feat(rnd): Split Execution Manager

.dockerignore

@@ -36,4 +36,3 @@ rnd/autogpt_builder/.env.example
 rnd/autogpt_builder/.env.local
 rnd/autogpt_server/.env
 rnd/autogpt_server/.venv/
-

rnd/autogpt_server/.env.example

@@ -11,6 +11,7 @@ REDIS_PASSWORD=password
 
 AUTH_ENABLED=false
 APP_ENV="local"
+PYRO_HOST=localhost
 SENTRY_DSN=
 
 ## ===== OPTIONAL API KEYS ===== ##

rnd/autogpt_server/Dockerfile

@@ -4,7 +4,6 @@ FROM python:3.11-slim-buster as server_base
 ENV PYTHONDONTWRITEBYTECODE 1
 ENV PYTHONUNBUFFERED 1
 
-
 WORKDIR /app
 
 RUN apt-get update \
@@ -17,7 +16,6 @@ RUN apt-get update \
     && make prefix=/usr all \
     && make prefix=/usr install
 
-
 ENV POETRY_VERSION=1.8.3 \
     POETRY_HOME="/opt/poetry" \
     POETRY_NO_INTERACTION=1 \
@@ -44,3 +42,10 @@ ENV PORT=8000
 ENV DATABASE_URL=""
 
 CMD ["poetry", "run", "rest"]
+
+FROM server_base as executor
+
+ENV PORT=8002
+ENV DATABASE_URL=""
+
+CMD ["poetry", "run", "executor"]

rnd/autogpt_server/README.md

@@ -183,6 +183,13 @@ A communication layer (`service.py`) is created to decouple the communication li
 
 Currently, the IPC is done using Pyro5 and abstracted in a way that allows a function decorated with `@expose` to be called from a different process.
 
+
+By default the daemons run on the following ports: 
+
+Execution Manager Daemon: 8002
+Execution Scheduler Daemon: 8003
+Rest Server Daemon: 8004
+
 ## Adding a New Agent Block
 
 To add a new agent block, you need to create a new class that inherits from `Block` and provides the following information:

rnd/autogpt_server/autogpt_server/app.py

@@ -26,10 +26,8 @@ def main(**kwargs):
 
     from autogpt_server.executor import ExecutionManager, ExecutionScheduler
     from autogpt_server.server import AgentServer, WebsocketServer
-    from autogpt_server.util.service import PyroNameServer
 
     run_processes(
-        PyroNameServer(),
         ExecutionManager(),
         ExecutionScheduler(),
         WebsocketServer(),

rnd/autogpt_server/autogpt_server/exec.py

@@ -0,0 +1,15 @@
+from autogpt_server.app import run_processes
+from autogpt_server.executor import ExecutionManager
+
+
+def main():
+    """
+    Run all the processes required for the AutoGPT-server REST API.
+    """
+    run_processes(
+        ExecutionManager(),
+    )
+
+
+if __name__ == "__main__":
+    main()

rnd/autogpt_server/autogpt_server/executor/manager.py

@@ -364,7 +364,7 @@ def validate_exec(
 def get_agent_server_client() -> "AgentServer":
     from autogpt_server.server.rest_api import AgentServer
 
-    return get_service_client(AgentServer)
+    return get_service_client(AgentServer, Config().agent_server_port)
 
 
 class Executor:
@@ -648,6 +648,7 @@ def callback(_):
 
 class ExecutionManager(AppService):
     def __init__(self):
+        super().__init__(port=Config().execution_manager_port)
         self.use_db = True
         self.pool_size = Config().num_graph_workers
         self.queue = ExecutionQueue[GraphExecution]()

rnd/autogpt_server/autogpt_server/executor/scheduler.py

@@ -9,6 +9,7 @@
 from autogpt_server.data.block import BlockInput
 from autogpt_server.executor.manager import ExecutionManager
 from autogpt_server.util.service import AppService, expose, get_service_client
+from autogpt_server.util.settings import Config
 
 logger = logging.getLogger(__name__)
 
@@ -19,13 +20,15 @@ def log(msg, **kwargs):
 
 class ExecutionScheduler(AppService):
     def __init__(self, refresh_interval=10):
+        super().__init__(port=Config().execution_scheduler_port)
         self.use_db = True
         self.last_check = datetime.min
         self.refresh_interval = refresh_interval
+        self.use_redis = False
 
