deployments.py

"""
Objects for specifying deployments and utilities for loading flows from deployments.
"""

import importlib
import json
import sys
from datetime import datetime
from pathlib import Path
from typing import Any, Dict, List, Optional
from uuid import UUID

import anyio
import pendulum
import yaml
from pydantic import BaseModel, Field, parse_obj_as, validator

from prefect.blocks.core import Block
from prefect.client.orion import OrionClient, get_client
from prefect.client.utilities import inject_client
from prefect.context import FlowRunContext, PrefectObjectRegistry
from prefect.exceptions import (
    BlockMissingCapabilities,
    ObjectAlreadyExists,
    ObjectNotFound,
)
from prefect.filesystems import LocalFileSystem
from prefect.flows import Flow
from prefect.infrastructure import Infrastructure, Process
from prefect.logging.loggers import flow_run_logger
from prefect.orion import schemas
from prefect.states import Scheduled
from prefect.tasks import Task
from prefect.utilities.asyncutils import run_sync_in_worker_thread, sync_compatible
from prefect.utilities.callables import ParameterSchema, parameter_schema
from prefect.utilities.dispatch import lookup_type
from prefect.utilities.filesystem import relative_path_to_current_platform, tmpchdir
from prefect.utilities.importtools import import_object
from prefect.utilities.slugify import slugify


@sync_compatible
@inject_client
async def run_deployment(
    name: str,
    client: OrionClient = None,
    parameters: dict = None,
    scheduled_time: datetime = None,
    flow_run_name: str = None,
    timeout: float = None,
    poll_interval: float = 5,
):
    """
    Create a flow run for a deployment and return it after completion or a timeout.

    This function will return when the created flow run enters any terminal state or
    the timeout is reached. If the timeout is reached and the flow run has not reached
    a terminal state, it will still be returned. When using a timeout, we suggest
    checking the state of the flow run if completion is important moving forward.

    Args:
        name: The deployment name in the form: '<flow-name>/<deployment-name>'
        parameters: Parameter overrides for this flow run. Merged with the deployment
            defaults
        scheduled_time: The time to schedule the flow run for, defaults to scheduling
            the flow run to start now.
        flow_run_name: A name for the created flow run
        timeout: The amount of time to wait for the flow run to complete before
            returning. Setting `timeout` to 0 will return the flow run immediately.
            Setting `timeout` to None will allow this function to poll indefinitely.
            Defaults to None
        poll_interval: The number of seconds between polls
    """
    if timeout is not None and timeout < 0:
        raise ValueError("`timeout` cannot be negative")

    if scheduled_time is None:
        scheduled_time = pendulum.now("UTC")

    parameters = parameters or {}

    deployment = await client.read_deployment_by_name(name)

    flow_run_ctx = FlowRunContext.get()
    if flow_run_ctx:
        # This was called from a flow. Link the flow run as a subflow.
        from prefect.engine import (
            Pending,
            _dynamic_key_for_task_run,
            collect_task_run_inputs,
        )

        task_inputs = {
            k: await collect_task_run_inputs(v) for k, v in parameters.items()
        }

        # Generate a task in the parent flow run to represent the result of the subflow
        dummy_task = Task(
            name=name,
            fn=lambda: None,
            version=deployment.version,
        )
        # Override the default task key to include the deployment name
        dummy_task.task_key = f"{__name__}.run_deployment.{slugify(name)}"
        parent_task_run = await client.create_task_run(
            task=dummy_task,
            flow_run_id=flow_run_ctx.flow_run.id,
            dynamic_key=_dynamic_key_for_task_run(flow_run_ctx, dummy_task),
            task_inputs=task_inputs,
            state=Pending(),
        )
        parent_task_run_id = parent_task_run.id
    else:
        parent_task_run_id = None

    flow_run = await client.create_flow_run_from_deployment(
        deployment.id,
        state=Scheduled(scheduled_time=scheduled_time),
        parameters=parameters,
        name=flow_run_name,
        parent_task_run_id=parent_task_run_id,
    )

