refactor: refactor dynamo serve part-1/N (#788)

Co-authored-by: ishandhanani <ishandhanani@gmail.com>

deploy/dynamo/sdk/README.md

@@ -35,7 +35,7 @@ The code for the pipeline looks like this:
 ```python
 # filename: pipeline.py
 
-from dynamo.sdk import service, dynamo_endpoint, depends, api
+from dynamo.sdk import service, dynamo_endpoint, depends
 from pydantic import BaseModel
 
 class RequestType(BaseModel):
@@ -93,7 +93,7 @@ dynamo serve pipeline:Frontend
 Once it's up and running, you can make a request to the pipeline using
 
 ```bash
-curl -X POST http://localhost:3000/generate \
+curl -X POST http://localhost:8000/generate \
     -H "Content-Type: application/json" \
     -d '{"text": "federer"}'
 ```

deploy/dynamo/sdk/src/dynamo/sdk/__init__.py

@@ -15,11 +15,10 @@
 
 from typing import Any
 
-from bentoml import api  # type: ignore
 from bentoml import on_shutdown as async_on_shutdown
 from bentoml._internal.context import server_context  # type: ignore
 
-from dynamo.sdk.lib.decorators import async_on_start, dynamo_api, dynamo_endpoint
+from dynamo.sdk.lib.decorators import async_on_start, dynamo_endpoint
 from dynamo.sdk.lib.dependency import depends
 from dynamo.sdk.lib.image import DYNAMO_IMAGE
 from dynamo.sdk.lib.service import service
@@ -28,11 +27,9 @@ dynamo_context: dict[str, Any] = {}
 
 __all__ = [
     "DYNAMO_IMAGE",
-    "api",
     "async_on_shutdown",
     "async_on_start",
     "depends",
-    "dynamo_api",
     "dynamo_context",
     "dynamo_endpoint",
     "server_context",

deploy/dynamo/sdk/src/dynamo/sdk/cli/allocator.py

@@ -17,120 +17,160 @@
 
 from __future__ import annotations
 
+import logging
 import os
-import warnings
 from typing import Any
 
 from _bentoml_sdk import Service
-from bentoml._internal.configuration.containers import BentoMLContainer
-from bentoml._internal.resource import system_resources
-from bentoml.exceptions import BentoMLConfigException
-from simple_di import Provide, inject
+from simple_di import inject
 
-NVIDIA_GPU = "nvidia.com/gpu"
+# Import our own resource module
+from dynamo.sdk.lib.resource import NVIDIA_GPU, GPUManager, system_resources
+
+logger = logging.getLogger(__name__)
+
+# Constants
 DYN_DISABLE_AUTO_GPU_ALLOCATION = "DYN_DISABLE_AUTO_GPU_ALLOCATION"
 DYN_DEPLOYMENT_ENV = "DYN_DEPLOYMENT_ENV"
 
 
+def format_memory_gb(memory_bytes: float) -> str:
+    """Convert memory from bytes to formatted GB string.
+    Args:
+        memory_bytes: Memory size in bytes
+    Returns:
+        Formatted string with memory size in GB with 1 decimal place
+    """
+    return f"{memory_bytes/1024/1024/1024:.1f}GB"
+
+
 class ResourceAllocator:
     def __init__(self) -> None:
+        """Initialize the resource allocator."""
         self.system_resources = system_resources()
+        self.gpu_manager = GPUManager()
         self.remaining_gpus = len(self.system_resources[NVIDIA_GPU])
+
+        # For compatibility with the old implementation
         self._available_gpus: list[tuple[float, float]] = [
             (1.0, 1.0)  # each item is (remaining, unit)
             for _ in range(self.remaining_gpus)
         ]
 
     def assign_gpus(self, count: float) -> list[int]:
-        if count > self.remaining_gpus:
-            warnings.warn(
-                f"Requested {count} GPUs, but only {self.remaining_gpus} are remaining. "
-                f"Serving may fail due to inadequate GPUs. Set {DYN_DISABLE_AUTO_GPU_ALLOCATION}=1 "
-                "to disable automatic allocation and allocate GPUs manually.",
-                ResourceWarning,
-                stacklevel=3,
-            )
-        self.remaining_gpus = int(max(0, self.remaining_gpus - count))
-        if count < 1:  # a fractional GPU
-            try:
-                # try to find the GPU used with the same fragment
-                gpu = next(
-                    i
-                    for i, v in enumerate(self._available_gpus)
-                    if v[0] > 0 and v[1] == count
-                )
-            except StopIteration:
-                try:
-                    gpu = next(
-                        i for i, v in enumerate(self._available_gpus) if v[0] == 1.0
-                    )
-                except StopIteration:
-                    gpu = len(self._available_gpus)
-                    self._available_gpus.append((1.0, count))
-            remaining, _ = self._available_gpus[gpu]
-            if (remaining := remaining - count) < count:
-                # can't assign to the next one, mark it as zero.
-                self._available_gpus[gpu] = (0.0, count)
-            else:
-                self._available_gpus[gpu] = (remaining, count)
-            return [gpu]
-        else:  # allocate n GPUs, n is a positive integer
-            if int(count) != count:
-                raise BentoMLConfigException(
-                    "Float GPUs larger than 1 is not supported"
-                )
-            count = int(count)
-            unassigned = [
-                gpu
-                for gpu, value in enumerate(self._available_gpus)
-                if value[0] > 0 and value[1] == 1.0
-            ]
-            if len(unassigned) < count:
-                warnings.warn(
-                    f"Not enough GPUs to be assigned, {count} is requested",
-                    ResourceWarning,
-                )
-                for _ in range(count - len(unassigned)):
-                    unassigned.append(len(self._available_gpus))
-                    self._available_gpus.append((1.0, 1.0))
-            for gpu in unassigned[:count]:
-                self._available_gpus[gpu] = (0.0, 1.0)
-            return unassigned[:count]
+        """
+        Assign GPUs for use.
+
+        Args:
+            count: Number of GPUs to assign (can be fractional)
+
+        Returns:
+            List of GPU indices that were assigned
+        """
+        # Use our GPU manager's assign_gpus method
+        return self.gpu_manager.assign_gpus(count)
+
+    def get_gpu_stats(self) -> list[dict[str, Any]]:
+        """Get detailed statistics for all GPUs."""
+        return self.gpu_manager.get_gpu_stats()
 
     @inject
     def get_resource_envs(
         self,
         service: Service[Any],
-        services: dict[str, Any] = Provide[BentoMLContainer.config.services],
     ) -> tuple[int, list[dict[str, str]]]:
+        """
+        Get resource environment variables for a service.
+
+        Args:
+            service: The service to get resource environment variables for
+
+        Returns:
+            Tuple of (number of workers, list of environment variables dictionaries)
+        """
+        logger.info(f"Getting resource envs for service {service.name}")
+        services = service.get_service_configs()
+        if service.name not in services:
+            logger.warning(f"No service configs found for {service.name}")
+            return 1, []  # Default to 1 worker, no special resources
+
         config = services[service.name]
+        logger.debug(f"Using config for {service.name}: {config}")
 
         num_gpus = 0
         num_workers = 1
         resource_envs: list[dict[str, str]] = []
+
+        # Check if service requires GPUs
         if "gpu" in (config.get("resources") or {}):
             num_gpus = config["resources"]["gpu"]  # type: ignore
+            logger.info(f"GPU requirement found: {num_gpus}")
+
+            # Check if we have enough GPUs
+            available_gpus = self.gpu_manager.get_available_gpus()
+            if num_gpus > len(available_gpus):
+                logger.warning(
+                    f"Requested {num_gpus} GPUs, but only {len(available_gpus)} are available. "
+                    f"Service may fail due to inadequate GPU resources."
+                )
+
+        # Determine number of workers
         if config.get("workers"):
-            if (workers := config["workers"]) == "cpu_count":
-                num_workers = int(self.system_resources["cpu"])
-                # don't assign gpus to workers
-                return num_workers, resource_envs
-            else:  # workers is a number
-                num_workers = workers
+            num_workers = config["workers"]
+            logger.info(f"Using configured worker count: {num_workers}")
+
+        # Handle GPU allocation
         if num_gpus and DYN_DISABLE_AUTO_GPU_ALLOCATION not in os.environ:
+            logger.info("GPU allocation enabled")
+
             if os.environ.get(DYN_DEPLOYMENT_ENV):
+                logger.info("K8s deployment detected")
                 # K8s replicas: Assumes DYNAMO_DEPLOYMENT_ENV is set
                 # each pod in replicaset will have separate GPU with same CUDA_VISIBLE_DEVICES
                 assigned = self.assign_gpus(num_gpus)
-                resource_envs = [
-                    {"CUDA_VISIBLE_DEVICES": ",".join(map(str, assigned))}
-                    for _ in range(num_workers)
-                ]
-            else:
-                # local deployment where we split all available GPUs across workers
+                logger.info(f"Assigned GPUs for K8s: {assigned}")
+
+                # Generate environment variables for each worker
                 for _ in range(num_workers):
+                    env_vars = {"CUDA_VISIBLE_DEVICES": ",".join(map(str, assigned))}
+                    resource_envs.append(env_vars)
+            else:
+                logger.info("Local deployment detected")
+                # Local deployment where we split all available GPUs across workers
+                for worker_id in range(num_workers):
                     assigned = self.assign_gpus(num_gpus)
-                    resource_envs.append(
-                        {"CUDA_VISIBLE_DEVICES": ",".join(map(str, assigned))}
-                    )
+                    logger.info(f"Assigned GPUs for worker {worker_id}: {assigned}")
+
+                    # Generate environment variables for this worker
+                    env_vars = {"CUDA_VISIBLE_DEVICES": ",".join(map(str, assigned))}
+
+                    # If we have comprehensive GPU stats, log them
+                    try:
+                        gpu_stats = [
+                            stat
+                            for stat in self.get_gpu_stats()
+                            if stat["index"] in assigned
+                        ]
+                        for stat in gpu_stats:
+                            logger.info(
+                                f"GPU {stat['index']} ({stat['name']}): "
+                                f"Memory: {format_memory_gb(stat['free_memory'])} free / "
+                                f"{format_memory_gb(stat['total_memory'])} total, "
+                                f"Utilization: {stat['gpu_utilization']}%, "
+                                f"Temperature: {stat['temperature']}°C"
+                            )
+                    except Exception as e:
+                        logger.debug(f"Failed to get GPU stats: {e}")
+
+                    resource_envs.append(env_vars)
+
+        logger.info(
+            f"Final resource allocation - workers: {num_workers}, envs: {resource_envs}"
+        )
         return num_workers, resource_envs
+
+    def reset_allocations(self):
+        """Reset all GPU allocations."""
