lightning-sdk 0.1.3__py3-none-any.whl → 0.1.47__py3-none-any.whl

lightning_sdk/mmt/base.py

@@ -0,0 +1,326 @@
+from abc import abstractmethod
+from typing import TYPE_CHECKING, Dict, List, Optional, Protocol, Tuple, Union
+
+if TYPE_CHECKING:
+    from lightning_sdk.job.base import MachineDict
+    from lightning_sdk.machine import Machine
+    from lightning_sdk.organization import Organization
+    from lightning_sdk.status import Status
+    from lightning_sdk.studio import Studio
+    from lightning_sdk.teamspace import Teamspace
+    from lightning_sdk.user import User
+
+from lightning_sdk.job.base import _BaseJob
+from lightning_sdk.utils.resolve import _resolve_deprecated_cluster
+
+
+class MMTMachine(Protocol):
+    """A single machine in multi-machine training."""
+
+    @property
+    def name(self) -> str:
+        """The Name of the individual machine. Usually corresponds to the rank."""
+        ...
+
+    @property
+    def machine(self) -> "Machine":
+        """The actual machine type this node is running on."""
+        ...
+
+    @property
+    def artifact_path(self) -> Optional[str]:
+        """The path to the artifacts of this job."""
+        ...
+
+    @property
+    def status(self) -> "Status":
+        """The status of this job."""
+        ...
+
+    @property
+    def logs(self) -> str:
+        """The logs of the given machine."""
+        ...
+
+    def dict(self) -> "MachineDict":
+        """Dict representation of the given machine."""
+        ...
+
+
+class _BaseMMT(_BaseJob):
+    """Base interface to all job types."""
+
+    @classmethod
+    def run(
+        cls,
+        name: str,
+        machine: "Machine",
+        num_machines: int,
+        command: Optional[str] = None,
+        studio: Union["Studio", str, None] = None,
+        image: Optional[str] = None,
+        teamspace: Union[str, "Teamspace", None] = None,
+        org: Union[str, "Organization", None] = None,
+        user: Union[str, "User", None] = None,
+        cloud_account: Optional[str] = None,
+        env: Optional[Dict[str, str]] = None,
+        interruptible: bool = False,
+        image_credentials: Optional[str] = None,
+        cloud_account_auth: bool = False,
+        artifacts_local: Optional[str] = None,
+        artifacts_remote: Optional[str] = None,
+        entrypoint: str = "sh -c",
+        cluster: Optional[str] = None,  # deprecated in favor of cloud_account
+    ) -> "_BaseMMT":
+        """Run async workloads using a docker image across multiple machines.
+
+        Args:
+            name: The name of the job. Needs to be unique within the teamspace.
+            machine: The machine type to run the job on. One of {", ".join(_MACHINE_VALUES)}.
+            num_machine: The number of machines to run on.
+            command: The command to run inside your job. Required if using a studio. Optional if using an image.
+                If not provided for images, will run the container entrypoint and default command.
+            studio: The studio env to run the job with. Mutually exclusive with image.
+            image: The docker image to run the job with. Mutually exclusive with studio.
+            teamspace: The teamspace the job should be associated with. Defaults to the current teamspace.
+            org: The organization owning the teamspace (if any). Defaults to the current organization.
+            user: The user owning the teamspace (if any). Defaults to the current user.
+            cloud_account: The cloud account to run the job on.
+                Defaults to the studio cloud account if running with studio compute env.
+                If not provided will fall back to the teamspaces default cloud account.
+            env: Environment variables to set inside the job.
+            interruptible: Whether the job should run on interruptible instances. They are cheaper but can be preempted.
+            image_credentials: The credentials used to pull the image. Required if the image is private.
+                This should be the name of the respective credentials secret created on the Lightning AI platform.
+            cloud_account_auth: Whether to authenticate with the cloud account to pull the image.
+                Required if the registry is part of a cloud provider (e.g. ECR).
+            artifacts_local: The path of inside the docker container, you want to persist images from.
+                CAUTION: When setting this to "/", it will effectively erase your container.
+                Only supported for jobs with a docker image compute environment.
+            artifacts_remote: The remote storage to persist your artifacts to.
