Spanner 및 PGAdapter 시작하기


목표

이 튜토리얼에서는 PostgreSQL 드라이버용 Spanner PGAdapter 로컬 프록시를 사용하여 다음 단계를 안내합니다.

  • Spanner 인스턴스와 데이터베이스 만들기
  • 데이터베이스에서 데이터 읽기, 쓰기 및 데이터에서 SQL 쿼리 실행
  • 데이터베이스 스키마 업데이트
  • 읽기-쓰기 트랜잭션을 사용하여 데이터 업데이트
  • 데이터베이스에 보조 색인 추가
  • 색인을 사용하여 데이터 읽기 및 데이터에서 SQL 쿼리 실행
  • 읽기 전용 트랜잭션을 사용하여 데이터 검색

비용

이 튜토리얼에서는 Google Cloud의 비용 청구 가능한 구성요소인 Spanner를 사용합니다. Spanner 사용 비용에 대한 자세한 내용은 가격 책정을 참조하세요.

시작하기 전에

설정에 설명된 단계를 완료하세요. 기본 Google Cloud 프로젝트 생성, 결제 사용 설정, Cloud Spanner API 사용 설정을 수행하고 Cloud Spanner API 사용에 필요한 사용자 인증 정보를 가져오기 위한 OAuth 2.0 설정을 완료해야 합니다.

특히 gcloud auth application-default login을 실행하여 사용자 인증 정보로 로컬 개발 환경을 설정해야 합니다.

로컬 PGAdapter 환경 준비

PostgreSQL 드라이버를 PGAdapter와 함께 사용하여 Spanner에 연결할 수 있습니다. PGAdapter는 PostgreSQL 네트워크 프로토콜을 Spanner gRPC 프로토콜로 변환하는 로컬 프록시입니다.

PGAdapter를 실행하려면 Java 또는 Docker가 필요합니다.

  1. 개발 머신에 아직 설치되어 있지 않다면 다음 중 하나를 설치합니다.

  2. 샘플 앱 저장소를 로컬 머신에 클론합니다.

    git clone https://github.com/GoogleCloudPlatform/pgadapter.git
    
  3. Spanner 샘플 코드가 있는 디렉터리로 변경합니다.

    psql

    cd pgadapter/samples/snippets/psql-snippets
    

    자바

    cd pgadapter/samples/snippets/java-snippets
    mvn package -DskipTests
    

    Go

    cd pgadapter/samples/snippets/golang-snippets
    

    Node.js

    cd pgadapter/samples/snippets/nodejs-snippets
    npm install
    

    Python

    cd pgadapter/samples/snippets/python-snippets
    python -m venv ./venv
    pip install -r requirements.txt
    cd samples
    

    C#

    cd pgadapter/samples/snippets/dotnet-snippets
    

인스턴스 만들기

Spanner를 처음 사용할 때는 인스턴스를 만들어야 합니다. 이 인스턴스는 Spanner 데이터베이스에서 사용하는 리소스를 할당한 것입니다. 인스턴스를 만들 때는 인스턴스 구성을 선택합니다. 이 구성에 따라 데이터 저장 위치와 사용할 노드 수가 결정되고, 또한 노드 수에 따라 인스턴스의 제공 리소스 및 스토리지 리소스 양이 결정됩니다.

us-central1 리전에 1개의 노드로 Spanner 인스턴스를 만들려면 다음 명령어를 실행합니다.

gcloud spanner instances create test-instance --config=regional-us-central1 \
    --description="Test Instance" --nodes=1

그러면 다음과 같은 특성을 가진 인스턴스가 생성됩니다.

  • 인스턴스 ID: test-instance
  • 표시 이름: Test Instance
  • 인스턴스 구성: regional-us-central1. 리전별 구성은 한 리전에 데이터를 저장하는 반면 멀티 리전 구성은 여러 리전에 데이터를 분산시킵니다. 자세한 내용은 인스턴스 정보를 참조하세요.
  • 노드 수: 1개. node_count에 따라 인스턴스의 데이터베이스에서 사용할 수 있는 제공 리소스 및 스토리지 리소스의 양이 달라집니다. 노드 및 처리 단위에서 자세히 알아보세요.