    flow_run_id = flow_run.id

    if timeout == 0:
        return flow_run

    with anyio.move_on_after(timeout):
        while True:
            flow_run = await client.read_flow_run(flow_run_id)
            flow_state = flow_run.state
            if flow_state and flow_state.is_final():
                return flow_run
            await anyio.sleep(poll_interval)

    return flow_run


@inject_client
async def load_flow_from_flow_run(
    flow_run: schemas.core.FlowRun, client: OrionClient, ignore_storage: bool = False
) -> Flow:
    """
    Load a flow from the location/script provided in a deployment's storage document.

    If `ignore_storage=True` is provided, no pull from remote storage occurs.  This flag
    is largely for testing, and assumes the flow is already available locally.
    """
    deployment = await client.read_deployment(flow_run.deployment_id)

    if not ignore_storage:
        if deployment.storage_document_id:
            storage_document = await client.read_block_document(
                deployment.storage_document_id
            )
            storage_block = Block._from_block_document(storage_document)
        else:
            basepath = deployment.path or Path(deployment.manifest_path).parent
            storage_block = LocalFileSystem(basepath=basepath)

        sys.path.insert(0, ".")
        await storage_block.get_directory(from_path=deployment.path, local_path=".")

    flow_run_logger(flow_run).debug(
        f"Loading flow for deployment {deployment.name!r}..."
    )

    import_path = relative_path_to_current_platform(deployment.entrypoint)

    # for backwards compat
    if deployment.manifest_path:
        with open(deployment.manifest_path, "r") as f:
            import_path = json.load(f)["import_path"]
            import_path = (
                Path(deployment.manifest_path).parent / import_path
            ).absolute()
    flow = await run_sync_in_worker_thread(import_object, str(import_path))
    return flow


def load_deployments_from_yaml(
    path: str,
) -> PrefectObjectRegistry:
    """
    Load deployments from a yaml file.
    """
    with open(path, "r") as f:
        contents = f.read()

    # Parse into a yaml tree to retrieve separate documents
    nodes = yaml.compose_all(contents)

    with PrefectObjectRegistry(capture_failures=True) as registry:
        for node in nodes:
            with tmpchdir(path):
                deployment_dict = yaml.safe_load(yaml.serialize(node))
                # The return value is not necessary, just instantiating the Deployment
                # is enough to get it recorded on the registry
                parse_obj_as(Deployment, deployment_dict)

    return registry


class Deployment(BaseModel):
    """
    A Prefect Deployment definition, used for specifying and building deployments.

    Args:
        name: A name for the deployment (required).
        version: An optional version for the deployment; defaults to the flow's version
        description: An optional description of the deployment; defaults to the flow's description
        tags: An optional list of tags to associate with this deployment; note that tags are
            used only for organizational purposes. For delegating work to agents, see `work_queue_name`.
        schedule: A schedule to run this deployment on, once registered
        work_queue_name: The work queue that will handle this deployment's runs
        flow: The name of the flow this deployment encapsulates
        parameters: A dictionary of parameter values to pass to runs created from this deployment
        infrastructure: An optional infrastructure block used to configure infrastructure for runs;
            if not provided, will default to running this deployment in Agent subprocesses
        infra_overrides: A dictionary of dot delimited infrastructure overrides that will be applied at
            runtime; for example `env.CONFIG_KEY=config_value` or `namespace='prefect'`
        storage: An optional remote storage block used to store and retrieve this workflow;
            if not provided, will default to referencing this flow by its local path
        path: The path to the working directory for the workflow, relative to remote storage or,
            if stored on a local filesystem, an absolute path
        entrypoint: The path to the entrypoint for the workflow, always relative to the `path`
        parameter_openapi_schema: The parameter schema of the flow, including defaults.