+                Should be of format <CONNECTION_TYPE>:<CONNECTION_NAME>:<PATH_WITHIN_CONNECTION>.
+                PATH_WITHIN_CONNECTION hereby is a path relative to the connection's root.
+                E.g. efs:data:some-path would result in an EFS connection named `data` and to the path `some-path`
+                within it.
+                Note that the connection needs to be added to the teamspace already in order for it to be found.
+                Only supported for jobs with a docker image compute environment.
+            entrypoint: The entrypoint of your docker container. Defaults to `sh -c` which
+                just runs the provided command in a standard shell.
+                To use the pre-defined entrypoint of the provided image, set this to an empty string.
+                Only applicable when submitting docker jobs.
+        """
+        from lightning_sdk.studio import Studio
+
+        cloud_account = _resolve_deprecated_cluster(cloud_account, cluster)
+
+        if num_machines <= 1:
+            raise ValueError("Multi-Machine training cannot be run with less than 2 Machines")
+
+        if not name:
+            raise ValueError("A job needs to have a name!")
+
+        if image is None:
+            if not isinstance(studio, Studio):
+                studio = Studio(
+                    name=studio, teamspace=teamspace, org=org, user=user, cloud_account=cloud_account, create_ok=False
+                )
+
+            # studio is a Studio instance at this point
+            if teamspace is None:
+                teamspace = studio.teamspace
+            else:
+                teamspace_name = teamspace if isinstance(teamspace, str) else teamspace.name
+
+                if studio.teamspace.name != teamspace_name:
+                    raise ValueError(
+                        "Studio teamspace does not match provided teamspace. "
+                        "Can only run jobs with Studio envs in the teamspace of that Studio."
+                    )
+
+            if cloud_account is None:
+                cloud_account = studio.cloud_account
+
+            if cloud_account != studio.cloud_account:
+                raise ValueError(
+                    "Studio cloud_account does not match provided cloud_account. "
+                    "Can only run jobs with Studio envs in the same cloud_account."
+                )
+
+            if image_credentials is not None:
+                raise ValueError("image_credentials is only supported when using a custom image")
+
+            if cloud_account_auth:
+                raise ValueError("cloud_account_auth is only supported when using a custom image")
+
+            if artifacts_local is not None or artifacts_remote is not None:
+                raise ValueError(
+                    "Specifying artifacts persistence is supported for docker images only. "
+                    "Other jobs will automatically persist artifacts to the teamspace distributed filesystem."
+                )
+
+            if entrypoint != "sh -c":
+                raise ValueError("Specifying the entrypoint has no effect for jobs with Studio envs.")
+
+        else:
+            if studio is not None:
+                raise RuntimeError(
+                    "image and studio are mutually exclusive as both define the environment to run the job in"
+                )
+
+            # they either need to specified both or none of them
+            if bool(artifacts_local) != bool(artifacts_remote):
+                raise ValueError("Artifact persistence requires both artifacts_local and artifacts_remote to be set")
+
+            if artifacts_remote and len(artifacts_remote.split(":")) != 3:
+                raise ValueError(
+                    "Artifact persistence requires exactly three arguments separated by colon of kind "
+                    f"<CONNECTION_TYPE>:<CONNECTION_NAME>:<PATH_WITHIN_CONNECTION>, got {artifacts_local}"
+                )
+
+        inst = cls(name=name, teamspace=teamspace, org=org, user=user, _fetch_job=False)
+        inst._submit(
+            num_machines=num_machines,
+            machine=machine,
+            cloud_account=cloud_account,
+            command=command,
+            studio=studio,
+            image=image,
+            env=env,
+            interruptible=interruptible,
+            image_credentials=image_credentials,
+            cloud_account_auth=cloud_account_auth,
+            artifacts_local=artifacts_local,
+            artifacts_remote=artifacts_remote,
+            entrypoint=entrypoint,
+        )
+        return inst
+
+    @abstractmethod
+    def _submit(
+        self,
+        num_machines: int,
+        machine: "Machine",
+        command: Optional[str] = None,
+        studio: Optional["Studio"] = None,
+        image: Optional[str] = None,
+        env: Optional[Dict[str, str]] = None,
+        interruptible: bool = False,
+        cloud_account: Optional[str] = None,
+        image_credentials: Optional[str] = None,
+        cloud_account_auth: bool = False,
+        artifacts_local: Optional[str] = None,
+        artifacts_remote: Optional[str] = None,
+        entrypoint: str = "sh -c",
+    ) -> None:
+        """Submit a new multi-machine job to the Lightning AI platform.
