modal 1.0.3.dev10__py3-none-any.whl → 1.2.3.dev7__py3-none-any.whl

modal/sandbox.py

@@ -1,8 +1,15 @@
 # Copyright Modal Labs 2022
 import asyncio
+import json
 import os
-from collections.abc import AsyncGenerator, Sequence
-from typing import TYPE_CHECKING, AsyncIterator, Literal, Optional, Union, overload
+import time
+import uuid
+from collections.abc import AsyncGenerator, Collection, Sequence
+from dataclasses import dataclass
+from typing import TYPE_CHECKING, Any, AsyncIterator, Literal, Optional, Union, overload
+
+from ._pty import get_pty_info
+from .config import config, logger
 
 if TYPE_CHECKING:
     import _typeshed
@@ -14,18 +21,20 @@ from modal._tunnel import Tunnel
 from modal.cloud_bucket_mount import _CloudBucketMount, cloud_bucket_mounts_to_proto
 from modal.mount import _Mount
 from modal.volume import _Volume
-from modal_proto import api_pb2
+from modal_proto import api_pb2, task_command_router_pb2 as sr_pb2
 
+from ._load_context import LoadContext
 from ._object import _get_environment_name, _Object
 from ._resolver import Resolver
 from ._resources import convert_fn_config_to_resources_config
 from ._utils.async_utils import TaskContext, synchronize_api
-from ._utils.grpc_utils import retry_transient_errors
+from ._utils.deprecation import deprecation_warning
 from ._utils.mount_utils import validate_network_file_systems, validate_volumes
+from ._utils.name_utils import check_object_name
+from ._utils.task_command_router_client import TaskCommandRouterClient
 from .client import _Client
-from .config import config
 from .container_process import _ContainerProcess
-from .exception import ExecutionError, InvalidError, SandboxTerminatedError, SandboxTimeoutError
+from .exception import AlreadyExistsError, ExecutionError, InvalidError, SandboxTerminatedError, SandboxTimeoutError
 from .file_io import FileWatchEvent, FileWatchEventType, _FileIO
 from .gpu import GPU_T
 from .image import _Image
@@ -49,44 +58,78 @@ _default_image: _Image = _Image.debian_slim()
 # e.g. 'runsc exec ...'. So we use 2**16 as the limit.
 ARG_MAX_BYTES = 2**16
 
+# This buffer extends the user-supplied timeout on ContainerExec-related RPCs. This was introduced to
+# give any in-flight status codes/IO data more time to reach the client before the stream is closed.
+CONTAINER_EXEC_TIMEOUT_BUFFER = 5
+
+
 if TYPE_CHECKING:
     import modal.app
 
 
-def _validate_exec_args(entrypoint_args: Sequence[str]) -> None:
+def _validate_exec_args(args: Sequence[str]) -> None:
     # Entrypoint args must be strings.
-    if not all(isinstance(arg, str) for arg in entrypoint_args):
+    if not all(isinstance(arg, str) for arg in args):
         raise InvalidError("All entrypoint arguments must be strings")
     # Avoid "[Errno 7] Argument list too long" errors.
-    total_arg_len = sum(len(arg) for arg in entrypoint_args)
+    total_arg_len = sum(len(arg) for arg in args)
     if total_arg_len > ARG_MAX_BYTES:
         raise InvalidError(
-            f"Total length of entrypoint arguments must be less than {ARG_MAX_BYTES} bytes (ARG_MAX). "
+            f"Total length of CMD arguments must be less than {ARG_MAX_BYTES} bytes (ARG_MAX). "
             f"Got {total_arg_len} bytes."
         )
 
 
+class DefaultSandboxNameOverride(str):
+    """A singleton class that represents the default sandbox name override.
+
+    It is used to indicate that the sandbox name should not be overridden.
+    """
+
+    def __repr__(self) -> str:
+        # NOTE: this must match the instance var name below in order for type stubs to work 😬
+        return "_DEFAULT_SANDBOX_NAME_OVERRIDE"
+
+
+_DEFAULT_SANDBOX_NAME_OVERRIDE = DefaultSandboxNameOverride()
+
+
+@dataclass(frozen=True)
+class SandboxConnectCredentials:
+    """Simple data structure storing credentials for making HTTP connections to a sandbox."""
+
+    url: str
+    token: str
+
+
 class _Sandbox(_Object, type_prefix="sb"):
     """A `Sandbox` object lets you interact with a running sandbox. This API is similar to Python's
     [asyncio.subprocess.Process](https://docs.python.org/3/library/asyncio-subprocess.html#asyncio.subprocess.Process).
 
-    Refer to the [guide](/docs/guide/sandbox) on how to spawn and use sandboxes.
+    Refer to the [guide](https://modal.com/docs/guide/sandbox) on how to spawn and use sandboxes.
     """
 
     _result: Optional[api_pb2.GenericResult]
     _stdout: _StreamReader[str]
     _stderr: _StreamReader[str]
     _stdin: _StreamWriter
-    _task_id: Optional[str] = None
-    _tunnels: Optional[dict[int, Tunnel]] = None
-    _enable_snapshot: bool = False
+    _task_id: Optional[str]
+    _tunnels: Optional[dict[int, Tunnel]]
+    _enable_snapshot: bool
+    _command_router_client: Optional[TaskCommandRouterClient]
+
+    @staticmethod
+    def _default_pty_info() -> api_pb2.PTYInfo:
+        return get_pty_info(shell=True, no_terminate_on_idle_stdin=True)
 
     @staticmethod
     def _new(
-        entrypoint_args: Sequence[str],
+        args: Sequence[str],
         image: _Image,
-        secrets: Sequence[_Secret],
-        timeout: Optional[int] = None,
+        secrets: Collection[_Secret],
+        name: Optional[str] = None,
+        timeout: int = 300,
+        idle_timeout: Optional[int] = None,
         workdir: Optional[str] = None,
         gpu: GPU_T = None,
         cloud: Optional[str] = None,
@@ -98,12 +141,16 @@ class _Sandbox(_Object, type_prefix="sb"):
         block_network: bool = False,
         cidr_allowlist: Optional[Sequence[str]] = None,
         volumes: dict[Union[str, os.PathLike], Union[_Volume, _CloudBucketMount]] = {},
-        pty_info: Optional[api_pb2.PTYInfo] = None,
+        pty: bool = False,
+        pty_info: Optional[api_pb2.PTYInfo] = None,  # deprecated
         encrypted_ports: Sequence[int] = [],
+        h2_ports: Sequence[int] = [],
         unencrypted_ports: Sequence[int] = [],
         proxy: Optional[_Proxy] = None,
+        experimental_options: Optional[dict[str, bool]] = None,
         _experimental_scheduler_placement: Optional[SchedulerPlacement] = None,
         enable_snapshot: bool = False,
+        verbose: bool = False,
     ) -> "_Sandbox":
         """mdmd:hidden"""
 