+        self.gpu_manager.reset_allocations()
+        # Reset legacy tracking
+        self._available_gpus = [(1.0, 1.0) for _ in range(self.remaining_gpus)]

deploy/dynamo/sdk/src/dynamo/sdk/lib/bento.py → deploy/dynamo/sdk/src/dynamo/sdk/cli/bento_util.py

@@ -117,7 +117,7 @@ class Bento(BaseBento):
             build_config.envs.extend(svc.envs)
             build_config.labels.update(svc.labels)
             if svc.image is not None:
-                image = svc.image
+                image = Image(base_image=svc.image)
         if not disable_image:
             image = populate_image_from_build_config(image, build_config, build_ctx)
         build_config = build_config.with_defaults()

deploy/dynamo/sdk/src/dynamo/sdk/cli/circus.py

+#  SPDX-FileCopyrightText: Copyright (c) 2020 Atalaya Tech. Inc
+#  SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#  SPDX-License-Identifier: Apache-2.0
+#  #
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#  #
+#  http://www.apache.org/licenses/LICENSE-2.0
+#  #
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#  Modifications Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES
+
+# Once planner v1 goes live - this will be be full of more granular APIs
+
+from __future__ import annotations
+
+import contextlib
+import os
+import pathlib
+import shlex
+import sys
+from dataclasses import dataclass
+from typing import Any, Callable
+
+import psutil
+from circus.arbiter import Arbiter as _Arbiter
+from circus.sockets import CircusSocket
+from circus.watcher import Watcher
+
+from .utils import ServiceProtocol
+
+
+class Arbiter(_Arbiter):
+    """Arbiter with cleanup support via exit_stack."""
+
+    def __init__(self, *args: Any, **kwargs: Any) -> None:
+        super().__init__(*args, **kwargs)
+        self.exit_stack = contextlib.ExitStack()
+
+    def start(self, cb: Callable[[Any], Any] | None = None) -> None:
+        """Start arbiter and enter context."""
+        self.exit_stack.__enter__()
+        fut = super().start(cb)
+        if exc := fut.exception():
+            raise exc
+
+    def stop(self) -> None:
+        """Stop arbiter and cleanup resources."""
+        self.exit_stack.__exit__(None, None, None)
+        return super().stop()
+
+
+@dataclass
+class CircusRunner:
+    """Simple server wrapper for arbiter lifecycle management."""
+
+    arbiter: Arbiter
+
+    def stop(self) -> None:
+        self.arbiter.stop()
+
+    @property
+    def running(self) -> bool:
+        return self.arbiter.running
+
+    def __enter__(self) -> CircusRunner:
+        return self
+
+    def __exit__(self, *_: Any) -> None:
+        self.stop()
+
+
+MAX_AF_UNIX_PATH_LENGTH = 103
+
+
+def create_circus_watcher(
+    name: str,
+    args: list[str],
+    *,
+    cmd: str = sys.executable,
+    use_sockets: bool = True,
+    **kwargs: Any,
+) -> Watcher:
+    return Watcher(
+        name=name,
+        cmd=shlex.quote(cmd) if psutil.POSIX else cmd,
+        args=args,
+        copy_env=True,
+        stop_children=True,
+        use_sockets=use_sockets,
+        graceful_timeout=86400,
+        respawn=False,  # TODO
+        **kwargs,
+    )
+
+
+def create_arbiter(
+    watchers: list[Watcher], *, threaded: bool = False, **kwargs: Any
+) -> Arbiter:
+    endpoint_port = int(os.environ.get("DYN_CIRCUS_ENDPOINT_PORT", "41234"))
+    pubsub_port = int(os.environ.get("DYN_CIRCUS_PUBSUB_PORT", "52345"))
+
+    return Arbiter(
+        watchers,
+        endpoint=f"tcp://127.0.0.1:{endpoint_port}",
+        pubsub_endpoint=f"tcp://127.0.0.1:{pubsub_port}",
+        check_delay=kwargs.pop("check_delay", 10),
+        **kwargs,
+    )
+
+
+def path_to_uri(path: str) -> str:
+    """
+    Convert a path to a URI.
+
+    Args:
+        path: Path to convert to URI.
+
+    Returns:
+        URI string. (quoted, absolute)
+    """
+    return pathlib.PurePosixPath(path).as_uri()
+
+
+def _get_server_socket(
+    service: ServiceProtocol,
+    uds_path: str,
+) -> tuple[str, CircusSocket]:
+    """Create a Unix Domain Socket for a service.
+
+    Args:
+        service: The service to create a socket for
+        uds_path: Base directory for Unix Domain Sockets
+        port_stack: Not used in POSIX implementation, kept for interface compatibility
+
+    Returns:
+        Tuple of (socket URI, CircusSocket object)
+
+    Raises:
+        AssertionError: If socket path exceeds maximum length
+    """
+    socket_path = os.path.join(uds_path, f"{id(service)}.sock")
+    assert (
+        len(socket_path) < MAX_AF_UNIX_PATH_LENGTH
+    ), f"Socket path '{socket_path}' exceeds maximum length of {MAX_AF_UNIX_PATH_LENGTH}"
+
+    return path_to_uri(socket_path), CircusSocket(name=service.name, path=socket_path)

deploy/dynamo/sdk/src/dynamo/sdk/cli/pipeline.py

@@ -36,7 +36,7 @@ from rich.console import Console
 from rich.syntax import Syntax
 from simple_di import Provide, inject
 
-from dynamo.sdk.lib.bento import Bento
+from dynamo.sdk.cli.bento_util import Bento
 
 if t.TYPE_CHECKING:
     from bentoml._internal.bento import BentoStore

deploy/dynamo/sdk/src/dynamo/sdk/cli/serve.py

@@ -47,7 +47,7 @@ def serve(
     service_name: str = typer.Option(
         "",
         help="Only serve the specified service. Don't serve any dependencies of this service.",
-        envvar="DYNAMO_SERVE_SERVICE_NAME",
+        envvar="DYNAMO_SERVICE_NAME",
     ),
     depends: List[str] = typer.Option(
         [],
@@ -92,8 +92,7 @@ def serve(
     """
 
     # Warning: internal
-    from bentoml._internal.service.loader import load
-
+    from dynamo.sdk.lib.loader import find_and_load_service
     from dynamo.sdk.lib.logging import configure_server_logging
     from dynamo.sdk.lib.service import LinkedServices
 
@@ -138,11 +137,12 @@ def serve(
     if sys.path[0] != working_dir_str:
         sys.path.insert(0, working_dir_str)
 
-    svc = load(bento_identifier=dynamo_pipeline, working_dir=working_dir_str)
-
+    svc = find_and_load_service(dynamo_pipeline, working_dir=working_dir)
+    logger.info(f"Loaded service: {svc.name}")
+    logger.info("Dependencies: %s", [dep.on.name for dep in svc.dependencies.values()])
     LinkedServices.remove_unused_edges()
 
-    from dynamo.sdk.cli.serving import serve_http  # type: ignore
+    from dynamo.sdk.cli.serving import serve_dynamo_graph  # type: ignore
 
     svc.inject_config()
 
@@ -155,11 +155,11 @@ def serve(
         )
     )
 
-    serve_http(
+    serve_dynamo_graph(
         dynamo_pipeline,
         working_dir=working_dir_str,
-        host=host,
-        port=port,
+        # host=host,
+        # port=port,
         dependency_map=runner_map_dict,
         service_name=service_name,
         enable_planner=enable_planner,

deploy/dynamo/sdk/src/dynamo/sdk/cli/serve_dynamo.py

@@ -22,11 +22,16 @@ import inspect
 import json
 import logging
 import os
+import signal
+import sys
+import time
 import typing as t
 from typing import Any
 
 import click
+import uvicorn
 import uvloop
+from fastapi.responses import StreamingResponse
 
 from dynamo.runtime import DistributedRuntime, dynamo_endpoint, dynamo_worker
 from dynamo.sdk import dynamo_context
@@ -35,6 +40,69 @@ from dynamo.sdk.lib.service import LinkedServices
 logger = logging.getLogger(__name__)
 
 
+def add_fastapi_routes(app, service, class_instance):
+    """
+    Add FastAPI routes for Dynamo endpoints marked with is_api=True.