다음과 같이 표시됩니다.

Creating instance...done.

샘플 파일 살펴보기

샘플 저장소에는 PGAdapter와 함께 Spanner를 사용하는 방법을 보여주는 샘플이 있습니다.

Spanner 사용 방법을 보여주는 samples/snippets 폴더를 살펴보세요. 코드는 새 데이터베이스를 만들고 사용하는 방법을 보여줍니다. 데이터는 스키마 및 데이터 모델 페이지에 나와 있는 스키마 예시를 사용합니다.

PGAdapter 시작

로컬 개발 머신에서 PGAdapter를 시작하고 생성한 인스턴스를 가리킵니다.

다음 명령어는 gcloud auth application-default login을 실행한 것으로 가정합니다.

Java 애플리케이션

wget https://storage.googleapis.com/pgadapter-jar-releases/pgadapter.tar.gz \
    && tar -xzvf pgadapter.tar.gz
java -jar pgadapter.jar -i test-instance

Docker

docker pull gcr.io/cloud-spanner-pg-adapter/pgadapter
docker run \
    --name pgadapter \
    --rm -d -p 5432:5432 \
    -v "$HOME/.config/gcloud":/gcloud:ro \
    --env CLOUDSDK_CONFIG=/gcloud \
    gcr.io/cloud-spanner-pg-adapter/pgadapter \
    -i test-instance -x

에뮬레이터

docker pull gcr.io/cloud-spanner-pg-adapter/pgadapter-emulator
docker run \
    --name pgadapter-emulator \
    --rm -d \
    -p 5432:5432 \
    -p 9010:9010 \
    -p 9020:9020 \
    gcr.io/cloud-spanner-pg-adapter/pgadapter-emulator

그러면 임베디드 Spanner 에뮬레이터로 PGAdapter가 시작됩니다. 이 임베디드 에뮬레이터는 연결할 Spanner 인스턴스 또는 데이터베이스를 자동으로 생성하므로 미리 수동으로 만들 필요가 없습니다.

프로덕션에서는 사이드카 컨테이너 또는 프로세스 내 종속 항목으로 PGAdapter를 실행하는 것이 좋습니다. 프로덕션에서 PGAdapter를 배포하는 방법에 대한 자세한 내용은 PGAdapter 실행 메서드 선택을 참조하세요.

데이터베이스 만들기

명령줄에서 다음을 실행하여 test-instance라는 인스턴스에 example-db라는 데이터베이스를 만듭니다.

gcloud spanner databases create example-db --instance=test-instance \
    --database-dialect=POSTGRESQL

다음과 같이 표시됩니다.

Creating database...done.

테이블 만들기

다음 코드는 데이터베이스에 테이블 두 개를 만듭니다.

psql

#!/bin/bash

# Set the connection variables for psql.
# The following statements use the existing value of the variable if it has
# already been set, and otherwise assigns a default value.
export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

# Create two tables in one batch.
psql << SQL
-- Create the singers table
CREATE TABLE singers (
  singer_id   bigint not null primary key,
  first_name  character varying(1024),
  last_name   character varying(1024),
  singer_info bytea,
  full_name   character varying(2048) GENERATED ALWAYS
          AS (first_name || ' ' || last_name) STORED
);

-- Create the albums table. This table is interleaved in the parent table
-- "singers".
CREATE TABLE albums (
  singer_id     bigint not null,
  album_id      bigint not null,
  album_title   character varying(1024),
  primary key (singer_id, album_id)
)
-- The 'interleave in parent' clause is a Spanner-specific extension to
-- open-source PostgreSQL.
INTERLEAVE IN PARENT singers ON DELETE CASCADE;
SQL

echo "Created Singers & Albums tables in database: [${PGDATABASE}]"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.SQLException;
import java.sql.Statement;

