Créer un graphe de propriétés cross-cloud sur un lakehouse ouvert et sans frontières

Le tutoriel suivant vous explique comment créer un seul graphe BigQuery qui unifie les silos de données de deux clouds différents à l'aide d'un lakehouse ouvert sans frontières et de points de terminaison de catalogue REST Apache Iceberg, sans déplacer les données.

Avant de commencer

Avant de commencer, configurez votre environnement et activez les API requises.

  1. Définissez votre projet et votre région, puis activez les API :

    export PROJECT_ID="your-gcp-project-id"
export REGION="us-east4"

gcloud config set project "$PROJECT_ID"

gcloud services enable \
  biglake.googleapis.com \
  bigquery.googleapis.com \
  secretmanager.googleapis.com \
  storage.googleapis.com

  2. Créez un environnement virtuel Python pour le chargeur :

    python3 -m venv iceberg-venv
source iceberg-venv/bin/activate
pip install --quiet "pyiceberg[pyarrow]"

Créer le spoke Google Cloud

Configurez un point de terminaison de catalogue REST Apache Iceberg ouvert, associé à un bucket Cloud Storage, et chargez-y trois tables Iceberg.

  1. Créez le bucket et le catalogue :

    export GCS_BUCKET="gs://${PROJECT_ID}-xcloud-lake"
export GCS_CATALOG="gcs_lake"

gcloud storage buckets create "$GCS_BUCKET" \
  --project="$PROJECT_ID" \
  --location="$REGION"

gcloud biglake iceberg catalogs create "$GCS_CATALOG" \
  --project="$PROJECT_ID" \
  --catalog-type=biglake \
  --primary-location="$REGION" \
  --default-location="$GCS_BUCKET"

  2. Enregistrez le script Python suivant sous le nom load_gcs.py pour alimenter les tables :

    import subprocess, pyarrow as pa
from pyiceberg.catalog.rest import RestCatalog
from pyiceberg.schema import Schema
from pyiceberg.types import NestedField, StringType, LongType, DoubleType
import os

PROJECT = os.environ["PROJECT_ID"]
CATALOG = os.environ["GCS_CATALOG"]
TOKEN = subprocess.check_output(
    ["gcloud", "auth", "application-default", "print-access-token"], text=True
).strip()

cat = RestCatalog(
    name=CATALOG,
    uri="https://biglake.googleapis.com/iceberg/v1/restcatalog",
    warehouse=f"bl://projects/{PROJECT}/catalogs/{CATALOG}",
    token=TOKEN,
    **{"header.x-goog-user-project": PROJECT,
       "header.X-Iceberg-Access-Delegation": "vended-credentials"},
)

cat.create_namespace_if_not_exists("retail")

def mk(name, schema, table):
    ident = ("retail", name)
    try: cat.drop_table(ident)
    except Exception: pass
    t = cat.create_table(ident, schema=schema)
    t.append(table)
    print(f"  {name}: {table.num_rows} rows")

# customers
mk("customers",
   Schema(NestedField(1, "customer_id", StringType()),
          NestedField(2, "name", StringType()),
          NestedField(3, "region", StringType())),
   pa.table({
       "customer_id": ["C1", "C2", "C3", "C4", "C5", "C6"],
       "name": ["Ana", "Ben", "Cara", "Dan", "Eve", "Finn"],
       "region": ["west", "west", "east", "east", "west", "south"],
   }))

# orders
mk("orders",
   Schema(NestedField(1, "order_id", StringType()),
          NestedField(2, "customer_id", StringType()),
          NestedField(3, "status", StringType()),
          NestedField(4, "amount", DoubleType())),
   pa.table({
       "order_id": ["O1","O2","O3","O4","O5","O6","O7","O8","O9","O10"],
       "customer_id": ["C1","C1","C2","C3","C3","C4","C5","C5","C6","C2"],
       "status": ["shipped"]*8 + ["pending","shipped"],
       "amount": [156.0,89.0,120.0,147.0,89.0,199.0,25.0,88.0,80.0,224.0],
   }))

# order_items
mk("order_items",
   Schema(NestedField(1, "order_item_id", StringType()),
          NestedField(2, "order_id", StringType()),
          NestedField(3, "product_id", StringType()),
          NestedField(4, "quantity", LongType()),
          NestedField(5, "amount", DoubleType())),
   pa.table({
       "order_item_id": [f"OI{i}" for i in range(1, 16)],
       "order_id":  ["O1","O1","O2","O3","O3","O4","O5","O6","O6","O7","O8","O9","O10","O10","O2"],
       "product_id": ["P1","P2","P3","P1","P5","P4","P3","P6","P8","P7","P1","P2","P4","P5","P6"],
       "quantity":  [1,2,1,1,3,1,1,1,4,2,1,1,1,2,1],
       "amount":    [120.0,36.0,89.0,120.0,27.0,199.0,89.0,25.0,88.0,80.0,120.0,18.0,199.0,18.0,25.0],
   }))

print("tables:", cat.list_tables("retail"))

  3. Exécutez le script pour charger les tables :

    python load_gcs.py

  4. Vérifiez les tables depuis BigQuery :

    bq --location="$REGION" query --use_legacy_sql=false \
  'SELECT customer_id, name, region
   FROM `'"$PROJECT_ID"'.gcs_lake.retail.customers`
   ORDER BY customer_id'

Créer le spoke AWS

Affichez un catalogue Databricks Unity dans BigQuery via un lakehouse ouvert et sans frontières.