@@ -129,6 +176,9 @@ class _Sandbox(_Object, type_prefix="sb"):
         cloud_bucket_mounts = [(k, v) for k, v in validated_volumes if isinstance(v, _CloudBucketMount)]
         validated_volumes = [(k, v) for k, v in validated_volumes if isinstance(v, _Volume)]
 
+        if pty:
+            pty_info = _Sandbox._default_pty_info()
+
         def _deps() -> list[_Object]:
             deps: list[_Object] = [image] + list(mounts) + list(secrets)
             for _, vol in validated_network_file_systems:
@@ -142,19 +192,28 @@ class _Sandbox(_Object, type_prefix="sb"):
                 deps.append(proxy)
             return deps
 
-        async def _load(self: _Sandbox, resolver: Resolver, _existing_object_id: Optional[str]):
+        async def _load(
+            self: _Sandbox, resolver: Resolver, load_context: LoadContext, _existing_object_id: Optional[str]
+        ):
             # Relies on dicts being ordered (true as of Python 3.6).
             volume_mounts = [
                 api_pb2.VolumeMount(
                     mount_path=path,
                     volume_id=volume.object_id,
                     allow_background_commits=True,
+                    read_only=volume._read_only,
                 )
                 for path, volume in validated_volumes
             ]
 
             open_ports = [api_pb2.PortSpec(port=port, unencrypted=False) for port in encrypted_ports]
             open_ports.extend([api_pb2.PortSpec(port=port, unencrypted=True) for port in unencrypted_ports])
+            open_ports.extend(
+                [
+                    api_pb2.PortSpec(port=port, unencrypted=False, tunnel_type=api_pb2.TUNNEL_TYPE_H2)
+                    for port in h2_ports
+                ]
+            )
 
             if block_network:
                 # If the network is blocked, cidr_allowlist is invalid as we don't allow any network access.
@@ -176,11 +235,12 @@ class _Sandbox(_Object, type_prefix="sb"):
 
             ephemeral_disk = None  # Ephemeral disk requests not supported on Sandboxes.
             definition = api_pb2.Sandbox(
-                entrypoint_args=entrypoint_args,
+                entrypoint_args=args,
                 image_id=image.object_id,
                 mount_ids=[mount.object_id for mount in mounts] + [mount.object_id for mount in image._mount_layers],
                 secret_ids=[secret.object_id for secret in secrets],
                 timeout_secs=timeout,
+                idle_timeout_secs=idle_timeout,
                 workdir=workdir,
                 resources=convert_fn_config_to_resources_config(
                     cpu=cpu, memory=memory, gpu=gpu, ephemeral_disk=ephemeral_disk
@@ -198,28 +258,37 @@ class _Sandbox(_Object, type_prefix="sb"):
                 network_access=network_access,
                 proxy_id=(proxy.object_id if proxy else None),
                 enable_snapshot=enable_snapshot,
+                verbose=verbose,
+                name=name,
+                experimental_options=experimental_options,
             )
 
-            # Note - `resolver.app_id` will be `None` for app-less sandboxes
-            create_req = api_pb2.SandboxCreateRequest(
-                app_id=resolver.app_id, definition=definition, environment_name=resolver.environment_name
-            )
-            create_resp = await retry_transient_errors(resolver.client.stub.SandboxCreate, create_req)
+            create_req = api_pb2.SandboxCreateRequest(app_id=load_context.app_id, definition=definition)
+            try:
+                create_resp = await load_context.client.stub.SandboxCreate(create_req)
+            except GRPCError as exc:
+                if exc.status == Status.ALREADY_EXISTS:
+                    raise AlreadyExistsError(exc.message)
+                raise exc
 
             sandbox_id = create_resp.sandbox_id
-            self._hydrate(sandbox_id, resolver.client, None)
+            self._hydrate(sandbox_id, load_context.client, None)
 
-        return _Sandbox._from_loader(_load, "Sandbox()", deps=_deps)
+        return _Sandbox._from_loader(_load, "Sandbox()", deps=_deps, load_context_overrides=LoadContext.empty())
 
     @staticmethod
     async def create(
-        *entrypoint_args: str,
-        app: Optional["modal.app._App"] = None,  # Optionally associate the sandbox with an app
-        environment_name: Optional[str] = None,  # Optionally override the default environment
+        *args: str,  # Set the CMD of the Sandbox, overriding any CMD of the container image.
+        # Associate the sandbox with an app. Required unless creating from a container.
+        app: Optional["modal.app._App"] = None,
+        name: Optional[str] = None,  # Optionally give the sandbox a name. Unique within an app.
         image: Optional[_Image] = None,  # The image to run as the container for the sandbox.
-        secrets: Sequence[_Secret] = (),  # Environment variables to inject into the sandbox.
+        env: Optional[dict[str, Optional[str]]] = None,  # Environment variables to set in the Sandbox.
+        secrets: Optional[Collection[_Secret]] = None,  # Secrets to inject into the Sandbox as environment variables.
         network_file_systems: dict[Union[str, os.PathLike], _NetworkFileSystem] = {},
-        timeout: Optional[int] = None,  # Maximum execution time of the sandbox in seconds.
+        timeout: int = 300,  # Maximum lifetime of the sandbox in seconds.
+        # The amount of time in seconds that a sandbox can be idle before being terminated.
+        idle_timeout: Optional[int] = None,
         workdir: Optional[str] = None,  # Working directory of the sandbox.
         gpu: GPU_T = None,
         cloud: Optional[str] = None,
@@ -237,23 +306,31 @@ class _Sandbox(_Object, type_prefix="sb"):
         volumes: dict[
             Union[str, os.PathLike], Union[_Volume, _CloudBucketMount]
         ] = {},  # Mount points for Modal Volumes and CloudBucketMounts
-        pty_info: Optional[api_pb2.PTYInfo] = None,
+        pty: bool = False,  # Enable a PTY for the Sandbox
         # List of ports to tunnel into the sandbox. Encrypted ports are tunneled with TLS.
         encrypted_ports: Sequence[int] = [],
+        # List of encrypted ports to tunnel into the sandbox, using HTTP/2.
+        h2_ports: Sequence[int] = [],
         # List of ports to tunnel into the sandbox without encryption.
         unencrypted_ports: Sequence[int] = [],
         # Reference to a Modal Proxy to use in front of this Sandbox.
         proxy: Optional[_Proxy] = None,
+        # Enable verbose logging for sandbox operations.
+        verbose: bool = False,
+        experimental_options: Optional[dict[str, bool]] = None,
         # Enable memory snapshots.
         _experimental_enable_snapshot: bool = False,
         _experimental_scheduler_placement: Optional[
             SchedulerPlacement
         ] = None,  # Experimental controls over fine-grained scheduling (alpha).
         client: Optional[_Client] = None,
+        environment_name: Optional[str] = None,  # *DEPRECATED* Optionally override the default environment
+        pty_info: Optional[api_pb2.PTYInfo] = None,  # *DEPRECATED* Use `pty` instead. `pty` will override `pty_info`.
     ) -> "_Sandbox":
         """
-        Create a new Sandbox to run untrusted, arbitrary code. The Sandbox's corresponding container
-        will be created asynchronously.
+        Create a new Sandbox to run untrusted, arbitrary code.
+
+        The Sandbox's corresponding container will be created asynchronously.
 