class CreateTables {
  static void createTables(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      try (Statement statement = connection.createStatement()) {
        // Create two tables in one batch.
        statement.addBatch(
            "create table singers ("
                + "  singer_id   bigint primary key not null,"
                + "  first_name  varchar(1024),"
                + "  last_name   varchar(1024),"
                + "  singer_info bytea,"
                + "  full_name   varchar(2048) generated always as (\n"
                + "      case when first_name is null then last_name\n"
                + "          when last_name  is null then first_name\n"
                + "          else first_name || ' ' || last_name\n"
                + "      end) stored"
                + ")");
        statement.addBatch(
            "create table albums ("
                + "  singer_id     bigint not null,"
                + "  album_id      bigint not null,"
                + "  album_title   varchar,"
                + "  primary key (singer_id, album_id)"
                + ") interleave in parent singers on delete cascade");
        statement.executeBatch();
        System.out.println("Created Singers & Albums tables in database: [" + database + "]");
      }
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func CreateTables(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	// Create two tables in one batch on Spanner.
	br := conn.SendBatch(ctx, &pgx.Batch{QueuedQueries: []*pgx.QueuedQuery{
		{SQL: "create table singers (" +
			"  singer_id   bigint primary key not null," +
			"  first_name  character varying(1024)," +
			"  last_name   character varying(1024)," +
			"  singer_info bytea," +
			"  full_name   character varying(2048) generated " +
			"  always as (first_name || ' ' || last_name) stored" +
			")"},
		{SQL: "create table albums (" +
			"  singer_id     bigint not null," +
			"  album_id      bigint not null," +
			"  album_title   character varying(1024)," +
			"  primary key (singer_id, album_id)" +
			") interleave in parent singers on delete cascade"},
	}})
	cmd, err := br.Exec()
	if err != nil {
		return err
	}
	if cmd.String() != "CREATE" {
		return fmt.Errorf("unexpected command tag: %v", cmd.String())
	}
	if err := br.Close(); err != nil {
		return err
	}
	fmt.Printf("Created Singers & Albums tables in database: [%s]\n", database)

	return nil
}

Node.js

import { Client } from 'pg';

async function createTables(host: string, port: number, database: string): Promise<void> {
  // Connect to Spanner through PGAdapter.
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  // Create two tables in one batch.
  await connection.query("start batch ddl");
  await connection.query("create table singers (" +
      "  singer_id   bigint primary key not null," +
      "  first_name  character varying(1024)," +
      "  last_name   character varying(1024)," +
      "  singer_info bytea," +
      "  full_name   character varying(2048) generated " +
      "  always as (first_name || ' ' || last_name) stored" +
      ")");
  await connection.query("create table albums (" +
      "  singer_id     bigint not null," +
      "  album_id      bigint not null," +
      "  album_title   character varying(1024)," +
      "  primary key (singer_id, album_id)" +
      ") interleave in parent singers on delete cascade");
  await connection.query("run batch");
  console.log(`Created Singers & Albums tables in database: [${database}]`);

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def create_tables(host: string, port: int, database: string):
    # Connect to Cloud Spanner using psycopg3 through PGAdapter.
    with psycopg.connect("host={host} port={port} "
                         "dbname={database} "
                         "sslmode=disable".format(host=host, port=port,
                                                  database=database)) as conn:
        # Enable autocommit to execute DDL statements, as psycopg otherwise
        # tries to use a read/write transaction.
        conn.autocommit = True

        # Use a pipeline to execute multiple DDL statements in one batch.
        with conn.pipeline():
            conn.execute("create table singers ("
                         + "  singer_id   bigint primary key not null,"
                         + "  first_name  character varying(1024),"
                         + "  last_name   character varying(1024),"
                         + "  singer_info bytea,"
                         + "  full_name   character varying(2048) generated "
                         + "  always as (first_name || ' ' || last_name) stored"
                         + ")")
            conn.execute("create table albums ("
                         + "  singer_id     bigint not null,"
                         + "  album_id      bigint not null,"
                         + "  album_title   character varying(1024),"
                         + "  primary key (singer_id, album_id)"
                         + ") interleave in parent singers on delete cascade")
        print("Created Singers & Albums tables in database: [{database}]"
              .format(database=database))

