modal 0.74.55__tar.gz → 0.74.57__tar.gz

{modal-0.74.55 → modal-0.74.57}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 0.74.55
+Version: 0.74.57
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-0.74.55 → modal-0.74.57}/modal/_functions.py

@@ -477,6 +477,7 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
         i6pn_enabled: bool = False,
         # Experimental: Clustered functions
         cluster_size: Optional[int] = None,
+        rdma: Optional[bool] = None,
         max_inputs: Optional[int] = None,
         ephemeral_disk: Optional[int] = None,
         # current default: first-party, future default: main-package
@@ -899,7 +900,7 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
 
                         function_definition_copy.resources.CopyFrom(
                             convert_fn_config_to_resources_config(
-                                cpu=cpu, memory=memory, gpu=_gpu, ephemeral_disk=ephemeral_disk
+                                cpu=cpu, memory=memory, gpu=_gpu, ephemeral_disk=ephemeral_disk, rdma=rdma
                             ),
                         )
                         ranked_function = api_pb2.FunctionData.RankedFunction(
@@ -914,7 +915,7 @@ class _Function(typing.Generic[P, ReturnType, OriginalReturnType], _Object, type
                     # assert isinstance(gpu, GPU_T)  # includes the case where gpu==None case
                     function_definition.resources.CopyFrom(
                         convert_fn_config_to_resources_config(
-                            cpu=cpu, memory=memory, gpu=gpu, ephemeral_disk=ephemeral_disk
+                            cpu=cpu, memory=memory, gpu=gpu, ephemeral_disk=ephemeral_disk, rdma=rdma
                         ),
                     )
 

{modal-0.74.55 → modal-0.74.57}/modal/_partial_function.py

@@ -77,6 +77,7 @@ class _PartialFunctionParams:
     max_concurrent_inputs: Optional[int] = None
     target_concurrent_inputs: Optional[int] = None
     build_timeout: Optional[int] = None
+    rdma: Optional[bool] = None
 
     def update(self, other: "_PartialFunctionParams") -> None:
         """Update self with params set in other."""
@@ -900,7 +901,7 @@ def _concurrent(
 
 
 # NOTE: clustered is currently exposed through modal.experimental, not the top-level namespace
-def _clustered(size: int, broadcast: bool = True):
+def _clustered(size: int, broadcast: bool = True, rdma: bool = False):
     """Provision clusters of colocated and networked containers for the Function.
 
     Parameters:
@@ -918,7 +919,7 @@ def _clustered(size: int, broadcast: bool = True):
         raise ValueError("cluster size must be greater than 0")
 
     flags = _PartialFunctionFlags.CLUSTERED
-    params = _PartialFunctionParams(cluster_size=size)
+    params = _PartialFunctionParams(cluster_size=size, rdma=rdma)
 
     def wrapper(
         obj: Union[_PartialFunction[P, ReturnType, ReturnType], Callable[P, ReturnType]],

{modal-0.74.55 → modal-0.74.57}/modal/_resources.py

@@ -13,6 +13,7 @@ def convert_fn_config_to_resources_config(
     memory: Optional[Union[int, tuple[int, int]]],
     gpu: GPU_T,
     ephemeral_disk: Optional[int],
+    rdma: Optional[bool] = None,
 ) -> api_pb2.Resources:
     gpu_config = parse_gpu_config(gpu)
     if cpu and isinstance(cpu, tuple):
@@ -48,4 +49,5 @@ def convert_fn_config_to_resources_config(
         memory_mb=memory_mb,
         memory_mb_max=memory_mb_max,
         ephemeral_disk_mb=ephemeral_disk,
+        rdma=rdma or False,
     )

{modal-0.74.55 → modal-0.74.57}/modal/app.py

@@ -747,6 +747,7 @@ class _App:
                     )
                 i6pn_enabled = i6pn or (f.flags & _PartialFunctionFlags.CLUSTERED)
                 cluster_size = f.params.cluster_size  # Experimental: Clustered functions
+                rdma = f.params.rdma
 
                 info = FunctionInfo(f.raw_f, serialized=serialized, name_override=name)
                 raw_f = f.raw_f
@@ -799,6 +800,7 @@ class _App:
                 target_concurrent_inputs = None
 
                 cluster_size = None  # Experimental: Clustered functions
+                rdma = None
                 i6pn_enabled = i6pn
 
             if info.function_name.endswith(".app"):
@@ -850,6 +852,7 @@ class _App:
                 scheduler_placement=scheduler_placement,
                 i6pn_enabled=i6pn_enabled,
                 cluster_size=cluster_size,  # Experimental: Clustered functions
+                rdma=rdma,
                 include_source=include_source if include_source is not None else self._include_source_default,
                 experimental_options={k: str(v) for k, v in (experimental_options or {}).items()},
                 _experimental_proxy_ip=_experimental_proxy_ip,

