gpustack-runtime 0.1.39__tar.gz → 0.1.39.post1__tar.gz

{gpustack_runtime-0.1.39 → gpustack_runtime-0.1.39.post1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpustack-runtime
-Version: 0.1.39
+Version: 0.1.39.post1
 Summary: GPUStack Runtime is library for detecting GPU resources and launching GPU workloads.
 Project-URL: Homepage, https://github.com/gpustack/runtime
 Project-URL: Bug Tracker, https://github.com/gpustack/gpustack/issues
@@ -15,7 +15,7 @@ Classifier: Programming Language :: Python :: 3.13
 Requires-Python: >=3.10
 Requires-Dist: argcomplete>=3.6.3
 Requires-Dist: docker>=7.1.0
-Requires-Dist: gpustack-runner>=0.1.23.post1
+Requires-Dist: gpustack-runner>=0.1.23.post2
 Requires-Dist: kubernetes>=33.1.0
 Requires-Dist: nvidia-ml-py>=13.580.65
 Requires-Dist: podman==5.6.0

{gpustack_runtime-0.1.39 → gpustack_runtime-0.1.39.post1}/gpustack_runtime/__main__.py

@@ -12,7 +12,6 @@ from . import deployer, detector
 from ._version import commit_id, version
 from .cmds import (
     CopyImagesSubCommand,
-    CreateRunnerWorkloadSubCommand,
     CreateWorkloadSubCommand,
     DeleteWorkloadsSubCommand,
     DeleteWorkloadSubCommand,
@@ -59,7 +58,6 @@ def main():
     subcommand_parser = parser.add_subparsers(
         help="gpustack-runtime command helpers",
     )
-    CreateRunnerWorkloadSubCommand.register(subcommand_parser)
     CreateWorkloadSubCommand.register(subcommand_parser)
     DeleteWorkloadSubCommand.register(subcommand_parser)
     DeleteWorkloadsSubCommand.register(subcommand_parser)

{gpustack_runtime-0.1.39 → gpustack_runtime-0.1.39.post1}/gpustack_runtime/_version.py

@@ -27,8 +27,8 @@ version_tuple: VERSION_TUPLE
 __commit_id__: COMMIT_ID
 commit_id: COMMIT_ID
 
-__version__ = version = '0.1.39'
-__version_tuple__ = version_tuple = (0, 1, 39)
+__version__ = version = '0.1.39.post1'
+__version_tuple__ = version_tuple = (0, 1, 39, 'post1')
 try:
     from ._version_appendix import git_commit
     __commit_id__ = commit_id = git_commit

gpustack_runtime-0.1.39.post1/gpustack_runtime/_version_appendix.py

@@ -0,0 +1 @@
+git_commit = "dea7237"

{gpustack_runtime-0.1.39 → gpustack_runtime-0.1.39.post1}/gpustack_runtime/cmds/__init__.py

@@ -1,7 +1,6 @@
 from __future__ import annotations
 
 from .deployer import (
-    CreateRunnerWorkloadSubCommand,
     CreateWorkloadSubCommand,
     DeleteWorkloadsSubCommand,
     DeleteWorkloadSubCommand,
@@ -23,7 +22,6 @@ from .images import (
 
 __all__ = [
     "CopyImagesSubCommand",
-    "CreateRunnerWorkloadSubCommand",
     "CreateWorkloadSubCommand",
     "DeleteWorkloadSubCommand",
     "DeleteWorkloadsSubCommand",

{gpustack_runtime-0.1.39 → gpustack_runtime-0.1.39.post1}/gpustack_runtime/cmds/deployer.py

@@ -92,238 +92,6 @@ _IGNORE_SENSITIVE_ENVS_SUFFIX = (
 )
 