  1. Dans votre éditeur SQL Databricks, créez les tables "products" et "suppliers" :

    CREATE SCHEMA IF NOT EXISTS `<CATALOG>`.retail;

CREATE OR REPLACE TABLE `<CATALOG>`.retail.suppliers (
  supplier_id STRING, name STRING, country STRING
) USING ICEBERG;

INSERT INTO `<CATALOG>`.retail.suppliers VALUES
  ('S1','Acme','USA'), ('S2','Globex','Germany'),
  ('S3','Initech','Japan'), ('S4','Umbrella','UK');

CREATE OR REPLACE TABLE `<CATALOG>`.retail.products (
  product_id STRING, name STRING, category STRING,
  supplier_id STRING, price DOUBLE
) USING ICEBERG;

INSERT INTO `<CATALOG>`.retail.products VALUES
  ('P1','Widget','Gadgets',  'S1',120.0),
  ('P2','Gizmo','Gadgets',   'S1', 18.0),
  ('P3','Sprocket','Parts',  'S2', 89.0),
  ('P4','Cog','Parts',       'S2',199.0),
  ('P5','Bolt','Parts',      'S3',  9.0),
  ('P6','Nut','Parts',       'S3', 25.0),
  ('P7','Gear','Machinery',  'S4', 40.0),
  ('P8','Axle','Machinery',  'S4', 22.0);

  2. Accordez un accès en lecture à votre compte principal de service :

    GRANT USE CATALOG ON CATALOG `<CATALOG>` TO `<SP_APPLICATION_ID>`;
GRANT USE SCHEMA, SELECT, EXTERNAL USE SCHEMA
  ON SCHEMA `<CATALOG>`.retail
  TO `<SP_APPLICATION_ID>`;

  3. Stockez les identifiants dans Secret Manager :

    export CLOUDSDK_API_ENDPOINT_OVERRIDES_SECRETMANAGER="https://secretmanager.${REGION}.rep.googleapis.com/"

printf '{"client_id":"<SP_CLIENT_ID>","client_secret":"<SP_CLIENT_SECRET>"}' \
  | gcloud secrets create dbx-fed-sp \
      --project="$PROJECT_ID" \
      --location="$REGION" \
      --data-file=-

  4. Créez le catalogue fédéré :

    export DBX_HOST="<your-workspace-host-without-https>"
export UC_CATALOG="<CATALOG>"

gcloud alpha biglake iceberg catalogs create dbx_fed_catalog \
  --project="$PROJECT_ID" \
  --catalog-type=federated \
  --federated-catalog-type=unity \
  --unity-instance-name="$DBX_HOST" \
  --unity-catalog-name="$UC_CATALOG" \
  --secret-name="projects/${PROJECT_ID}/locations/${REGION}/secrets/dbx-fed-sp" \
  --primary-location="$REGION" \
  --refresh-interval=330s

  5. Accordez au compte de service BigLake l'accès au secret :

    export BLIRC_SA="<paste-the-blirc-...-service-account-from-the-output>"
export CLOUDSDK_API_ENDPOINT_OVERRIDES_SECRETMANAGER="https://secretmanager.${REGION}.rep.googleapis.com/"

gcloud secrets add-iam-policy-binding dbx-fed-sp \
  --project="$PROJECT_ID" \
  --location="$REGION" \
  --member="serviceAccount:${BLIRC_SA}" \
  --role="roles/secretmanager.secretAccessor"

  6. Vérifiez la synchronisation :

    bq --location="$REGION" ls dbx_fed_catalog.retail

bq --location="$REGION" query --use_legacy_sql=false \
  'SELECT product_id, name, category, supplier_id, price
   FROM `'"$PROJECT_ID"'.dbx_fed_catalog.retail.products`
   ORDER BY product_id'

Créer le graphique de propriétés cross-cloud

Créez un graphique dont les tables de nœuds et d'arêtes s'étendent sur les deux clouds.

  1. Créez un ensemble de données dans votre région :

    bq --location="$REGION" mk --dataset "${PROJECT_ID}:xcloud_graph"