modal-0.74.57/modal/cli/cluster.py

@@ -0,0 +1,82 @@
+# Copyright Modal Labs 2022
+from typing import Optional, Union
+
+import typer
+from rich.console import Console
+from rich.text import Text
+
+from modal._object import _get_environment_name
+from modal._pty import get_pty_info
+from modal._utils.async_utils import synchronizer
+from modal.cli.utils import ENV_OPTION, display_table, is_tty, timestamp_to_local
+from modal.client import _Client
+from modal.config import config
+from modal.container_process import _ContainerProcess
+from modal.environments import ensure_env
+from modal.stream_type import StreamType
+from modal_proto import api_pb2
+
+cluster_cli = typer.Typer(
+    name="cluster", help="Manage and connect to running multi-node clusters.", no_args_is_help=True
+)
+
+
+@cluster_cli.command("list")
+@synchronizer.create_blocking
+async def list_(env: Optional[str] = ENV_OPTION, json: bool = False):
+    """List all clusters that are currently running."""
+    env = ensure_env(env)
+    client = await _Client.from_env()
+    environment_name = _get_environment_name(env)
+    res: api_pb2.ClusterListResponse = await client.stub.ClusterList(
+        api_pb2.ClusterListRequest(environment_name=environment_name)
+    )
+
+    column_names = ["Cluster ID", "App ID", "Start Time", "Nodes"]
+    rows: list[list[Union[Text, str]]] = []
+    res.clusters.sort(key=lambda c: c.started_at, reverse=True)
+
+    for c in res.clusters:
+        rows.append(
+            [
+                c.cluster_id,
+                c.app_id,
+                timestamp_to_local(c.started_at, json) if c.started_at else "Pending",
+                str(len(c.task_ids)),
+            ]
+        )
+
+    display_table(column_names, rows, json=json, title=f"Active Multi-node Clusters in environment: {environment_name}")
+
+
+@cluster_cli.command("shell")
+@synchronizer.create_blocking
+async def shell(
+    cluster_id: str = typer.Argument(help="Cluster ID"),
+    rank: int = typer.Option(default=0, help="Rank of the node to shell into"),
+):
+    """Open a shell to a multi-node cluster node."""
+    client = await _Client.from_env()
+    res: api_pb2.ClusterGetResponse = await client.stub.ClusterGet(api_pb2.ClusterGetRequest(cluster_id=cluster_id))
+    if len(res.cluster.task_ids) <= rank:
+        raise typer.Abort(f"No node with rank {rank} in cluster {cluster_id}")
+    task_id = res.cluster.task_ids[rank]
+    console = Console()
+    is_main = "(main)" if rank == 0 else ""
+    console.print(
+        f"Opening shell to node {rank} {is_main} of cluster {cluster_id} (container {task_id})", style="green"
+    )
+
+    pty = is_tty()
+    req = api_pb2.ContainerExecRequest(
+        task_id=task_id,
+        command=["/bin/bash"],
+        pty_info=get_pty_info(shell=True) if pty else None,
+        runtime_debug=config.get("function_runtime_debug"),
+    )
+    exec_res: api_pb2.ContainerExecResponse = await client.stub.ContainerExec(req)
+    if pty:
+        await _ContainerProcess(exec_res.exec_id, client).attach()
+    else:
+        # TODO: redirect stderr to its own stream?
+        await _ContainerProcess(exec_res.exec_id, client, stdout=StreamType.STDOUT, stderr=StreamType.STDOUT).wait()

{modal-0.74.55 → modal-0.74.57}/modal/cli/entry_point.py

@@ -10,6 +10,7 @@ from modal._utils.async_utils import synchronizer
 
 from . import run
 from .app import app_cli
+from .cluster import cluster_cli
 from .config import config_cli
 from .container import container_cli
 from .dict import dict_cli
@@ -92,6 +93,8 @@ entrypoint_cli_typer.add_typer(launch_cli)
 # Deployments
 entrypoint_cli_typer.add_typer(app_cli, rich_help_panel="Deployments")
 entrypoint_cli_typer.add_typer(container_cli, rich_help_panel="Deployments")
+# TODO: cluster is hidden while multi-node is in beta/experimental
+entrypoint_cli_typer.add_typer(cluster_cli, rich_help_panel="Deployments", hidden=True)
 