    Examples:

        Create a new deployment using configuration defaults for an imported flow:

        >>> from my_project.flows import my_flow
        >>> from prefect.deployments import Deployment
        >>>
        >>> deployment = Deployment.build_from_flow(
        ...     flow=my_flow,
        ...     name="example",
        ...     version="1",
        ...     tags=["demo"],
        >>> )
        >>> deployment.apply()

        Create a new deployment with custom storage and an infrastructure override:

        >>> from my_project.flows import my_flow
        >>> from prefect.deployments import Deployment
        >>> from prefect.filesystems import S3

        >>> storage = S3.load("dev-bucket") # load a pre-defined block
        >>> deployment = Deployment.build_from_flow(
        ...     flow=my_flow,
        ...     name="s3-example",
        ...     version="2",
        ...     tags=["aws"],
        ...     storage=storage,
        ...     infra_overrides=dict("env.PREFECT_LOGGING_LEVEL"="DEBUG"),
        >>> )
        >>> deployment.apply()

    """

    class Config:
        validate_assignment = True
        extra = "forbid"

    @property
    def _editable_fields(self) -> List[str]:
        editable_fields = [
            "name",
            "description",
            "version",
            "work_queue_name",
            "tags",
            "parameters",
            "schedule",
            "infra_overrides",
        ]

        # if infrastructure is baked as a pre-saved block, then
        # editing its fields will not update anything
        if self.infrastructure._block_document_id:
            return editable_fields
        else:
            return editable_fields + ["infrastructure"]

    @property
    def location(self) -> str:
        """
        The 'location' that this deployment points to is given by `path` alone
        in the case of no remote storage, and otherwise by `storage.basepath / path`.

        The underlying flow entrypoint is interpreted relative to this location.
        """
        location = ""
        if self.storage:
            location = (
                self.storage.basepath + "/"
                if not self.storage.basepath.endswith("/")
                else ""
            )
        if self.path:
            location += self.path
        return location

    @sync_compatible
    async def to_yaml(self, path: Path) -> None:
        yaml_dict = self._yaml_dict()
        schema = self.schema()

        with open(path, "w") as f:
            # write header
            f.write(
                f"###\n### A complete description of a Prefect Deployment for flow {self.flow_name!r}\n###\n"
            )

            # write editable fields
            for field in self._editable_fields:
                # write any comments
                if schema["properties"][field].get("yaml_comment"):
                    f.write(f"# {schema['properties'][field]['yaml_comment']}\n")
                # write the field
                yaml.dump({field: yaml_dict[field]}, f, sort_keys=False)

            # write non-editable fields
            f.write("\n###\n### DO NOT EDIT BELOW THIS LINE\n###\n")
            yaml.dump(
                {k: v for k, v in yaml_dict.items() if k not in self._editable_fields},
                f,
                sort_keys=False,
            )

    def _yaml_dict(self) -> dict:
        """
        Returns a YAML-compatible representation of this deployment as a dictionary.
        """
        # avoids issues with UUIDs showing up in YAML
        all_fields = json.loads(
            self.json(
                exclude={
                    "storage": {"_filesystem", "filesystem", "_remote_file_system"}
                }
            )
        )
        if all_fields["storage"]:
            all_fields["storage"][
                "_block_type_slug"
            ] = self.storage.get_block_type_slug()
        if all_fields["infrastructure"]:
            all_fields["infrastructure"][
                "_block_type_slug"
            ] = self.infrastructure.get_block_type_slug()
        return all_fields

    # top level metadata
    name: str = Field(..., description="The name of the deployment.")
    description: Optional[str] = Field(
        default=None, description="An optional description of the deployment."
    )
    version: Optional[str] = Field(
        default=None, description="An optional version for the deployment."
    )
    tags: List[str] = Field(
        default_factory=list,
        description="One of more tags to apply to this deployment.",
    )
    schedule: schemas.schedules.SCHEDULE_TYPES = None
    flow_name: Optional[str] = Field(default=None, description="The name of the flow.")
    work_queue_name: Optional[str] = Field(
        "default",
        description="The work queue for the deployment.",
        yaml_comment="The work queue that will handle this deployment's runs",
    )