+
+    Args:
+        app: FastAPI app instance
+        service: Dynamo service instance
+        class_instance: Instance of the service class
+    """
+
+    added_routes = []
+    for name, endpoint in service.get_dynamo_endpoints().items():
+        if endpoint.is_api:
+            path = name if name.startswith("/") else f"/{name}"
+            # Bind the method to the class instance
+            bound_method = endpoint.func.__get__(class_instance)
+
+            # Check if the method is a generator or async generator
+            is_streaming = inspect.isasyncgenfunction(
+                bound_method
+            ) or inspect.isgeneratorfunction(bound_method)
+
+            # Set up appropriate response model and response class
+            if is_streaming:
+                logger.info(f"Registering streaming endpoint {path}")
+                app.add_api_route(
+                    path,
+                    bound_method,
+                    methods=["POST"],
+                    response_class=StreamingResponse,
+                )
+            else:
+                logger.info(f"Registering regular endpoint {path}")
+                app.add_api_route(
+                    path,
+                    bound_method,
+                    methods=["POST"],
+                )
+
+            added_routes.append(path)
+            logger.info(f"Added API route {path} to FastAPI app")
+    return added_routes
+
+
+class GracefulExit(SystemExit):
+    """Exception to signal a graceful exit."""
+
+    pass
+
+
+def setup_signal_handlers():
+    """Setup signal handlers for graceful shutdown."""
+
+    def signal_handler(sig, frame):
+        logger.info(f"Received signal {sig}, initiating graceful shutdown")
+        raise GracefulExit(0)
+
+    # Register SIGINT and SIGTERM handlers
+    signal.signal(signal.SIGINT, signal_handler)
+    signal.signal(signal.SIGTERM, signal_handler)
+    signal.signal(signal.SIGQUIT, signal_handler)
+
+
 @click.command()
 @click.argument("bento_identifier", type=click.STRING, required=False, default=".")
 @click.option("--service-name", type=click.STRING, required=False, default="")
@@ -68,6 +136,10 @@ def main(
 
     from dynamo.sdk.lib.logging import configure_server_logging
 
+    # Setup signal handlers for graceful shutdown
+    setup_signal_handlers()
+
+    run_id = service_name
     dynamo_context["service_name"] = service_name
     dynamo_context["runner_map"] = runner_map
     dynamo_context["worker_id"] = worker_id
@@ -168,7 +240,7 @@ def main(
                 # Run startup hooks before setting up endpoints
                 for name, member in vars(class_instance.__class__).items():
                     if callable(member) and getattr(
-                        member, "__bentoml_startup_hook__", False
+                        member, "__dynamo_startup_hook__", False
                     ):
                         logger.debug(f"Running startup hook: {name}")
                         result = getattr(class_instance, name)()
@@ -188,13 +260,75 @@ def main(
                     logger.info(f"Serving {service.name} with lease: {lease.id()}")
                 result = await endpoints[0].serve_endpoint(twm[0], lease)
 
+            except GracefulExit:
+                logger.info(f"[{run_id}] Gracefully shutting down {service.name}")
+                # Add any specific cleanup needed
+                return None
             except Exception as e:
                 logger.error(f"Error in Dynamo component setup: {str(e)}")
                 raise
 
-        uvloop.install()
-        asyncio.run(worker())
+        # if the service has a FastAPI app, add the worker as an event handler
+        def web_worker():
+            try:
+                if not service.app:
+                    return
+
+                # Create the class instance
+                class_instance = service.inner()
+                # TODO: init hooks
+                # Add API routes to the FastAPI app
+                added_routes = add_fastapi_routes(service.app, service, class_instance)
+
+                if added_routes:
+                    # Configure uvicorn with graceful shutdown
+                    config = uvicorn.Config(
+                        service.app, host="0.0.0.0", port=8000, log_level="info"
+                    )
+                    server = uvicorn.Server(config)
+
+                    # Start the server with graceful shutdown handling
+                    logger.info(
+                        f"Starting FastAPI server on 0.0.0.0:8000 with routes: {added_routes}"
+                    )
+                    server.run()
+                else:
+                    logger.warning("No API routes found, not starting FastAPI server")
+                    # Keep the process running until interrupted
+                    logger.info("Service is running, press Ctrl+C to stop")
+                    while True:
+                        try:
+                            # Sleep in small increments to respond to signals quickly
+                            time.sleep(0.1)
+                        except (KeyboardInterrupt, GracefulExit):
+                            logger.info("Gracefully shutting down FastAPI process")
+                            break
+            except GracefulExit:
+                logger.info("Gracefully shutting down FastAPI service")
+            except Exception as e:
+                logger.error(f"Error in web worker: {str(e)}")
+                raise
+
+        try:
+            uvloop.install()
+            if service.app:
+                web_worker()
+            else:
+                asyncio.run(worker())
+        except GracefulExit:
+            logger.info("Exiting gracefully")
+            sys.exit(0)
+        except KeyboardInterrupt:
+            logger.info("Interrupted, shutting down gracefully")
+            sys.exit(0)
 
 
 if __name__ == "__main__":
-    main()
+    try:
+        main()
+    except (GracefulExit, KeyboardInterrupt):
+        logger.info("Exiting gracefully")
+        sys.exit(0)
+    except Exception as e:
+        logger.error(f"Error in main: {str(e)}")
+        sys.exit(1)

deploy/dynamo/sdk/src/dynamo/sdk/cli/serving.py

@@ -18,112 +18,40 @@
 from __future__ import annotations
 
 import contextlib
-import ipaddress
 import json
 import logging
 import os
 import pathlib
-import platform
 import shutil
-import socket
 import tempfile
-import typing as t
-from typing import Any, Dict, Optional, Protocol, TypeVar
+from typing import Any, Dict, Optional, TypeVar
 
-# WARNING: internal
+# TODO: WARNING: internal but only for type checking in the deploy path i believe
 from _bentoml_sdk import Service
-
-# WARNING: internal
-from bentoml._internal.container import BentoMLContainer
-
-# WARNING: internal
-from bentoml._internal.utils.circus import Server
-from bentoml.exceptions import BentoMLConfigException
 from circus.sockets import CircusSocket
 from circus.watcher import Watcher
-from simple_di import Provide, inject
+from simple_di import inject
+
+from dynamo.sdk.cli.circus import CircusRunner
 
 from .allocator import ResourceAllocator
+from .circus import _get_server_socket
 from .utils import (
     DYN_LOCAL_STATE_DIR,
-    path_to_uri,
+    ServiceProtocol,
     reserve_free_port,
     save_dynamo_state,
 )
 
-
-# Define a Protocol for services to ensure type safety
-class ServiceProtocol(Protocol):
-    name: str
-    inner: Any
-    models: list[Any]
-    bento: Any
-
-    def is_dynamo_component(self) -> bool:
-        ...
-
-    def dynamo_address(self) -> tuple[str, str]:
-        ...
+# WARNING: internal
 
 
 # Use Protocol as the base for type alias
 AnyService = TypeVar("AnyService", bound=ServiceProtocol)
 
-POSIX = os.name == "posix"
-WINDOWS = os.name == "nt"
-IS_WSL = "microsoft-standard" in platform.release()
-API_SERVER_NAME = "_bento_api_server"
 
-MAX_AF_UNIX_PATH_LENGTH = 103
 logger = logging.getLogger(__name__)
 
-if POSIX and not IS_WSL:
-
-    def _get_server_socket(
-        service: ServiceProtocol,
-        uds_path: str,
-        port_stack: contextlib.ExitStack,
-    ) -> tuple[str, CircusSocket]:
-        from circus.sockets import CircusSocket
-
-        socket_path = os.path.join(uds_path, f"{id(service)}.sock")
-        assert len(socket_path) < MAX_AF_UNIX_PATH_LENGTH
-        return path_to_uri(socket_path), CircusSocket(
-            name=service.name, path=socket_path
-        )
-
-elif WINDOWS or IS_WSL:
-
-    def _get_server_socket(
-        service: ServiceProtocol,
-        uds_path: str,
-        port_stack: contextlib.ExitStack,
-    ) -> tuple[str, CircusSocket]:
-        from circus.sockets import CircusSocket
-
-        runner_port = port_stack.enter_context(reserve_free_port())
-        runner_host = "127.0.0.1"
-
-        return f"tcp://{runner_host}:{runner_port}", CircusSocket(
-            name=service.name,
-            host=runner_host,
-            port=runner_port,
-        )
-
-else:
-
-    def _get_server_socket(
-        service: ServiceProtocol,
-        uds_path: str,
-        port_stack: contextlib.ExitStack,
-    ) -> tuple[str, CircusSocket]:
-        from bentoml.exceptions import BentoMLException
-
-        raise BentoMLException("Unsupported platform")
-
-
-# WARNING: internal
-_BENTO_WORKER_SCRIPT = "_bentoml_impl.worker.service"
 _DYNAMO_WORKER_SCRIPT = "dynamo.sdk.cli.serve_dynamo"
 
 
@@ -140,63 +68,19 @@ def _get_dynamo_worker_script(bento_identifier: str, svc_name: str) -> list[str]
     return args
 
 
-def _get_bento_worker_script(bento_identifier: str, svc_name: str) -> list[str]:
-    args = [
-        "-m",
-        _BENTO_WORKER_SCRIPT,
-        bento_identifier,
-        "--service-name",
-        svc_name,
-        "--fd",
-        f"$(circus.sockets.{svc_name})",
-        "--worker-id",
-        "$(CIRCUS.WID)",
-    ]
-
-    return args
-
-
-def create_dependency_watcher(
-    bento_identifier: str,
-    svc: ServiceProtocol,
-    uds_path: str,
-    port_stack: contextlib.ExitStack,
-    scheduler: ResourceAllocator,
-    working_dir: Optional[str] = None,
-    env: Optional[Dict[str, str]] = None,
-) -> tuple[Watcher, CircusSocket, str]:
-    from bentoml.serving import create_watcher
-
-    num_workers, resource_envs = scheduler.get_resource_envs(svc)
-    uri, socket = _get_server_socket(svc, uds_path, port_stack)
-    args = _get_bento_worker_script(bento_identifier, svc.name)
-    if resource_envs:
-        args.extend(["--worker-env", json.dumps(resource_envs)])
-
-    watcher = create_watcher(
-        name=f"service_{svc.name}",
-        args=args,
-        numprocesses=num_workers,
-        working_dir=working_dir,
-        env=env,
-    )
-    return watcher, socket, uri
-
-
 def create_dynamo_watcher(
     bento_identifier: str,
     svc: ServiceProtocol,
     uds_path: str,
-    port_stack: contextlib.ExitStack,
     scheduler: ResourceAllocator,
     working_dir: Optional[str] = None,
     env: Optional[Dict[str, str]] = None,
 ) -> tuple[Watcher, CircusSocket, str]:
     """Create a watcher for a Dynamo service in the dependency graph"""
-    from bentoml.serving import create_watcher
+    from dynamo.sdk.cli.circus import create_circus_watcher
 
     num_workers, resource_envs = scheduler.get_resource_envs(svc)
-    uri, socket = _get_server_socket(svc, uds_path, port_stack)
+    uri, socket = _get_server_socket(svc, uds_path)
     args = _get_dynamo_worker_script(bento_identifier, svc.name)
     if resource_envs:
         args.extend(["--worker-env", json.dumps(resource_envs)])
@@ -226,7 +110,7 @@ def create_dynamo_watcher(
     namespace, _ = svc.dynamo_address()
 
     # Create the watcher with updated environment
-    watcher = create_watcher(
+    watcher = create_circus_watcher(
         name=f"{namespace}_{svc.name}",
         args=args,
         numprocesses=num_workers,
@@ -240,23 +124,15 @@ def create_dynamo_watcher(
 
 
 @inject(squeeze_none=True)
-def serve_http(
+def serve_dynamo_graph(
     bento_identifier: str | AnyService,
     working_dir: str | None = None,
-    host: str = Provide[BentoMLContainer.http.host],
-    port: int = Provide[BentoMLContainer.http.port],
     dependency_map: dict[str, str] | None = None,
     service_name: str = "",
     enable_planner: bool = False,
-) -> Server:
-    # WARNING: internal
-    from _bentoml_impl.loader import load
-
-    # WARNING: internal
-    from bentoml._internal.utils.circus import create_standalone_arbiter
-    from bentoml.serving import create_watcher
-    from circus.sockets import CircusSocket
-
+) -> CircusRunner:
+    from dynamo.sdk.cli.circus import create_arbiter, create_circus_watcher
+    from dynamo.sdk.lib.loader import find_and_load_service
     from dynamo.sdk.lib.logging import configure_server_logging
 
     from .allocator import ResourceAllocator
@@ -275,7 +151,7 @@ def serve_http(
         # use cwd
         bento_path = pathlib.Path(".")