+
+        Args:
+            num_machines: The number of machines to run on.
+            machine: The machine type to run the job on. One of {", ".join(_MACHINE_VALUES)}.
+            command: The command to run inside your job. Required if using a studio. Optional if using an image.
+                If not provided for images, will run the container entrypoint and default command.
+            studio: The studio env to run the job with. Mutually exclusive with image.
+            image: The docker image to run the job with. Mutually exclusive with studio.
+            env: Environment variables to set inside the job.
+            interruptible: Whether the job should run on interruptible instances. They are cheaper but can be preempted.
+            cloud_account: The cloud account to run the job on.
+                Defaults to the studio cloud account if running with studio compute env.
+                If not provided will fall back to the teamspaces default cloud account.
+            image_credentials: The credentials used to pull the image. Required if the image is private.
+                This should be the name of the respective credentials secret created on the Lightning AI platform.
+            cloud_account_auth: Whether to authenticate with the cloud account to pull the image.
+                Required if the registry is part of a cloud provider (e.g. ECR).
+            artifacts_local: The path of inside the docker container, you want to persist images from.
+                CAUTION: When setting this to "/", it will effectively erase your container.
+                Only supported for jobs with a docker image compute environment.
+            artifacts_remote: The remote storage to persist your artifacts to.
+                Should be of format <CONNECTION_TYPE>:<CONNECTION_NAME>:<PATH_WITHIN_CONNECTION>.
+                PATH_WITHIN_CONNECTION hereby is a path relative to the connection's root.
+                E.g. efs:data:some-path would result in an EFS connection named `data` and to the path `some-path`
+                within it.
+                Note that the connection needs to be added to the teamspace already in order for it to be found.
+                Only supported for jobs with a docker image compute environment.
+            entrypoint: The entrypoint of your docker container. Defaults to sh -c.
+                To use the pre-defined entrypoint of the provided image, set this to an empty string.
+                Only applicable when submitting docker jobs.
+        """
+
+    @property
+    @abstractmethod
+    def machines(self) -> Tuple[MMTMachine, ...]:
+        """Returns the sub-jobs for each individual instance."""
+
+    @property
+    @abstractmethod
+    def machine(self) -> "Machine":
+        """Returns the machine type this job is running on."""
+
+    @abstractmethod
+    def stop(self) -> None:
+        """Stops the job."""
+
+    @abstractmethod
+    def delete(self) -> None:
+        """Deletes the job.
+
+        Caution: This also deletes all artifacts and snapshots associated with the job.
+        """
+
+    @property
+    @abstractmethod
+    def status(self) -> "Status":
+        """The current status of the job."""
+
+    @property
+    @abstractmethod
+    def artifact_path(self) -> Optional[str]:
+        """Path to the artifacts created by the job within the distributed teamspace filesystem."""
+
+    @property
+    @abstractmethod
+    def snapshot_path(self) -> Optional[str]:
+        """Path to the studio snapshot used to create the job within the distributed teamspace filesystem."""
+
+    @property
+    def share_path(self) -> Optional[str]:
+        """Path to the jobs share path."""
+        return None
+
+    @property
+    def name(self) -> str:
+        """The job's name."""
+        return self._name
+
+    @property
+    def teamspace(self) -> "Teamspace":
+        """The teamspace the job is part of."""
+        return self._teamspace
+
+    @property
+    def logs(self) -> str:
+        """Logs of the rank 0 machine."""
+        return self.machines[0].logs
+
+    def dict(
+        self
+    ) -> Dict[str, Union[str, "Studio", "Status", "Machine", None, List[Dict[str, Union[str, "Status", "Machine"]]]]]:
+        """Dict representation of this job."""