    # flow data
    parameters: Dict[str, Any] = Field(default_factory=dict)
    manifest_path: Optional[str] = Field(
        default=None,
        description="The path to the flow's manifest file, relative to the chosen storage.",
    )
    infrastructure: Infrastructure = Field(default_factory=Process)
    infra_overrides: Dict[str, Any] = Field(
        default_factory=dict,
        description="Overrides to apply to the base infrastructure block at runtime.",
    )
    storage: Optional[Block] = Field(
        None,
        help="The remote storage to use for this workflow.",
    )
    path: Optional[str] = Field(
        default=None,
        description="The path to the working directory for the workflow, relative to remote storage or an absolute path.",
    )
    entrypoint: Optional[str] = Field(
        default=None,
        description="The path to the entrypoint for the workflow, relative to the `path`.",
    )
    parameter_openapi_schema: ParameterSchema = Field(
        default_factory=ParameterSchema,
        description="The parameter schema of the flow, including defaults.",
    )

    @validator("infrastructure", pre=True)
    def infrastructure_must_have_capabilities(cls, value):
        if isinstance(value, dict):
            if "_block_type_slug" in value:
                # Replace private attribute with public for dispatch
                value["block_type_slug"] = value.pop("_block_type_slug")
            block = Block(**value)
        elif value is None:
            return value
        else:
            block = value

        if "run-infrastructure" not in block.get_block_capabilities():
            raise ValueError(
                "Infrastructure block must have 'run-infrastructure' capabilities."
            )
        return block

    @validator("storage", pre=True)
    def storage_must_have_capabilities(cls, value):
        if isinstance(value, dict):
            block_type = lookup_type(Block, value.pop("_block_type_slug"))
            block = block_type(**value)
        elif value is None:
            return value
        else:
            block = value

        capabilities = block.get_block_capabilities()
        if "get-directory" not in capabilities:
            raise ValueError(
                "Remote Storage block must have 'get-directory' capabilities."
            )
        return block

    @validator("parameter_openapi_schema", pre=True)
    def handle_openapi_schema(cls, value):
        """
        This method ensures setting a value of `None` is handled gracefully.
        """
        if value is None:
            return ParameterSchema()
        return value

    @classmethod
    @sync_compatible
    async def load_from_yaml(cls, path: str):
        with open(str(path), "r") as f:
            data = yaml.safe_load(f)

            # load blocks from server to ensure secret values are properly hydrated
            if data["storage"]:
                block_doc_name = data["storage"].get("_block_document_name")
                # if no doc name, this block is not stored on the server
                if block_doc_name:
                    block_slug = data["storage"]["_block_type_slug"]
                    block = await Block.load(f"{block_slug}/{block_doc_name}")
                    data["storage"] = block

            if data["infrastructure"]:
                block_doc_name = data["infrastructure"].get("_block_document_name")
                # if no doc name, this block is not stored on the server
                if block_doc_name:
                    block_slug = data["infrastructure"]["_block_type_slug"]
                    block = await Block.load(f"{block_slug}/{block_doc_name}")
                    data["infrastructure"] = block

            return cls(**data)

    @sync_compatible
    async def load(self) -> bool:
        """
        Queries the API for a deployment with this name for this flow, and if found, prepopulates
        any settings that were not set at initialization.

        Returns a boolean specifying whether a load was successful or not.