     else:
-        svc = load(bento_identifier, working_dir)
+        svc = find_and_load_service(bento_identifier, working_dir)
         bento_id = str(bento_identifier)
         bento_path = pathlib.Path(working_dir or ".")
 
@@ -294,7 +170,7 @@ def serve_http(
     if service_name and service_name != svc.name:
         svc = svc.find_dependent_by_name(service_name)
     num_workers, resource_envs = allocator.get_resource_envs(svc)
-    uds_path = tempfile.mkdtemp(prefix="bentoml-uds-")
+    uds_path = tempfile.mkdtemp(prefix="dynamo-uds-")
     try:
         if not service_name and not standalone:
             with contextlib.ExitStack() as port_stack:
@@ -303,74 +179,28 @@ def serve_http(
                         continue
                     if name in dependency_map:
                         continue
-
-                    # Check if this is a Dynamo service
-                    if (
+                    if not (
                         hasattr(dep_svc, "is_dynamo_component")
                         and dep_svc.is_dynamo_component()
                     ):
-                        new_watcher, new_socket, uri = create_dynamo_watcher(
-                            bento_id,
-                            dep_svc,
-                            uds_path,
-                            port_stack,
-                            allocator,
-                            str(bento_path.absolute()),
-                            env=env,
-                        )
-                        namespace, _ = dep_svc.dynamo_address()
-                    else:
-                        # Regular BentoML service
-                        new_watcher, new_socket, uri = create_dependency_watcher(
-                            bento_id,
-                            dep_svc,
-                            uds_path,
-                            port_stack,
-                            allocator,
-                            str(bento_path.absolute()),
-                            env=env,
+                        raise RuntimeError(
+                            f"Service {dep_svc.name} is not a Dynamo component"
                         )
-
+                    new_watcher, new_socket, uri = create_dynamo_watcher(
+                        bento_id,
+                        dep_svc,
+                        uds_path,
+                        allocator,
+                        str(bento_path.absolute()),
+                        env=env,
+                    )
+                    namespace, _ = dep_svc.dynamo_address()
                     watchers.append(new_watcher)
                     sockets.append(new_socket)
                     dependency_map[name] = uri
                 # reserve one more to avoid conflicts
                 port_stack.enter_context(reserve_free_port())
 
-        try:
-            ipaddr = ipaddress.ip_address(host)
-            if ipaddr.version == 4:
-                family = socket.AF_INET
-            elif ipaddr.version == 6:
-                family = socket.AF_INET6
-            else:
-                raise BentoMLConfigException(
-                    f"Unsupported host IP address version: {ipaddr.version}"
-                )
-        except ValueError as e:
-            raise BentoMLConfigException(f"Invalid host IP address: {host}") from e
-
-        if not svc.is_dynamo_component():
-            sockets.append(
-                CircusSocket(
-                    name=API_SERVER_NAME,
-                    host=host,
-                    port=port,
-                    family=family,
-                )
-            )
-
-        server_args = [
-            "-m",
-            _BENTO_WORKER_SCRIPT,
-            bento_identifier,
-            "--fd",
-            f"$(circus.sockets.{API_SERVER_NAME})",
-            "--service-name",
-            svc.name,
-            "--worker-id",
-            "$(CIRCUS.WID)",
-        ]
         dynamo_args = [
             "-m",
             _DYNAMO_WORKER_SCRIPT,
@@ -380,12 +210,7 @@ def serve_http(
             "--worker-id",
             "$(CIRCUS.WID)",
         ]
-        if resource_envs:
-            server_args.extend(["--worker-env", json.dumps(resource_envs)])
-
-        scheme = "http"
 
-        # Check if this is a Dynamo service
         if hasattr(svc, "is_dynamo_component") and svc.is_dynamo_component():
             # resource_envs is the resource allocation (ie CUDA_VISIBLE_DEVICES) for each worker created by the allocator
             # these resource_envs are passed to each individual worker's environment which is set in serve_dynamo
@@ -411,7 +236,7 @@ def serve_http(
                 except json.JSONDecodeError as e:
                     logger.warning(f"Failed to parse DYNAMO_SERVICE_ENVS: {e}")
 
-            watcher = create_watcher(
+            watcher = create_circus_watcher(
                 name=f"{namespace}_{svc.name}",
                 args=dynamo_args,
                 numprocesses=num_workers,
@@ -422,20 +247,6 @@ def serve_http(
             logger.info(
                 f"Created watcher for {svc.name} with {num_workers} workers in the {namespace} namespace"
             )
-        else:
-            watchers.append(
-                create_watcher(
-                    name="service",
-                    args=server_args,
-                    working_dir=str(bento_path.absolute()),
-                    numprocesses=num_workers,
-                    env=env,
-                )
-            )
-            logger.info(f"Created watcher for service with {num_workers} workers")
-
-        log_host = "localhost" if host in ["0.0.0.0", "::"] else host
-        dependency_map[svc.name] = f"{scheme}://{log_host}:{port}"
 
         # inject runner map now
         inject_env = {"BENTOML_RUNNER_MAP": json.dumps(dependency_map)}
@@ -446,12 +257,12 @@ def serve_http(
             else:
                 watcher.env.update(inject_env)
 
-        arbiter_kwargs: dict[str, t.Any] = {
+        arbiter_kwargs: dict[str, Any] = {
             "watchers": watchers,
             "sockets": sockets,
         }
 
-        arbiter = create_standalone_arbiter(**arbiter_kwargs)
+        arbiter = create_arbiter(**arbiter_kwargs)
         arbiter.exit_stack.callback(shutil.rmtree, uds_path, ignore_errors=True)
         if enable_planner:
             arbiter.exit_stack.callback(
@@ -499,7 +310,7 @@ def serve_http(
                 ),
             ),
         )
-        return Server(url=f"{scheme}://{log_host}:{port}", arbiter=arbiter)
+        return CircusRunner(arbiter=arbiter)
     except Exception:
         shutil.rmtree(uds_path, ignore_errors=True)
         raise

deploy/dynamo/sdk/src/dynamo/sdk/cli/utils.py

@@ -25,10 +25,9 @@ import os
 import pathlib
 import random
 import socket
-import typing as t
+from typing import Any, DefaultDict, Dict, Iterator, Optional, Protocol, TextIO, Union
 
 import click
-import psutil
 import yaml
 from click import Command, Context
 
@@ -41,10 +40,24 @@ logger = logging.getLogger(__name__)
 DYN_LOCAL_STATE_DIR = "DYN_LOCAL_STATE_DIR"
 
 
+# Define a Protocol for services to ensure type safety
+class ServiceProtocol(Protocol):
+    name: str
+    inner: Any
+    models: list[Any]
+    bento: Any
+
+    def is_dynamo_component(self) -> bool:
+        ...
+
+    def dynamo_address(self) -> tuple[str, str]:
+        ...