 
-class CreateRunnerWorkloadSubCommand(SubCommand):
-    """
-    Command to create a runner workload deployment.
-    """
-
-    backend: str
-    device: str
-    command_script: str | None
-    port: int
-    host_network: bool
-    check: bool
-    namespace: str
-    service: str
-    version: str
-    name: str
-    volume: str
-    extra_args: list[str]
-
-    @staticmethod
-    def register(parser: _SubParsersAction):
-        deploy_parser = parser.add_parser(
-            "create-runner",
-            help="Create a runner workload deployment",
-        )
-
-        deploy_parser.add_argument(
-            "--backend",
-            type=str,
-            help="Backend to use (default: detect from current environment)",
-            choices=supported_backends(),
-        )
-
-        deploy_parser.add_argument(
-            "--device",
-            type=str,
-            help="Device to use, multiple devices join by comma (default: all devices)",
-            default="all",
-        )
-
-        deploy_parser.add_argument(
-            "--command-script-file",
-            type=str,
-            help="Path of command script for the workload",
-        )
-
-        deploy_parser.add_argument(
-            "--port",
-            type=int,
-            help="Port to expose",
-        )
-
-        deploy_parser.add_argument(
-            "--host-network",
-            action="store_true",
-            help="Use host network (default: False)",
-            default=False,
-        )
-
-        deploy_parser.add_argument(
-            "--check",
-            action="store_true",
-            help="Enable health check, needs --port (default: False)",
-            default=False,
-        )
-
-        deploy_parser.add_argument(
-            "--namespace",
-            type=str,
-            help="Namespace of the runner",
-        )
-
-        deploy_parser.add_argument(
-            "service",
-            type=str,
-            help="Service of the runner",
-        )
-
-        deploy_parser.add_argument(
-            "version",
-            type=str,
-            help="Version of the runner",
-        )
-
-        deploy_parser.add_argument(
-            "volume",
-            type=str,
-            help="Volume to mount",
-        )
-
-        deploy_parser.add_argument(
-            "extra_args",
-            nargs=REMAINDER,
-            help="Extra arguments for the runner",
-        )
-
-        deploy_parser.set_defaults(func=CreateRunnerWorkloadSubCommand)
-
-    def __init__(self, args: Namespace):
-        self.backend = args.backend
-        self.device = args.device
-        self.command_script = None
-        self.port = args.port
-        self.host_network = args.host_network
-        self.check = args.check
-        self.namespace = args.namespace
-        self.service = args.service
-        self.version = args.version
-        self.name = f"{args.service}-{args.version}".lower().replace(".", "-")
-        self.volume = args.volume
-        self.extra_args = args.extra_args
-
-        if not self.name or not self.volume:
-            msg = "The name and volume arguments are required."
-            raise ValueError(msg)
-
-        if args.command_script_file:
-            command_script_file = Path(args.command_script_file)
-            if not command_script_file.is_file():
-                msg = f"The command script file '{command_script_file}' does not exist."
-                raise ValueError(msg)
-            self.command_script = command_script_file.read_text(
-                encoding="utf-8",
-            ).strip()
-
-    def run(self):
-        env = [
-            ContainerEnv(
-                name=name,
-                value=value,
-            )
-            for name, value in os.environ.items()
-            if not name.startswith(_IGNORE_ENVS_PREFIX)
-            and not name.endswith(_IGNORE_ENVS_SUFFIX)
-        ]
-        if self.backend:
-            resources = ContainerResources(
-                **{
-                    v: self.device
-                    for k, v in envs.GPUSTACK_RUNTIME_DETECT_BACKEND_MAP_RESOURCE_KEY.items()
-                    if k == self.backend
-                },
-            )
-        else:
-            resources = ContainerResources(
-                **{
-                    envs.GPUSTACK_RUNTIME_DEPLOY_AUTOMAP_RESOURCE_KEY: self.device,
-                },
-            )
-        mounts = [
-            ContainerMount(
-                path=self.volume,
-            ),
-        ]
-        execution = ContainerExecution(
-            command_script=self.command_script,
-            args=self.extra_args,
-        )
-        ports = (
-            [
-                ContainerPort(
-                    internal=self.port,
-                ),
-            ]
-            if self.port
-            else None
-        )
-        checks = (
-            [
-                ContainerCheck(
-                    delay=60,
-                    interval=10,
-                    timeout=5,
-                    retries=6,
-                    tcp=ContainerCheckTCP(port=self.port),
-                    teardown=True,
-                ),
-            ]
-            if self.check and self.port
-            else None
-        )
-        plan = WorkloadPlan(
-            name=self.name,
-            namespace=self.namespace,
-            host_network=self.host_network,
-            containers=[
-                Container(
-                    restart_policy=(
-                        ContainerRestartPolicyEnum.NEVER
-                        if not self.check
-                        else ContainerRestartPolicyEnum.ALWAYS
-                    ),
-                    image=f"gpustack/runner:{self.backend if self.backend else 'Host'}X.Y-{self.service}{self.version}",
-                    name=self.name,
-                    envs=env,
-                    resources=resources,
-                    mounts=mounts,
-                    execution=execution,
-                    ports=ports,
-                    checks=checks,
-                ),
-            ],
-        )
-        create_workload(plan)
-        print(f"Created workload '{self.name}'.")
-
-        while True:
-            st = get_workload(
-                name=self.name,
-                namespace=self.namespace,
-            )
-            if st and st.state not in (
-                WorkloadStatusStateEnum.PENDING,
-                WorkloadStatusStateEnum.INITIALIZING,
-            ):
-                break
-            time.sleep(1)
-
-        print("\033[2J\033[H", end="")
-
-        async def stream_logs():
-            logs_result = await async_logs_workload(
-                name=self.name,
-                namespace=self.namespace,
-                tail=-1,
-                follow=True,
-            )
-            async for line in logs_result:
-                print(line.decode("utf-8").rstrip())
-
-        asyncio.run(stream_logs())
-
-
 class CreateWorkloadSubCommand(SubCommand):
     """
     Command to create a workload deployment.
@@ -358,8 +126,7 @@ class CreateWorkloadSubCommand(SubCommand):
         deploy_parser.add_argument(
             "--device",
             type=str,
-            help="Device to use, multiple devices join by comma (default: all devices)",
-            default="all",
+            help="Device to use, multiple devices join by comma, all for all devices",
         )
 