        Raises:
            - ValueError: if both name and flow name are not set
        """
        if not self.name or not self.flow_name:
            raise ValueError("Both a deployment name and flow name must be provided.")
        async with get_client() as client:
            try:
                deployment = await client.read_deployment_by_name(
                    f"{self.flow_name}/{self.name}"
                )
                if deployment.storage_document_id:
                    storage = Block._from_block_document(
                        await client.read_block_document(deployment.storage_document_id)
                    )

                excluded_fields = self.__fields_set__.union(
                    {"infrastructure", "storage"}
                )
                for field in set(self.__fields__.keys()) - excluded_fields:
                    new_value = getattr(deployment, field)
                    setattr(self, field, new_value)

                if "infrastructure" not in self.__fields_set__:
                    if deployment.infrastructure_document_id:
                        self.infrastructure = Block._from_block_document(
                            await client.read_block_document(
                                deployment.infrastructure_document_id
                            )
                        )
                if "storage" not in self.__fields_set__:
                    if deployment.storage_document_id:
                        self.storage = Block._from_block_document(
                            await client.read_block_document(
                                deployment.storage_document_id
                            )
                        )
            except ObjectNotFound:
                return False
        return True

    @sync_compatible
    async def update(self, ignore_none: bool = False, **kwargs):
        """
        Performs an in-place update with the provided settings.

        Args:
            ignore_none: if True, all `None` values are ignored when performing the update
        """
        unknown_keys = set(kwargs.keys()) - set(self.dict().keys())
        if unknown_keys:
            raise ValueError(
                f"Received unexpected attributes: {', '.join(unknown_keys)}"
            )
        for key, value in kwargs.items():
            if ignore_none and value is None:
                continue
            setattr(self, key, value)

    @sync_compatible
    async def upload_to_storage(
        self, storage_block: str = None, ignore_file: str = ".prefectignore"
    ) -> Optional[int]:
        """
        Uploads the workflow this deployment represents using a provided storage block;
        if no block is provided, defaults to configuring self for local storage.

        Args:
            storage_block: a string reference a remote storage block slug `$type/$name`; if provided,
                used to upload the workflow's project
            ignore_file: an optional path to a `.prefectignore` file that specifies filename patterns
                to ignore when uploading to remote storage; if not provided, looks for `.prefectignore`
                in the current working directory
        """
        deployment_path = None
        file_count = None
        if storage_block:
            storage = await Block.load(storage_block)

            if "put-directory" not in storage.get_block_capabilities():
                raise BlockMissingCapabilities(
                    f"Storage block {storage!r} missing 'put-directory' capability."
                )

            self.storage = storage

            # upload current directory to storage location
            file_count = await self.storage.put_directory(
                ignore_file=ignore_file, to_path=self.path
            )
        elif self.storage:
            if "put-directory" not in self.storage.get_block_capabilities():
                raise BlockMissingCapabilities(
                    f"Storage block {self.storage!r} missing 'put-directory' capability."
                )

            file_count = await self.storage.put_directory(
                ignore_file=ignore_file, to_path=self.path
            )

        # persists storage now in case it contains secret values
        if self.storage and not self.storage._block_document_id:
            await self.storage._save(is_anonymous=True)

        return file_count

    @sync_compatible
    async def apply(
        self, upload: bool = False, work_queue_concurrency: int = None
    ) -> UUID:
        """
        Registers this deployment with the API and returns the deployment's ID.

        Args:
            upload: if True, deployment files are automatically uploaded to remote storage
            work_queue_concurrency: If provided, sets the concurrency limit on the deployment's work queue
        """
        if not self.name or not self.flow_name:
            raise ValueError("Both a deployment name and flow name must be set.")
        async with get_client() as client:
            # prep IDs
            flow_id = await client.create_flow_from_name(self.flow_name)

            infrastructure_document_id = self.infrastructure._block_document_id
            if not infrastructure_document_id:
                # if not building off a block, will create an anonymous block
                self.infrastructure = self.infrastructure.copy()
                infrastructure_document_id = await self.infrastructure._save(
                    is_anonymous=True,
                )

            if upload:
                await self.upload_to_storage()

            if self.work_queue_name and work_queue_concurrency is not None:
                try:
                    res = await client.create_work_queue(name=self.work_queue_name)
                except ObjectAlreadyExists:
                    res = await client.read_work_queue_by_name(
                        name=self.work_queue_name
                    )
                await client.update_work_queue(
                    res.id, concurrency_limit=work_queue_concurrency
                )