+
+
 class DynamoCommandGroup(click.Group):
     """Simplified version of BentoMLCommandGroup for Dynamo CLI"""
 
-    def __init__(self, *args: t.Any, **kwargs: t.Any) -> None:
+    def __init__(self, *args: Any, **kwargs: Any) -> None:
         self.aliases = kwargs.pop("aliases", [])
         super().__init__(*args, **kwargs)
         self._commands: dict[str, list[str]] = {}
@@ -101,26 +114,19 @@ class DynamoCommandGroup(click.Group):
 def reserve_free_port(
     host: str = "localhost",
     port: int | None = None,
-    prefix: t.Optional[str] = None,
+    prefix: Optional[str] = None,
     max_retry: int = 50,
     enable_so_reuseport: bool = False,
-) -> t.Iterator[int]:
+) -> Iterator[int]:
     """
     detect free port and reserve until exit the context
     """
-    import psutil
-
     sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
     if enable_so_reuseport:
-        if psutil.WINDOWS:
-            sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
-        elif psutil.MACOS or psutil.FREEBSD:
-            sock.setsockopt(socket.SOL_SOCKET, 0x10000, 1)  # SO_REUSEPORT_LB
-        else:
-            sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEPORT, 1)
+        sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEPORT, 1)
+        if sock.getsockopt(socket.SOL_SOCKET, socket.SO_REUSEPORT) == 0:
+            raise RuntimeError("Failed to set SO_REUSEPORT.") from None
 
-            if sock.getsockopt(socket.SOL_SOCKET, socket.SO_REUSEPORT) == 0:
-                raise RuntimeError("Failed to set SO_REUSEPORT.") from None
     if prefix is not None:
         prefix_num = int(prefix) * 10 ** (5 - len(prefix))
         suffix_range = min(65535 - prefix_num, 10 ** (5 - len(prefix)))
@@ -147,29 +153,11 @@ def reserve_free_port(
         sock.close()
 
 
-def path_to_uri(path: str) -> str:
-    """
-    Convert a path to a URI.
-
-    Args:
-        path: Path to convert to URI.
-
-    Returns:
-        URI string. (quoted, absolute)
-    """
-    path = os.path.abspath(path)
-    if psutil.WINDOWS:
-        return pathlib.PureWindowsPath(path).as_uri()
-    if psutil.POSIX:
-        return pathlib.PurePosixPath(path).as_uri()
-    raise ValueError("Unsupported OS")
-
-
 def save_dynamo_state(
     namespace: str,
     circus_endpoint: str,
-    components: dict[str, t.Any],
-    environment: dict[str, t.Any],
+    components: dict[str, Any],
+    environment: dict[str, Any],
 ):
     state_dir = os.environ.get(
         DYN_LOCAL_STATE_DIR, os.path.expanduser("~/.dynamo/state")
@@ -192,7 +180,7 @@ def save_dynamo_state(
     logger.warning(f"Saved state to {state_file}")
 
 
-def _parse_service_arg(arg_name: str, arg_value: str) -> tuple[str, str, t.Any]:
+def _parse_service_arg(arg_name: str, arg_value: str) -> tuple[str, str, Any]:
     """Parse a single CLI argument into service name, key, and value."""
 
     parts = arg_name.split(".")
@@ -205,7 +193,7 @@ def _parse_service_arg(arg_name: str, arg_value: str) -> tuple[str, str, t.Any]:
         and nested_keys[0] == "ServiceArgs"
         and nested_keys[1] == "envs"
     ):
-        value: t.Union[str, int, float, bool, dict, list] = arg_value
+        value: Union[str, int, float, bool, dict, list] = arg_value
     else:
         # Parse value based on type for non-env vars
         try:
@@ -228,12 +216,10 @@ def _parse_service_arg(arg_name: str, arg_value: str) -> tuple[str, str, t.Any]:
     return service, nested_keys[0], result
 
 
-def _parse_service_args(args: list[str]) -> t.Dict[str, t.Any]:
-    service_configs: t.DefaultDict[str, t.Dict[str, t.Any]] = collections.defaultdict(
-        dict
-    )
+def _parse_service_args(args: list[str]) -> Dict[str, Any]:
+    service_configs: DefaultDict[str, Dict[str, Any]] = collections.defaultdict(dict)
 
-    def deep_update(d: dict, key: str, value: t.Any):
+    def deep_update(d: dict, key: str, value: Any):
         """
         Recursively updates nested dictionaries. We use this to process arguments like
 
@@ -283,9 +269,9 @@ def _parse_service_args(args: list[str]) -> t.Dict[str, t.Any]:
 
 
 def resolve_service_config(
-    config_file: pathlib.Path | t.TextIO | None = None,
+    config_file: pathlib.Path | TextIO | None = None,
     args: list[str] | None = None,
-) -> dict[str, dict[str, t.Any]]:
+) -> dict[str, dict[str, Any]]:
     """Resolve service configuration from file and command line arguments.
 
     Args:
@@ -295,7 +281,7 @@ def resolve_service_config(
     Returns:
         Dictionary mapping service names to their configurations
     """
-    service_configs: dict[str, dict[str, t.Any]] = {}
+    service_configs: dict[str, dict[str, Any]] = {}
 
     # Check for deployment config first
     if "DYN_DEPLOYMENT_CONFIG" in os.environ:

deploy/dynamo/sdk/src/dynamo/sdk/lib/decorators.py

@@ -19,18 +19,17 @@ import typing as t
 from functools import wraps
 from typing import Any, get_type_hints
 
-import bentoml
 from pydantic import BaseModel
 
 
 class DynamoEndpoint:
     """Decorator class for Dynamo endpoints"""
 
-    def __init__(self, func: t.Callable, name: str | None = None):
+    def __init__(self, func: t.Callable, name: str | None = None, is_api: bool = False):
         self.func = func
         self.name = name or func.__name__
         self.is_dynamo_endpoint = True
-
+        self.is_api = is_api
         # Extract request type from hints
         hints = get_type_hints(func)
         args = list(hints.items())
@@ -60,11 +59,13 @@ class DynamoEndpoint:
 
 def dynamo_endpoint(
     name: str | None = None,
+    is_api: bool = False,
 ) -> t.Callable[[t.Callable], DynamoEndpoint]:
     """Decorator for Dynamo endpoints.
 
     Args:
         name: Optional name for the endpoint. Defaults to function name.
+        is_api: Whether to expose the endpoint as an API. Defaults to False.
 
     Example:
         @dynamo_endpoint()
@@ -77,25 +78,13 @@ def dynamo_endpoint(
     """
 
     def decorator(func: t.Callable) -> DynamoEndpoint:
-        return DynamoEndpoint(func, name)
+        return DynamoEndpoint(func, name, is_api)
 
     return decorator
 
 
-def dynamo_api(func: t.Callable) -> t.Callable:
-    """Decorator for BentoML API endpoints.
-
-    Args:
-        func: The function to be decorated.
-
-    Returns:
-        The decorated function.
-    """
-    return bentoml.api(func)
-
-
 def async_on_start(func: t.Callable) -> t.Callable:
     """Decorator for async onstart functions."""
     # Mark the function as a startup hook
-    setattr(func, "__bentoml_startup_hook__", True)
-    return bentoml.on_startup(func)
+    setattr(func, "__dynamo_startup_hook__", True)
+    return func

deploy/dynamo/sdk/src/dynamo/sdk/lib/exceptions.py

+#  SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#  SPDX-License-Identifier: Apache-2.0
+#  #
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#  #
+#  http://www.apache.org/licenses/LICENSE-2.0
+#  #
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+
+from __future__ import annotations
+
+from http import HTTPStatus
+
+
+class DynamoException(Exception):
+    """Base class for all Dynamo SDK Exception."""
+
+    error_code = HTTPStatus.INTERNAL_SERVER_ERROR
+    error_mapping: dict[HTTPStatus, type[DynamoException]] = {}
+
+    def __init_subclass__(cls) -> None:
+        if "error_code" in cls.__dict__:
+            cls.error_mapping[cls.error_code] = cls
+
+    def __init__(self, message: str, error_code: HTTPStatus | None = None):
+        super().__init__(message)
+        self.message = message
+        self.error_code = error_code or self.error_code

deploy/dynamo/sdk/src/dynamo/sdk/lib/image.py

@@ -13,13 +13,6 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 
-# wrapper over bento images to handle Dynamo base image
-
 import os
 
-import bentoml
-
-# TODO: "dynamo:latest-vllm-dev" image will not be available to image builder in k8s
-# so We'd consider publishing the base image for releases to public nvcr.io registry.
-image_name = os.getenv("DYNAMO_IMAGE", "dynamo:latest-vllm-dev")
-DYNAMO_IMAGE = bentoml.images.Image(base_image=image_name)
+DYNAMO_IMAGE = os.getenv("DYNAMO_IMAGE", "dynamo:latest-vllm")

deploy/dynamo/sdk/src/dynamo/sdk/lib/loader.py

+#  SPDX-FileCopyrightText: Copyright (c) 2020 Atalaya Tech. Inc
+#  SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#  SPDX-License-Identifier: Apache-2.0
+#  #
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#  #
+#  http://www.apache.org/licenses/LICENSE-2.0
+#  #
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#  Modifications Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES
+
+from __future__ import annotations
+
+import importlib
+import logging
+import os
+import sys
+from typing import Optional, TypeVar
+
+from dynamo.sdk.lib.service import DynamoService
+
+logger = logging.getLogger(__name__)
+T = TypeVar("T", bound=object)
+
+
+def find_and_load_service(
+    import_str: str,
+    working_dir: Optional[str] = None,
+) -> DynamoService:
+    """Load a DynamoService instance from source code by providing an import string.