+        studio = self.studio
+
+        return {
+            "name": self.name,
+            "teamspace": f"{self.teamspace.owner.name}/{self.teamspace.name}",
+            "studio": studio.name if studio else None,
+            "image": self.image,
+            "command": self.command,
+            "status": self.status,
+            "machine": self.machine,
+            "machines": [
+                {"name": d["name"], "status": d["status"], "machine": d["machine"]}
+                for d in (x.dict() for x in self.machines)
+            ],
+        }
+
+    @abstractmethod
+    def _update_internal_job(self) -> None:
+        pass

lightning_sdk/mmt/mmt.py

@@ -0,0 +1,306 @@
+from contextlib import suppress
+from functools import lru_cache
+from typing import TYPE_CHECKING, Any, Dict, Optional, Tuple, Union
+
+from lightning_sdk.api.user_api import UserApi
+from lightning_sdk.job.job import _has_jobs_v2
+from lightning_sdk.mmt.base import MMTMachine, _BaseMMT
+from lightning_sdk.mmt.v1 import _MMTV1
+from lightning_sdk.mmt.v2 import _MMTV2
+from lightning_sdk.utils.resolve import _setup_logger
+
+if TYPE_CHECKING:
+    from lightning_sdk.machine import Machine
+    from lightning_sdk.organization import Organization
+    from lightning_sdk.status import Status
+    from lightning_sdk.studio import Studio
+    from lightning_sdk.teamspace import Teamspace
+    from lightning_sdk.user import User
+
+_logger = _setup_logger(__name__)
+
+
+@lru_cache(maxsize=None)
+def _has_mmt_v2() -> bool:
+    # users need both mmtv2 and jobsv2 flags in order for mmtv2 to work correctly
+    if not _has_jobs_v2():
+        return False
+
+    api = UserApi()
+    try:
+        return api._get_feature_flags().mmt_v2
+    except Exception:
+        return False
+
+
+class MMT(_BaseMMT):
+    """Class to submit and manage multi-machine jobs on the Lightning AI Platform."""
+
+    _force_v1: (
+        bool
+    ) = False  # required for studio plugin still working correctly as v2 currently does not support the studio env
+
+    def __init__(
+        self,
+        name: str,
+        teamspace: Union[str, "Teamspace", None] = None,
+        org: Union[str, "Organization", None] = None,
+        user: Union[str, "User", None] = None,
+        *,
+        _fetch_job: bool = True,
+    ) -> None:
+        """Fetch already existing jobs.
+
+        Args:
+            name: the name of the job
+            teamspace: the teamspace the job is part of
+            org: the name of the organization owning the :param`teamspace` in case it is owned by an org
+            user: the name of the user owning the :param`teamspace`
+                in case it is owned directly by a user instead of an org.
+        """
+        internal_mmt_cls = _MMTV2 if _has_mmt_v2() and not self._force_v1 else _MMTV1
+
+        self._internal_mmt = internal_mmt_cls(
+            name=name,
+            teamspace=teamspace,
+            org=org,
+            user=user,
+            _fetch_job=_fetch_job,
+        )
+
+    @classmethod
+    def run(
+        cls,
+        name: str,
+        num_machines: int,
+        machine: "Machine",
+        command: Optional[str] = None,
+        studio: Union["Studio", str, None] = None,
+        image: Union[str, None] = None,
+        teamspace: Union[str, "Teamspace", None] = None,
+        org: Union[str, "Organization", None] = None,
+        user: Union[str, "User", None] = None,
+        cloud_account: Optional[str] = None,
+        env: Optional[Dict[str, str]] = None,
+        interruptible: bool = False,
+        image_credentials: Optional[str] = None,
+        cloud_account_auth: bool = False,
+        artifacts_local: Optional[str] = None,
+        artifacts_remote: Optional[str] = None,
+        entrypoint: str = "sh -c",
+        cluster: Optional[str] = None,  # deprecated in favor of cloud_account
+    ) -> "MMT":
+        """Run async workloads using a docker image across multiple machines.
+
+        Args:
+            name: The name of the job. Needs to be unique within the teamspace.
+            machine: The machine type to run the job on. One of {", ".join(_MACHINE_VALUES)}.
+            num_machine: The number of machines to run on.
+            command: The command to run inside your job. Required if using a studio. Optional if using an image.
