manta-node 0.5b0.dev81__py3-none-any.whl → 0.5b0.dev83__py3-none-any.whl

manta_node/domain/task_lifecycle.py

@@ -63,6 +63,13 @@ class TaskLifecycleService:
         }
         return status_map.get(container_status, TaskStatus.FAILED)
 
+    @staticmethod
+    def status_from_exit(container_status: str, exit_code: int) -> TaskStatus:
+        """For image-mode tasks: exit code is the authoritative completion signal."""
+        if container_status == "exited":
+            return TaskStatus.COMPLETED if exit_code == 0 else TaskStatus.FAILED
+        return TaskLifecycleService.calculate_status_from_container(container_status)
+
     @staticmethod
     def should_restart_task(
         current_iteration: int,

manta_node/infrastructure/container/__init__.py

@@ -1,9 +1,11 @@
 """Container infrastructure - Docker runtime management."""
 
 from .docker_adapter import DockerRaiser
+from .image_executor import ImageExecutor
 from .manager import ContainerManager
 
 __all__ = [
     "ContainerManager",
     "DockerRaiser",
+    "ImageExecutor",
 ]

manta_node/infrastructure/container/image_executor.py

@@ -0,0 +1,86 @@
+from datetime import datetime
+from typing import Dict, Optional, Tuple
+
+from aiodocker.docker import DockerContainer
+
+from manta_common.build.common.tasks import TaskStatus
+from manta_common.conversions import ID
+from manta_common.errors import MantaDockerError
+from manta_common.traces import Tracer
+
+from ...domain.task_lifecycle import TaskLifecycleService
+from .docker_adapter import DockerRaiser
+
+
+class ImageExecutor(DockerRaiser):
+    """
+    Runs plain Docker images with native ENTRYPOINT/CMD.
+    No zipapp, no manta SDK, no gRPC inside the container.
+    """
+
+    tracer: Tracer
+    containers: Dict[Tuple[ID, ID], DockerContainer]
+
+    async def run_image(
+        self,
+        swarm_id: ID,
+        task_id: ID,
+        image: str,
+        environment: Optional[dict] = None,
+    ) -> datetime:
+        """
+        Pull (if not present locally) and start a Docker image with its native ENTRYPOINT.
+
+        Returns
+        -------
+        datetime
+            Container start time
+        """
+        try:
+            await self.async_client.images.inspect(image)
+        except Exception:
+            self.tracer.info(f"Image {image} not found locally, pulling...")
+            await self.pull_image_or_raise(image)
+
+        env = [f"{k}={v}" for k, v in environment.items()] if environment else None
+        config = {
+            "Image": image,
+            "HostConfig": {
+                "NetworkMode": "host",
+                "ExtraHosts": ["host.docker.internal:host-gateway"],
+            },
+            "Env": env,
+            "Detach": True,
+            "LogConfig": {
+                "type": "json-file",
+                "config": {"max-size": "10m", "max-file": "3"},
+            },
+            "Labels": {
+                "com.docker.compose.project": "manta_tasks",
+            },
+        }
+
+        start_time = datetime.now()
+        self.tracer.debug(f"Starting image {image} for task {task_id}")
+        self.containers[(swarm_id, task_id)] = await self.run_or_raise(config)
+        self.tracer.info(f"Task {task_id} image container started.")
+        return start_time
+
+    async def get_image_task_status(self, swarm_id: ID, task_id: ID) -> TaskStatus:
+        """
+        Exit-code-aware status for image-mode tasks (no SDK inside the container).
+        """
+        container = self.containers.get((swarm_id, task_id))
+        if container is None:
+            raise MantaDockerError("Container not found", metadata=self.tracer.metadata)
+
+        inspect = await container.show()
+        container_status = inspect["State"]["Status"]
+        exit_code = inspect["State"].get("ExitCode", 0)
+
+        status = TaskLifecycleService.status_from_exit(container_status, exit_code)
+        self.tracer.debug(
+            f"Task {task_id} image status: {container_status} "
+            f"(exit={exit_code}) → {status}"
+        )
+        return status

manta_node/task_manager.py

@@ -3,11 +3,11 @@ from collections import deque
 from datetime import datetime
 from pathlib import Path
 from tempfile import TemporaryDirectory
-from typing import List, Optional, Tuple
+from typing import List, Optional, Set, Tuple
 
 from manta_common.build.common.system import EnvId, PairId
 from manta_common.build.common.tasks import MqttTask, TaskStatus, TaskUpdate
-from manta_common.conversions import ID
+from manta_common.conversions import ID, TupleID
 from manta_common.errors import (
     MantaError,
     MantaGRPCError,
@@ -16,6 +16,7 @@ from manta_common.errors import (
 )
 from manta_common.traces import Tracer
 from .domain.task_lifecycle import TaskLifecycleService
+from .infrastructure.container.image_executor import ImageExecutor
 from .infrastructure.container.manager import ContainerManager
 from .infrastructure.grpc.client import NodeClient
 from .tasks import TaskProgression, TaskProgressions
@@ -23,7 +24,7 @@ from .tasks import TaskProgression, TaskProgressions
 __all__ = ["TaskManager"]
 