         **Usage**
 
@@ -264,14 +341,33 @@ class _Sandbox(_Object, type_prefix="sb"):
         sandbox.wait()
         ```
         """
+        if environment_name is not None:
+            deprecation_warning(
+                (2025, 7, 16),
+                "Passing `environment_name` to `Sandbox.create` is deprecated and will be removed in a future release. "
+                "A sandbox's environment is determined by the app it is associated with.",
+            )
+
+        if pty_info is not None:
+            deprecation_warning(
+                (2025, 9, 12),
+                "The `pty_info` parameter is deprecated and will be removed in a future release. "
+                "Set the `pty` parameter to `True` instead.",
+            )
+
+        secrets = secrets or []
+        if env:
+            secrets = [*secrets, _Secret.from_dict(env)]
+
         return await _Sandbox._create(
-            *entrypoint_args,
+            *args,
             app=app,
-            environment_name=environment_name,
+            name=name,
             image=image,
             secrets=secrets,
             network_file_systems=network_file_systems,
             timeout=timeout,
+            idle_timeout=idle_timeout,
             workdir=workdir,
             gpu=gpu,
             cloud=cloud,
@@ -281,71 +377,79 @@ class _Sandbox(_Object, type_prefix="sb"):
             block_network=block_network,
             cidr_allowlist=cidr_allowlist,
             volumes=volumes,
-            pty_info=pty_info,
+            pty=pty,
             encrypted_ports=encrypted_ports,
+            h2_ports=h2_ports,
             unencrypted_ports=unencrypted_ports,
             proxy=proxy,
+            experimental_options=experimental_options,
             _experimental_enable_snapshot=_experimental_enable_snapshot,
             _experimental_scheduler_placement=_experimental_scheduler_placement,
             client=client,
+            verbose=verbose,
+            pty_info=pty_info,
         )
 
     @staticmethod
     async def _create(
-        *entrypoint_args: str,
-        app: Optional["modal.app._App"] = None,  # Optionally associate the sandbox with an app
-        environment_name: Optional[str] = None,  # Optionally override the default environment
-        image: Optional[_Image] = None,  # The image to run as the container for the sandbox.
-        secrets: Sequence[_Secret] = (),  # Environment variables to inject into the sandbox.
+        *args: str,
+        app: Optional["modal.app._App"] = None,
+        name: Optional[str] = None,
+        image: Optional[_Image] = None,
+        env: Optional[dict[str, Optional[str]]] = None,
+        secrets: Optional[Collection[_Secret]] = None,
         mounts: Sequence[_Mount] = (),
         network_file_systems: dict[Union[str, os.PathLike], _NetworkFileSystem] = {},
-        timeout: Optional[int] = None,  # Maximum execution time of the sandbox in seconds.
-        workdir: Optional[str] = None,  # Working directory of the sandbox.
+        timeout: int = 300,
+        idle_timeout: Optional[int] = None,
+        workdir: Optional[str] = None,
         gpu: GPU_T = None,
         cloud: Optional[str] = None,
-        region: Optional[Union[str, Sequence[str]]] = None,  # Region or regions to run the sandbox on.
-        # Specify, in fractional CPU cores, how many CPU cores to request.
-        # Or, pass (request, limit) to additionally specify a hard limit in fractional CPU cores.
-        # CPU throttling will prevent a container from exceeding its specified limit.
+        region: Optional[Union[str, Sequence[str]]] = None,
         cpu: Optional[Union[float, tuple[float, float]]] = None,
-        # Specify, in MiB, a memory request which is the minimum memory required.
-        # Or, pass (request, limit) to additionally specify a hard limit in MiB.
         memory: Optional[Union[int, tuple[int, int]]] = None,
-        block_network: bool = False,  # Whether to block network access
-        # List of CIDRs the sandbox is allowed to access. If None, all CIDRs are allowed.
+        block_network: bool = False,
         cidr_allowlist: Optional[Sequence[str]] = None,
-        volumes: dict[
-            Union[str, os.PathLike], Union[_Volume, _CloudBucketMount]
-        ] = {},  # Mount points for Modal Volumes and CloudBucketMounts
-        pty_info: Optional[api_pb2.PTYInfo] = None,
-        # List of ports to tunnel into the sandbox. Encrypted ports are tunneled with TLS.
+        volumes: dict[Union[str, os.PathLike], Union[_Volume, _CloudBucketMount]] = {},
+        pty: bool = False,
         encrypted_ports: Sequence[int] = [],
-        # List of ports to tunnel into the sandbox without encryption.
+        h2_ports: Sequence[int] = [],
         unencrypted_ports: Sequence[int] = [],
-        # Reference to a Modal Proxy to use in front of this Sandbox.
         proxy: Optional[_Proxy] = None,
-        # Enable memory snapshots.
+        experimental_options: Optional[dict[str, bool]] = None,
         _experimental_enable_snapshot: bool = False,
-        _experimental_scheduler_placement: Optional[
-            SchedulerPlacement
-        ] = None,  # Experimental controls over fine-grained scheduling (alpha).
+        _experimental_scheduler_placement: Optional[SchedulerPlacement] = None,
         client: Optional[_Client] = None,
+        verbose: bool = False,
+        pty_info: Optional[api_pb2.PTYInfo] = None,
     ):
-        # This method exposes some internal arguments (currently `mounts`) which are not in the public API
-        # `mounts` is currently only used by modal shell (cli) to provide a function's mounts to the
-        # sandbox that runs the shell session
+        """Private method used internally.
+
+        This method exposes some internal arguments (currently `mounts`) which are not in the public API.
+        `mounts` is currently only used by modal shell (cli) to provide a function's mounts to the
+        sandbox that runs the shell session.
+        """
         from .app import _App
 