+                If not provided for images, will run the container entrypoint and default command.
+            studio: The studio env to run the job with. Mutually exclusive with image.
+            image: The docker image to run the job with. Mutually exclusive with studio.
+            teamspace: The teamspace the job should be associated with. Defaults to the current teamspace.
+            org: The organization owning the teamspace (if any). Defaults to the current organization.
+            user: The user owning the teamspace (if any). Defaults to the current user.
+            cloud_account: The cloud account to run the job on.
+                Defaults to the studio cloud account if running with studio compute env.
+                If not provided will fall back to the teamspaces default cloud account.
+            env: Environment variables to set inside the job.
+            interruptible: Whether the job should run on interruptible instances. They are cheaper but can be preempted.
+            image_credentials: The credentials used to pull the image. Required if the image is private.
+                This should be the name of the respective credentials secret created on the Lightning AI platform.
+            cloud_account_auth: Whether to authenticate with the cloud account to pull the image.
+                Required if the registry is part of a cloud provider (e.g. ECR).
+            artifacts_local: The path of inside the docker container, you want to persist images from.
+                CAUTION: When setting this to "/", it will effectively erase your container.
+                Only supported for jobs with a docker image compute environment.
+            artifacts_remote: The remote storage to persist your artifacts to.
+                Should be of format <CONNECTION_TYPE>:<CONNECTION_NAME>:<PATH_WITHIN_CONNECTION>.
+                PATH_WITHIN_CONNECTION hereby is a path relative to the connection's root.
+                E.g. efs:data:some-path would result in an EFS connection named `data` and to the path `some-path`
+                within it.
+                Note that the connection needs to be added to the teamspace already in order for it to be found.
+                Only supported for jobs with a docker image compute environment.
+            entrypoint: The entrypoint of your docker container. Defaults to sh -c.
+                To use the pre-defined entrypoint of the provided image, set this to an empty string.
+                Only applicable when submitting docker jobs.
+        """
+        ret_val = super().run(
+            name=name,
+            num_machines=num_machines,
+            machine=machine,
+            command=command,
+            studio=studio,
+            image=image,
+            teamspace=teamspace,
+            org=org,
+            user=user,
+            cloud_account=cloud_account,
+            env=env,
+            interruptible=interruptible,
+            image_credentials=image_credentials,
+            cloud_account_auth=cloud_account_auth,
+            artifacts_local=artifacts_local,
+            artifacts_remote=artifacts_remote,
+            cluster=cluster,  # deprecated in favor of cloud_account
+        )
+        # required for typing with "MMT"
+        assert isinstance(ret_val, cls)
+
+        msg = "Multi-Machine Job was successfully launched."
+
+        with suppress(NotImplementedError):
+            msg += f" View it at {ret_val.link}"
+
+        _logger.info(msg)
+        return ret_val
+
+    def _submit(
+        self,
+        num_machines: int,
+        machine: "Machine",
+        command: Optional[str] = None,
+        studio: Optional["Studio"] = None,
+        image: Optional[str] = None,
+        env: Optional[Dict[str, str]] = None,
+        interruptible: bool = False,
+        cloud_account: Optional[str] = None,
+        image_credentials: Optional[str] = None,
+        cloud_account_auth: bool = False,
+        artifacts_local: Optional[str] = None,
+        artifacts_remote: Optional[str] = None,
+        entrypoint: str = "sh -c",
+    ) -> "MMT":
+        """Submit a new multi-machine job to the Lightning AI platform.
+
+        Args:
+            num_machines: The number of machines to run on.
+            machine: The machine type to run the job on. One of {", ".join(_MACHINE_VALUES)}.
+            command: The command to run inside your job. Required if using a studio. Optional if using an image.
+                If not provided for images, will run the container entrypoint and default command.
+            studio: The studio env to run the job with. Mutually exclusive with image.
+            image: The docker image to run the job with. Mutually exclusive with studio.
+            env: Environment variables to set inside the job.
+            interruptible: Whether the job should run on interruptible instances. They are cheaper but can be preempted.
+            cloud_account: The cloud account to run the job on.
+                Defaults to the studio cloud account if running with studio compute env.
+                If not provided will fall back to the teamspaces default cloud account.