 
-class TaskManager(ContainerManager):
+class TaskManager(ContainerManager, ImageExecutor):
     """
     The Task Manager is responsible to distribute tasks depending on device resources.
     It is composed with a Queue of Tasks.
@@ -54,6 +55,7 @@ class TaskManager(ContainerManager):
         # Initialize the task queue
         self.task_queue = deque()
         self.future_task = None
+        self.image_task_ids: Set[Tuple[TupleID, TupleID]] = set()
 
         ContainerManager.__init__(self)
 
@@ -155,6 +157,11 @@ class TaskManager(ContainerManager):
             updated.append(task_progression)
         return updated
 
+    async def get_task_status(self, swarm_id: ID, task_id: ID) -> TaskStatus:
+        if (swarm_id.tid, task_id.tid) in self.image_task_ids:
+            return await self.get_image_task_status(swarm_id, task_id)
+        return await ContainerManager.get_task_status(self, swarm_id, task_id)
+
     async def from_pending_to_running(self, task_progression: TaskProgression) -> bool:
         """
         Change a task progression from pending status (supposed) to running status
@@ -219,56 +226,69 @@ class TaskManager(ContainerManager):
         )
         self.tracer.debug(f"Creating task {task_id} for {swarm_id}")
 
-        # Create a temporary file and write the module code
-        # (even if the file does not exist, it will be created)
-        module_file = self.module_folder / f"{swarm_id.oid}_{task_id.oid}.pyz"
-        module_file.unlink(missing_ok=True)
-        module_file.touch(exist_ok=True)
-        module_file.write_bytes(task.payload)
-
-        container_file = "/usr/src/module.pyz"
-
-        network = None
-        if task.network != str(None):
-            if task.auth_key != str(None) or task.login_server != str(None):
-                self.tracer.debug(
-                    f"Task {task_id} has auth_key {task.auth_key} and login_server {task.login_server}"
-                )
-                network = await self.ensure_tailscale_service(
-                    swarm_id,
-                    task.auth_key,
-                    task.login_server,
-                    task.alias,
-                    self.module_folder,
-                )
-            else:
-                self.tracer.error(
-                    f"Task {task_id} has network {task.network} but no auth_key or login_server"
-                )
-                raise MantaNodeError(
-                    f"Task {task_id} has network {task.network} but no auth_key or login_server",
-                    metadata=self.tracer.metadata,
-                )
-
-        # Deploy the task
-        start_time = await self.run_container(
-            swarm_id=swarm_id,
-            task_id=task_id,
-            image=task.image,
-            command=f"python {container_file}",
-            volumes={module_file.absolute(): {"bind": container_file, "mode": "ro"}},
-            environment={
-                "TASK_ID": task_id.xid,
-                "SWARM_ID": swarm_id.xid,
-                "RPC_PORT": self.ls_port,
-                "RPC_HOST": "host.docker.internal",
-            },
-            gpu=task.gpu,
-            network=network,
-            alias=task.alias,
-        )
+        if not task.payload:
+            # Image execution path: native ENTRYPOINT, no zipapp, no gRPC inside
+            module_file = self.module_folder / f"{swarm_id.oid}_{task_id.oid}.img"
+            start_time = await self.run_image(
+                swarm_id=swarm_id,
+                task_id=task_id,
+                image=task.image,
+                environment={
+                    "TASK_ID": task_id.xid,
+                    "SWARM_ID": swarm_id.xid,
+                },
+            )
+            self.image_task_ids.add((swarm_id.tid, task_id.tid))
+        else:
+            # Zipapp execution path (existing)
+            module_file = self.module_folder / f"{swarm_id.oid}_{task_id.oid}.pyz"
+            module_file.unlink(missing_ok=True)
+            module_file.touch(exist_ok=True)
+            module_file.write_bytes(task.payload)
+
+            container_file = "/usr/src/module.pyz"
+
+            network = None
+            if task.network != str(None):
+                if task.auth_key != str(None) or task.login_server != str(None):
+                    self.tracer.debug(
+                        f"Task {task_id} has auth_key {task.auth_key} and login_server {task.login_server}"
+                    )
+                    network = await self.ensure_tailscale_service(
+                        swarm_id,
+                        task.auth_key,
+                        task.login_server,
+                        task.alias,
+                        self.module_folder,
+                    )
+                else:
+                    self.tracer.error(
+                        f"Task {task_id} has network {task.network} but no auth_key or login_server"
+                    )
+                    raise MantaNodeError(
+                        f"Task {task_id} has network {task.network} but no auth_key or login_server",
+                        metadata=self.tracer.metadata,
+                    )
+
+            start_time = await self.run_container(
+                swarm_id=swarm_id,
+                task_id=task_id,
+                image=task.image,
+                command=f"python {container_file}",
+                volumes={
+                    module_file.absolute(): {"bind": container_file, "mode": "ro"}
+                },
+                environment={
+                    "TASK_ID": task_id.xid,
+                    "SWARM_ID": swarm_id.xid,
+                    "RPC_PORT": self.ls_port,
+                    "RPC_HOST": "host.docker.internal",
+                },
+                gpu=task.gpu,
+                network=network,
+                alias=task.alias,
+            )
 