+
+    Args:
+        import_str: String in format "module[:attribute]" or "path/to/file.py[:attribute]"
+            Examples:
+                "graphs:disagg:Frontend"
+                "fraud_detector:svc"
+                "./path/to/service.py:MyService"
+                "fraud_detector"  # Will find the root service if only one exists
+        working_dir: Optional directory to use as base for imports. Defaults to cwd.
+
+    Returns:
+        The loaded DynamoService instance
+
+    Raises:
+        ImportError: If module cannot be imported
+        ValueError: If service cannot be found or multiple root services exist
+    """
+    logger.info(f"Loading service from import string: {import_str}")
+    logger.info(f"Working directory: {working_dir or os.getcwd()}")
+
+    sys_path_modified = False
+    prev_cwd = None
+
+    if working_dir is not None:
+        prev_cwd = os.getcwd()
+        working_dir = os.path.realpath(os.path.expanduser(working_dir))
+        logger.info(f"Changing working directory to: {working_dir}")
+        os.chdir(working_dir)
+    else:
+        working_dir = os.getcwd()
+
+    if working_dir not in sys.path:
+        logger.info(f"Adding {working_dir} to sys.path")
+        sys.path.insert(0, working_dir)
+        sys_path_modified = True
+
+    try:
+        return _do_import(import_str, working_dir)
+    finally:
+        if sys_path_modified and working_dir:
+            logger.info(f"Removing {working_dir} from sys.path")
+            sys.path.remove(working_dir)
+        if prev_cwd is not None:
+            logger.info(f"Restoring working directory to: {prev_cwd}")
+            os.chdir(prev_cwd)
+
+
+def _do_import(import_str: str, working_dir: str) -> DynamoService:
+    """Internal function to handle the actual import logic"""
+
+    import_path, _, attrs_str = import_str.partition(":")
+    logger.info(f"Parsed import string - path: {import_path}, attributes: {attrs_str}")
+
+    if not import_path:
+        raise ValueError(
+            f'Invalid import string "{import_str}", must be in format '
+            '"<module>:<attribute>" or "<module>"'
+        )
+
+    # Handle file path vs module name imports
+    if os.path.isfile(import_path):
+        logger.info(f"Importing from file path: {import_path}")
+        import_path = os.path.realpath(import_path)
+        if not import_path.startswith(working_dir):
+            raise ImportError(
+                f'Module "{import_path}" not found in working directory "{working_dir}"'
+            )
+
+        file_name, ext = os.path.splitext(import_path)
+        if ext != ".py":
+            raise ImportError(
+                f'Invalid module extension "{ext}", only ".py" files are supported'
+            )
+
+        # Build module name from path components
+        module_parts = []
+        path = file_name
+        while True:
+            path, name = os.path.split(path)
+            module_parts.append(name)
+            if (
+                not os.path.exists(os.path.join(path, "__init__.py"))
+                or path == working_dir
+            ):
+                break
+        module_name = ".".join(module_parts[::-1])
+        logger.info(f"Constructed module name from path: {module_name}")
+    else:
+        logger.info(f"Importing from module name: {import_path}")
+        module_name = import_path
+
+    try:
+        logger.info(f"Attempting to import module: {module_name}")
+        module = importlib.import_module(module_name)
+    except ImportError as e:
+        raise ImportError(f'Failed to import module "{module_name}": {e}')
+
+    # If no specific attribute given, find the root service
+    if not attrs_str:
+        logger.info("No attributes specified, searching for root service")
+        services = [
+            (name, obj)
+            for name, obj in module.__dict__.items()
+            if isinstance(obj, DynamoService)
+        ]
+        logger.info(f"Found {len(services)} DynamoService instances")
+
+        if not services:
+            raise ValueError(
+                f"No DynamoService instances found in module '{module_name}'"
+            )
+
+        # Find root services (those that aren't dependencies of other services)
+        dependents = set()
+        for _, svc in services:
+            for dep in svc.dependencies.values():
+                if dep.on is not None:
+                    dependents.add(dep.on)
+
+        root_services = [(n, s) for n, s in services if s not in dependents]
+        logger.info(f"Found {len(root_services)} root services")
+
+        if not root_services:
+            raise ValueError(
+                f"No root DynamoService found in module '{module_name}'. "
+                "All services are dependencies of other services."
+            )
+        if len(root_services) > 1:
+            names = [n for n, _ in root_services]
+            raise ValueError(
+                f"Multiple root services found in module '{module_name}': {names}. "
+                "Please specify which service to use with '<module>:<service_name>'"
+            )
+
+        _, instance = root_services[0]
+        logger.info(f"Selected root service: {instance}")
+    else:
+        # Navigate through dot-separated attributes
+        logger.info(f"Navigating attributes: {attrs_str}")
+        instance = module
+        for attr in attrs_str.split("."):
+            try:
+                if isinstance(instance, DynamoService):
+                    logger.info(f"Following dependency link: {attr}")
+                    instance = instance.dependencies[attr].on
+                else:
+                    logger.info(f"Getting attribute: {attr}")
+                    instance = getattr(instance, attr)
+            except (AttributeError, KeyError):
+                raise ValueError(f'Attribute "{attr}" not found in "{module_name}"')
+
+    if not isinstance(instance, DynamoService):
+        raise ValueError(
+            f'Object "{attrs_str}" in module "{module_name}" is not a DynamoService'
+        )
+
+    # Set import string for debugging/logging
+    if not hasattr(instance, "_import_str"):
+        import_str_val = f"{module_name}:{attrs_str}" if attrs_str else module_name
+        logger.info(f"Setting _import_str to: {import_str_val}")
+        object.__setattr__(instance, "_import_str", import_str_val)
+
+    return instance

deploy/dynamo/sdk/src/dynamo/sdk/lib/resource.py

+#  SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#  SPDX-License-Identifier: Apache-2.0
+#  #
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#  #
+#  http://www.apache.org/licenses/LICENSE-2.0
+#  #
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+
+# TODO: this should be used for planner as well and should leverage proper nvml bindings
+
+from __future__ import annotations
+
+import logging
+import typing as t
+from dataclasses import dataclass
+
+import psutil
+
+try:
+    import pynvml
+
+    PYNVML_AVAILABLE = True
+except (ImportError, ModuleNotFoundError):
+    PYNVML_AVAILABLE = False
+
+logger = logging.getLogger(__name__)
+
+# Constants
+NVIDIA_GPU = "nvidia.com/gpu"
+
+
+class ResourceError(Exception):
+    """Base exception for resource-related errors."""
+
+    pass
+
+
+@dataclass
+class GPUProcess:
+    """Information about a process running on a GPU."""
+
+    pid: int
+    used_memory: int  # in bytes
+    name: str = ""
+
+    def __post_init__(self):
+        """Get process name if available."""
+        try:
+            self.name = psutil.Process(self.pid).name()
+        except (psutil.NoSuchProcess, psutil.AccessDenied):
+            pass
+
+
+class GPUInfo:
+    """Information about a specific GPU device."""
+
+    def __init__(self, index: int, total_memory: int, name: str, uuid: str):
+        self.index = index
+        self.total_memory = total_memory  # in bytes
+        self.name = name
+        self.uuid = uuid
+        self.available = True  # Can be set to False if GPU is reserved/in use
+        self.temperature = 0  # in Celsius
+        self.utilization = 0  # in percent (0-100)
+        self.processes: list[GPUProcess] = []
+
+    def __repr__(self) -> str:
+        return f"GPUInfo(index={self.index}, name='{self.name}', total_memory={self.total_memory/1024/1024:.0f}MB, available={self.available})"
+
+
+class GPUManager:
+    """
+    Manages GPU resources using NVML.
+
+    This class provides methods to:
+    - Discover available GPUs
+    - Query GPU properties and status
+    - Track GPU processes
+    - Allocate and release GPUs
+    - Generate CUDA_VISIBLE_DEVICES environment variables
+    """
+
+    def __init__(self):
+        """Initialize the GPU manager."""
+        self.gpus: list[GPUInfo] = []
+        self._initialized = False
+        # List to track fractional GPU allocations
+        # Each item is (gpu_index, fraction_used, fraction_size)
+        # E.g. (0, 0.5, 0.5) means GPU 0 has 0.5 used with fraction size of 0.5
+        self._gpu_fractions: list[tuple[int, float, float]] = []
+        self._init_nvml()
+
+    def _init_nvml(self):
+        """Initialize NVML and discover GPUs."""
+        if not PYNVML_AVAILABLE:
+            logger.warning("PyNVML not available. GPU functionality will be limited.")
+            return
+
+        try:
+            pynvml.nvmlInit()
+            self._initialized = True
+            self._discover_gpus()
+        except (
+            pynvml.NVMLError_LibraryNotFound,
+            pynvml.NVMLError_DriverNotLoaded,
+            OSError,
+        ) as e:
+            logger.warning(f"Failed to initialize NVML: {e}")
+            self._initialized = False
+
+    def __del__(self):
+        """Clean up NVML."""
+        if self._initialized:
+            try:
+                pynvml.nvmlShutdown()
+            except Exception:  # pylint: disable=broad-except
+                pass
+
+    def _discover_gpus(self):
+        """Discover available GPUs and their properties."""