     @property
     def execution_manager_client(self) -> ExecutionManager:
-        return get_service_client(ExecutionManager)
+        return get_service_client(ExecutionManager, Config().execution_manager_port)
 
     def run_service(self):
         scheduler = BackgroundScheduler()

rnd/autogpt_server/autogpt_server/rest.py

@@ -1,16 +1,13 @@
 from autogpt_server.app import run_processes
-from autogpt_server.executor import ExecutionManager, ExecutionScheduler
+from autogpt_server.executor import ExecutionScheduler
 from autogpt_server.server import AgentServer
-from autogpt_server.util.service import PyroNameServer
 
 
 def main():
     """
     Run all the processes required for the AutoGPT-server REST API.
     """
     run_processes(
-        PyroNameServer(),
-        ExecutionManager(),
         ExecutionScheduler(),
         AgentServer(),
     )

rnd/autogpt_server/autogpt_server/server/rest_api.py

@@ -21,7 +21,7 @@
 from autogpt_server.server.model import CreateGraph, SetGraphActiveVersion
 from autogpt_server.util.lock import KeyedMutex
 from autogpt_server.util.service import AppService, expose, get_service_client
-from autogpt_server.util.settings import Settings
+from autogpt_server.util.settings import Config, Settings
 
 from .utils import get_user_id
 
@@ -34,6 +34,7 @@ class AgentServer(AppService):
     _test_dependency_overrides = {}
 
     def __init__(self, event_queue: AsyncEventQueue | None = None):
+        super().__init__(port=Config().agent_server_port)
         self.event_queue = event_queue or AsyncRedisEventQueue()
 
     @asynccontextmanager
@@ -239,11 +240,11 @@ async def wrapper(*args, **kwargs):
 
     @property
     def execution_manager_client(self) -> ExecutionManager:
-        return get_service_client(ExecutionManager)
+        return get_service_client(ExecutionManager, Config().execution_manager_port)
 
     @property
     def execution_scheduler_client(self) -> ExecutionScheduler:
-        return get_service_client(ExecutionScheduler)
+        return get_service_client(ExecutionScheduler, Config().execution_scheduler_port)
 
     @classmethod
     def handle_internal_http_error(cls, request: Request, exc: Exception):

rnd/autogpt_server/autogpt_server/util/service.py

@@ -1,12 +1,13 @@
 import asyncio
 import logging
+import os
 import threading
 import time
 from abc import abstractmethod
 from typing import Any, Callable, Coroutine, Type, TypeVar, cast
 
+import Pyro5.api
 from Pyro5 import api as pyro
-from Pyro5 import nameserver
 
 from autogpt_server.data import db
 from autogpt_server.data.queue import AsyncEventQueue, AsyncRedisEventQueue
@@ -42,25 +43,16 @@ def wrapper(*args, **kwargs):
     return pyro.expose(wrapper)  # type: ignore
 
 
-class PyroNameServer(AppProcess):
-    def run(self):
-        nameserver.start_ns_loop(host=pyro_host, port=9090)
-
-    @conn_retry
-    def _wait_for_ns(self):
-        pyro.locate_ns(host="localhost", port=9090)
-
-    def health_check(self):
-        self._wait_for_ns()
-        logger.info(f"{__class__.__name__} is ready")
-
-
 class AppService(AppProcess):
     shared_event_loop: asyncio.AbstractEventLoop
     event_queue: AsyncEventQueue = AsyncRedisEventQueue()
     use_db: bool = False
     use_redis: bool = False
 
+    def __init__(self, port):
+        self.port = port
+        self.uri = None
+
     @classmethod
     @property
     def service_name(cls) -> str:
@@ -108,11 +100,10 @@ def cleanup(self):
 
     @conn_retry
     def __start_pyro(self):
-        daemon = pyro.Daemon(host=pyro_host)
-        ns = pyro.locate_ns(host=pyro_host, port=9090)
-        uri = daemon.register(self)
-        ns.register(self.service_name, uri)
-        logger.info(f"[{self.service_name}] Connected to Pyro; URI = {uri}")
+        host = Config().pyro_host
+        daemon = Pyro5.api.Daemon(host=host, port=self.port)
+        self.uri = daemon.register(self, objectId=self.service_name)
+        logger.info(f"[{self.service_name}] Connected to Pyro; URI = {self.uri}")
         daemon.requestLoop()
 
     def __start_async_loop(self):
@@ -122,16 +113,19 @@ def __start_async_loop(self):
 AS = TypeVar("AS", bound=AppService)
 