C#

using Npgsql;

namespace dotnet_snippets;

public static class CreateTablesSample
{
    public static void CreateTables(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        // Create two tables in one batch.
        var batch = connection.CreateBatch();
        batch.BatchCommands.Add(new NpgsqlBatchCommand(
            "create table singers ("
            + "  singer_id   bigint primary key not null,"
            + "  first_name  varchar(1024),"
            + "  last_name   varchar(1024),"
            + "  singer_info bytea,"
            + "  full_name   varchar(2048) generated always as (\n"
            + "      case when first_name is null then last_name\n"
            + "          when last_name  is null then first_name\n"
            + "          else first_name || ' ' || last_name\n"
            + "      end) stored"
            + ")"));
        batch.BatchCommands.Add(new NpgsqlBatchCommand(
            "create table albums ("
            + "  singer_id     bigint not null,"
            + "  album_id      bigint not null,"
            + "  album_title   varchar,"
            + "  primary key (singer_id, album_id)"
            + ") interleave in parent singers on delete cascade"));
        batch.ExecuteNonQuery();
        Console.WriteLine($"Created Singers & Albums tables in database: [{database}]");
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./create_tables.sh example-db

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar createtables example-db

Go

go run sample_runner.go createtables example-db

Node.js

npm start createtables example-db

Python

python create_tables.py example-db

C#

dotnet run createtables example-db

다음 단계는 데이터베이스에 데이터 쓰기입니다.

연결 만들기

읽기 또는 쓰기를 수행하려면 먼저 PGAdapter에 연결을 만들어야 합니다. Spanner와의 모든 상호작용은 Connection을 거쳐야 합니다. 데이터베이스 이름은 연결 문자열에 지정됩니다.

psql

#!/bin/bash

# Set the connection variables for psql.
# The following statements use the existing value of the variable if it has
# already been set, and otherwise assigns a default value.
export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

# Connect to Cloud Spanner using psql through PGAdapter
# and execute a simple query.
psql -c "select 'Hello world!' as hello"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.ResultSet;
import java.sql.SQLException;

class CreateConnection {
  static void createConnection(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      try (ResultSet resultSet =
          connection.createStatement().executeQuery("select 'Hello world!' as hello")) {
        while (resultSet.next()) {
          System.out.printf("Greeting from Cloud Spanner PostgreSQL: %s\n", resultSet.getString(1));
        }
      }
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func CreateConnection(host string, port int, database string) error {
	ctx := context.Background()
	// Connect to Cloud Spanner using pgx through PGAdapter.
	// 'sslmode=disable' is optional, but adding it reduces the connection time,
	// as pgx will then skip first trying to create an SSL connection.
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	row := conn.QueryRow(ctx, "select 'Hello world!' as hello")
	var msg string
	if err := row.Scan(&msg); err != nil {
		return err
	}
	fmt.Printf("Greeting from Cloud Spanner PostgreSQL: %s\n", msg)

	return nil
}

Node.js

import { Client } from 'pg';

async function createConnection(host: string, port: number, database: string): Promise<void> {
  // Connect to Spanner through PGAdapter.
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  const result = await connection.query("select 'Hello world!' as hello");
  console.log(`Greeting from Cloud Spanner PostgreSQL: ${result.rows[0]['hello']}`);

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def create_connection(host: string, port: int, database: string):
    # Connect to Cloud Spanner using psycopg3 through PGAdapter.
    # 'sslmode=disable' is optional, but adding it reduces the connection time,
    # as psycopg3 will then skip first trying to create an SSL connection.
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        conn.autocommit = True
        with conn.cursor() as cur:
            cur.execute("select 'Hello world!' as hello")
            print("Greeting from Cloud Spanner PostgreSQL:", cur.fetchone()[0])

C#

using Npgsql;

namespace dotnet_snippets;

public static class CreateConnectionSample
{
    public static void CreateConnection(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        using var cmd = new NpgsqlCommand("select 'Hello World!' as hello", connection);
        using var reader = cmd.ExecuteReader();
        while (reader.Read())
        {
            var greeting = reader.GetString(0);
            Console.WriteLine($"Greeting from Cloud Spanner PostgreSQL: {greeting}");
        }
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./create_connection.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar createconnection example-db

Go

go run sample_runner.go createconnection example-db

Node.js

npm start createconnection example-db

Python

python create_connection.py example-db

C#

dotnet run createconnection example-db

DML을 사용하여 데이터 쓰기

읽기-쓰기 트랜잭션에서 데이터 조작 언어(DML)를 사용하여 데이터를 삽입할 수 있습니다.