-        # Update the task information
         self.tasks.append(
             TaskProgression(
                 task_update=TaskUpdate(
@@ -511,9 +531,6 @@ class TaskManager(ContainerManager):
             task_progression.swarm_id,
             task_progression.task_id,
         )
-        self.tasks.pop(
-            (
-                task_progression.swarm_id.tid,
-                task_progression.task_id.tid,
-            )
-        )
+        env_id = (task_progression.swarm_id.tid, task_progression.task_id.tid)
+        self.tasks.pop(env_id)
+        self.image_task_ids.discard(env_id)

{manta_node-0.5b0.dev81.dist-info → manta_node-0.5b0.dev83.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: manta_node
-Version: 0.5b0.dev81
+Version: 0.5b0.dev83
 Summary: Node side software of the Manta platform. It is responsible for the communication with the Manta Core and the execution of the tasks.
 Author-email: Hugo Miralles <hugomiralles@manta-tech.io>, Matthew Thompson <matthew.thompson@manta-tech.io>
 License:                     GNU AFFERO GENERAL PUBLIC LICENSE

{manta_node-0.5b0.dev81.dist-info → manta_node-0.5b0.dev83.dist-info}/RECORD

@@ -1,7 +1,7 @@
 manta_node/__init__.py,sha256=bgGWXH3yQVhNf0v1ZGUBAy5Y51fqb0Y-EaLnocbcino,153
 manta_node/__main__.py,sha256=0VzbY1KfezaARZDuCBc7Th1GO_-BnKTK8eDEXhgP5J4,1731
 manta_node/node_orchestrator.py,sha256=0VMBsZSMQiIe9L-Bd7Jn4avNpNyDCMa1kyPSTgRiOes,17336
-manta_node/task_manager.py,sha256=LFFE83JreNa9YejFa8JTRFkZgVOjFREqdYfQREYD2S4,17406
+manta_node/task_manager.py,sha256=8Yy8ShoKbMemvm3UMPt9xJnoZsxiCptqdWnhAKJnhl4,18498
 manta_node/tasks.py,sha256=W__XjU17AX8VqOtgy8HRBmxWtVufTkFvri8s8Y3R394,6721
 manta_node/utils.py,sha256=gkYRB7fylk1jy1sm7nKU7o8eadGxn0AX4IjCHT6Wzuw,1179
 manta_node/cli/__init__.py,sha256=arzO3mXDJ-E2yeeytbZfN7mCswwxdQ43_QjyeL2E5yE,134
@@ -16,12 +16,13 @@ manta_node/cli/commands/start.py,sha256=49kvB6p01nBTp0yqLX36RiXeEccGYTcl9YMcB2sI
 manta_node/cli/commands/status.py,sha256=9YYUaelmQJWqh6FbVuy6y2y1lvjoPIkA2d18JU0PHls,6430
 manta_node/cli/commands/stop.py,sha256=mcO8AMtaPkpNFEy6lVIe4fvNFJSEkheFdxkj1ySiYaI,7719
 manta_node/domain/__init__.py,sha256=uyJXDf0QgfGCsXLMggqKCB5QV9oneYJ4iM-REaJGXDM,217
-manta_node/domain/task_lifecycle.py,sha256=RRRXYvOraID0mg5TrInbQLeyUGzSeU6HIGUvvo6rbTw,2856
+manta_node/domain/task_lifecycle.py,sha256=bnmarFdgrRoGkDeDlDj6OSOoCT8-qg00OMrny30stNI,3248
 manta_node/infrastructure/__init__.py,sha256=yM60jKhdaD18BOsLb-0aR9D2XxGvrTnV92oOT7Kor2w,297