 # Storage
 entrypoint_cli_typer.add_typer(dict_cli, rich_help_panel="Storage")

{modal-0.74.55 → modal-0.74.57}/modal/client.pyi

@@ -27,7 +27,7 @@ class _Client:
     _snapshotted: bool
 
     def __init__(
-        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.74.55"
+        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.74.57"
     ): ...
     def is_closed(self) -> bool: ...
     @property
@@ -86,7 +86,7 @@ class Client:
     _snapshotted: bool
 
     def __init__(
-        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.74.55"
+        self, server_url: str, client_type: int, credentials: typing.Optional[tuple[str, str]], version: str = "0.74.57"
     ): ...
     def is_closed(self) -> bool: ...
     @property

{modal-0.74.55 → modal-0.74.57}/modal/functions.pyi

@@ -94,6 +94,7 @@ class Function(
         restrict_modal_access: bool = False,
         i6pn_enabled: bool = False,
         cluster_size: typing.Optional[int] = None,
+        rdma: typing.Optional[bool] = None,
         max_inputs: typing.Optional[int] = None,
         ephemeral_disk: typing.Optional[int] = None,
         include_source: typing.Optional[bool] = None,
@@ -222,11 +223,11 @@ class Function(
 
     _call_generator_nowait: ___call_generator_nowait_spec[typing_extensions.Self]
 
-    class __remote_spec(typing_extensions.Protocol[ReturnType_INNER, P_INNER, SUPERSELF]):
+    class __remote_spec(typing_extensions.Protocol[P_INNER, ReturnType_INNER, SUPERSELF]):
         def __call__(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> ReturnType_INNER: ...
         async def aio(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> ReturnType_INNER: ...
 
-    remote: __remote_spec[modal._functions.ReturnType, modal._functions.P, typing_extensions.Self]
+    remote: __remote_spec[modal._functions.P, modal._functions.ReturnType, typing_extensions.Self]
 
     class __remote_gen_spec(typing_extensions.Protocol[SUPERSELF]):
         def __call__(self, *args, **kwargs) -> typing.Generator[typing.Any, None, None]: ...
@@ -241,12 +242,12 @@ class Function(
         self, *args: modal._functions.P.args, **kwargs: modal._functions.P.kwargs
     ) -> modal._functions.OriginalReturnType: ...
 
-    class ___experimental_spawn_spec(typing_extensions.Protocol[ReturnType_INNER, P_INNER, SUPERSELF]):
+    class ___experimental_spawn_spec(typing_extensions.Protocol[P_INNER, ReturnType_INNER, SUPERSELF]):
         def __call__(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> FunctionCall[ReturnType_INNER]: ...
         async def aio(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> FunctionCall[ReturnType_INNER]: ...
 
     _experimental_spawn: ___experimental_spawn_spec[
-        modal._functions.ReturnType, modal._functions.P, typing_extensions.Self
+        modal._functions.P, modal._functions.ReturnType, typing_extensions.Self
     ]
 
     class ___spawn_map_inner_spec(typing_extensions.Protocol[P_INNER, SUPERSELF]):
@@ -255,11 +256,11 @@ class Function(
 
     _spawn_map_inner: ___spawn_map_inner_spec[modal._functions.P, typing_extensions.Self]
 
-    class __spawn_spec(typing_extensions.Protocol[ReturnType_INNER, P_INNER, SUPERSELF]):
+    class __spawn_spec(typing_extensions.Protocol[P_INNER, ReturnType_INNER, SUPERSELF]):
         def __call__(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> FunctionCall[ReturnType_INNER]: ...
         async def aio(self, *args: P_INNER.args, **kwargs: P_INNER.kwargs) -> FunctionCall[ReturnType_INNER]: ...
 
-    spawn: __spawn_spec[modal._functions.ReturnType, modal._functions.P, typing_extensions.Self]
+    spawn: __spawn_spec[modal._functions.P, modal._functions.ReturnType, typing_extensions.Self]
 
     def get_raw_f(self) -> collections.abc.Callable[..., typing.Any]: ...
 

{modal-0.74.55 → modal-0.74.57}/modal.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: modal
-Version: 0.74.55
+Version: 0.74.57
 Summary: Python client library for Modal
 Author-email: Modal Labs <support@modal.com>
 License: Apache-2.0

{modal-0.74.55 → modal-0.74.57}/modal.egg-info/SOURCES.txt

@@ -127,6 +127,7 @@ modal/cli/__init__.py
 modal/cli/_download.py
 modal/cli/_traceback.py
 modal/cli/app.py
+modal/cli/cluster.py
 modal/cli/config.py
 modal/cli/container.py
 modal/cli/dict.py

{modal-0.74.55 → modal-0.74.57}/modal_version/_version_generated.py

@@ -1,4 +1,4 @@
 # Copyright Modal Labs 2025
 
 # Note: Reset this value to -1 whenever you make a minor `0.X` release of the client.
-build_number = 55  # git: aa1b399
+build_number = 57  # git: 747bcf5