Migrer des versions de processeur

Cette page décrit la procédure à suivre pour migrer les versions de processeurs entraînés par Document AI d'un projet à un autre, ainsi que le schéma et les exemples de l'ensemble de données de la source vers le processeur de destination. Ces étapes automatisent le processus d'importation de la version du processeur, de son déploiement et de sa définition comme version par défaut dans le projet de destination.

Avant de commencer

• Obtenez un ID de projet Google Cloud .
• disposer d'un ID de processeur Document AI ;
• disposer d'un compte Cloud Storage ;
• Utiliser Python: notebook Jupyter (Vertex AI)
• Vous devez disposer d'autorisations pour accorder l'accès au compte de service dans les projets source et de destination.

Procédure par étapes

La procédure est décrite dans les étapes suivantes.

Étape 1: Identifiez le compte de service associé au notebook Vertex AI

!gcloud config list account

Sortie :

[core]
account = [e-mail]

Your active configuration is: [default]

Étape 2: Accordez les autorisations requises au compte de service

Dans le projet Google Cloud qui est la destination prévue de la migration, ajoutez le compte de service acquis à l'étape précédente en tant que principal et attribuez-lui les deux rôles suivants:

• Administrateur Document AI
• Administrateur de l'espace de stockage

Pour en savoir plus, consultez les pages Attribuer des rôles à des comptes de service et Clés de chiffrement gérées par le client (CMEK).

Pour que la migration fonctionne, le compte de service utilisé pour exécuter ce notebook doit disposer des éléments suivants:

• Rôles dans les projets source et de destination pour créer le bucket de l'ensemble de données ou le créer s'il n'existe pas, ainsi que des autorisations de lecture et d'écriture pour tous les objets.
• Rôle "Éditeur Document AI" dans le projet source, comme décrit dans la section Importer une version de processeur.

Téléchargez une clé JSON pour le compte de service afin de pouvoir vous authentifier et vous autoriser en tant que compte de service. Pour en savoir plus, consultez la page Clés de compte de service.

Suivant :

1. Accédez au compte de service.
2. Sélectionnez le compte de service destiné à effectuer cette tâche.
3. Accédez à l'onglet Clés, cliquez sur Add Key, puis sélectionnez Créer une clé.
4. Sélectionnez le type de clé (de préférence JSON).

5. Cliquez sur Create, puis téléchargez le fichier dans un chemin spécifique.

6. Mettez à jour le chemin d'accès dans la variable service_account_key dans l'extrait suivant.

service_account_key='path_to_sa_key.json'

from google.oauth2 import service_account
from google.cloud import storage

# Authenticate the service account
credentials = service_account.Credentials.from_service_account_file(
    service_account_key
)

# pass this credentials variable to all client initializations
# storage_client = storage.Client(credentials=credentials)
# docai_client = documentai.DocumentProcessorServiceClient(credentials=credentials)

Étape 3 : Importez les bibliothèques

import time
from pathlib import Path
from typing import Optional, Tuple
from google.cloud.documentai_v1beta3.services.document_service import pagers
from google.api_core.client_options import ClientOptions
from google.api_core.operation import Operation
from google.cloud import documentai_v1beta3 as documentai
from google.cloud import storage
from tqdm import tqdm

Étape 4: Saisissez les informations

• source_project_id: indiquez l'ID du projet source.
• source_location: indiquez l'emplacement du processeur source (us ou eu).
• source_processor_id: indiquez l' Google Cloud ID du processeur Document AI.
• source_processor_version_to_import: indiquez l'ID de version du processeur Document AI pour la version entraînée. Google Cloud
• migrate_dataset:indiquez cette valeur sous la forme True ou False. Si vous souhaitez migrer un ensemble de données du processeur source vers le processeur de destination, indiquez True, sinon False. La valeur par défaut est False.
• source_exported_gcs_path: indiquez le chemin d'accès Cloud Storage pour stocker les fichiers JSON.
• destination_project_id: indiquez l'ID du projet de destination.
• destination_processor_id : indiquez l'ID du processeur d'IA Document : "" ou processor_id du projet de destination. Google Cloud

source_project_id = "source-project-id"
source_location = "processor-location"
source_processor_id = "source-processor-id"
source_processor_version_to_import = "source-processor-version-id"
migrate_dataset = False  # Either True or False
source_exported_gcs_path = (
    "gs://bucket/path/to/export_dataset/"
)
destination_project_id = "< destination-project-id >"
# Give an empty string if you wish to create a new processor
destination_processor_id = ""