         deploy_parser.add_argument(
@@ -456,20 +223,22 @@ class CreateWorkloadSubCommand(SubCommand):
             if not name.startswith(_IGNORE_ENVS_PREFIX)
             and not name.endswith(_IGNORE_ENVS_SUFFIX)
         ]
-        if self.backend:
-            resources = ContainerResources(
-                **{
-                    v: self.device
-                    for k, v in envs.GPUSTACK_RUNTIME_DETECT_BACKEND_MAP_RESOURCE_KEY.items()
-                    if k == self.backend
-                },
-            )
-        else:
-            resources = ContainerResources(
-                **{
-                    envs.GPUSTACK_RUNTIME_DEPLOY_AUTOMAP_RESOURCE_KEY: self.device,
-                },
-            )
+        resources = None
+        if self.device:
+            if self.backend:
+                resources = ContainerResources(
+                    **{
+                        v: self.device
+                        for k, v in envs.GPUSTACK_RUNTIME_DETECT_BACKEND_MAP_RESOURCE_KEY.items()
+                        if k == self.backend
+                    },
+                )
+            else:
+                resources = ContainerResources(
+                    **{
+                        envs.GPUSTACK_RUNTIME_DEPLOY_AUTOMAP_RESOURCE_KEY: self.device,
+                    },
+                )
         mounts = [
             ContainerMount(
                 path=self.volume,

{gpustack_runtime-0.1.39 → gpustack_runtime-0.1.39.post1}/gpustack_runtime/deployer/docker.py

