=[
+            TargetReplica(
+                sub_component_type=SubComponentType.PREFILL, desired_replicas=4
+            )
+        ],
+    )
+
+    # Send first request
+    await client.send_scale_request(request1)
+    first_client = client._client
+
+    # Send second request
+    await client.send_scale_request(request2)
+    second_client = client._client
+
+    # Should be the same client instance
+    assert first_client is second_client
+
+
+# ============================================================================
+# GlobalPlannerConnector Tests
+# ============================================================================
+
+
+@pytest.fixture
+def connector_runtime():
+    """Mock runtime for GlobalPlannerConnector"""
+    return MagicMock()
+
+
+@pytest.fixture
+def connector(connector_runtime):
+    """Create GlobalPlannerConnector instance"""
+    return GlobalPlannerConnector(
+        runtime=connector_runtime,
+        dynamo_namespace="test-ns",
+        global_planner_namespace="global-ns",
+        model_name="test-model",
+    )
+
+
+@pytest.mark.asyncio
+async def test_connector_initialization(connector, connector_runtime):
+    """Test GlobalPlannerConnector initialization and async_init"""
+    assert connector.dynamo_namespace == "test-ns"
+    assert connector.global_planner_namespace == "global-ns"
+    assert connector.remote_client is None
+
+    with patch(
+        "dynamo.planner.global_planner_connector.RemotePlannerClient"
+    ) as mock_client_class:
+        mock_client = MagicMock()
+        mock_client_class.return_value = mock_client
+        await connector._async_init()
+        mock_client_class.assert_called_once_with(
+            connector_runtime, "global-ns", "GlobalPlanner"
+        )
+        assert connector.remote_client == mock_client
+
+
+@pytest.mark.asyncio
+async def test_connector_set_replicas_success(connector):
+    """Test GlobalPlannerConnector scaling with enum conversion and predicted load"""
+    target_replicas = [
+        TargetReplica(
+            sub_component_type=SubComponentType.PREFILL,
+            component_name="prefill-svc",
+            desired_replicas=3,
+        ),
+        TargetReplica(
+            sub_component_type=SubComponentType.DECODE,
+            component_name="decode-svc",
+            desired_replicas=5,
+        ),
+    ]
+
+    with patch.dict(
+        os.environ, {"DYN_PARENT_DGD_K8S_NAME": "dgd", "POD_NAMESPACE": "ns"}
+    ):
+        mock_response = ScaleResponse(
+            status=ScaleStatus.SUCCESS,
+            message="OK",
+            current_replicas={"prefill": 3, "decode": 5},
+        )
+        mock_client = AsyncMock()
+        mock_client.send_scale_request = AsyncMock(return_value=mock_response)
+        connector.remote_client = mock_client
+        connector.set_predicted_load(100.0, 512.0, 256.0)
+
+        await connector.set_component_replicas(target_replicas, blocking=False)
+
+        # Verify request structure and enum to string conversion
+        request = mock_client.send_scale_request.call_args[0][0]
+        assert request.caller_namespace == "test-ns"
+        assert request.blocking is False
+        assert request.predicted_load["num_requests"] == 100.0
+        assert len(request.target_replicas) == 2
+        assert request.target_replicas[0].sub_component_type == "prefill"
+        assert isinstance(request.target_replicas[0].sub_component_type, str)
+
+
+@pytest.mark.asyncio
+async def test_connector_error_handling(connector):
+    """Test GlobalPlannerConnector error handling"""
+    # Empty list
+    with pytest.raises(EmptyTargetReplicasError):
+        await connector.set_component_replicas([])
+
+    # Uninitialized
+    target = [
+        TargetReplica(
+            sub_component_type=SubComponentType.PREFILL,
+            component_name="p",
+            desired_replicas=1,
+        )
+    ]
+    with pytest.raises(RuntimeError, match="not initialized"):
+        await connector.set_component_replicas(target)
+
+    # Error response
+    with patch.dict(os.environ, {"DYN_PARENT_DGD_K8S_NAME": "d", "POD_NAMESPACE": "n"}):
+        mock_response = ScaleResponse(
+            status=ScaleStatus.ERROR, message="Failed", current_replicas={}
+        )
+        mock_client = AsyncMock()
+        mock_client.send_scale_request = AsyncMock(return_value=mock_response)
+        connector.remote_client = mock_client
+        with pytest.raises(RuntimeError, match="GlobalPlanner scaling failed"):
+            await connector.set_component_replicas(target)
+
+
+@pytest.mark.asyncio
+async def test_connector_unsupported_and_noop_operations(connector):
+    """Test unsupported and no-op operations"""
+    # Unsupported
+    with pytest.raises(NotImplementedError, match="batch operations"):
+        await connector.add_component(SubComponentType.PREFILL)
+    with pytest.raises(NotImplementedError, match="batch operations"):
+        await connector.remove_component(SubComponentType.DECODE)
+
+    # No-op operations
+    await connector.validate_deployment(
+        prefill_component_name="p", decode_component_name="d"
+    )
+    await connector.wait_for_deployment_ready()
+
+
+def test_connector_model_name_and_predicted_load(connector_runtime):
+    """Test GlobalPlannerConnector model name and predicted load tracking"""
+    # With model name
+    c1 = GlobalPlannerConnector(connector_runtime, "ns", "gns", "GP", model_name="test")
+    assert c1.get_model_name() == "test"
+
+    # Without model name
+    c2 = GlobalPlannerConnector(connector_runtime, "ns", "gns", "GP", model_name=None)
+    assert c2.get_model_name() == "managed-remotely"
+
+    # Predicted load
+    c1.set_predicted_load(42.0, 256.0, 128.0)
+    assert c1.last_predicted_load == {"num_requests": 42.0, "isl": 256.0, "osl": 128.0}