Étape 5: Exécutez le code

import time
from pathlib import Path
from typing import Optional, Tuple
from google.cloud.documentai_v1beta3.services.document_service import pagers
from google.api_core.client_options import ClientOptions
from google.api_core.operation import Operation
from google.cloud import documentai_v1beta3 as documentai
from google.cloud import storage
from tqdm import tqdm

source_project_id = "source-project-id"
source_location = "processor-location"
source_processor_id = "source-processor-id"
source_processor_version_to_import = "source-processor-version-id"
migrate_dataset = False # Either True or False
source_exported_gcs_path = (
    "gs://bucket/path/to/export_dataset/"
)
destination_project_id = "< destination-project-id >"
# Give empty string if you wish to create a new processor
destination_processor_id = ""

exported_bucket_name = source_exported_gcs_path.split("/")[2]
exported_bucket_path_prefix = "/".join(source_exported_gcs_path.split("/")[3:])
destination_location = source_location

def sample_get_processor(project_id: str, processor_id: str, location: str)->Tuple[str, str]:
    """
    This function returns Processor Display Name and Type of Processor from source project

    Args:
        project_id (str): Project ID
        processor_id (str): Document AI Processor ID
        location (str): Processor Location

    Returns:
        Tuple[str, str]: Returns Processor Display name and type
    """
    client = documentai.DocumentProcessorServiceClient()
    print(
        f"Fetching processor({processor_id}) details from source project ({project_id})"
    )
    name = f"projects/{project_id}/locations/{location}/processors/{processor_id}"
    request = documentai.GetProcessorRequest(
        name=name,
    )
    response = client.get_processor(request=request)
    print(f"Processor Name: {response.name}")
    print(f"Processor Display Name: {response.display_name}")
    print(f"Processor Type: {response.type_}")
    return response.display_name, response.type_

def sample_create_processor(project_id: str, location: str, display_name: str, processor_type: str)->documentai.Processor:
    """It will create Processor in Destination project

    Args:
        project_id (str): Project ID
        location (str): Location fo processor
        display_name (str): Processor Display Name
        processor_type (str): Google Cloud Document AI Processor type

    Returns:
        documentai.Processor: Returns details abouts newly created processor
    """

    client = documentai.DocumentProcessorServiceClient()
    request = documentai.CreateProcessorRequest(
        parent=f"projects/{project_id}/locations/{location}",
        processor={
            "type_": processor_type,
            "display_name": display_name,
        },
    )
    print(f"Creating Processor in project: {project_id} in location: {location}")
    print(f"Display Name: {display_name} & Processor Type: {processor_type}")
    res = client.create_processor(request=request)
    return res

def initialize_dataset(project_id: str, processor_id: str, location: str)-> Operation:
    """It will configure dataset for target processor in destination project

    Args:
        project_id (str): Project ID
        processor_id (str): DocuemntAI Processor ID
        location (str): Processor Location

    Returns:
        Operation: An object representing a long-running operation
    """

    # opts = ClientOptions(api_endpoint=f"{location}-documentai.googleapis.com")
    client = documentai.DocumentServiceClient()  # client_options=opts
    dataset = documentai.types.Dataset(
        name=f"projects/{project_id}/locations/{location}/processors/{processor_id}/dataset",
        state=3,
        unmanaged_dataset_config={},
        spanner_indexing_config={},
    )
    request = documentai.types.UpdateDatasetRequest(dataset=dataset)
    print(
        f"Configuring Dataset in  project: {project_id} for processor: {processor_id}"
    )
    response = client.update_dataset(request=request)
    return response

def get_dataset_schema(project_id: str, processor_id: str, location: str)->documentai.DatasetSchema:
    """It helps to fetch processor schema