이 샘플은 PostgreSQL 드라이버를 사용하여 Spanner에서 DML 문을 실행하는 방법을 보여줍니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

psql -c "INSERT INTO singers (singer_id, first_name, last_name) VALUES
                             (12, 'Melissa', 'Garcia'),
                             (13, 'Russel', 'Morales'),
                             (14, 'Jacqueline', 'Long'),
                             (15, 'Dylan', 'Shaw')"

echo "4 records inserted"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.PreparedStatement;
import java.sql.SQLException;
import java.util.Arrays;
import java.util.List;

class WriteDataWithDml {
  static class Singer {
    private final long singerId;
    private final String firstName;
    private final String lastName;

    Singer(final long id, final String first, final String last) {
      this.singerId = id;
      this.firstName = first;
      this.lastName = last;
    }
  }

  static void writeDataWithDml(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      // Add 4 rows in one statement.
      // JDBC always uses '?' as a parameter placeholder.
      try (PreparedStatement preparedStatement =
          connection.prepareStatement(
              "INSERT INTO singers (singer_id, first_name, last_name) VALUES "
                  + "(?, ?, ?), "
                  + "(?, ?, ?), "
                  + "(?, ?, ?), "
                  + "(?, ?, ?)")) {

        final List<Singer> singers =
            Arrays.asList(
                new Singer(/* SingerId = */ 12L, "Melissa", "Garcia"),
                new Singer(/* SingerId = */ 13L, "Russel", "Morales"),
                new Singer(/* SingerId = */ 14L, "Jacqueline", "Long"),
                new Singer(/* SingerId = */ 15L, "Dylan", "Shaw"));

        // Note that JDBC parameters start at index 1.
        int paramIndex = 0;
        for (Singer singer : singers) {
          preparedStatement.setLong(++paramIndex, singer.singerId);
          preparedStatement.setString(++paramIndex, singer.firstName);
          preparedStatement.setString(++paramIndex, singer.lastName);
        }

        int updateCount = preparedStatement.executeUpdate();
        System.out.printf("%d records inserted.\n", updateCount);
      }
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func WriteDataWithDml(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	tag, err := conn.Exec(ctx,
		"INSERT INTO singers (singer_id, first_name, last_name) "+
			"VALUES ($1, $2, $3), ($4, $5, $6), "+
			"       ($7, $8, $9), ($10, $11, $12)",
		12, "Melissa", "Garcia",
		13, "Russel", "Morales",
		14, "Jacqueline", "Long",
		15, "Dylan", "Shaw")
	if err != nil {
		return err
	}
	fmt.Printf("%v records inserted\n", tag.RowsAffected())

	return nil
}

Node.js

import { Client } from 'pg';

async function writeDataWithDml(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  const result = await connection.query("INSERT INTO singers (singer_id, first_name, last_name) " +
      "VALUES ($1, $2, $3), ($4, $5, $6), " +
      "       ($7, $8, $9), ($10, $11, $12)",
       [12, "Melissa", "Garcia",
        13, "Russel", "Morales",
        14, "Jacqueline", "Long",
        15, "Dylan", "Shaw"])
  console.log(`${result.rowCount} records inserted`);

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def write_data_with_dml(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        conn.autocommit = True
        with conn.cursor() as cur:
            cur.execute("INSERT INTO singers (singer_id, first_name, last_name)"
                        " VALUES (%s, %s, %s), (%s, %s, %s), "
                        "        (%s, %s, %s), (%s, %s, %s)",
                        (12, "Melissa", "Garcia",
                         13, "Russel", "Morales",
                         14, "Jacqueline", "Long",
                         15, "Dylan", "Shaw",))
            print("%d records inserted" % cur.rowcount)

