Source code for azure.mgmt.cosmosdb.operations._gremlin_resources_operations

# pylint: disable=too-many-lines
# coding=utf-8
# --------------------------------------------------------------------------
# Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License. See License.txt in the project root for license information.
# Code generated by Microsoft (R) AutoRest Code Generator.
# Changes may cause incorrect behavior and will be lost if the code is regenerated.
# --------------------------------------------------------------------------
import sys
from typing import Any, Callable, Dict, IO, Iterable, Optional, TypeVar, Union, cast, overload
import urllib.parse

from azure.core.exceptions import (
    ClientAuthenticationError,
    HttpResponseError,
    ResourceExistsError,
    ResourceNotFoundError,
    ResourceNotModifiedError,
    map_error,
)
from azure.core.paging import ItemPaged
from azure.core.pipeline import PipelineResponse
from azure.core.pipeline.transport import HttpResponse
from azure.core.polling import LROPoller, NoPolling, PollingMethod
from azure.core.rest import HttpRequest
from azure.core.tracing.decorator import distributed_trace
from azure.core.utils import case_insensitive_dict
from azure.mgmt.core.exceptions import ARMErrorFormat
from azure.mgmt.core.polling.arm_polling import ARMPolling

from .. import models as _models
from .._serialization import Serializer
from .._vendor import _convert_request, _format_url_section

if sys.version_info >= (3, 8):
    from typing import Literal  # pylint: disable=no-name-in-module, ungrouped-imports
else:
    from typing_extensions import Literal  # type: ignore  # pylint: disable=ungrouped-imports
T = TypeVar("T")
ClsType = Optional[Callable[[PipelineResponse[HttpRequest, HttpResponse], T, Dict[str, Any]], Any]]

_SERIALIZER = Serializer()
_SERIALIZER.client_side_validation = False


def build_list_gremlin_databases_request(
    resource_group_name: str, account_name: str, subscription_id: str, **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="GET", url=_url, params=_params, headers=_headers, **kwargs)


def build_get_gremlin_database_request(
    resource_group_name: str, account_name: str, database_name: str, subscription_id: str, **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="GET", url=_url, params=_params, headers=_headers, **kwargs)


def build_create_update_gremlin_database_request(
    resource_group_name: str, account_name: str, database_name: str, subscription_id: str, **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None))
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    if content_type is not None:
        _headers["Content-Type"] = _SERIALIZER.header("content_type", content_type, "str")
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="PUT", url=_url, params=_params, headers=_headers, **kwargs)


def build_delete_gremlin_database_request(
    resource_group_name: str, account_name: str, database_name: str, subscription_id: str, **kwargs: Any
) -> HttpRequest:
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    return HttpRequest(method="DELETE", url=_url, params=_params, **kwargs)


def build_get_gremlin_database_throughput_request(
    resource_group_name: str, account_name: str, database_name: str, subscription_id: str, **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/throughputSettings/default",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="GET", url=_url, params=_params, headers=_headers, **kwargs)


def build_update_gremlin_database_throughput_request(
    resource_group_name: str, account_name: str, database_name: str, subscription_id: str, **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None))
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/throughputSettings/default",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    if content_type is not None:
        _headers["Content-Type"] = _SERIALIZER.header("content_type", content_type, "str")
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="PUT", url=_url, params=_params, headers=_headers, **kwargs)


def build_migrate_gremlin_database_to_autoscale_request(
    resource_group_name: str, account_name: str, database_name: str, subscription_id: str, **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/throughputSettings/default/migrateToAutoscale",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="POST", url=_url, params=_params, headers=_headers, **kwargs)


def build_migrate_gremlin_database_to_manual_throughput_request(
    resource_group_name: str, account_name: str, database_name: str, subscription_id: str, **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/throughputSettings/default/migrateToManualThroughput",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="POST", url=_url, params=_params, headers=_headers, **kwargs)


def build_list_gremlin_graphs_request(
    resource_group_name: str, account_name: str, database_name: str, subscription_id: str, **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="GET", url=_url, params=_params, headers=_headers, **kwargs)


def build_get_gremlin_graph_request(
    resource_group_name: str,
    account_name: str,
    database_name: str,
    graph_name: str,
    subscription_id: str,
    **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
        "graphName": _SERIALIZER.url("graph_name", graph_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="GET", url=_url, params=_params, headers=_headers, **kwargs)


def build_create_update_gremlin_graph_request(
    resource_group_name: str,
    account_name: str,
    database_name: str,
    graph_name: str,
    subscription_id: str,
    **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None))
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
        "graphName": _SERIALIZER.url("graph_name", graph_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    if content_type is not None:
        _headers["Content-Type"] = _SERIALIZER.header("content_type", content_type, "str")
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="PUT", url=_url, params=_params, headers=_headers, **kwargs)


def build_delete_gremlin_graph_request(
    resource_group_name: str,
    account_name: str,
    database_name: str,
    graph_name: str,
    subscription_id: str,
    **kwargs: Any
) -> HttpRequest:
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
        "graphName": _SERIALIZER.url("graph_name", graph_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    return HttpRequest(method="DELETE", url=_url, params=_params, **kwargs)


def build_get_gremlin_graph_throughput_request(
    resource_group_name: str,
    account_name: str,
    database_name: str,
    graph_name: str,
    subscription_id: str,
    **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/throughputSettings/default",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
        "graphName": _SERIALIZER.url("graph_name", graph_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="GET", url=_url, params=_params, headers=_headers, **kwargs)


def build_update_gremlin_graph_throughput_request(
    resource_group_name: str,
    account_name: str,
    database_name: str,
    graph_name: str,
    subscription_id: str,
    **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None))
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/throughputSettings/default",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
        "graphName": _SERIALIZER.url("graph_name", graph_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    if content_type is not None:
        _headers["Content-Type"] = _SERIALIZER.header("content_type", content_type, "str")
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="PUT", url=_url, params=_params, headers=_headers, **kwargs)


def build_migrate_gremlin_graph_to_autoscale_request(
    resource_group_name: str,
    account_name: str,
    database_name: str,
    graph_name: str,
    subscription_id: str,
    **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/throughputSettings/default/migrateToAutoscale",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
        "graphName": _SERIALIZER.url("graph_name", graph_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="POST", url=_url, params=_params, headers=_headers, **kwargs)


def build_migrate_gremlin_graph_to_manual_throughput_request(
    resource_group_name: str,
    account_name: str,
    database_name: str,
    graph_name: str,
    subscription_id: str,
    **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/throughputSettings/default/migrateToManualThroughput",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
        "graphName": _SERIALIZER.url("graph_name", graph_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="POST", url=_url, params=_params, headers=_headers, **kwargs)


def build_retrieve_continuous_backup_information_request(
    resource_group_name: str,
    account_name: str,
    database_name: str,
    graph_name: str,
    subscription_id: str,
    **kwargs: Any
) -> HttpRequest:
    _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {})
    _params = case_insensitive_dict(kwargs.pop("params", {}) or {})

    api_version: Literal["2022-11-15-preview"] = kwargs.pop(
        "api_version", _params.pop("api-version", "2022-11-15-preview")
    )
    content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None))
    accept = _headers.pop("Accept", "application/json")