@@ -1,12 +1,14 @@
 from __future__ import annotations
 
 import contextlib
+import io
 import json
 import logging
 import operator
 import os
 import socket
 import sys
+import tarfile
 from dataclasses import dataclass, field
 from functools import lru_cache, reduce
 from math import ceil
@@ -303,10 +305,6 @@ class DockerDeployer(Deployer):
     """
     Client for interacting with the Docker daemon.
     """
-    _container_ephemeral_files_dir: Path | None = None
-    """
-    Directory for ephemeral files inside containers, internal use only.
-    """
     _mutate_create_options: Callable[[dict[str, Any]], dict[str, Any]] | None = None
     """
     Function to handle mirrored deployment, internal use only.
@@ -383,48 +381,6 @@ class DockerDeployer(Deployer):
 
         return wrapper
 
-    @staticmethod
-    def _create_ephemeral_files(
-        workload: DockerWorkloadPlan,
-    ) -> dict[tuple[int, str], str]:
-        """
-        Create ephemeral files as local file for the workload.
-
-        Returns:
-            A mapping from (container index, configured path) to actual filename.
-
-        Raises:
-            OperationError:
-                If the ephemeral files fail to create.
-
-        """
-        # Map (container index, configured path) to actual filename.
-        ephemeral_filename_mapping: dict[tuple[int, str], str] = {}
-        ephemeral_files: list[tuple[str, str, int]] = []
-        for ci, c in enumerate(workload.containers):
-            for fi, f in enumerate(c.files or []):
-                if f.content is not None:
-                    fn = f"{workload.name}-{ci}-{fi}"
-                    ephemeral_filename_mapping[(ci, f.path)] = fn
-                    ephemeral_files.append((fn, f.content, f.mode))
-        if not ephemeral_filename_mapping:
-            return ephemeral_filename_mapping
-
-        # Create ephemeral files directory if not exists.
-        try:
-            for fn, fc, fm in ephemeral_files:
-                fp = envs.GPUSTACK_RUNTIME_DOCKER_EPHEMERAL_FILES_DIR.joinpath(fn)
-                with fp.open("w", encoding="utf-8") as f:
-                    f.write(fc)
-                    f.flush()
-                fp.chmod(fm)
-                logger.debug("Created local file %s with mode %s", fp, oct(fm))
-        except OSError as e:
-            msg = "Failed to create ephemeral files"
-            raise OperationError(msg) from e
-
-        return ephemeral_filename_mapping
-
     def _create_ephemeral_volumes(self, workload: DockerWorkloadPlan) -> dict[str, str]:
         """
         Create ephemeral volumes for the workload.
@@ -717,12 +673,10 @@ class DockerDeployer(Deployer):
         else:
             return d_container
 
+    @staticmethod
     def _append_container_mounts(
-        self,
         create_options: dict[str, Any],
         c: Container,
-        ci: int,
-        ephemeral_filename_mapping: dict[tuple[int, str] : str],
         ephemeral_volume_name_mapping: dict[str, str],
     ):
         """
@@ -738,17 +692,7 @@ class DockerDeployer(Deployer):
                     target="",
                 )
 
-                if f.content is not None:
-                    # Ephemeral file, use from local ephemeral files directory.
-                    if (ci, f.path) not in ephemeral_filename_mapping:
-                        continue
-                    fn = ephemeral_filename_mapping[(ci, f.path)]
-                    path = str(
-                        self._container_ephemeral_files_dir.joinpath(fn),
-                    )
-                    binding["Source"] = path
-                    binding["Target"] = f"/{f.path.lstrip('/')}"
-                elif f.path:
+                if f.content is None and f.path:
                     # Host file, bind directly.
                     binding["Source"] = f.path
                     binding["Target"] = f.path
@@ -860,10 +804,39 @@ class DockerDeployer(Deployer):
 
         return healthcheck
 
+    @staticmethod
+    def _upload_ephemeral_files(
+        c: Container,
+        container: docker.models.containers.Container,
+    ):
+        if not c.files:
+            return
+
+        f_tar = io.BytesIO()
+        with tarfile.open(fileobj=f_tar, mode="w") as tar:
+            for f in c.files:
+                if f.content is None or not f.path:
+                    continue
+                fc_bytes = f.content.encode("utf-8")
+                info = tarfile.TarInfo(name=f.path.lstrip("/"))
+                info.size = len(fc_bytes)
+                info.mode = f.mode
+                tar.addfile(tarinfo=info, fileobj=io.BytesIO(fc_bytes))
+        if f_tar.getbuffer().nbytes == 0:
+            return
+
+        f_tar.seek(0)
+        uploaded = container.put_archive(
+            path="/",
+            data=f_tar.getvalue(),
+        )
+        if not uploaded:
+            msg = f"Failed to upload ephemeral files to container {container.name}"
+            raise OperationError(msg)
+
     def _create_containers(
         self,
         workload: DockerWorkloadPlan,
-        ephemeral_filename_mapping: dict[tuple[int, str] : str],
         ephemeral_volume_name_mapping: dict[str, str],
         pause_container: docker.models.containers.Container,
     ) -> (
@@ -1106,8 +1079,6 @@ class DockerDeployer(Deployer):
             self._append_container_mounts(
                 create_options,
                 c,
-                ci,
-                ephemeral_filename_mapping,
                 ephemeral_volume_name_mapping,
             )
 