+        if not self._initialized:
+            return
+
+        try:
+            device_count = pynvml.nvmlDeviceGetCount()
+            self.gpus = []
+
+            for i in range(device_count):
+                handle = pynvml.nvmlDeviceGetHandleByIndex(i)
+                name = pynvml.nvmlDeviceGetName(handle)
+                memory_info = pynvml.nvmlDeviceGetMemoryInfo(handle)
+                uuid = pynvml.nvmlDeviceGetUUID(handle)
+
+                gpu_info = GPUInfo(
+                    index=i, total_memory=memory_info.total, name=name, uuid=uuid
+                )
+
+                # Get additional GPU information if available
+                try:
+                    gpu_info.temperature = pynvml.nvmlDeviceGetTemperature(
+                        handle, pynvml.NVML_TEMPERATURE_GPU
+                    )
+                except pynvml.NVMLError:
+                    logger.debug(f"Could not get temperature for GPU {i}")
+
+                try:
+                    utilization = pynvml.nvmlDeviceGetUtilizationRates(handle)
+                    gpu_info.utilization = utilization.gpu
+                except pynvml.NVMLError:
+                    logger.debug(f"Could not get utilization for GPU {i}")
+
+                # Get processes running on GPU
+                try:
+                    processes = pynvml.nvmlDeviceGetComputeRunningProcesses(handle)
+                    gpu_info.processes = [
+                        GPUProcess(pid=p.pid, used_memory=p.usedGpuMemory)
+                        for p in processes
+                    ]
+                except pynvml.NVMLError:
+                    logger.debug(f"Could not get processes for GPU {i}")
+
+                self.gpus.append(gpu_info)
+
+            logger.info(f"Discovered {len(self.gpus)} GPUs")
+        except pynvml.NVMLError as e:
+            logger.warning(f"Error discovering GPUs: {e}")
+
+    def update_gpu_stats(self):
+        """Update GPU statistics (utilization, memory, temperature, etc.)."""
+        if not self._initialized:
+            return
+
+        for gpu in self.gpus:
+            try:
+                handle = pynvml.nvmlDeviceGetHandleByIndex(gpu.index)
+
+                # Update memory info
+                memory_info = pynvml.nvmlDeviceGetMemoryInfo(handle)
+                gpu.total_memory = memory_info.total
+
+                # Update temperature
+                try:
+                    gpu.temperature = pynvml.nvmlDeviceGetTemperature(
+                        handle, pynvml.NVML_TEMPERATURE_GPU
+                    )
+                except pynvml.NVMLError:
+                    pass
+
+                # Update utilization
+                try:
+                    utilization = pynvml.nvmlDeviceGetUtilizationRates(handle)
+                    gpu.utilization = utilization.gpu
+                except pynvml.NVMLError:
+                    pass
+
+                # Update processes
+                try:
+                    processes = pynvml.nvmlDeviceGetComputeRunningProcesses(handle)
+                    gpu.processes = [
+                        GPUProcess(pid=p.pid, used_memory=p.usedGpuMemory)
+                        for p in processes
+                    ]
+                except pynvml.NVMLError:
+                    pass
+
+            except pynvml.NVMLError as e:
+                logger.warning(f"Error updating GPU {gpu.index} stats: {e}")
+
+    def get_gpu_count(self) -> int:
+        """Return the number of available GPUs."""
+        return len(self.gpus)
+
+    def get_available_gpus(self) -> list[int]:
+        """Return a list of available GPU indices."""
+        return [gpu.index for gpu in self.gpus if gpu.available]
+
+    def get_gpu_memory(self, index: int) -> tuple[int, int]:
+        """
+        Return (total memory, free memory) in bytes for a specific GPU.
+
+        Args:
+            index: GPU index
+
+        Returns:
+            Tuple of (total memory, free memory) in bytes
+        """
+        if not self._initialized or index >= len(self.gpus):
+            return (0, 0)
+
+        try:
+            handle = pynvml.nvmlDeviceGetHandleByIndex(index)
+            memory_info = pynvml.nvmlDeviceGetMemoryInfo(handle)
+            return (memory_info.total, memory_info.free)
+        except pynvml.NVMLError as e:
+            logger.warning(f"Error getting GPU memory for GPU {index}: {e}")
+            return (0, 0)
+
+    def get_gpu_utilization(self, index: int) -> int:
+        """
+        Return GPU utilization percentage for a specific GPU.
+
+        Args:
+            index: GPU index
+
+        Returns:
+            GPU utilization percentage (0-100)
+        """
+        if not self._initialized or index >= len(self.gpus):
+            return 0
+
+        try:
+            handle = pynvml.nvmlDeviceGetHandleByIndex(index)
+            utilization = pynvml.nvmlDeviceGetUtilizationRates(handle)
+            return utilization.gpu  # Returns GPU utilization percentage (0-100)
+        except pynvml.NVMLError as e:
+            logger.warning(f"Error getting GPU utilization for GPU {index}: {e}")
+            return 0
+
+    def get_gpu_temperature(self, index: int) -> int:
+        """
+        Return GPU temperature for a specific GPU.
+
+        Args:
+            index: GPU index
+
+        Returns:
+            GPU temperature in Celsius
+        """
+        if not self._initialized or index >= len(self.gpus):
+            return 0
+
+        try:
+            handle = pynvml.nvmlDeviceGetHandleByIndex(index)
+            return pynvml.nvmlDeviceGetTemperature(handle, pynvml.NVML_TEMPERATURE_GPU)
+        except pynvml.NVMLError as e:
+            logger.warning(f"Error getting GPU temperature for GPU {index}: {e}")
+            return 0
+
+    def get_gpu_processes(self, index: int) -> list[GPUProcess]:
+        """
+        Return processes running on a specific GPU.
+
+        Args:
+            index: GPU index
+
+        Returns:
+            List of processes running on the GPU
+        """
+        if not self._initialized or index >= len(self.gpus):
+            return []
+
+        try:
+            handle = pynvml.nvmlDeviceGetHandleByIndex(index)
+            processes = pynvml.nvmlDeviceGetComputeRunningProcesses(handle)
+            return [
+                GPUProcess(pid=p.pid, used_memory=p.usedGpuMemory) for p in processes
+            ]
+        except pynvml.NVMLError as e:
+            logger.warning(f"Error getting GPU processes for GPU {index}: {e}")
+            return []
+
+    def assign_gpus(self, count: float) -> list[int]:
+        """
+        Assign GPUs for use. It can handle fractional GPU requests.
+
+        Args:
+            count: Number of GPUs to assign (can be fractional)
+
+        Returns:
+            List of GPU indices that were assigned
+        """
+        available_gpus = self.get_available_gpus()
+
+        if count > len(available_gpus):
+            logger.warning(
+                f"Requested {count} GPUs, but only {len(available_gpus)} are available. "
+                "Service may fail due to inadequate GPU resources."
+            )
+
+        # Handle fractional GPU allocation
+        if count < 1:
+            # Try to find a GPU with the same fraction size
+            try:
+                # Find a GPU where we've already used the same fraction size
+                gpu_idx, used_fraction = next(
+                    (idx, used)
+                    for idx, used, frac_size in self._gpu_fractions
+                    if frac_size == count and used < 1.0
+                )
+
+                # Update the usage for this GPU
+                for i, (idx, used, frac_size) in enumerate(self._gpu_fractions):
+                    if idx == gpu_idx and frac_size == count:
+                        new_used = used + count
+                        if new_used > 1.0:
+                            new_used = 1.0  # Cap at 1.0
+                        self._gpu_fractions[i] = (idx, new_used, frac_size)
+                        break
+
+                return [gpu_idx]
+            except StopIteration:
+                # No existing fraction of this size, find a free GPU
+                if available_gpus:
+                    gpu_idx = available_gpus[0]
+                    self._gpu_fractions.append((gpu_idx, count, count))
+                    return [gpu_idx]
+                else:
+                    # No available GPUs, return the first GPU (or log warning)
+                    if self.gpus:
+                        logger.warning("No available GPUs, using GPU 0 by default")
+                        self._gpu_fractions.append((0, count, count))
+                        return [0]
+                    else:
+                        logger.error("No GPUs available for allocation")
+                        return []
+
+        # Integer GPU allocation
+        if count >= 1:
+            if int(count) != count:
+                raise ResourceError(
+                    "Fractional GPU count greater than 1 is not supported"
+                )
+
+            count_int = int(count)
+            assigned_gpus = available_gpus[:count_int]
+
+            # Mark these GPUs as fully used
+            for gpu_idx in assigned_gpus:
+                # Check if this GPU is already in _gpu_fractions
+                if not any(idx == gpu_idx for idx, _, _ in self._gpu_fractions):
+                    self._gpu_fractions.append((gpu_idx, 1.0, 1.0))
+                else:
+                    # Update the existing entry
+                    for i, (idx, _, frac_size) in enumerate(self._gpu_fractions):
+                        if idx == gpu_idx:
+                            self._gpu_fractions[i] = (idx, 1.0, frac_size)
+
+                # Mark this GPU as unavailable for future requests
+                for gpu in self.gpus:
+                    if gpu.index == gpu_idx:
+                        gpu.available = False
+
+            return assigned_gpus
+
+        return []
+
+    def get_best_gpu_for_memory(self, required_memory: int) -> int:
+        """
+        Return the index of the GPU with the most available memory that meets the requirement.
+
+        Args:
+            required_memory: Required memory in bytes
+
+        Returns:
+            GPU index, or -1 if no suitable GPU was found
+        """
+        if not self._initialized:
+            return -1
+
+        best_gpu = -1
+        max_free = 0
+
+        for gpu in self.gpus:
+            if not gpu.available:
+                continue
+
+            _, free = self.get_gpu_memory(gpu.index)
+            if free > required_memory and free > max_free:
+                max_free = free
+                best_gpu = gpu.index
+
+        return best_gpu
+
+    def reset_allocations(self):
+        """Reset all GPU allocations."""
+        self._gpu_fractions = []
+        for gpu in self.gpus:
+            gpu.available = True
+
+    def get_gpu_stats(self) -> list[dict[str, t.Any]]:
+        """
+        Get detailed statistics for all GPUs.