    # Construct URL
    _url = kwargs.pop(
        "template_url",
        "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/retrieveContinuousBackupInformation",
    )  # pylint: disable=line-too-long
    path_format_arguments = {
        "subscriptionId": _SERIALIZER.url("subscription_id", subscription_id, "str", min_length=1),
        "resourceGroupName": _SERIALIZER.url(
            "resource_group_name", resource_group_name, "str", max_length=90, min_length=1
        ),
        "accountName": _SERIALIZER.url(
            "account_name", account_name, "str", max_length=50, min_length=3, pattern=r"^[a-z0-9]+(-[a-z0-9]+)*"
        ),
        "databaseName": _SERIALIZER.url("database_name", database_name, "str"),
        "graphName": _SERIALIZER.url("graph_name", graph_name, "str"),
    }

    _url: str = _format_url_section(_url, **path_format_arguments)  # type: ignore

    # Construct parameters
    _params["api-version"] = _SERIALIZER.query("api_version", api_version, "str")

    # Construct headers
    if content_type is not None:
        _headers["Content-Type"] = _SERIALIZER.header("content_type", content_type, "str")
    _headers["Accept"] = _SERIALIZER.header("accept", accept, "str")

    return HttpRequest(method="POST", url=_url, params=_params, headers=_headers, **kwargs)