-        environment_name = _get_environment_name(environment_name)
+        _validate_exec_args(args)
+        if name is not None:
+            check_object_name(name, "Sandbox")
+
+        if block_network and (encrypted_ports or h2_ports or unencrypted_ports):
+            raise InvalidError("Cannot specify open ports when `block_network` is enabled")
 
-        _validate_exec_args(entrypoint_args)
+        secrets = secrets or []
+        if env:
+            secrets = [*secrets, _Secret.from_dict(env)]
 
         # TODO(erikbern): Get rid of the `_new` method and create an already-hydrated object
         obj = _Sandbox._new(
-            entrypoint_args,
+            args,
             image=image or _default_image,
             secrets=secrets,
+            name=name,
             timeout=timeout,
+            idle_timeout=idle_timeout,
             workdir=workdir,
             gpu=gpu,
             cloud=cloud,
@@ -357,12 +461,16 @@ class _Sandbox(_Object, type_prefix="sb"):
             block_network=block_network,
             cidr_allowlist=cidr_allowlist,
             volumes=volumes,
+            pty=pty,
             pty_info=pty_info,
             encrypted_ports=encrypted_ports,
+            h2_ports=h2_ports,
             unencrypted_ports=unencrypted_ports,
             proxy=proxy,
+            experimental_options=experimental_options,
             _experimental_scheduler_placement=_experimental_scheduler_placement,
             enable_snapshot=_experimental_enable_snapshot,
+            verbose=verbose,
         )
         obj._enable_snapshot = _experimental_enable_snapshot
 
@@ -381,6 +489,7 @@ class _Sandbox(_Object, type_prefix="sb"):
             app_id = app.app_id
             app_client = app._client
         elif (container_app := _App._get_container_app()) is not None:
+            # implicit app/client provided by running in a modal Function
             app_id = container_app.app_id
             app_client = container_app._client
         else:
@@ -393,21 +502,47 @@ class _Sandbox(_Object, type_prefix="sb"):
                 "```",
             )
 
-        client = client or app_client or await _Client.from_env()
+        client = client or app_client
 
-        resolver = Resolver(client, environment_name=environment_name, app_id=app_id)
-        await resolver.load(obj)
+        resolver = Resolver()
+        load_context = LoadContext(client=client, app_id=app_id)
+        await resolver.load(obj, load_context)
         return obj
 
     def _hydrate_metadata(self, handle_metadata: Optional[Message]):
-        self._stdout: _StreamReader[str] = StreamReader[str](
+        self._stdout = StreamReader(
             api_pb2.FILE_DESCRIPTOR_STDOUT, self.object_id, "sandbox", self._client, by_line=True
         )
-        self._stderr: _StreamReader[str] = StreamReader[str](
+        self._stderr = StreamReader(
             api_pb2.FILE_DESCRIPTOR_STDERR, self.object_id, "sandbox", self._client, by_line=True
         )
         self._stdin = StreamWriter(self.object_id, "sandbox", self._client)
         self._result = None
+        self._task_id = None
+        self._tunnels = None
+        self._enable_snapshot = False
+        self._command_router_client = None
+
+    @staticmethod
+    async def from_name(
+        app_name: str,
+        name: str,
+        *,
+        environment_name: Optional[str] = None,
+        client: Optional[_Client] = None,
+    ) -> "_Sandbox":
+        """Get a running Sandbox by name from a deployed App.
+
+        Raises a modal.exception.NotFoundError if no running sandbox is found with the given name.
+        A Sandbox's name is the `name` argument passed to `Sandbox.create`.
+        """
+        if client is None:
+            client = await _Client.from_env()
+        env_name = _get_environment_name(environment_name)
+
+        req = api_pb2.SandboxGetFromNameRequest(sandbox_name=name, app_name=app_name, environment_name=env_name)
+        resp = await client.stub.SandboxGetFromName(req)
+        return _Sandbox._new_hydrated(resp.sandbox_id, client, None)
 
     @staticmethod
     async def from_id(sandbox_id: str, client: Optional[_Client] = None) -> "_Sandbox":
@@ -419,7 +554,7 @@ class _Sandbox(_Object, type_prefix="sb"):
             client = await _Client.from_env()
 
         req = api_pb2.SandboxWaitRequest(sandbox_id=sandbox_id, timeout=0)
-        resp = await retry_transient_errors(client.stub.SandboxWait, req)
+        resp = await client.stub.SandboxWait(req)
 
         obj = _Sandbox._new_hydrated(sandbox_id, client, None)
 
@@ -428,11 +563,25 @@ class _Sandbox(_Object, type_prefix="sb"):
 
         return obj
 
-    async def set_tags(self, tags: dict[str, str], *, client: Optional[_Client] = None):
+    async def get_tags(self) -> dict[str, str]:
+        """Fetches any tags (key-value pairs) currently attached to this Sandbox from the server."""
+        req = api_pb2.SandboxTagsGetRequest(sandbox_id=self.object_id)
+        try:
+            resp = await self._client.stub.SandboxTagsGet(req)
+        except GRPCError as exc:
+            raise InvalidError(exc.message) if exc.status == Status.INVALID_ARGUMENT else exc
+
+        return {tag.tag_name: tag.tag_value for tag in resp.tags}
+
+    async def set_tags(self, tags: dict[str, str], *, client: Optional[_Client] = None) -> None:
         """Set tags (key-value pairs) on the Sandbox. Tags can be used to filter results in `Sandbox.list`."""
         environment_name = _get_environment_name()
-        if client is None:
-            client = await _Client.from_env()
+        if client is not None:
+            deprecation_warning(
+                (2025, 9, 18),
+                "The `client` parameter is deprecated. Set `client` when creating the Sandbox instead "
+                "(in e.g. `Sandbox.create()`/`.from_id()`/`.from_name()`).",
+            )
 
         tags_list = [api_pb2.SandboxTag(tag_name=name, tag_value=value) for name, value in tags.items()]
 
@@ -442,7 +591,7 @@ class _Sandbox(_Object, type_prefix="sb"):
             tags=tags_list,
         )
         try:
-            await retry_transient_errors(client.stub.SandboxTagsSet, req)
+            await self._client.stub.SandboxTagsSet(req)
         except GRPCError as exc:
             raise InvalidError(exc.message) if exc.status == Status.INVALID_ARGUMENT else exc
 