+
+        Returns:
+            List of dictionaries with GPU statistics
+        """
+        self.update_gpu_stats()
+
+        stats = []
+        for gpu in self.gpus:
+            total_memory, free_memory = self.get_gpu_memory(gpu.index)
+            stats.append(
+                {
+                    "index": gpu.index,
+                    "name": gpu.name,
+                    "uuid": gpu.uuid,
+                    "total_memory": total_memory,
+                    "free_memory": free_memory,
+                    "used_memory": total_memory - free_memory,
+                    "memory_utilization": (total_memory - free_memory)
+                    / total_memory
+                    * 100
+                    if total_memory > 0
+                    else 0,
+                    "gpu_utilization": gpu.utilization,
+                    "temperature": gpu.temperature,
+                    "process_count": len(gpu.processes),
+                    "processes": [
+                        {
+                            "pid": process.pid,
+                            "name": process.name,
+                            "used_memory": process.used_memory,
+                        }
+                        for process in gpu.processes
+                    ],
+                    "available": gpu.available,
+                }
+            )
+
+        return stats
+
+
+def system_resources() -> dict[str, t.Any]:
+    """
+    Get available system resources (CPU and GPU).
+
+    Returns:
+        Dictionary of resources with keys 'cpu' and 'nvidia.com/gpu'
+    """
+    resources = {}
+
+    # Get GPU resources
+    gpu_manager = GPUManager()
+    resources[NVIDIA_GPU] = gpu_manager.get_available_gpus()
+
+    return resources

deploy/dynamo/sdk/src/dynamo/sdk/lib/service.py

@@ -25,6 +25,7 @@ from typing import Any, Dict, List, Optional, Set, Tuple, TypeVar, Union
 from _bentoml_sdk import Service, ServiceConfig
 from _bentoml_sdk.images import Image
 from _bentoml_sdk.service.config import validate
+from fastapi import FastAPI
 
 from dynamo.sdk.lib.decorators import DynamoEndpoint
 
@@ -86,9 +87,11 @@ class DynamoService(Service[T]):
         image: Optional[Image] = None,
         envs: Optional[list[dict[str, Any]]] = None,
         dynamo_config: Optional[DynamoConfig] = None,
+        app: Optional[FastAPI] = None,
     ):
         service_name = inner.__name__
         service_args = self._get_service_args(service_name)
+        self.app = app
 
         if service_args:
             # Validate and merge service args with existing config
@@ -224,14 +227,91 @@ class DynamoService(Service[T]):
                     }
                     os.environ["DYNAMO_SERVICE_ENVS"] = json.dumps(envs_config)
 
+    def inject_config(self) -> None:
+        """Inject configuration from environment into service configs.
+
+        This reads from DYNAMO_SERVICE_CONFIG environment variable and merges
+        the configuration with any existing service config.
+        """
+        # Get service configs from environment
+        service_config_str = os.environ.get("DYNAMO_SERVICE_CONFIG")
+        if not service_config_str:
+            logger.debug("No DYNAMO_SERVICE_CONFIG found in environment")
+            return
+
+        try:
+            service_configs = json.loads(service_config_str)
+            logger.debug(f"Loaded service configs: {service_configs}")
+        except json.JSONDecodeError as e:
+            logger.error(f"Failed to parse DYNAMO_SERVICE_CONFIG: {e}")
+            return
+
+        # Store the entire config at class level
+        if not hasattr(DynamoService, "_global_service_configs"):
+            setattr(DynamoService, "_global_service_configs", {})
+        DynamoService._global_service_configs = service_configs
+
+        # Process ServiceArgs for all services
+        all_services = self.all_services()
+        logger.debug(f"Processing configs for services: {list(all_services.keys())}")
+
+        for name, svc in all_services.items():
+            if name in service_configs:
+                svc_config = service_configs[name]
+                # Extract ServiceArgs if present
+                if "ServiceArgs" in svc_config:
+                    logger.debug(
+                        f"Found ServiceArgs for {name}: {svc_config['ServiceArgs']}"
+                    )
+                    if not hasattr(svc, "_service_args"):
+                        object.__setattr__(svc, "_service_args", {})
+                    svc._service_args = svc_config["ServiceArgs"]
+                else:
+                    logger.debug(f"No ServiceArgs found for {name}")
+                    # Set default config
+                    if not hasattr(svc, "_service_args"):
+                        object.__setattr__(svc, "_service_args", {"workers": 1})
+
+    def get_service_configs(self) -> Dict[str, Dict[str, Any]]:
+        """Get the service configurations for resource allocation.
+
+        Returns:
+            Dict mapping service names to their configs
+        """
+        # Get all services in the dependency chain
+        all_services = self.all_services()
+        result = {}
+
+        # If we have global configs, use them to build service configs
+        if hasattr(DynamoService, "_global_service_configs"):
+            for name, svc in all_services.items():
+                # Start with default config
+                config = {"workers": 1}
+
+                # If service has specific args, use them
+                if hasattr(svc, "_service_args"):
+                    config.update(svc._service_args)
+
+                # If there are global configs for this service, get ServiceArgs
+                if name in DynamoService._global_service_configs:
+                    svc_config = DynamoService._global_service_configs[name]
+                    if "ServiceArgs" in svc_config:
+                        config.update(svc_config["ServiceArgs"])
+
+                result[name] = config
+                logger.debug(f"Built config for {name}: {config}")
+
+        return result
+
 
 def service(
     inner: Optional[type[T]] = None,
     /,
     *,
-    image: Optional[Image] = None,
+    image: Optional[str] = None,
     envs: Optional[list[dict[str, Any]]] = None,
     dynamo: Optional[Union[Dict[str, Any], DynamoConfig]] = None,
+    app: Optional[FastAPI] = None,
     **kwargs: Any,
 ) -> Any:
     """Enhanced service decorator that supports Dynamo configuration
@@ -262,6 +342,7 @@ def service(
             image=image,
             envs=envs or [],
             dynamo_config=dynamo_config,
+            app=app,
         )
 
     return decorator(inner) if inner is not None else decorator

deploy/dynamo/sdk/src/dynamo/sdk/tests/pipeline.py

@@ -13,13 +13,15 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# This is a simple example of a pipeline that uses Dynamo to deploy a backend, middle, and frontend service. Use this to test
-# changes made to CLI, SDK, etc
+# This is a simple example of a pipeline that uses Dynamo to deploy a backend, middle, and frontend service.
+# Use this to test changes made to CLI, SDK, etc
 
 
+from fastapi import FastAPI
+from fastapi.responses import StreamingResponse
 from pydantic import BaseModel
 
-from dynamo.sdk import api, depends, dynamo_endpoint, service
+from dynamo.sdk import depends, dynamo_endpoint, service
 
 """
 Pipeline Architecture:
@@ -54,6 +56,9 @@ class ResponseType(BaseModel):
 GPU_ENABLED = False
 
 
+app = FastAPI(title="Hello World!")
+
+
 @service(
     resources={"cpu": "1"},
     traffic={"timeout": 30},
@@ -130,7 +135,12 @@ class Middle:
                 yield f"Frontend: {back_resp}"
 
 
-@service(resources={"cpu": "1"}, traffic={"timeout": 60})
+@service(
+    resources={"cpu": "1"},
+    traffic={"timeout": 60},
+    dynamo={"enabled": True, "namespace": "inference"},
+    app=app,
+)
 class Frontend:
     middle = depends(Middle)
     backend = depends(Backend)
@@ -138,13 +148,13 @@ class Frontend:
     def __init__(self) -> None:
         print("Starting frontend")
 
-    @api
-    async def generate(self, text):
+    @dynamo_endpoint(is_api=True)
+    async def generate(self, request: RequestType):
         """Stream results from the pipeline."""
-        print(f"Frontend received: {text}")
-        print(f"Frontend received type: {type(text)}")
-        txt = RequestType(text=text)
-        print(f"Frontend sending: {type(txt)}")
-        async for mid_resp in self.middle.generate(txt.model_dump_json()):
-            print(f"Frontend received mid_resp: {mid_resp}")
-            yield f"Frontend: {mid_resp}"
+        print(f"Frontend received: {request.text}")
+
+        async def content_generator():
+            async for response in self.middle.generate(request.model_dump_json()):
+                yield f"Frontend: {response}"
+
+        return StreamingResponse(content_generator())

deploy/dynamo/sdk/src/dynamo/sdk/tests/test_e2e.py

@@ -94,7 +94,7 @@ async def test_pipeline(setup_and_teardown):
         try:
             async with aiohttp.ClientSession() as session:
                 async with session.post(
-                    "http://localhost:3000/generate",
+                    "http://localhost:8000/generate",
                     json={"text": "federer-is-the-greatest-tennis-player-of-all-time"},
                     headers={"accept": "text/event-stream"},
                 ) as resp:

examples/hello_world/README.md

@@ -79,7 +79,7 @@ The `dynamo serve` command deploys the entire service graph, automatically handl
 
 ```bash
 curl -X 'POST' \
-  'http://localhost:3000/generate' \
+  'http://localhost:8000/generate' \
   -H 'accept: text/event-stream' \
   -H 'Content-Type: application/json' \
   -d '{
@@ -173,10 +173,10 @@ Once you create the Dynamo deployment, a pod prefixed with `yatai-dynamonim-imag
 
 ```bash
 # Forward the service port to localhost
-kubectl -n ${KUBE_NS} port-forward svc/${HELM_RELEASE}-frontend 3000:3000
+kubectl -n ${KUBE_NS} port-forward svc/${HELM_RELEASE}-frontend 8000:8000
 
 # Test the API endpoint
-curl -X 'POST' 'http://localhost:3000/generate' \
+curl -X 'POST' 'http://localhost:8000/generate' \
     -H 'accept: text/event-stream' \
     -H 'Content-Type: application/json' \
     -d '{"text": "test"}'