            # we assume storage was already saved
            storage_document_id = getattr(self.storage, "_block_document_id", None)

            deployment_id = await client.create_deployment(
                flow_id=flow_id,
                name=self.name,
                work_queue_name=self.work_queue_name,
                version=self.version,
                schedule=self.schedule,
                parameters=self.parameters,
                description=self.description,
                tags=self.tags,
                manifest_path=self.manifest_path,  # allows for backwards YAML compat
                path=self.path,
                entrypoint=self.entrypoint,
                infra_overrides=self.infra_overrides,
                storage_document_id=storage_document_id,
                infrastructure_document_id=infrastructure_document_id,
                parameter_openapi_schema=self.parameter_openapi_schema.dict(),
            )

            return deployment_id

    @classmethod
    @sync_compatible
    async def build_from_flow(
        cls,
        flow: Flow,
        name: str,
        output: str = None,
        skip_upload: bool = False,
        ignore_file: str = ".prefectignore",
        apply: bool = False,
        load_existing: bool = True,
        **kwargs,
    ) -> "Deployment":
        """
        Configure a deployment for a given flow.

        Args:
            flow: A flow function to deploy
            name: A name for the deployment
            output (optional): if provided, the full deployment specification will be written as a YAML
                file in the location specified by `output`
            skip_upload: if True, deployment files are not automatically uploaded to remote storage
            ignore_file: an optional path to a `.prefectignore` file that specifies filename patterns
                to ignore when uploading to remote storage; if not provided, looks for `.prefectignore`
                in the current working directory
            apply: if True, the deployment is automatically registered with the API
            load_existing: if True, load any settings that may already be configured for the named deployment
                server-side (e.g., schedules, default parameter values, etc.)
            **kwargs: other keyword arguments to pass to the constructor for the `Deployment` class
        """
        if not name:
            raise ValueError("A deployment name must be provided.")

        # note that `deployment.load` only updates settings that were *not*
        # provided at initialization
        deployment = cls(name=name, **kwargs)
        deployment.flow_name = flow.name
        if not deployment.entrypoint:
            ## first see if an entrypoint can be determined
            flow_file = getattr(flow, "__globals__", {}).get("__file__")
            mod_name = getattr(flow, "__module__", None)
            if not flow_file:
                if not mod_name:
                    # todo, check if the file location was manually set already
                    raise ValueError("Could not determine flow's file location.")
                module = importlib.import_module(mod_name)
                flow_file = getattr(module, "__file__", None)
                if not flow_file:
                    raise ValueError("Could not determine flow's file location.")

            # set entrypoint
            entry_path = Path(flow_file).absolute().relative_to(Path(".").absolute())
            deployment.entrypoint = f"{entry_path}:{flow.fn.__name__}"

        if load_existing:
            await deployment.load()

        # set a few attributes for this flow object
        deployment.parameter_openapi_schema = parameter_schema(flow)

        if not deployment.version:
            deployment.version = flow.version
        if not deployment.description:
            deployment.description = flow.description

        # proxy for whether infra is docker-based
        is_docker_based = hasattr(deployment.infrastructure, "image")

        if not deployment.storage and not is_docker_based and not deployment.path:
            deployment.path = str(Path(".").absolute())
        elif not deployment.storage and is_docker_based:
            # only update if a path is not already set
            if not deployment.path:
                deployment.path = "/opt/prefect/flows"

        if not skip_upload:
            if (
                deployment.storage
                and "put-directory" in deployment.storage.get_block_capabilities()
            ):
                await deployment.upload_to_storage(ignore_file=ignore_file)

        if output:
            await deployment.to_yaml(output)

        if apply:
            await deployment.apply()

        return deployment