@@ -454,7 +603,7 @@ class _Sandbox(_Object, type_prefix="sb"):
         """
         await self._get_task_id()  # Ensure the sandbox has started
         req = api_pb2.SandboxSnapshotFsRequest(sandbox_id=self.object_id, timeout=timeout)
-        resp = await retry_transient_errors(self._client.stub.SandboxSnapshotFs, req)
+        resp = await self._client.stub.SandboxSnapshotFs(req)
 
         if resp.result.status != api_pb2.GenericResult.GENERIC_STATUS_SUCCESS:
             raise ExecutionError(resp.result.exception)
@@ -462,12 +611,13 @@ class _Sandbox(_Object, type_prefix="sb"):
         image_id = resp.image_id
         metadata = resp.image_metadata
 
-        async def _load(self: _Image, resolver: Resolver, existing_object_id: Optional[str]):
+        async def _load(self: _Image, resolver: Resolver, load_context: LoadContext, existing_object_id: Optional[str]):
             # no need to hydrate again since we do it eagerly below
             pass
 
         rep = "Image()"
-        image = _Image._from_loader(_load, rep, hydrate_lazily=True)
+        # TODO: use ._new_hydrated instead
+        image = _Image._from_loader(_load, rep, hydrate_lazily=True, load_context_overrides=LoadContext.empty())
         image._hydrate(image_id, self._client, metadata)  # hydrating eagerly since we have all of the data
 
         return image
@@ -479,8 +629,9 @@ class _Sandbox(_Object, type_prefix="sb"):
 
         while True:
             req = api_pb2.SandboxWaitRequest(sandbox_id=self.object_id, timeout=10)
-            resp = await retry_transient_errors(self._client.stub.SandboxWait, req)
+            resp = await self._client.stub.SandboxWait(req)
             if resp.result.status:
+                logger.debug(f"Sandbox {self.object_id} wait completed with status {resp.result.status}")
                 self._result = resp.result
 
                 if resp.result.status == api_pb2.GenericResult.GENERIC_STATUS_TIMEOUT:
@@ -496,7 +647,7 @@ class _Sandbox(_Object, type_prefix="sb"):
 
         Returns a dictionary of `Tunnel` objects which are keyed by the container port.
 
-        NOTE: Previous to client [v0.64.153](/docs/reference/changelog#064153-2024-09-30), this
+        NOTE: Previous to client [v0.64.153](https://modal.com/docs/reference/changelog#064153-2024-09-30), this
         returned a list of `TunnelData` objects.
         """
 
@@ -504,7 +655,7 @@ class _Sandbox(_Object, type_prefix="sb"):
             return self._tunnels
 
         req = api_pb2.SandboxGetTunnelsRequest(sandbox_id=self.object_id, timeout=timeout)
-        resp = await retry_transient_errors(self._client.stub.SandboxGetTunnels, req)
+        resp = await self._client.stub.SandboxGetTunnels(req)
 
         # If we couldn't get the tunnels in time, report the timeout.
         if resp.result.status == api_pb2.GenericResult.GENERIC_STATUS_TIMEOUT:
@@ -517,14 +668,42 @@ class _Sandbox(_Object, type_prefix="sb"):
 
         return self._tunnels
 
+    async def create_connect_token(
+        self, user_metadata: Optional[Union[str, dict[str, Any]]] = None
+    ) -> SandboxConnectCredentials:
+        """
+        [Alpha] Create a token for making HTTP connections to the Sandbox.
+
+        Also accepts an optional user_metadata string or dict to associate with the token. This metadata
+        will be added to the headers by the proxy when forwarding requests to the Sandbox."""
+        if user_metadata is not None and isinstance(user_metadata, dict):
+            try:
+                user_metadata = json.dumps(user_metadata)
+            except Exception as e:
+                raise InvalidError(f"Failed to serialize user_metadata: {e}")
+
+        req = api_pb2.SandboxCreateConnectTokenRequest(sandbox_id=self.object_id, user_metadata=user_metadata)
+        resp = await self._client.stub.SandboxCreateConnectToken(req)
+        return SandboxConnectCredentials(resp.url, resp.token)
+
+    async def reload_volumes(self) -> None:
+        """Reload all Volumes mounted in the Sandbox.
+
+        Added in v1.1.0.
+        """
+        task_id = await self._get_task_id()
+        await self._client.stub.ContainerReloadVolumes(
+            api_pb2.ContainerReloadVolumesRequest(
+                task_id=task_id,
+            ),
+        )
+
     async def terminate(self) -> None:
         """Terminate Sandbox execution.
 
         This is a no-op if the Sandbox has already finished running."""
 
-        await retry_transient_errors(
-            self._client.stub.SandboxTerminate, api_pb2.SandboxTerminateRequest(sandbox_id=self.object_id)
-        )
+        await self._client.stub.SandboxTerminate(api_pb2.SandboxTerminateRequest(sandbox_id=self.object_id))
 
     async def poll(self) -> Optional[int]:
         """Check if the Sandbox has finished running.
@@ -533,7 +712,7 @@ class _Sandbox(_Object, type_prefix="sb"):
         """
 
         req = api_pb2.SandboxWaitRequest(sandbox_id=self.object_id, timeout=0)
-        resp = await retry_transient_errors(self._client.stub.SandboxWait, req)
+        resp = await self._client.stub.SandboxWait(req)
 
         if resp.result.status:
             self._result = resp.result
@@ -548,132 +727,329 @@ class _Sandbox(_Object, type_prefix="sb"):
                 await asyncio.sleep(0.5)
         return self._task_id
 
+    async def _get_command_router_client(self, task_id: str) -> Optional[TaskCommandRouterClient]:
+        if self._command_router_client is None:
+            # Attempt to initialize a router client. Returns None if the new exec path not enabled
+            # for this sandbox.
+            self._command_router_client = await TaskCommandRouterClient.try_init(self._client, task_id)
+        return self._command_router_client
+
     @overload
     async def exec(
         self,
-        *cmds: str,
-        pty_info: Optional[api_pb2.PTYInfo] = None,
+        *args: str,
         stdout: StreamType = StreamType.PIPE,
         stderr: StreamType = StreamType.PIPE,
         timeout: Optional[int] = None,
         workdir: Optional[str] = None,
-        secrets: Sequence[_Secret] = (),
+        env: Optional[dict[str, Optional[str]]] = None,
+        secrets: Optional[Collection[_Secret]] = None,
         text: Literal[True] = True,
         bufsize: Literal[-1, 1] = -1,
+        pty: bool = False,
+        pty_info: Optional[api_pb2.PTYInfo] = None,
         _pty_info: Optional[api_pb2.PTYInfo] = None,
     ) -> _ContainerProcess[str]: ...
 