    Args:
        project_id (str): Project ID
        processor_id (str): DocumentAI Processor ID
        location (str): Processor Location

    Returns:
        documentai.DatasetSchema: Return deails about Processor Dataset Schema
    """

    # Create a client
    processor_name = (
        f"projects/{project_id}/locations/{location}/processors/{processor_id}"
    )
    client = documentai.DocumentServiceClient()
    request = documentai.GetDatasetSchemaRequest(
        name=processor_name + "/dataset/datasetSchema"
    )
    # Make the request
    print(f"Fetching schema from source processor: {processor_id}")
    response = client.get_dataset_schema(request=request)
    return response

def upload_dataset_schema(schema: documentai.DatasetSchema)->documentai.DatasetSchema:
    """It helps to update the schema in destination processor

    Args:
        schema (documentai.DatasetSchema): Document AI Processor Schema details & Metadata

    Returns:
        documentai.DatasetSchema: Returns Dataset Schema object
    """

    client = documentai.DocumentServiceClient()
    request = documentai.UpdateDatasetSchemaRequest(dataset_schema=schema)
    print("Updating Schema in destination processor")
    res = client.update_dataset_schema(request=request)
    return res

def store_document_as_json(document: str, bucket_name: str, file_name: str)->None:
    """It helps to upload data to Cloud Storage and stores as a blob

    Args:
        document (str): Processor response in json string format
        bucket_name (str): Cloud Storage bucket name
        file_name (str): Cloud Storage blob uri
    """

    print(f"\tUploading file to Cloud Storage gs://{bucket_name}/{file_name}")
    storage_client = storage.Client()
    process_result_bucket = storage_client.get_bucket(bucket_name)
    document_blob = storage.Blob(
        name=str(Path(file_name)), bucket=process_result_bucket
    )
    document_blob.upload_from_string(document, content_type="application/json")

def list_documents(project_id: str, location: str, processor: str, page_size: Optional[int]=100, page_token: Optional[str]="")->pagers.ListDocumentsPager:
    """This function helps to list the samples present in processor dataset

    Args:
        project_id (str): Project ID
        location (str): Processor Location 
        processor (str): DocumentAI Processor ID
        page_size (Optional[int], optional): The maximum number of documents to return. Defaults to 100.
        page_token (Optional[str], optional): A page token, received from a previous ListDocuments call. Defaults to "".

    Returns:
        pagers.ListDocumentsPager: Returns all details about documents present in Processor Dataset
    """
    client = documentai.DocumentServiceClient()
    dataset = (
        f"projects/{project_id}/locations/{location}/processors/{processor}/dataset"
    )
    request = documentai.types.ListDocumentsRequest(
        dataset=dataset,
        page_token=page_token,
        page_size=page_size,
        return_total_size=True,
    )
    print(f"Listingll  documents/Samples present in processor: {processor}")
    operation = client.list_documents(request)
    return operation

def get_document(project_id: str, location: str, processor: str, doc_id: documentai.DocumentId)->documentai.GetDocumentResponse:
    """It will fetch data for individual sample/document present in dataset

    Args:
        project_id (str): Project ID
        location (str): Processor Location
        processor (str): Document AI Processor ID
        doc_id (documentai.DocumentId): Document identifier

    Returns:
        documentai.GetDocumentResponse: Returns data related to doc_id
    """

    client = documentai.DocumentServiceClient()
    dataset = (
        f"projects/{project_id}/locations/{location}/processors/{processor}/dataset"
    )
    request = documentai.GetDocumentRequest(dataset=dataset, document_id=doc_id)
    operation = client.get_document(request)
    return operation

def import_documents(project_id: str, processor_id: str, location: str, gcs_path: str)->Operation:
    """It helps to import samples/docuemnts from Cloud Storage path to processor via API call

    Args:
        project_id (str): Project ID
        processor_id (str): Document AI Processor ID
        location (str): Processor Location
        gcs_path (str): Cloud Storage path uri prefix 