[docs]class GremlinResourcesOperations: # pylint: disable=too-many-public-methods """ .. warning:: **DO NOT** instantiate this class directly. Instead, you should access the following operations through :class:`~azure.mgmt.cosmosdb.CosmosDBManagementClient`'s :attr:`gremlin_resources` attribute. """ models = _models def __init__(self, *args, **kwargs): input_args = list(args) self._client = input_args.pop(0) if input_args else kwargs.pop("client") self._config = input_args.pop(0) if input_args else kwargs.pop("config") self._serialize = input_args.pop(0) if input_args else kwargs.pop("serializer") self._deserialize = input_args.pop(0) if input_args else kwargs.pop("deserializer")
[docs] @distributed_trace def list_gremlin_databases( self, resource_group_name: str, account_name: str, **kwargs: Any ) -> Iterable["_models.GremlinDatabaseGetResults"]: """Lists the Gremlin databases under an existing Azure Cosmos DB database account. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :keyword callable cls: A custom type or function that will be passed the direct response :return: An iterator like instance of either GremlinDatabaseGetResults or the result of cls(response) :rtype: ~azure.core.paging.ItemPaged[~azure.mgmt.cosmosdb.models.GremlinDatabaseGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[_models.GremlinDatabaseListResult] = kwargs.pop("cls", None) error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) def prepare_request(next_link=None): if not next_link: request = build_list_gremlin_databases_request( resource_group_name=resource_group_name, account_name=account_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self.list_gremlin_databases.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) else: # make call to next link with the client's api-version _parsed_next_link = urllib.parse.urlparse(next_link) _next_request_params = case_insensitive_dict( { key: [urllib.parse.quote(v) for v in value] for key, value in urllib.parse.parse_qs(_parsed_next_link.query).items() } ) _next_request_params["api-version"] = self._config.api_version request = HttpRequest( "GET", urllib.parse.urljoin(next_link, _parsed_next_link.path), params=_next_request_params ) request = _convert_request(request) request.url = self._client.format_url(request.url) request.method = "GET" return request def extract_data(pipeline_response): deserialized = self._deserialize("GremlinDatabaseListResult", pipeline_response) list_of_elem = deserialized.value if cls: list_of_elem = cls(list_of_elem) # type: ignore return None, iter(list_of_elem) def get_next(next_link=None): request = prepare_request(next_link) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) return pipeline_response return ItemPaged(get_next, extract_data)
list_gremlin_databases.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases" }
[docs] @distributed_trace def get_gremlin_database( self, resource_group_name: str, account_name: str, database_name: str, **kwargs: Any ) -> _models.GremlinDatabaseGetResults: """Gets the Gremlin databases under an existing Azure Cosmos DB database account with the provided name. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :keyword callable cls: A custom type or function that will be passed the direct response :return: GremlinDatabaseGetResults or the result of cls(response) :rtype: ~azure.mgmt.cosmosdb.models.GremlinDatabaseGetResults :raises ~azure.core.exceptions.HttpResponseError: """ error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[_models.GremlinDatabaseGetResults] = kwargs.pop("cls", None) request = build_get_gremlin_database_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self.get_gremlin_database.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = self._deserialize("GremlinDatabaseGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized
get_gremlin_database.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}" } def _create_update_gremlin_database_initial( self, resource_group_name: str, account_name: str, database_name: str, create_update_gremlin_database_parameters: Union[_models.GremlinDatabaseCreateUpdateParameters, IO], **kwargs: Any ) -> Optional[_models.GremlinDatabaseGetResults]: error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {}) _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None)) cls: ClsType[Optional[_models.GremlinDatabaseGetResults]] = kwargs.pop("cls", None) content_type = content_type or "application/json" _json = None _content = None if isinstance(create_update_gremlin_database_parameters, (IO, bytes)): _content = create_update_gremlin_database_parameters else: _json = self._serialize.body( create_update_gremlin_database_parameters, "GremlinDatabaseCreateUpdateParameters" ) request = build_create_update_gremlin_database_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, subscription_id=self._config.subscription_id, api_version=api_version, content_type=content_type, json=_json, content=_content, template_url=self._create_update_gremlin_database_initial.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200, 202]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = None response_headers = {} if response.status_code == 200: deserialized = self._deserialize("GremlinDatabaseGetResults", pipeline_response) if response.status_code == 202: response_headers["azure-AsyncOperation"] = self._deserialize( "str", response.headers.get("azure-AsyncOperation") ) response_headers["location"] = self._deserialize("str", response.headers.get("location")) if cls: return cls(pipeline_response, deserialized, response_headers) return deserialized _create_update_gremlin_database_initial.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}" } @overload def begin_create_update_gremlin_database( self, resource_group_name: str, account_name: str, database_name: str, create_update_gremlin_database_parameters: _models.GremlinDatabaseCreateUpdateParameters, *, content_type: str = "application/json", **kwargs: Any ) -> LROPoller[_models.GremlinDatabaseGetResults]: """Create or update an Azure Cosmos DB Gremlin database. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param create_update_gremlin_database_parameters: The parameters to provide for the current Gremlin database. Required. :type create_update_gremlin_database_parameters: ~azure.mgmt.cosmosdb.models.GremlinDatabaseCreateUpdateParameters :keyword content_type: Body Parameter content-type. Content type parameter for JSON body. Default value is "application/json". :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either GremlinDatabaseGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.GremlinDatabaseGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ @overload def begin_create_update_gremlin_database( self, resource_group_name: str, account_name: str, database_name: str, create_update_gremlin_database_parameters: IO, *, content_type: str = "application/json", **kwargs: Any ) -> LROPoller[_models.GremlinDatabaseGetResults]: """Create or update an Azure Cosmos DB Gremlin database. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param create_update_gremlin_database_parameters: The parameters to provide for the current Gremlin database. Required. :type create_update_gremlin_database_parameters: IO :keyword content_type: Body Parameter content-type. Content type parameter for binary body. Default value is "application/json". :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either GremlinDatabaseGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.GremlinDatabaseGetResults] :raises ~azure.core.exceptions.HttpResponseError: """
[docs] @distributed_trace def begin_create_update_gremlin_database( self, resource_group_name: str, account_name: str, database_name: str, create_update_gremlin_database_parameters: Union[_models.GremlinDatabaseCreateUpdateParameters, IO], **kwargs: Any ) -> LROPoller[_models.GremlinDatabaseGetResults]: """Create or update an Azure Cosmos DB Gremlin database. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param create_update_gremlin_database_parameters: The parameters to provide for the current Gremlin database. Is either a GremlinDatabaseCreateUpdateParameters type or a IO type. Required. :type create_update_gremlin_database_parameters: ~azure.mgmt.cosmosdb.models.GremlinDatabaseCreateUpdateParameters or IO :keyword content_type: Body Parameter content-type. Known values are: 'application/json'. Default value is None. :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either GremlinDatabaseGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.GremlinDatabaseGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {}) _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None)) cls: ClsType[_models.GremlinDatabaseGetResults] = kwargs.pop("cls", None) polling: Union[bool, PollingMethod] = kwargs.pop("polling", True) lro_delay = kwargs.pop("polling_interval", self._config.polling_interval) cont_token: Optional[str] = kwargs.pop("continuation_token", None) if cont_token is None: raw_result = self._create_update_gremlin_database_initial( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, create_update_gremlin_database_parameters=create_update_gremlin_database_parameters, api_version=api_version, content_type=content_type, cls=lambda x, y, z: x, headers=_headers, params=_params, **kwargs ) kwargs.pop("error_map", None) def get_long_running_output(pipeline_response): deserialized = self._deserialize("GremlinDatabaseGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized if polling is True: polling_method: PollingMethod = cast(PollingMethod, ARMPolling(lro_delay, **kwargs)) elif polling is False: polling_method = cast(PollingMethod, NoPolling()) else: polling_method = polling if cont_token: return LROPoller.from_continuation_token( polling_method=polling_method, continuation_token=cont_token, client=self._client, deserialization_callback=get_long_running_output, ) return LROPoller(self._client, raw_result, get_long_running_output, polling_method) # type: ignore