     @overload
     async def exec(
         self,
-        *cmds: str,
-        pty_info: Optional[api_pb2.PTYInfo] = None,
+        *args: str,
         stdout: StreamType = StreamType.PIPE,
         stderr: StreamType = StreamType.PIPE,
         timeout: Optional[int] = None,
         workdir: Optional[str] = None,
-        secrets: Sequence[_Secret] = (),
+        env: Optional[dict[str, Optional[str]]] = None,
+        secrets: Optional[Collection[_Secret]] = None,
         text: Literal[False] = False,
         bufsize: Literal[-1, 1] = -1,
+        pty: bool = False,
+        pty_info: Optional[api_pb2.PTYInfo] = None,
         _pty_info: Optional[api_pb2.PTYInfo] = None,
     ) -> _ContainerProcess[bytes]: ...
 
     async def exec(
         self,
-        *cmds: str,
-        pty_info: Optional[api_pb2.PTYInfo] = None,  # Deprecated: internal use only
+        *args: str,
         stdout: StreamType = StreamType.PIPE,
         stderr: StreamType = StreamType.PIPE,
         timeout: Optional[int] = None,
         workdir: Optional[str] = None,
-        secrets: Sequence[_Secret] = (),
+        env: Optional[dict[str, Optional[str]]] = None,  # Environment variables to set during command execution.
+        secrets: Optional[
+            Collection[_Secret]
+        ] = None,  # Secrets to inject as environment variables during command execution.
         # Encode output as text.
         text: bool = True,
         # Control line-buffered output.
         # -1 means unbuffered, 1 means line-buffered (only available if `text=True`).
         bufsize: Literal[-1, 1] = -1,
-        # Internal option to set terminal size and metadata
-        _pty_info: Optional[api_pb2.PTYInfo] = None,
+        pty: bool = False,  # Enable a PTY for the command
+        _pty_info: Optional[api_pb2.PTYInfo] = None,  # *DEPRECATED* Use `pty` instead. `pty` will override `pty_info`.
+        pty_info: Optional[api_pb2.PTYInfo] = None,  # *DEPRECATED* Use `pty` instead. `pty` will override `pty_info`.
     ):
         """Execute a command in the Sandbox and return a ContainerProcess handle.
 
-        See the [`ContainerProcess`](/docs/reference/modal.container_process#modalcontainer_processcontainerprocess)
+        See the [`ContainerProcess`](https://modal.com/docs/reference/modal.container_process#modalcontainer_processcontainerprocess)
         docs for more information.
 
         **Usage**
 
-        ```python
-        app = modal.App.lookup("my-app", create_if_missing=True)
-
-        sandbox = modal.Sandbox.create("sleep", "infinity", app=app)
-
-        process = sandbox.exec("bash", "-c", "for i in $(seq 1 10); do echo foo $i; sleep 0.5; done")
-
+        ```python fixture:sandbox
+        process = sandbox.exec("bash", "-c", "for i in $(seq 1 3); do echo foo $i; sleep 0.1; done")
         for line in process.stdout:
             print(line)
         ```
         """
+        if pty_info is not None or _pty_info is not None:
+            deprecation_warning(
+                (2025, 9, 12),
+                "The `_pty_info` and `pty_info` parameters are deprecated and will be removed in a future release. "
+                "Set the `pty` parameter to `True` instead.",
+            )
+        pty_info = _pty_info or pty_info
+        if pty:
+            pty_info = self._default_pty_info()
+
+        return await self._exec(
+            *args,
+            pty_info=pty_info,
+            stdout=stdout,
+            stderr=stderr,
+            timeout=timeout,
+            workdir=workdir,
+            env=env,
+            secrets=secrets,
+            text=text,
+            bufsize=bufsize,
+        )
+
+    async def _exec(
+        self,
+        *args: str,
+        pty_info: Optional[api_pb2.PTYInfo] = None,
+        stdout: StreamType = StreamType.PIPE,
+        stderr: StreamType = StreamType.PIPE,
+        timeout: Optional[int] = None,
+        workdir: Optional[str] = None,
+        env: Optional[dict[str, Optional[str]]] = None,
+        secrets: Optional[Collection[_Secret]] = None,
+        text: bool = True,
+        bufsize: Literal[-1, 1] = -1,
+    ) -> Union[_ContainerProcess[bytes], _ContainerProcess[str]]:
+        """Private method used internally.
 
+        This method exposes some internal arguments (currently `pty_info`) which are not in the public API.
+        """
         if workdir is not None and not workdir.startswith("/"):
             raise InvalidError(f"workdir must be an absolute path, got: {workdir}")
-        _validate_exec_args(cmds)
+        _validate_exec_args(args)
+
+        secrets = secrets or []
+        if env:
+            secrets = [*secrets, _Secret.from_dict(env)]
 
         # Force secret resolution so we can pass the secret IDs to the backend.
         secret_coros = [secret.hydrate(client=self._client) for secret in secrets]
         await TaskContext.gather(*secret_coros)
 