    Returns:
        Operation: An object representing a long-running operation
    """

    client = documentai.DocumentServiceClient()
    dataset = (
        f"projects/{project_id}/locations/{location}/processors/{processor_id}/dataset"
    )
    request = documentai.ImportDocumentsRequest(
        dataset=dataset,
        batch_documents_import_configs=[
            {
                "dataset_split": "DATASET_SPLIT_TRAIN",
                "batch_input_config": {
                    "gcs_prefix": {"gcs_uri_prefix": gcs_path + "train/"}
                },
            },
            {
                "dataset_split": "DATASET_SPLIT_TEST",
                "batch_input_config": {
                    "gcs_prefix": {"gcs_uri_prefix": gcs_path + "test/"}
                },
            },
            {
                "dataset_split": "DATASET_SPLIT_UNASSIGNED",
                "batch_input_config": {
                    "gcs_prefix": {"gcs_uri_prefix": gcs_path + "unassigned/"}
                },
            },
        ],
    )
    print(
        f"Importing Documents/samples from {gcs_path} to corresponding tran_test_unassigned sections"
    )
    response = client.import_documents(request=request)
    return response

def import_processor_version(source_processor_version_name: str, destination_processor_name: str)->Operation:
    """It helps to import processor version from source processor to destanation processor

    Args:
        source_processor_version_name (str): source processor name in this format projects/{project}/locations/{location}/processors/{processor}
        destination_processor_name (str): destination processor name in this format projects/{project}/locations/{location}/processors/{processor}

    Returns:
        Operation: An object representing a long-running operation
    """

    from google.cloud import documentai_v1beta3

    # provide the source version(to copy) processor details in the following format
    client = documentai_v1beta3.DocumentProcessorServiceClient()
    # provide the new processor name in the parent variable in format 'projects/{project_number}/locations/{location}/processors/{new_processor_id}'
    import google.cloud.documentai_v1beta3 as documentai

    op_import_version_req = (
        documentai.types.document_processor_service.ImportProcessorVersionRequest(
            processor_version_source=source_processor_version_name,
            parent=destination_processor_name,
        )
    )

    print("Importing processor from source to destination")
    print(f"\tSource: {source_processor_version_name}")
    print(f"\tDestination: {destination_processor_name}")
    # copying the processor
    operation = client.import_processor_version(request=op_import_version_req)
    print(operation.metadata)
    print("Waitin for operation to complete...")
    operation.result()
    return operation

def deploy_and_set_default_processor_version(
    project_id: str, location: str, processor_id: str, processor_version_id: str
)->None:
    """It helps to deploy to imported processor version and set it as default version

    Args:
        project_id (str): Project ID
        location (str): Processor Location
        processor_id (str): Document AI Processor ID
        processor_version_id (str): Document AI Processor Version ID
    """

    # Construct the resource name of the processor version
    processor_name = (
        f"projects/{project_id}/locations/{location}/processors/{processor_id}"
    )
    default_processor_version_name = f"projects/{project_id}/locations/{location}/processors/{processor_id}/processorVersions/{processor_version_id}"

    # Initialize the Document AI client
    client_options = ClientOptions(api_endpoint=f"{location}-documentai.googleapis.com")
    client = documentai.DocumentProcessorServiceClient(client_options=client_options)

    # Deploy the processor version
    operation = client.deploy_processor_version(name=default_processor_version_name)
    print(f"Deploying processor version: {operation.operation.name}")
    print("Waiting for operation to complete...")
    result = operation.result()
    print("Processor version deployed")
    # Set the deployed version as the default version
    request = documentai.SetDefaultProcessorVersionRequest(
        processor=processor_name,
        default_processor_version=default_processor_version_name,
    )
    operation = client.set_default_processor_version(request=request)
    print(f"Setting default processor version: {operation.operation.name}")
    operation.result()
    print(f"Default processor version set {default_processor_version_name}")

def main(destination_processor_id: str, migrate_dataset: bool = False)->None:
    """Entry function to perform Processor Migration from Source Project to Destination project