begin_create_update_gremlin_database.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}" } def _delete_gremlin_database_initial( # pylint: disable=inconsistent-return-statements self, resource_group_name: str, account_name: str, database_name: str, **kwargs: Any ) -> None: error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[None] = kwargs.pop("cls", None) request = build_delete_gremlin_database_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self._delete_gremlin_database_initial.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [202, 204]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) response_headers = {} if response.status_code == 202: response_headers["azure-AsyncOperation"] = self._deserialize( "str", response.headers.get("azure-AsyncOperation") ) response_headers["location"] = self._deserialize("str", response.headers.get("location")) if cls: return cls(pipeline_response, None, response_headers) _delete_gremlin_database_initial.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}" }
[docs] @distributed_trace def begin_delete_gremlin_database( self, resource_group_name: str, account_name: str, database_name: str, **kwargs: Any ) -> LROPoller[None]: """Deletes an existing Azure Cosmos DB Gremlin database. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either None or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[None] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[None] = kwargs.pop("cls", None) polling: Union[bool, PollingMethod] = kwargs.pop("polling", True) lro_delay = kwargs.pop("polling_interval", self._config.polling_interval) cont_token: Optional[str] = kwargs.pop("continuation_token", None) if cont_token is None: raw_result = self._delete_gremlin_database_initial( # type: ignore resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, api_version=api_version, cls=lambda x, y, z: x, headers=_headers, params=_params, **kwargs ) kwargs.pop("error_map", None) def get_long_running_output(pipeline_response): # pylint: disable=inconsistent-return-statements if cls: return cls(pipeline_response, None, {}) if polling is True: polling_method: PollingMethod = cast(PollingMethod, ARMPolling(lro_delay, **kwargs)) elif polling is False: polling_method = cast(PollingMethod, NoPolling()) else: polling_method = polling if cont_token: return LROPoller.from_continuation_token( polling_method=polling_method, continuation_token=cont_token, client=self._client, deserialization_callback=get_long_running_output, ) return LROPoller(self._client, raw_result, get_long_running_output, polling_method) # type: ignore
begin_delete_gremlin_database.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}" }
[docs] @distributed_trace def get_gremlin_database_throughput( self, resource_group_name: str, account_name: str, database_name: str, **kwargs: Any ) -> _models.ThroughputSettingsGetResults: """Gets the RUs per second of the Gremlin database under an existing Azure Cosmos DB database account with the provided name. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :keyword callable cls: A custom type or function that will be passed the direct response :return: ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults :raises ~azure.core.exceptions.HttpResponseError: """ error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[_models.ThroughputSettingsGetResults] = kwargs.pop("cls", None) request = build_get_gremlin_database_throughput_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self.get_gremlin_database_throughput.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized
get_gremlin_database_throughput.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/throughputSettings/default" } def _update_gremlin_database_throughput_initial( self, resource_group_name: str, account_name: str, database_name: str, update_throughput_parameters: Union[_models.ThroughputSettingsUpdateParameters, IO], **kwargs: Any ) -> Optional[_models.ThroughputSettingsGetResults]: error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {}) _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None)) cls: ClsType[Optional[_models.ThroughputSettingsGetResults]] = kwargs.pop("cls", None) content_type = content_type or "application/json" _json = None _content = None if isinstance(update_throughput_parameters, (IO, bytes)): _content = update_throughput_parameters else: _json = self._serialize.body(update_throughput_parameters, "ThroughputSettingsUpdateParameters") request = build_update_gremlin_database_throughput_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, subscription_id=self._config.subscription_id, api_version=api_version, content_type=content_type, json=_json, content=_content, template_url=self._update_gremlin_database_throughput_initial.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200, 202]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = None response_headers = {} if response.status_code == 200: deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if response.status_code == 202: response_headers["azure-AsyncOperation"] = self._deserialize( "str", response.headers.get("azure-AsyncOperation") ) response_headers["location"] = self._deserialize("str", response.headers.get("location")) if cls: return cls(pipeline_response, deserialized, response_headers) return deserialized _update_gremlin_database_throughput_initial.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/throughputSettings/default" } @overload def begin_update_gremlin_database_throughput( self, resource_group_name: str, account_name: str, database_name: str, update_throughput_parameters: _models.ThroughputSettingsUpdateParameters, *, content_type: str = "application/json", **kwargs: Any ) -> LROPoller[_models.ThroughputSettingsGetResults]: """Update RUs per second of an Azure Cosmos DB Gremlin database. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param update_throughput_parameters: The RUs per second of the parameters to provide for the current Gremlin database. Required. :type update_throughput_parameters: ~azure.mgmt.cosmosdb.models.ThroughputSettingsUpdateParameters :keyword content_type: Body Parameter content-type. Content type parameter for JSON body. Default value is "application/json". :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ @overload def begin_update_gremlin_database_throughput( self, resource_group_name: str, account_name: str, database_name: str, update_throughput_parameters: IO, *, content_type: str = "application/json", **kwargs: Any ) -> LROPoller[_models.ThroughputSettingsGetResults]: """Update RUs per second of an Azure Cosmos DB Gremlin database. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param update_throughput_parameters: The RUs per second of the parameters to provide for the current Gremlin database. Required. :type update_throughput_parameters: IO :keyword content_type: Body Parameter content-type. Content type parameter for binary body. Default value is "application/json". :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults] :raises ~azure.core.exceptions.HttpResponseError: """
[docs] @distributed_trace def begin_update_gremlin_database_throughput( self, resource_group_name: str, account_name: str, database_name: str, update_throughput_parameters: Union[_models.ThroughputSettingsUpdateParameters, IO], **kwargs: Any ) -> LROPoller[_models.ThroughputSettingsGetResults]: """Update RUs per second of an Azure Cosmos DB Gremlin database. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param update_throughput_parameters: The RUs per second of the parameters to provide for the current Gremlin database. Is either a ThroughputSettingsUpdateParameters type or a IO type. Required. :type update_throughput_parameters: ~azure.mgmt.cosmosdb.models.ThroughputSettingsUpdateParameters or IO :keyword content_type: Body Parameter content-type. Known values are: 'application/json'. Default value is None. :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {}) _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None)) cls: ClsType[_models.ThroughputSettingsGetResults] = kwargs.pop("cls", None) polling: Union[bool, PollingMethod] = kwargs.pop("polling", True) lro_delay = kwargs.pop("polling_interval", self._config.polling_interval) cont_token: Optional[str] = kwargs.pop("continuation_token", None) if cont_token is None: raw_result = self._update_gremlin_database_throughput_initial( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, update_throughput_parameters=update_throughput_parameters, api_version=api_version, content_type=content_type, cls=lambda x, y, z: x, headers=_headers, params=_params, **kwargs ) kwargs.pop("error_map", None) def get_long_running_output(pipeline_response): deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized if polling is True: polling_method: PollingMethod = cast(PollingMethod, ARMPolling(lro_delay, **kwargs)) elif polling is False: polling_method = cast(PollingMethod, NoPolling()) else: polling_method = polling if cont_token: return LROPoller.from_continuation_token( polling_method=polling_method, continuation_token=cont_token, client=self._client, deserialization_callback=get_long_running_output, ) return LROPoller(self._client, raw_result, get_long_running_output, polling_method) # type: ignore