  2. Créez le graphique :

    CREATE OR REPLACE PROPERTY GRAPH `PROJECT_ID.xcloud_graph.retail_xcloud`
NODE TABLES (
  `PROJECT_ID.gcs_lake.retail.customers` AS Customer
    KEY (customer_id)
    LABEL Customer PROPERTIES (customer_id, name, region),
  `PROJECT_ID.gcs_lake.retail.orders` AS Orders
    KEY (order_id)
    LABEL OrderNode PROPERTIES (order_id, status, amount),
  `PROJECT_ID.gcs_lake.retail.order_items` AS OrderItems
    KEY (order_item_id)
    LABEL OrderItem PROPERTIES (order_item_id, quantity, amount),
  `PROJECT_ID.dbx_fed_catalog.retail.products` AS Products
    KEY (product_id)
    LABEL Product PROPERTIES (product_id, name, category, price),
  `PROJECT_ID.dbx_fed_catalog.retail.suppliers` AS Suppliers
    KEY (supplier_id)
    LABEL Supplier PROPERTIES (supplier_id, name, country)
)
EDGE TABLES (
  `PROJECT_ID.gcs_lake.retail.orders` AS Placed
    KEY (order_id)
    SOURCE KEY (customer_id) REFERENCES Customer (customer_id)
    DESTINATION KEY (order_id) REFERENCES Orders (order_id)
    LABEL PLACED,
  `PROJECT_ID.gcs_lake.retail.order_items` AS Contains_Item
    KEY (order_item_id)
    SOURCE KEY (order_id) REFERENCES Orders (order_id)
    DESTINATION KEY (order_item_id) REFERENCES OrderItems (order_item_id)
    LABEL CONTAINS_ITEM,
  `PROJECT_ID.gcs_lake.retail.order_items` AS Is_Product
    KEY (order_item_id)
    SOURCE KEY (order_item_id) REFERENCES OrderItems (order_item_id)
    DESTINATION KEY (product_id) REFERENCES Products (product_id)
    LABEL IS_PRODUCT,
  `PROJECT_ID.dbx_fed_catalog.retail.products` AS Supplied_By
    KEY (product_id)
    SOURCE KEY (product_id) REFERENCES Products (product_id)
    DESTINATION KEY (supplier_id) REFERENCES Suppliers (supplier_id)
    LABEL SUPPLIED_BY
)

Interroger le graphique cross-cloud

Parcourez le graphique pour interroger les relations entre les deux clouds.

  1. Exécutez une requête qui traverse les relations existant sur les deux clouds :

    GRAPH `PROJECT_ID.xcloud_graph.retail_xcloud`
MATCH (c:Customer)-[:PLACED]->
      (o:OrderNode)-[:CONTAINS_ITEM]->
      (li:OrderItem)-[:IS_PRODUCT]->
      (p:Product)-[:SUPPLIED_BY]->
      (s:Supplier)
RETURN
  c.name AS customer,
  o.order_id AS order_id,
  p.name AS product,
  p.category AS category,
  s.name AS supplier,
  s.country AS supplier_country,
  li.quantity AS qty
ORDER BY customer, order_id
LIMIT 50;

  2. Exécutez une requête pour identifier les pays qui approvisionnent chaque client :

    GRAPH `PROJECT_ID.xcloud_graph.retail_xcloud`
MATCH (c:Customer)-[:PLACED]->
      (:OrderNode)-[:CONTAINS_ITEM]->
      (:OrderItem)-[:IS_PRODUCT]->
      (:Product)-[:SUPPLIED_BY]->
      (s:Supplier)
RETURN DISTINCT c.name AS customer, s.country AS supplier_country
ORDER BY customer, supplier_country;

Effectuer un nettoyage

Pour éviter que les ressources utilisées dans cette démonstration soient facturées sur votre compte Google Cloud , procédez comme suit :

Pour éviter des frais récurrents, supprimez les ressources que vous avez créées.

  1. Supprimez l'ensemble de données de graphe :

    bq rm -r -f -d "${PROJECT_ID}:xcloud_graph"

  2. Supprimez le catalogue fédéré :

    gcloud alpha biglake iceberg catalogs delete dbx_fed_catalog --project="$PROJECT_ID"

  3. Enregistrez le script suivant sous le nom drop_gcs.py pour supprimer les tables et l'espace de noms GCS :

    import subprocess, os
from pyiceberg.catalog.rest import RestCatalog

PROJECT = os.environ["PROJECT_ID"]
CATALOG = os.environ["GCS_CATALOG"]
TOKEN = subprocess.check_output(
    ["gcloud", "auth", "application-default", "print-access-token"], text=True
).strip()

cat = RestCatalog(
    name=CATALOG,
    uri="https://biglake.googleapis.com/iceberg/v1/restcatalog",
    warehouse=f"bl://projects/{PROJECT}/catalogs/{CATALOG}",
    token=TOKEN,
    **{"header.x-goog-user-project": PROJECT},
)
for tbl in cat.list_tables("retail"):
    cat.drop_table(tbl)
cat.drop_namespace("retail")
print("emptied", CATALOG)

  4. Exécutez le script :

    python drop_gcs.py

  5. Supprimez le catalogue Cloud Storage :

    gcloud biglake iceberg catalogs delete "$GCS_CATALOG" --project="$PROJECT_ID"

  6. Supprimez le secret :

    export CLOUDSDK_API_ENDPOINT_OVERRIDES_SECRETMANAGER="https://secretmanager.${REGION}.rep.googleapis.com/"
gcloud secrets delete dbx-fed-sp --project="$PROJECT_ID" --location="$REGION"

  7. Supprimez le bucket :

    gcloud storage rm -r "$GCS_BUCKET"

Étapes suivantes