C#

using Npgsql;

namespace dotnet_snippets;

public static class WriteDataWithDmlSample
{
    readonly struct Singer
    {
        public Singer(long singerId, string firstName, string lastName)
        {
            SingerId = singerId;
            FirstName = firstName;
            LastName = lastName;
        }

        public long SingerId { get; }
        public string FirstName { get; }
        public string LastName { get; }
    }

    public static void WriteDataWithDml(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();
        // Add 4 rows in one statement.
        using var cmd = new NpgsqlCommand("INSERT INTO singers (singer_id, first_name, last_name) VALUES "
                                          + "($1, $2, $3), "
                                          + "($4, $5, $6), "
                                          + "($7, $8, $9), "
                                          + "($10, $11, $12)", connection);
        List<Singer> singers =
        [
            new Singer(/* SingerId = */ 12L, "Melissa", "Garcia"),
            new Singer(/* SingerId = */ 13L, "Russel", "Morales"),
            new Singer(/* SingerId = */ 14L, "Jacqueline", "Long"),
            new Singer(/* SingerId = */ 15L, "Dylan", "Shaw")
        ];
        foreach (var singer in singers)
        {
            cmd.Parameters.Add(new NpgsqlParameter { Value = singer.SingerId });
            cmd.Parameters.Add(new NpgsqlParameter { Value = singer.FirstName });
            cmd.Parameters.Add(new NpgsqlParameter { Value = singer.LastName });
        }
        var updateCount = cmd.ExecuteNonQuery();
        Console.WriteLine($"{updateCount} records inserted.");
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./write_data_with_dml.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar writeusingdml example-db

Go

go run sample_runner.go writeusingdml example-db

Node.js

npm start writeusingdml example-db

Python

python write_data_with_dml.py example-db

C#

dotnet run writeusingdml example-db

다음과 같은 응답이 표시됩니다.

 4 records inserted.

DML 배치로 데이터 쓰기

PGAdapter는 DML 일괄 실행을 지원합니다. 한 번의 일괄 작업으로 여러 DML 문을 보내면 Spanner로의 왕복 횟수가 줄어들고 애플리케이션 성능이 향상됩니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

# Create a prepared insert statement and execute this prepared
# insert statement three times in one SQL string. The single
# SQL string with three insert statements will be executed as
# a single DML batch on Spanner.
psql -c "PREPARE insert_singer AS
           INSERT INTO singers (singer_id, first_name, last_name)
           VALUES (\$1, \$2, \$3)" \
     -c "EXECUTE insert_singer (16, 'Sarah', 'Wilson');
         EXECUTE insert_singer (17, 'Ethan', 'Miller');
         EXECUTE insert_singer (18, 'Maya', 'Patel');"

echo "3 records inserted"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.PreparedStatement;
import java.sql.SQLException;
import java.util.Arrays;
import java.util.List;

class WriteDataWithDmlBatch {
  static class Singer {
    private final long singerId;
    private final String firstName;
    private final String lastName;

    Singer(final long id, final String first, final String last) {
      this.singerId = id;
      this.firstName = first;
      this.lastName = last;
    }
  }

  static void writeDataWithDmlBatch(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      // Add multiple rows in one DML batch.
      // JDBC always uses '?' as a parameter placeholder.
      try (PreparedStatement preparedStatement =
          connection.prepareStatement(
              "INSERT INTO singers (singer_id, first_name, last_name) VALUES (?, ?, ?)")) {
        final List<Singer> singers =
            Arrays.asList(
                new Singer(/* SingerId = */ 16L, "Sarah", "Wilson"),
                new Singer(/* SingerId = */ 17L, "Ethan", "Miller"),
                new Singer(/* SingerId = */ 18L, "Maya", "Patel"));

        for (Singer singer : singers) {
          // Note that JDBC parameters start at index 1.
          int paramIndex = 0;
          preparedStatement.setLong(++paramIndex, singer.singerId);
          preparedStatement.setString(++paramIndex, singer.firstName);
          preparedStatement.setString(++paramIndex, singer.lastName);
          preparedStatement.addBatch();
        }