+            image_credentials: The credentials used to pull the image. Required if the image is private.
+                This should be the name of the respective credentials secret created on the Lightning AI platform.
+            cloud_account_auth: Whether to authenticate with the cloud account to pull the image.
+                Required if the registry is part of a cloud provider (e.g. ECR).
+            artifacts_local: The path of inside the docker container, you want to persist images from.
+                CAUTION: When setting this to "/", it will effectively erase your container.
+                Only supported for jobs with a docker image compute environment.
+            artifacts_remote: The remote storage to persist your artifacts to.
+                Should be of format <CONNECTION_TYPE>:<CONNECTION_NAME>:<PATH_WITHIN_CONNECTION>.
+                PATH_WITHIN_CONNECTION hereby is a path relative to the connection's root.
+                E.g. efs:data:some-path would result in an EFS connection named `data` and to the path `some-path`
+                within it.
+                Note that the connection needs to be added to the teamspace already in order for it to be found.
+                Only supported for jobs with a docker image compute environment.
+            entrypoint: The entrypoint of your docker container. Defaults to `sh -c` which
+                just runs the provided command in a standard shell.
+                To use the pre-defined entrypoint of the provided image, set this to an empty string.
+                Only applicable when submitting docker jobs.
+        """
+        self._job = self._internal_mmt._submit(
+            num_machines=num_machines,
+            machine=machine,
+            cloud_account=cloud_account,
+            command=command,
+            studio=studio,
+            image=image,
+            env=env,
+            interruptible=interruptible,
+            image_credentials=image_credentials,
+            cloud_account_auth=cloud_account_auth,
+            artifacts_local=artifacts_local,
+            artifacts_remote=artifacts_remote,
+        )
+        return self
+
+    def stop(self) -> None:
+        """Stops the job."""
+        return self._internal_mmt.stop()
+
+    def delete(self) -> None:
+        """Deletes the job.
+
+        Caution: This also deletes all artifacts and snapshots associated with the job.
+        """
+        return self._internal_mmt.delete()
+
+    @property
+    def status(self) -> "Status":
+        """The current status of the job (accumulated over all machines)."""
+        return self._internal_mmt.status
+
+    @property
+    def machines(self) -> Tuple[MMTMachine, ...]:
+        """Returns the sub-jobs for each individual instance."""
+        return self._internal_mmt.machines
+
+    @property
+    def machine(self) -> "Machine":
+        """Returns the machine type this job is running on."""
+        return self._internal_mmt.machine
+
+    @property
+    def artifact_path(self) -> Optional[str]:
+        """Path to the artifacts created by the job within the distributed teamspace filesystem."""
+        return self._internal_mmt.artifact_path
+
+    @property
+    def snapshot_path(self) -> Optional[str]:
+        """Path to the studio snapshot used to create the job within the distributed teamspace filesystem."""
+        return self._internal_mmt.snapshot_path
+
+    @property
+    def share_path(self) -> Optional[str]:
+        """Path to the jobs share path."""
+        return None
+
+    def _update_internal_job(self) -> None:
+        return self._internal_mmt._update_internal_job()
+
+    @property
+    def name(self) -> str:
+        """The job's name."""
+        return self._internal_mmt.name
+
+    @property
+    def teamspace(self) -> "Teamspace":
+        """The teamspace the job is part of."""
+        return self._internal_mmt._teamspace
+
+    @property
+    def link(self) -> str:
+        return self._internal_mmt.link
+
+    @property
+    def studio(self) -> Optional["Studio"]:
+        """The studio used to submit the MMT."""
+        return self._internal_mmt.studio
+
+    @property
+    def image(self) -> Optional[str]:
+        """The image used to submit the MMT."""
+        return self._internal_mmt.image
+
+    @property
+    def command(self) -> str:
+        """The command the MMT is running."""
+        return self._internal_mmt.command
+
+    def __getattr__(self, key: str) -> Any:
+        """Forward the attribute lookup to the internal job implementation."""
+        try:
+            return getattr(super(), key)
+        except AttributeError:
+            return getattr(self._internal_mmt, key)
+
+    @property
+    def _guaranteed_job(self) -> Any:
+        return self._internal_mmt._guaranteed_job