@@ -1149,6 +1120,10 @@ class DockerDeployer(Deployer):
                     detach=detach,
                     **create_options,
                 )
+
+                # Upload ephemeral files into the container.
+                self._upload_ephemeral_files(c, d_container)
+
             except docker.errors.APIError as e:
                 msg = f"Failed to create container {container_name}{_detail_api_call_error(e)}"
                 raise OperationError(msg) from e
@@ -1198,9 +1173,6 @@ class DockerDeployer(Deployer):
     def __init__(self):
         super().__init__(_NAME)
         self._client = self._get_client()
-        self._container_ephemeral_files_dir = (
-            envs.GPUSTACK_RUNTIME_DOCKER_EPHEMERAL_FILES_DIR
-        )
 
     def _prepare_create(self):
         """
@@ -1434,25 +1406,6 @@ class DockerDeployer(Deployer):
 
         self._mutate_create_options = mutate_create_options
 
-        # Extract ephemeral files dir mutation if any.
-        if mirrored_mounts:
-            e_target = str(envs.GPUSTACK_RUNTIME_DOCKER_EPHEMERAL_FILES_DIR)
-            b_source = ""
-            b_target = ""
-            for m in mirrored_mounts:
-                c_target = m.get("Destination", "///")
-                if (
-                    e_target == c_target or e_target.startswith(f"{c_target}/")
-                ) and len(c_target) >= len(b_target):
-                    b_source = m.get("Source")
-                    b_target = c_target
-            if b_source:
-                result = Path(b_source)
-                if e_target != b_target:
-                    b_subpath = e_target.removeprefix(b_target)
-                    result = result.joinpath(b_subpath.lstrip("/"))
-                self._container_ephemeral_files_dir = result
-
     def _find_self_container(
         self,
         self_container_id: str,
@@ -1536,12 +1489,6 @@ class DockerDeployer(Deployer):
         if logger.isEnabledFor(logging.DEBUG):
             logger.debug("Creating workload:\n%s", workload.to_yaml())
 
-        # Create ephemeral file if needed,
-        # (container index, configured path): <actual filename>
-        ephemeral_filename_mapping: dict[tuple[int, str] : str] = (
-            self._create_ephemeral_files(workload)
-        )
-
         # Create ephemeral volumes if needed,
         # <configured volume name>: <actual volume name>
         ephemeral_volume_name_mapping: dict[str, str] = self._create_ephemeral_volumes(
@@ -1554,7 +1501,6 @@ class DockerDeployer(Deployer):
         # Create init/run containers.
         init_containers, run_containers = self._create_containers(
             workload,
-            ephemeral_filename_mapping,
             ephemeral_volume_name_mapping,
             pause_container,
         )
@@ -1696,17 +1642,6 @@ class DockerDeployer(Deployer):
             msg = f"Failed to delete volumes for workload {name}{_detail_api_call_error(e)}"
             raise OperationError(msg) from e
 
-        # Remove all ephemeral files for the workload.
-        try:
-            for fp in envs.GPUSTACK_RUNTIME_DOCKER_EPHEMERAL_FILES_DIR.glob(
-                f"{name}-*",
-            ):
-                if fp.is_file():
-                    fp.unlink(missing_ok=True)
-        except OSError as e:
-            msg = f"Failed to delete ephemeral files for workload {name}"
-            raise OperationError(msg) from e
-
         return workload
 
     @_supported