         task_id = await self._get_task_id()
+        kwargs = {
+            "task_id": task_id,
+            "pty_info": pty_info,
+            "stdout": stdout,
+            "stderr": stderr,
+            "timeout": timeout,
+            "workdir": workdir,
+            "secret_ids": [secret.object_id for secret in secrets],
+            "text": text,
+            "bufsize": bufsize,
+            "runtime_debug": config.get("function_runtime_debug"),
+        }
+        # NB: This must come after the task ID is set, since the sandbox must be
+        # scheduled before we can create a router client.
+        if (command_router_client := await self._get_command_router_client(task_id)) is not None:
+            kwargs["command_router_client"] = command_router_client
+            return await self._exec_through_command_router(*args, **kwargs)
+        else:
+            return await self._exec_through_server(*args, **kwargs)
+
+    async def _exec_through_server(
+        self,
+        *args: str,
+        task_id: str,
+        pty_info: Optional[api_pb2.PTYInfo] = None,
+        stdout: StreamType = StreamType.PIPE,
+        stderr: StreamType = StreamType.PIPE,
+        timeout: Optional[int] = None,
+        workdir: Optional[str] = None,
+        secret_ids: Optional[Collection[str]] = None,
+        text: bool = True,
+        bufsize: Literal[-1, 1] = -1,
+        runtime_debug: bool = False,
+    ) -> Union[_ContainerProcess[bytes], _ContainerProcess[str]]:
+        """Execute a command through the Modal server."""
         req = api_pb2.ContainerExecRequest(
             task_id=task_id,
-            command=cmds,
-            pty_info=_pty_info or pty_info,
-            runtime_debug=config.get("function_runtime_debug"),
+            command=args,
+            pty_info=pty_info,
+            runtime_debug=runtime_debug,
             timeout_secs=timeout or 0,
             workdir=workdir,
-            secret_ids=[secret.object_id for secret in secrets],
+            secret_ids=secret_ids,
         )
-        resp = await retry_transient_errors(self._client.stub.ContainerExec, req)
+        resp = await self._client.stub.ContainerExec(req)
         by_line = bufsize == 1
-        return _ContainerProcess(resp.exec_id, self._client, stdout=stdout, stderr=stderr, text=text, by_line=by_line)
+        exec_deadline = time.monotonic() + int(timeout) + CONTAINER_EXEC_TIMEOUT_BUFFER if timeout else None
+        logger.debug(f"Created ContainerProcess for exec_id {resp.exec_id} on Sandbox {self.object_id}")
+        return _ContainerProcess(
+            resp.exec_id,
+            task_id,
+            self._client,
+            stdout=stdout,
+            stderr=stderr,
+            text=text,
+            exec_deadline=exec_deadline,
+            by_line=by_line,
+        )
+
+    async def _exec_through_command_router(
+        self,
+        *args: str,
+        task_id: str,
+        command_router_client: TaskCommandRouterClient,
+        pty_info: Optional[api_pb2.PTYInfo] = None,
+        stdout: StreamType = StreamType.PIPE,
+        stderr: StreamType = StreamType.PIPE,
+        timeout: Optional[int] = None,
+        workdir: Optional[str] = None,
+        secret_ids: Optional[Collection[str]] = None,
+        text: bool = True,
+        bufsize: Literal[-1, 1] = -1,
+        runtime_debug: bool = False,
+    ) -> Union[_ContainerProcess[bytes], _ContainerProcess[str]]:
+        """Execute a command through a task command router running on the Modal worker."""
+
+        # Generate a random process ID to use as a combination of idempotency key/process identifier.
+        process_id = str(uuid.uuid4())
+        if stdout == StreamType.PIPE:
+            stdout_config = sr_pb2.TaskExecStdoutConfig.TASK_EXEC_STDOUT_CONFIG_PIPE
+        elif stdout == StreamType.DEVNULL:
+            stdout_config = sr_pb2.TaskExecStdoutConfig.TASK_EXEC_STDOUT_CONFIG_DEVNULL
+        elif stdout == StreamType.STDOUT:
+            # TODO(saltzm): This is a behavior change from the old implementation. We should
+            # probably implement the old behavior of printing to stdout before moving out of beta.
+            raise NotImplementedError(
+                "Currently the STDOUT stream type is not supported when using exec "
+                "through a task command router, which is currently in beta."
+            )
+        else:
+            raise ValueError("Unsupported StreamType for stdout")
+
+        if stderr == StreamType.PIPE:
+            stderr_config = sr_pb2.TaskExecStderrConfig.TASK_EXEC_STDERR_CONFIG_PIPE
+        elif stderr == StreamType.DEVNULL:
+            stderr_config = sr_pb2.TaskExecStderrConfig.TASK_EXEC_STDERR_CONFIG_DEVNULL
+        elif stderr == StreamType.STDOUT:
+            stderr_config = sr_pb2.TaskExecStderrConfig.TASK_EXEC_STDERR_CONFIG_STDOUT
+        else:
+            raise ValueError("Unsupported StreamType for stderr")
+
+        # Start the process.
+        start_req = sr_pb2.TaskExecStartRequest(
+            task_id=task_id,
+            exec_id=process_id,
+            command_args=args,
+            stdout_config=stdout_config,
+            stderr_config=stderr_config,
+            timeout_secs=timeout,
+            workdir=workdir,
+            secret_ids=secret_ids,
+            pty_info=pty_info,
+            runtime_debug=runtime_debug,
+        )
+        _ = await command_router_client.exec_start(start_req)
+
+        return _ContainerProcess(
+            process_id,
+            task_id,
+            self._client,
+            command_router_client=command_router_client,
+            stdout=stdout,
+            stderr=stderr,
+            text=text,
+            by_line=bufsize == 1,
+            exec_deadline=time.monotonic() + int(timeout) if timeout else None,
+        )
 
     async def _experimental_snapshot(self) -> _SandboxSnapshot:
         await self._get_task_id()
         snap_req = api_pb2.SandboxSnapshotRequest(sandbox_id=self.object_id)
-        snap_resp = await retry_transient_errors(self._client.stub.SandboxSnapshot, snap_req)
+        snap_resp = await self._client.stub.SandboxSnapshot(snap_req)
 
         snapshot_id = snap_resp.snapshot_id
 
         # wait for the snapshot to succeed. this is implemented as a second idempotent rpc
         # because the snapshot itself may take a while to complete.
         wait_req = api_pb2.SandboxSnapshotWaitRequest(snapshot_id=snapshot_id, timeout=55.0)
-        wait_resp = await retry_transient_errors(self._client.stub.SandboxSnapshotWait, wait_req)
+        wait_resp = await self._client.stub.SandboxSnapshotWait(wait_req)
         if wait_resp.result.status != api_pb2.GenericResult.GENERIC_STATUS_SUCCESS:
             raise ExecutionError(wait_resp.result.exception)
 
-        async def _load(self: _SandboxSnapshot, resolver: Resolver, existing_object_id: Optional[str]):
+        async def _load(
+            self: _SandboxSnapshot, resolver: Resolver, load_context: LoadContext, existing_object_id: Optional[str]
+        ):
             # we eagerly hydrate the sandbox snapshot below
             pass
 
         rep = "SandboxSnapshot()"
-        obj = _SandboxSnapshot._from_loader(_load, rep, hydrate_lazily=True)
+        # TODO: use ._new_hydrated instead
+        obj = _SandboxSnapshot._from_loader(_load, rep, hydrate_lazily=True, load_context_overrides=LoadContext.empty())
         obj._hydrate(snapshot_id, self._client, None)
 
         return obj
 
     @staticmethod
-    async def _experimental_from_snapshot(snapshot: _SandboxSnapshot, client: Optional[_Client] = None):
+    async def _experimental_from_snapshot(
+        snapshot: _SandboxSnapshot,
+        client: Optional[_Client] = None,
+        *,
+        name: Optional[str] = _DEFAULT_SANDBOX_NAME_OVERRIDE,
+    ):
         client = client or await _Client.from_env()
 