    Args:
        destination_processor_id (str): Either empty string or processor id in desination project
    """
    # Checking processor id of destination project
    if destination_processor_id == "":
        # Fetching Processor Display Name and Type of Processor from source project
        display_name, processor_type = sample_get_processor(
            source_project_id, source_processor_id, source_location
        )
        # Creating Processor in Destination project
        des_processor = sample_create_processor(
            destination_project_id, destination_location, display_name, processor_type
        )
        print(des_processor)
        destination_processor_id = des_processor.name.split("/")[-1]
        # configuring dataset for target processor in destination project
        r = initialize_dataset(
            destination_project_id, destination_processor_id, destination_location
        )
    # fetching processor schema from source processor
    exported_schema = get_dataset_schema(
        source_project_id, source_processor_id, source_location
    )
    exported_schema.name = f"projects/{destination_project_id}/locations/{destination_location}/processors/{destination_processor_id}/dataset/datasetSchema"
    # Copying schema from source processor to desination processor 
    import_schema = upload_dataset_schema(exported_schema)
    if migrate_dataset == True: # to migrate dataset from source to destination processor
        print("Migrating Dataset from source to destination processor")
        # Fetching/listing the samples/JSONs present in source processor dataset
        results = list_documents(source_project_id, source_location, source_processor_id)
        document_list = results.document_metadata
        while len(document_list) != results.total_size:
            page_token = results.next_page_token
            results = list_documents(
                source_project_id,
                source_location,
                source_processor_id,
                page_token=page_token,
            )
            document_list.extend(results.document_metadata)
        print("Exporting Dataset...")
        for doc in tqdm(document_list):
            doc_id = doc.document_id
            split_type = doc.dataset_type
            if split_type == 3:
                split = "unassigned"
            elif split_type == 2:
                split = "test"
            elif split_type == 1:
                split = "train"
            else:
                split = "unknown"
            file_name = doc.display_name
            # fetching/downloading data for individual sample/document present in dataset
            res = get_document(
                source_project_id, source_location, source_processor_id, doc_id
            )
            output_file_name = (
                f"{exported_bucket_path_prefix.strip('/')}/{split}/{file_name}.json"
            )
            # Converting Document AI Proto object to JSON string
            json_data = documentai.Document.to_json(res.document)
            # Uploading JSON data to specified Cloud Storage path
            store_document_as_json(json_data, exported_bucket_name, output_file_name)

        print(f"Importing dataset to {destination_processor_id}")
        gcs_path = source_exported_gcs_path.strip("/") + "/"
        project = destination_project_id
        location = destination_location
        processor = destination_processor_id
        # importing samples/docuemnts from Cloud Storage path to destination processor
        res = import_documents(project, processor, location, gcs_path)
        print(f"Waiting for {len(document_list)*1.5} seconds")
        time.sleep(len(document_list) * 1.5)
    else:
        print("\tSkipping Dataset Migration actions like, exporting source dataset to Cloud Storage and importing dataset to destination processor")

    # Checking for source processor vesion, if id provided then it will be imported to destination processor
    if source_processor_version_to_import != "":
        print(f"Importing Processor Version {source_processor_version_to_import}")
        source_version = f"projects/{source_project_id}/locations/{source_location}/processors/{source_processor_id}/processorVersions/{source_processor_version_to_import}"
        destination_version = f"projects/{destination_project_id}/locations/{destination_location}/processors/{destination_processor_id}"
        # source_version = f"projects/{source_project_id}/locations/us/processors/a82fc086440d7ea1/processorVersions/f1eeed93aad5e317"  # Data for testing
        # Importing processor version from source processor to destanation processor
        operation = import_processor_version(source_version, destination_version)
        name = operation.metadata.common_metadata.resource
        destination_processor_version_id = name.split("/")[-1]
        # deploying newly imported processor version and set it as default version in desination project
        deploy_and_set_default_processor_version(
            destination_project_id,
            destination_location,
            destination_processor_id,
            destination_processor_version_id,
        )

main(destination_processor_id, migrate_dataset)
print("Process Completed!!!")

Étape 6: Vérifier les détails de la sortie

Accédez au projet de destination et vérifiez la création de l'outil de traitement, la disponibilité de l'ensemble de données et la nouvelle version de l'outil de traitement en tant que version par défaut.