 manta_node/infrastructure/config/__init__.py,sha256=jbRzb1c0RohpF83WOwtDpe19dBXVfbZYF3SoOaESh6k,617
 manta_node/infrastructure/config/node_config_manager.py,sha256=iAzMWPjs2AWTxwDLAdCK8GpMMw_14QOzX11hvQMnSvY,30414
-manta_node/infrastructure/container/__init__.py,sha256=9I9g5y2ncEAa7PhEsSfL2nFOltfFr-GsBGgn8B2vIUY,199
+manta_node/infrastructure/container/__init__.py,sha256=EO4SndUwmXrHczM4qkleUbxC7lTGz3c3Hd3iGrPg6D8,262
 manta_node/infrastructure/container/docker_adapter.py,sha256=Sxprl0RpjT8epxTzYUJxSyHko3IBcus2ihneq4kRH7Q,6980
+manta_node/infrastructure/container/image_executor.py,sha256=PfLCqzpn8HVsxpPEpZmoQYnFQj4W91jrZUjcntnyzoQ,2881
 manta_node/infrastructure/container/manager.py,sha256=qpHwBOsDdiv10i5Qlx9PUn5UHr1oy8ejRGalTCi6AOQ,17375
 manta_node/infrastructure/filesystem/__init__.py,sha256=nf6Fakp3wuFKjSAm_oihI6WE_yyUY6Uelr7A1GwSvXs,145
 manta_node/infrastructure/filesystem/dataset_manager.py,sha256=lCcsLdDHM-rRfhndpD9fQ56rPKYcuel-p3qx8nUxffE,14894
@@ -34,9 +35,9 @@ manta_node/infrastructure/metrics/collector.py,sha256=nhFPAbjA04ye3UaGN95_zdpf9h
 manta_node/infrastructure/metrics/metrics_collector.py,sha256=jhEocAPWS6xgbNvcBBiEB9cAN_9hEGssZJXoysQzWOk,10518
 manta_node/infrastructure/mqtt/__init__.py,sha256=riXo1sXgO4YV6DHCmWRqK61BAmuGLdHiVdQFunAJadA,140
 manta_node/infrastructure/mqtt/command_handler.py,sha256=bDH8k3Vkbbb7q3P68VamhMbtLJ4vSSivqZP3IfMBqIQ,11833
-manta_node-0.5b0.dev81.dist-info/licenses/LICENSE,sha256=P5HojupjKoAusNtHqWOvO-YFolwLFWQ7Uq0XVKK9nFQ,35291
-manta_node-0.5b0.dev81.dist-info/METADATA,sha256=Cgjc8gVbusw56H8XU8OoWDQi4j1SF1tIggNFF5RWuls,44653
-manta_node-0.5b0.dev81.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
-manta_node-0.5b0.dev81.dist-info/entry_points.txt,sha256=C_zcw6mdmztNxVUZGLZ83UH5TwcSuqb2C-rB4e2CQoc,56
-manta_node-0.5b0.dev81.dist-info/top_level.txt,sha256=nbT5J5WETuXM9UPoQckl29uUOq8wMYdRb9QDx5kAFeo,11
-manta_node-0.5b0.dev81.dist-info/RECORD,,
+manta_node-0.5b0.dev83.dist-info/licenses/LICENSE,sha256=P5HojupjKoAusNtHqWOvO-YFolwLFWQ7Uq0XVKK9nFQ,35291
+manta_node-0.5b0.dev83.dist-info/METADATA,sha256=hcm1pKCE158stT99YW-PdsVoFDNWbbmKtsj2sdFIXu0,44653
+manta_node-0.5b0.dev83.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
+manta_node-0.5b0.dev83.dist-info/entry_points.txt,sha256=C_zcw6mdmztNxVUZGLZ83UH5TwcSuqb2C-rB4e2CQoc,56
+manta_node-0.5b0.dev83.dist-info/top_level.txt,sha256=nbT5J5WETuXM9UPoQckl29uUOq8wMYdRb9QDx5kAFeo,11
+manta_node-0.5b0.dev83.dist-info/RECORD,,