begin_update_gremlin_database_throughput.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/throughputSettings/default" } def _migrate_gremlin_database_to_autoscale_initial( self, resource_group_name: str, account_name: str, database_name: str, **kwargs: Any ) -> Optional[_models.ThroughputSettingsGetResults]: error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[Optional[_models.ThroughputSettingsGetResults]] = kwargs.pop("cls", None) request = build_migrate_gremlin_database_to_autoscale_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self._migrate_gremlin_database_to_autoscale_initial.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200, 202]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = None response_headers = {} if response.status_code == 200: deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if response.status_code == 202: response_headers["azure-AsyncOperation"] = self._deserialize( "str", response.headers.get("azure-AsyncOperation") ) response_headers["location"] = self._deserialize("str", response.headers.get("location")) if cls: return cls(pipeline_response, deserialized, response_headers) return deserialized _migrate_gremlin_database_to_autoscale_initial.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/throughputSettings/default/migrateToAutoscale" }
[docs] @distributed_trace def begin_migrate_gremlin_database_to_autoscale( self, resource_group_name: str, account_name: str, database_name: str, **kwargs: Any ) -> LROPoller[_models.ThroughputSettingsGetResults]: """Migrate an Azure Cosmos DB Gremlin database from manual throughput to autoscale. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[_models.ThroughputSettingsGetResults] = kwargs.pop("cls", None) polling: Union[bool, PollingMethod] = kwargs.pop("polling", True) lro_delay = kwargs.pop("polling_interval", self._config.polling_interval) cont_token: Optional[str] = kwargs.pop("continuation_token", None) if cont_token is None: raw_result = self._migrate_gremlin_database_to_autoscale_initial( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, api_version=api_version, cls=lambda x, y, z: x, headers=_headers, params=_params, **kwargs ) kwargs.pop("error_map", None) def get_long_running_output(pipeline_response): deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized if polling is True: polling_method: PollingMethod = cast(PollingMethod, ARMPolling(lro_delay, **kwargs)) elif polling is False: polling_method = cast(PollingMethod, NoPolling()) else: polling_method = polling if cont_token: return LROPoller.from_continuation_token( polling_method=polling_method, continuation_token=cont_token, client=self._client, deserialization_callback=get_long_running_output, ) return LROPoller(self._client, raw_result, get_long_running_output, polling_method) # type: ignore
begin_migrate_gremlin_database_to_autoscale.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/throughputSettings/default/migrateToAutoscale" } def _migrate_gremlin_database_to_manual_throughput_initial( self, resource_group_name: str, account_name: str, database_name: str, **kwargs: Any ) -> Optional[_models.ThroughputSettingsGetResults]: error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[Optional[_models.ThroughputSettingsGetResults]] = kwargs.pop("cls", None) request = build_migrate_gremlin_database_to_manual_throughput_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self._migrate_gremlin_database_to_manual_throughput_initial.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200, 202]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = None response_headers = {} if response.status_code == 200: deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if response.status_code == 202: response_headers["azure-AsyncOperation"] = self._deserialize( "str", response.headers.get("azure-AsyncOperation") ) response_headers["location"] = self._deserialize("str", response.headers.get("location")) if cls: return cls(pipeline_response, deserialized, response_headers) return deserialized _migrate_gremlin_database_to_manual_throughput_initial.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/throughputSettings/default/migrateToManualThroughput" }
[docs] @distributed_trace def begin_migrate_gremlin_database_to_manual_throughput( self, resource_group_name: str, account_name: str, database_name: str, **kwargs: Any ) -> LROPoller[_models.ThroughputSettingsGetResults]: """Migrate an Azure Cosmos DB Gremlin database from autoscale to manual throughput. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[_models.ThroughputSettingsGetResults] = kwargs.pop("cls", None) polling: Union[bool, PollingMethod] = kwargs.pop("polling", True) lro_delay = kwargs.pop("polling_interval", self._config.polling_interval) cont_token: Optional[str] = kwargs.pop("continuation_token", None) if cont_token is None: raw_result = self._migrate_gremlin_database_to_manual_throughput_initial( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, api_version=api_version, cls=lambda x, y, z: x, headers=_headers, params=_params, **kwargs ) kwargs.pop("error_map", None) def get_long_running_output(pipeline_response): deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized if polling is True: polling_method: PollingMethod = cast(PollingMethod, ARMPolling(lro_delay, **kwargs)) elif polling is False: polling_method = cast(PollingMethod, NoPolling()) else: polling_method = polling if cont_token: return LROPoller.from_continuation_token( polling_method=polling_method, continuation_token=cont_token, client=self._client, deserialization_callback=get_long_running_output, ) return LROPoller(self._client, raw_result, get_long_running_output, polling_method) # type: ignore
begin_migrate_gremlin_database_to_manual_throughput.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/throughputSettings/default/migrateToManualThroughput" }
[docs] @distributed_trace def list_gremlin_graphs( self, resource_group_name: str, account_name: str, database_name: str, **kwargs: Any ) -> Iterable["_models.GremlinGraphGetResults"]: """Lists the Gremlin graph under an existing Azure Cosmos DB database account. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :keyword callable cls: A custom type or function that will be passed the direct response :return: An iterator like instance of either GremlinGraphGetResults or the result of cls(response) :rtype: ~azure.core.paging.ItemPaged[~azure.mgmt.cosmosdb.models.GremlinGraphGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[_models.GremlinGraphListResult] = kwargs.pop("cls", None) error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) def prepare_request(next_link=None): if not next_link: request = build_list_gremlin_graphs_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self.list_gremlin_graphs.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) else: # make call to next link with the client's api-version _parsed_next_link = urllib.parse.urlparse(next_link) _next_request_params = case_insensitive_dict( { key: [urllib.parse.quote(v) for v in value] for key, value in urllib.parse.parse_qs(_parsed_next_link.query).items() } ) _next_request_params["api-version"] = self._config.api_version request = HttpRequest( "GET", urllib.parse.urljoin(next_link, _parsed_next_link.path), params=_next_request_params ) request = _convert_request(request) request.url = self._client.format_url(request.url) request.method = "GET" return request def extract_data(pipeline_response): deserialized = self._deserialize("GremlinGraphListResult", pipeline_response) list_of_elem = deserialized.value if cls: list_of_elem = cls(list_of_elem) # type: ignore return None, iter(list_of_elem) def get_next(next_link=None): request = prepare_request(next_link) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) return pipeline_response return ItemPaged(get_next, extract_data)
list_gremlin_graphs.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs" }
[docs] @distributed_trace def get_gremlin_graph( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, **kwargs: Any ) -> _models.GremlinGraphGetResults: """Gets the Gremlin graph under an existing Azure Cosmos DB database account. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :keyword callable cls: A custom type or function that will be passed the direct response :return: GremlinGraphGetResults or the result of cls(response) :rtype: ~azure.mgmt.cosmosdb.models.GremlinGraphGetResults :raises ~azure.core.exceptions.HttpResponseError: """ error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[_models.GremlinGraphGetResults] = kwargs.pop("cls", None) request = build_get_gremlin_graph_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self.get_gremlin_graph.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = self._deserialize("GremlinGraphGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized
get_gremlin_graph.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}" } def _create_update_gremlin_graph_initial( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, create_update_gremlin_graph_parameters: Union[_models.GremlinGraphCreateUpdateParameters, IO], **kwargs: Any ) -> Optional[_models.GremlinGraphGetResults]: error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {}) _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None)) cls: ClsType[Optional[_models.GremlinGraphGetResults]] = kwargs.pop("cls", None) content_type = content_type or "application/json" _json = None _content = None if isinstance(create_update_gremlin_graph_parameters, (IO, bytes)): _content = create_update_gremlin_graph_parameters else: _json = self._serialize.body(create_update_gremlin_graph_parameters, "GremlinGraphCreateUpdateParameters") request = build_create_update_gremlin_graph_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, subscription_id=self._config.subscription_id, api_version=api_version, content_type=content_type, json=_json, content=_content, template_url=self._create_update_gremlin_graph_initial.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200, 202]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = None response_headers = {} if response.status_code == 200: deserialized = self._deserialize("GremlinGraphGetResults", pipeline_response) if response.status_code == 202: response_headers["azure-AsyncOperation"] = self._deserialize( "str", response.headers.get("azure-AsyncOperation") ) response_headers["location"] = self._deserialize("str", response.headers.get("location")) if cls: return cls(pipeline_response, deserialized, response_headers) return deserialized _create_update_gremlin_graph_initial.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}" } @overload def begin_create_update_gremlin_graph( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, create_update_gremlin_graph_parameters: _models.GremlinGraphCreateUpdateParameters, *, content_type: str = "application/json", **kwargs: Any ) -> LROPoller[_models.GremlinGraphGetResults]: """Create or update an Azure Cosmos DB Gremlin graph. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :param create_update_gremlin_graph_parameters: The parameters to provide for the current Gremlin graph. Required. :type create_update_gremlin_graph_parameters: ~azure.mgmt.cosmosdb.models.GremlinGraphCreateUpdateParameters :keyword content_type: Body Parameter content-type. Content type parameter for JSON body. Default value is "application/json". :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either GremlinGraphGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.GremlinGraphGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ @overload def begin_create_update_gremlin_graph( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, create_update_gremlin_graph_parameters: IO, *, content_type: str = "application/json", **kwargs: Any ) -> LROPoller[_models.GremlinGraphGetResults]: """Create or update an Azure Cosmos DB Gremlin graph. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :param create_update_gremlin_graph_parameters: The parameters to provide for the current Gremlin graph. Required. :type create_update_gremlin_graph_parameters: IO :keyword content_type: Body Parameter content-type. Content type parameter for binary body. Default value is "application/json". :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either GremlinGraphGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.GremlinGraphGetResults] :raises ~azure.core.exceptions.HttpResponseError: """
[docs] @distributed_trace def begin_create_update_gremlin_graph( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, create_update_gremlin_graph_parameters: Union[_models.GremlinGraphCreateUpdateParameters, IO], **kwargs: Any ) -> LROPoller[_models.GremlinGraphGetResults]: """Create or update an Azure Cosmos DB Gremlin graph. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :param create_update_gremlin_graph_parameters: The parameters to provide for the current Gremlin graph. Is either a GremlinGraphCreateUpdateParameters type or a IO type. Required. :type create_update_gremlin_graph_parameters: ~azure.mgmt.cosmosdb.models.GremlinGraphCreateUpdateParameters or IO :keyword content_type: Body Parameter content-type. Known values are: 'application/json'. Default value is None. :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either GremlinGraphGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.GremlinGraphGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {}) _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None)) cls: ClsType[_models.GremlinGraphGetResults] = kwargs.pop("cls", None) polling: Union[bool, PollingMethod] = kwargs.pop("polling", True) lro_delay = kwargs.pop("polling_interval", self._config.polling_interval) cont_token: Optional[str] = kwargs.pop("continuation_token", None) if cont_token is None: raw_result = self._create_update_gremlin_graph_initial( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, create_update_gremlin_graph_parameters=create_update_gremlin_graph_parameters, api_version=api_version, content_type=content_type, cls=lambda x, y, z: x, headers=_headers, params=_params, **kwargs ) kwargs.pop("error_map", None) def get_long_running_output(pipeline_response): deserialized = self._deserialize("GremlinGraphGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized if polling is True: polling_method: PollingMethod = cast(PollingMethod, ARMPolling(lro_delay, **kwargs)) elif polling is False: polling_method = cast(PollingMethod, NoPolling()) else: polling_method = polling if cont_token: return LROPoller.from_continuation_token( polling_method=polling_method, continuation_token=cont_token, client=self._client, deserialization_callback=get_long_running_output, ) return LROPoller(self._client, raw_result, get_long_running_output, polling_method) # type: ignore
begin_create_update_gremlin_graph.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}" } def _delete_gremlin_graph_initial( # pylint: disable=inconsistent-return-statements self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, **kwargs: Any ) -> None: error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[None] = kwargs.pop("cls", None) request = build_delete_gremlin_graph_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self._delete_gremlin_graph_initial.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [202, 204]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) response_headers = {} if response.status_code == 202: response_headers["azure-AsyncOperation"] = self._deserialize( "str", response.headers.get("azure-AsyncOperation") ) response_headers["location"] = self._deserialize("str", response.headers.get("location")) if cls: return cls(pipeline_response, None, response_headers) _delete_gremlin_graph_initial.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}" }
[docs] @distributed_trace def begin_delete_gremlin_graph( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, **kwargs: Any ) -> LROPoller[None]: """Deletes an existing Azure Cosmos DB Gremlin graph. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either None or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[None] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[None] = kwargs.pop("cls", None) polling: Union[bool, PollingMethod] = kwargs.pop("polling", True) lro_delay = kwargs.pop("polling_interval", self._config.polling_interval) cont_token: Optional[str] = kwargs.pop("continuation_token", None) if cont_token is None: raw_result = self._delete_gremlin_graph_initial( # type: ignore resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, api_version=api_version, cls=lambda x, y, z: x, headers=_headers, params=_params, **kwargs ) kwargs.pop("error_map", None) def get_long_running_output(pipeline_response): # pylint: disable=inconsistent-return-statements if cls: return cls(pipeline_response, None, {}) if polling is True: polling_method: PollingMethod = cast(PollingMethod, ARMPolling(lro_delay, **kwargs)) elif polling is False: polling_method = cast(PollingMethod, NoPolling()) else: polling_method = polling if cont_token: return LROPoller.from_continuation_token( polling_method=polling_method, continuation_token=cont_token, client=self._client, deserialization_callback=get_long_running_output, ) return LROPoller(self._client, raw_result, get_long_running_output, polling_method) # type: ignore
begin_delete_gremlin_graph.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}" }
[docs] @distributed_trace def get_gremlin_graph_throughput( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, **kwargs: Any ) -> _models.ThroughputSettingsGetResults: """Gets the Gremlin graph throughput under an existing Azure Cosmos DB database account with the provided name. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :keyword callable cls: A custom type or function that will be passed the direct response :return: ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults :raises ~azure.core.exceptions.HttpResponseError: """ error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[_models.ThroughputSettingsGetResults] = kwargs.pop("cls", None) request = build_get_gremlin_graph_throughput_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self.get_gremlin_graph_throughput.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized
get_gremlin_graph_throughput.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/throughputSettings/default" } def _update_gremlin_graph_throughput_initial( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, update_throughput_parameters: Union[_models.ThroughputSettingsUpdateParameters, IO], **kwargs: Any ) -> Optional[_models.ThroughputSettingsGetResults]: error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {}) _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None)) cls: ClsType[Optional[_models.ThroughputSettingsGetResults]] = kwargs.pop("cls", None) content_type = content_type or "application/json" _json = None _content = None if isinstance(update_throughput_parameters, (IO, bytes)): _content = update_throughput_parameters else: _json = self._serialize.body(update_throughput_parameters, "ThroughputSettingsUpdateParameters") request = build_update_gremlin_graph_throughput_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, subscription_id=self._config.subscription_id, api_version=api_version, content_type=content_type, json=_json, content=_content, template_url=self._update_gremlin_graph_throughput_initial.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200, 202]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = None response_headers = {} if response.status_code == 200: deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if response.status_code == 202: response_headers["azure-AsyncOperation"] = self._deserialize( "str", response.headers.get("azure-AsyncOperation") ) response_headers["location"] = self._deserialize("str", response.headers.get("location")) if cls: return cls(pipeline_response, deserialized, response_headers) return deserialized _update_gremlin_graph_throughput_initial.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/throughputSettings/default" } @overload def begin_update_gremlin_graph_throughput( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, update_throughput_parameters: _models.ThroughputSettingsUpdateParameters, *, content_type: str = "application/json", **kwargs: Any ) -> LROPoller[_models.ThroughputSettingsGetResults]: """Update RUs per second of an Azure Cosmos DB Gremlin graph. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :param update_throughput_parameters: The RUs per second of the parameters to provide for the current Gremlin graph. Required. :type update_throughput_parameters: ~azure.mgmt.cosmosdb.models.ThroughputSettingsUpdateParameters :keyword content_type: Body Parameter content-type. Content type parameter for JSON body. Default value is "application/json". :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ @overload def begin_update_gremlin_graph_throughput( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, update_throughput_parameters: IO, *, content_type: str = "application/json", **kwargs: Any ) -> LROPoller[_models.ThroughputSettingsGetResults]: """Update RUs per second of an Azure Cosmos DB Gremlin graph. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :param update_throughput_parameters: The RUs per second of the parameters to provide for the current Gremlin graph. Required. :type update_throughput_parameters: IO :keyword content_type: Body Parameter content-type. Content type parameter for binary body. Default value is "application/json". :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults] :raises ~azure.core.exceptions.HttpResponseError: """
[docs] @distributed_trace def begin_update_gremlin_graph_throughput( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, update_throughput_parameters: Union[_models.ThroughputSettingsUpdateParameters, IO], **kwargs: Any ) -> LROPoller[_models.ThroughputSettingsGetResults]: """Update RUs per second of an Azure Cosmos DB Gremlin graph. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :param update_throughput_parameters: The RUs per second of the parameters to provide for the current Gremlin graph. Is either a ThroughputSettingsUpdateParameters type or a IO type. Required. :type update_throughput_parameters: ~azure.mgmt.cosmosdb.models.ThroughputSettingsUpdateParameters or IO :keyword content_type: Body Parameter content-type. Known values are: 'application/json'. Default value is None. :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {}) _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None)) cls: ClsType[_models.ThroughputSettingsGetResults] = kwargs.pop("cls", None) polling: Union[bool, PollingMethod] = kwargs.pop("polling", True) lro_delay = kwargs.pop("polling_interval", self._config.polling_interval) cont_token: Optional[str] = kwargs.pop("continuation_token", None) if cont_token is None: raw_result = self._update_gremlin_graph_throughput_initial( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, update_throughput_parameters=update_throughput_parameters, api_version=api_version, content_type=content_type, cls=lambda x, y, z: x, headers=_headers, params=_params, **kwargs ) kwargs.pop("error_map", None) def get_long_running_output(pipeline_response): deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized if polling is True: polling_method: PollingMethod = cast(PollingMethod, ARMPolling(lro_delay, **kwargs)) elif polling is False: polling_method = cast(PollingMethod, NoPolling()) else: polling_method = polling if cont_token: return LROPoller.from_continuation_token( polling_method=polling_method, continuation_token=cont_token, client=self._client, deserialization_callback=get_long_running_output, ) return LROPoller(self._client, raw_result, get_long_running_output, polling_method) # type: ignore
begin_update_gremlin_graph_throughput.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/throughputSettings/default" } def _migrate_gremlin_graph_to_autoscale_initial( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, **kwargs: Any ) -> Optional[_models.ThroughputSettingsGetResults]: error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[Optional[_models.ThroughputSettingsGetResults]] = kwargs.pop("cls", None) request = build_migrate_gremlin_graph_to_autoscale_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self._migrate_gremlin_graph_to_autoscale_initial.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200, 202]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = None response_headers = {} if response.status_code == 200: deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if response.status_code == 202: response_headers["azure-AsyncOperation"] = self._deserialize( "str", response.headers.get("azure-AsyncOperation") ) response_headers["location"] = self._deserialize("str", response.headers.get("location")) if cls: return cls(pipeline_response, deserialized, response_headers) return deserialized _migrate_gremlin_graph_to_autoscale_initial.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/throughputSettings/default/migrateToAutoscale" }
[docs] @distributed_trace def begin_migrate_gremlin_graph_to_autoscale( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, **kwargs: Any ) -> LROPoller[_models.ThroughputSettingsGetResults]: """Migrate an Azure Cosmos DB Gremlin graph from manual throughput to autoscale. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[_models.ThroughputSettingsGetResults] = kwargs.pop("cls", None) polling: Union[bool, PollingMethod] = kwargs.pop("polling", True) lro_delay = kwargs.pop("polling_interval", self._config.polling_interval) cont_token: Optional[str] = kwargs.pop("continuation_token", None) if cont_token is None: raw_result = self._migrate_gremlin_graph_to_autoscale_initial( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, api_version=api_version, cls=lambda x, y, z: x, headers=_headers, params=_params, **kwargs ) kwargs.pop("error_map", None) def get_long_running_output(pipeline_response): deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized if polling is True: polling_method: PollingMethod = cast(PollingMethod, ARMPolling(lro_delay, **kwargs)) elif polling is False: polling_method = cast(PollingMethod, NoPolling()) else: polling_method = polling if cont_token: return LROPoller.from_continuation_token( polling_method=polling_method, continuation_token=cont_token, client=self._client, deserialization_callback=get_long_running_output, ) return LROPoller(self._client, raw_result, get_long_running_output, polling_method) # type: ignore
begin_migrate_gremlin_graph_to_autoscale.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/throughputSettings/default/migrateToAutoscale" } def _migrate_gremlin_graph_to_manual_throughput_initial( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, **kwargs: Any ) -> Optional[_models.ThroughputSettingsGetResults]: error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[Optional[_models.ThroughputSettingsGetResults]] = kwargs.pop("cls", None) request = build_migrate_gremlin_graph_to_manual_throughput_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, subscription_id=self._config.subscription_id, api_version=api_version, template_url=self._migrate_gremlin_graph_to_manual_throughput_initial.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200, 202]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = None response_headers = {} if response.status_code == 200: deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if response.status_code == 202: response_headers["azure-AsyncOperation"] = self._deserialize( "str", response.headers.get("azure-AsyncOperation") ) response_headers["location"] = self._deserialize("str", response.headers.get("location")) if cls: return cls(pipeline_response, deserialized, response_headers) return deserialized _migrate_gremlin_graph_to_manual_throughput_initial.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/throughputSettings/default/migrateToManualThroughput" }