-        restore_req = api_pb2.SandboxRestoreRequest(snapshot_id=snapshot.object_id)
-        restore_resp: api_pb2.SandboxRestoreResponse = await retry_transient_errors(
-            client.stub.SandboxRestore, restore_req
-        )
+        if name is not None and name != _DEFAULT_SANDBOX_NAME_OVERRIDE:
+            check_object_name(name, "Sandbox")
+
+        if name is _DEFAULT_SANDBOX_NAME_OVERRIDE:
+            restore_req = api_pb2.SandboxRestoreRequest(
+                snapshot_id=snapshot.object_id,
+                sandbox_name_override_type=api_pb2.SandboxRestoreRequest.SANDBOX_NAME_OVERRIDE_TYPE_UNSPECIFIED,
+            )
+        elif name is None:
+            restore_req = api_pb2.SandboxRestoreRequest(
+                snapshot_id=snapshot.object_id,
+                sandbox_name_override_type=api_pb2.SandboxRestoreRequest.SANDBOX_NAME_OVERRIDE_TYPE_NONE,
+            )
+        else:
+            restore_req = api_pb2.SandboxRestoreRequest(
+                snapshot_id=snapshot.object_id,
+                sandbox_name_override=name,
+                sandbox_name_override_type=api_pb2.SandboxRestoreRequest.SANDBOX_NAME_OVERRIDE_TYPE_STRING,
+            )
+        try:
+            restore_resp: api_pb2.SandboxRestoreResponse = await client.stub.SandboxRestore(restore_req)
+        except GRPCError as exc:
+            if exc.status == Status.ALREADY_EXISTS:
+                raise AlreadyExistsError(exc.message)
+            raise exc
+
         sandbox = await _Sandbox.from_id(restore_resp.sandbox_id, client)
 
         task_id_req = api_pb2.SandboxGetTaskIdRequest(
             sandbox_id=restore_resp.sandbox_id, wait_until_ready=True, timeout=55.0
         )
-        resp = await retry_transient_errors(client.stub.SandboxGetTaskId, task_id_req)
+        resp = await client.stub.SandboxGetTaskId(task_id_req)
         if resp.task_result.status not in [
             api_pb2.GenericResult.GENERIC_STATUS_UNSPECIFIED,
             api_pb2.GenericResult.GENERIC_STATUS_SUCCESS,
@@ -700,9 +1076,9 @@ class _Sandbox(_Object, type_prefix="sb"):
         path: str,
         mode: Union["_typeshed.OpenTextMode", "_typeshed.OpenBinaryMode"] = "r",
     ):
-        """Open a file in the Sandbox and return a FileIO handle.
+        """[Alpha] Open a file in the Sandbox and return a FileIO handle.
 
-        See the [`FileIO`](/docs/reference/modal.file_io#modalfile_iofileio) docs for more information.
+        See the [`FileIO`](https://modal.com/docs/reference/modal.file_io#modalfile_iofileio) docs for more information.
 
         **Usage**
 
@@ -717,17 +1093,17 @@ class _Sandbox(_Object, type_prefix="sb"):
         return await _FileIO.create(path, mode, self._client, task_id)
 
     async def ls(self, path: str) -> list[str]:
-        """List the contents of a directory in the Sandbox."""
+        """[Alpha] List the contents of a directory in the Sandbox."""
         task_id = await self._get_task_id()
         return await _FileIO.ls(path, self._client, task_id)
 
     async def mkdir(self, path: str, parents: bool = False) -> None:
-        """Create a new directory in the Sandbox."""
+        """[Alpha] Create a new directory in the Sandbox."""
         task_id = await self._get_task_id()
         return await _FileIO.mkdir(path, self._client, task_id, parents)
 
     async def rm(self, path: str, recursive: bool = False) -> None:
-        """Remove a file or directory in the Sandbox."""
+        """[Alpha] Remove a file or directory in the Sandbox."""
         task_id = await self._get_task_id()
         return await _FileIO.rm(path, self._client, task_id, recursive)
 
@@ -738,7 +1114,7 @@ class _Sandbox(_Object, type_prefix="sb"):
         recursive: Optional[bool] = None,
         timeout: Optional[int] = None,
     ) -> AsyncIterator[FileWatchEvent]:
-        """Watch a file or directory in the Sandbox for changes."""
+        """[Alpha] Watch a file or directory in the Sandbox for changes."""
         task_id = await self._get_task_id()
         async for event in _FileIO.watch(path, self._client, task_id, filter, recursive, timeout):
             yield event
@@ -746,7 +1122,7 @@ class _Sandbox(_Object, type_prefix="sb"):
     @property
     def stdout(self) -> _StreamReader[str]:
         """
-        [`StreamReader`](/docs/reference/modal.io_streams#modalio_streamsstreamreader) for
+        [`StreamReader`](https://modal.com/docs/reference/modal.io_streams#modalio_streamsstreamreader) for
         the sandbox's stdout stream.
         """
 
@@ -754,7 +1130,7 @@ class _Sandbox(_Object, type_prefix="sb"):
 
     @property
     def stderr(self) -> _StreamReader[str]:
-        """[`StreamReader`](/docs/reference/modal.io_streams#modalio_streamsstreamreader) for
+        """[`StreamReader`](https://modal.com/docs/reference/modal.io_streams#modalio_streamsstreamreader) for
         the Sandbox's stderr stream.
         """
 
@@ -763,7 +1139,7 @@ class _Sandbox(_Object, type_prefix="sb"):
     @property
     def stdin(self) -> _StreamWriter:
         """
-        [`StreamWriter`](/docs/reference/modal.io_streams#modalio_streamsstreamwriter) for
+        [`StreamWriter`](https://modal.com/docs/reference/modal.io_streams#modalio_streamsstreamwriter) for
         the Sandbox's stdin stream.
         """
 
@@ -808,7 +1184,7 @@ class _Sandbox(_Object, type_prefix="sb"):
 
             # Fetches a batch of sandboxes.
             try:
-                resp = await retry_transient_errors(client.stub.SandboxList, req)
+                resp = await client.stub.SandboxList(req)
             except GRPCError as exc:
                 raise InvalidError(exc.message) if exc.status == Status.INVALID_ARGUMENT else exc