 
-def get_service_client(service_type: Type[AS]) -> AS:
+def get_service_client(service_type: Type[AS], port: int) -> AS:
     service_name = service_type.service_name
 
     class DynamicClient:
         @conn_retry
         def __init__(self):
-            ns = pyro.locate_ns()
-            uri = ns.lookup(service_name)
-            self.proxy = pyro.Proxy(uri)
+            host = os.environ.get(f"{service_name.upper()}_HOST", "localhost")
+            uri = f"PYRO:{service_type.service_name}@{host}:{port}"
+            logger.debug(f"Connecting to service [{service_name}]. URI = {uri}")
+            self.proxy = Pyro5.api.Proxy(uri)
+            # Attempt to bind to ensure the connection is established
             self.proxy._pyroBind()
+            logger.debug(f"Successfully connected to service [{service_name}]")
 
         def __getattr__(self, name: str) -> Callable[..., Any]:
             return getattr(self.proxy, name)

rnd/autogpt_server/autogpt_server/util/settings.py

@@ -72,6 +72,21 @@ class Config(UpdateTrackingModel["Config"], BaseSettings):
         extra="allow",
     )
 
+    execution_manager_port: int = Field(
+        default=8002,
+        description="The port for execution manager daemon to run on",
+    )
+
+    execution_scheduler_port: int = Field(
+        default=8003,
+        description="The port for execution scheduler daemon to run on",
+    )
+
+    agent_server_port: int = Field(
+        default=8004,
+        description="The port for agent server daemon to run on",
+    )
+
     @classmethod
     def settings_customise_sources(
         cls,

rnd/autogpt_server/autogpt_server/util/test.py

@@ -8,7 +8,6 @@
 from autogpt_server.executor import ExecutionManager, ExecutionScheduler
 from autogpt_server.server import AgentServer
 from autogpt_server.server.rest_api import get_user_id
-from autogpt_server.util.service import PyroNameServer
 
 log = print
 
@@ -48,7 +47,6 @@ async def get(self):
 
 class SpinTestServer:
     def __init__(self):
-        self.name_server = PyroNameServer()
         self.exec_manager = ExecutionManager()
         self.in_memory_queue = InMemoryAsyncEventQueue()
         self.agent_server = AgentServer(event_queue=self.in_memory_queue)
@@ -59,7 +57,6 @@ def test_get_user_id():
         return "3e53486c-cf57-477e-ba2a-cb02dc828e1a"
 
     async def __aenter__(self):
-        self.name_server.__enter__()
         self.setup_dependency_overrides()
         self.agent_server.__enter__()
         self.exec_manager.__enter__()
@@ -76,7 +73,6 @@ async def __aexit__(self, exc_type, exc_val, exc_tb):
         self.scheduler.__exit__(exc_type, exc_val, exc_tb)
         self.exec_manager.__exit__(exc_type, exc_val, exc_tb)
         self.agent_server.__exit__(exc_type, exc_val, exc_tb)
-        self.name_server.__exit__(exc_type, exc_val, exc_tb)
 
     def setup_dependency_overrides(self):
         # Override get_user_id for testing

rnd/autogpt_server/pyproject.toml

@@ -66,6 +66,7 @@ build-backend = "poetry.core.masonry.api"
 app = "autogpt_server.app:main"
 rest = "autogpt_server.rest:main"
 ws = "autogpt_server.ws:main"
+executor = "autogpt_server.exec:main"
 cli = "autogpt_server.cli:main"
 format = "linter:format"
 lint = "linter:lint"

rnd/autogpt_server/test/executor/test_scheduler.py

@@ -4,6 +4,7 @@
 from autogpt_server.executor import ExecutionScheduler
 from autogpt_server.usecases.sample import create_test_graph, create_test_user
 from autogpt_server.util.service import get_service_client
+from autogpt_server.util.settings import Config
 from autogpt_server.util.test import SpinTestServer
 
 
@@ -13,7 +14,9 @@ async def test_agent_schedule(server: SpinTestServer):
     test_user = await create_test_user()
     test_graph = await graph.create_graph(create_test_graph(), user_id=test_user.id)
 
-    scheduler = get_service_client(ExecutionScheduler)
+    scheduler = get_service_client(
+        ExecutionScheduler, Config().execution_scheduler_port
+    )
 
     schedules = scheduler.get_execution_schedules(test_graph.id, test_user.id)
     assert len(schedules) == 0

rnd/autogpt_server/test/util/test_service.py

@@ -5,6 +5,7 @@
 
 class TestService(AppService):
     def __init__(self):
+        super().__init__(port=8005)
         self.use_redis = False
 
     def run_service(self):
@@ -29,7 +30,7 @@ async def add_async(a: int, b: int) -> int:
 @pytest.mark.asyncio(scope="session")
 async def test_service_creation(server):
     with TestService():
-        client = get_service_client(TestService)
+        client = get_service_client(TestService, 8005)
         assert client.add(5, 3) == 8
         assert client.subtract(10, 4) == 6
         assert client.fun_with_async(5, 3) == 8