[docs] @distributed_trace def begin_migrate_gremlin_graph_to_manual_throughput( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, **kwargs: Any ) -> LROPoller[_models.ThroughputSettingsGetResults]: """Migrate an Azure Cosmos DB Gremlin graph from autoscale to manual throughput. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either ThroughputSettingsGetResults or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.ThroughputSettingsGetResults] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = kwargs.pop("headers", {}) or {} _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) cls: ClsType[_models.ThroughputSettingsGetResults] = kwargs.pop("cls", None) polling: Union[bool, PollingMethod] = kwargs.pop("polling", True) lro_delay = kwargs.pop("polling_interval", self._config.polling_interval) cont_token: Optional[str] = kwargs.pop("continuation_token", None) if cont_token is None: raw_result = self._migrate_gremlin_graph_to_manual_throughput_initial( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, api_version=api_version, cls=lambda x, y, z: x, headers=_headers, params=_params, **kwargs ) kwargs.pop("error_map", None) def get_long_running_output(pipeline_response): deserialized = self._deserialize("ThroughputSettingsGetResults", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized if polling is True: polling_method: PollingMethod = cast(PollingMethod, ARMPolling(lro_delay, **kwargs)) elif polling is False: polling_method = cast(PollingMethod, NoPolling()) else: polling_method = polling if cont_token: return LROPoller.from_continuation_token( polling_method=polling_method, continuation_token=cont_token, client=self._client, deserialization_callback=get_long_running_output, ) return LROPoller(self._client, raw_result, get_long_running_output, polling_method) # type: ignore
begin_migrate_gremlin_graph_to_manual_throughput.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/throughputSettings/default/migrateToManualThroughput" } def _retrieve_continuous_backup_information_initial( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, location: Union[_models.ContinuousBackupRestoreLocation, IO], **kwargs: Any ) -> Optional[_models.BackupInformation]: error_map = { 401: ClientAuthenticationError, 404: ResourceNotFoundError, 409: ResourceExistsError, 304: ResourceNotModifiedError, } error_map.update(kwargs.pop("error_map", {}) or {}) _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {}) _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None)) cls: ClsType[Optional[_models.BackupInformation]] = kwargs.pop("cls", None) content_type = content_type or "application/json" _json = None _content = None if isinstance(location, (IO, bytes)): _content = location else: _json = self._serialize.body(location, "ContinuousBackupRestoreLocation") request = build_retrieve_continuous_backup_information_request( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, subscription_id=self._config.subscription_id, api_version=api_version, content_type=content_type, json=_json, content=_content, template_url=self._retrieve_continuous_backup_information_initial.metadata["url"], headers=_headers, params=_params, ) request = _convert_request(request) request.url = self._client.format_url(request.url) pipeline_response: PipelineResponse = self._client._pipeline.run( # pylint: disable=protected-access request, stream=False, **kwargs ) response = pipeline_response.http_response if response.status_code not in [200, 202]: map_error(status_code=response.status_code, response=response, error_map=error_map) raise HttpResponseError(response=response, error_format=ARMErrorFormat) deserialized = None if response.status_code == 200: deserialized = self._deserialize("BackupInformation", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized _retrieve_continuous_backup_information_initial.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/retrieveContinuousBackupInformation" } @overload def begin_retrieve_continuous_backup_information( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, location: _models.ContinuousBackupRestoreLocation, *, content_type: str = "application/json", **kwargs: Any ) -> LROPoller[_models.BackupInformation]: """Retrieves continuous backup information for a gremlin graph. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :param location: The name of the continuous backup restore location. Required. :type location: ~azure.mgmt.cosmosdb.models.ContinuousBackupRestoreLocation :keyword content_type: Body Parameter content-type. Content type parameter for JSON body. Default value is "application/json". :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either BackupInformation or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.BackupInformation] :raises ~azure.core.exceptions.HttpResponseError: """ @overload def begin_retrieve_continuous_backup_information( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, location: IO, *, content_type: str = "application/json", **kwargs: Any ) -> LROPoller[_models.BackupInformation]: """Retrieves continuous backup information for a gremlin graph. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :param location: The name of the continuous backup restore location. Required. :type location: IO :keyword content_type: Body Parameter content-type. Content type parameter for binary body. Default value is "application/json". :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either BackupInformation or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.BackupInformation] :raises ~azure.core.exceptions.HttpResponseError: """
[docs] @distributed_trace def begin_retrieve_continuous_backup_information( self, resource_group_name: str, account_name: str, database_name: str, graph_name: str, location: Union[_models.ContinuousBackupRestoreLocation, IO], **kwargs: Any ) -> LROPoller[_models.BackupInformation]: """Retrieves continuous backup information for a gremlin graph. :param resource_group_name: The name of the resource group. The name is case insensitive. Required. :type resource_group_name: str :param account_name: Cosmos DB database account name. Required. :type account_name: str :param database_name: Cosmos DB database name. Required. :type database_name: str :param graph_name: Cosmos DB graph name. Required. :type graph_name: str :param location: The name of the continuous backup restore location. Is either a ContinuousBackupRestoreLocation type or a IO type. Required. :type location: ~azure.mgmt.cosmosdb.models.ContinuousBackupRestoreLocation or IO :keyword content_type: Body Parameter content-type. Known values are: 'application/json'. Default value is None. :paramtype content_type: str :keyword callable cls: A custom type or function that will be passed the direct response :keyword str continuation_token: A continuation token to restart a poller from a saved state. :keyword polling: By default, your polling method will be ARMPolling. Pass in False for this operation to not poll, or pass in your own initialized polling object for a personal polling strategy. :paramtype polling: bool or ~azure.core.polling.PollingMethod :keyword int polling_interval: Default waiting time between two polls for LRO operations if no Retry-After header is present. :return: An instance of LROPoller that returns either BackupInformation or the result of cls(response) :rtype: ~azure.core.polling.LROPoller[~azure.mgmt.cosmosdb.models.BackupInformation] :raises ~azure.core.exceptions.HttpResponseError: """ _headers = case_insensitive_dict(kwargs.pop("headers", {}) or {}) _params = case_insensitive_dict(kwargs.pop("params", {}) or {}) api_version: Literal["2022-11-15-preview"] = kwargs.pop( "api_version", _params.pop("api-version", self._config.api_version) ) content_type: Optional[str] = kwargs.pop("content_type", _headers.pop("Content-Type", None)) cls: ClsType[_models.BackupInformation] = kwargs.pop("cls", None) polling: Union[bool, PollingMethod] = kwargs.pop("polling", True) lro_delay = kwargs.pop("polling_interval", self._config.polling_interval) cont_token: Optional[str] = kwargs.pop("continuation_token", None) if cont_token is None: raw_result = self._retrieve_continuous_backup_information_initial( resource_group_name=resource_group_name, account_name=account_name, database_name=database_name, graph_name=graph_name, location=location, api_version=api_version, content_type=content_type, cls=lambda x, y, z: x, headers=_headers, params=_params, **kwargs ) kwargs.pop("error_map", None) def get_long_running_output(pipeline_response): deserialized = self._deserialize("BackupInformation", pipeline_response) if cls: return cls(pipeline_response, deserialized, {}) return deserialized if polling is True: polling_method: PollingMethod = cast( PollingMethod, ARMPolling(lro_delay, lro_options={"final-state-via": "location"}, **kwargs) ) elif polling is False: polling_method = cast(PollingMethod, NoPolling()) else: polling_method = polling if cont_token: return LROPoller.from_continuation_token( polling_method=polling_method, continuation_token=cont_token, client=self._client, deserialization_callback=get_long_running_output, ) return LROPoller(self._client, raw_result, get_long_running_output, polling_method) # type: ignore
begin_retrieve_continuous_backup_information.metadata = { "url": "/subscriptions/{subscriptionId}/resourceGroups/{resourceGroupName}/providers/Microsoft.DocumentDB/databaseAccounts/{accountName}/gremlinDatabases/{databaseName}/graphs/{graphName}/retrieveContinuousBackupInformation" }