        int[] updateCounts = preparedStatement.executeBatch();
        System.out.printf("%d records inserted.\n", Arrays.stream(updateCounts).sum());
      }
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func WriteDataWithDmlBatch(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	sql := "INSERT INTO singers (singer_id, first_name, last_name) " +
		"VALUES ($1, $2, $3)"
	batch := &pgx.Batch{}
	batch.Queue(sql, 16, "Sarah", "Wilson")
	batch.Queue(sql, 17, "Ethan", "Miller")
	batch.Queue(sql, 18, "Maya", "Patel")
	br := conn.SendBatch(ctx, batch)
	_, err = br.Exec()
	if err := br.Close(); err != nil {
		return err
	}

	if err != nil {
		return err
	}
	fmt.Printf("%v records inserted\n", batch.Len())

	return nil
}

Node.js

import { Client } from 'pg';

async function writeDataWithDmlBatch(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  // node-postgres does not support PostgreSQL pipeline mode, so we must use the
  // `start batch dml` / `run batch` statements to execute a DML batch.
  const sql = "INSERT INTO singers (singer_id, first_name, last_name) VALUES ($1, $2, $3)";
  await connection.query("start batch dml");
  await connection.query(sql, [16, "Sarah", "Wilson"]);
  await connection.query(sql, [17, "Ethan", "Miller"]);
  await connection.query(sql, [18, "Maya", "Patel"]);
  const result = await connection.query("run batch");
  // RUN BATCH returns the update counts as an array of strings, with one element for each
  // DML statement in the batch. This calculates the total number of affected rows from that array.
  const updateCount = result.rows[0]["UPDATE_COUNTS"]
      .map((s: string) => parseInt(s))
      .reduce((c: number, current: number) => c + current, 0);
  console.log(`${updateCount} records inserted`);

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def write_data_with_dml_batch(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        conn.autocommit = True
        with conn.cursor() as cur:
            cur.executemany("INSERT INTO singers "
                            "(singer_id, first_name, last_name) "
                            "VALUES (%s, %s, %s)",
                            [(16, "Sarah", "Wilson",),
                             (17, "Ethan", "Miller",),
                             (18, "Maya", "Patel",), ])
            print("%d records inserted" % cur.rowcount)

C#

using Npgsql;

namespace dotnet_snippets;

public static class WriteDataWithDmlBatchSample
{
    readonly struct Singer
    {
        public Singer(long singerId, string firstName, string lastName)
        {
            SingerId = singerId;
            FirstName = firstName;
            LastName = lastName;
        }

        public long SingerId { get; }
        public string FirstName { get; }
        public string LastName { get; }
    }

    public static void WriteDataWithDmlBatch(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        // Add multiple rows in one DML batch.
        const string sql = "INSERT INTO singers (singer_id, first_name, last_name) VALUES ($1, $2, $3)";
        List<Singer> singers =
        [
            new Singer(/* SingerId = */ 16L, "Sarah", "Wilson"),
            new Singer(/* SingerId = */ 17L, "Ethan", "Miller"),
            new Singer(/* SingerId = */ 18L, "Maya", "Patel")
        ];
        using var batch = new NpgsqlBatch(connection);
        foreach (var singer in singers)
        {
            batch.BatchCommands.Add(new NpgsqlBatchCommand
            {
                CommandText = sql,
                Parameters =
                {
                    new NpgsqlParameter {Value = singer.SingerId},
                    new NpgsqlParameter {Value = singer.FirstName},
                    new NpgsqlParameter {Value = singer.LastName}
                }
            });
        }
        var updateCount = batch.ExecuteNonQuery();
        Console.WriteLine($"{updateCount} records inserted.");
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./write_data_with_dml_batch.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar writeusingdmlbatch example-db

Go

go run sample_runner.go writeusingdmlbatch example-db

Node.js

npm start writeusingdmlbatch example-db

Python

python write_data_with_dml_batch.py example-db

C#

dotnet run writeusingdmlbatch example-db

다음과 같이 표시됩니다.

3 records inserted.

변형을 사용하여 데이터 쓰기

변형을 사용하여 데이터를 삽입할 수도 있습니다.

PGAdapter는 PostgreSQL COPY 명령어를 변형으로 변환합니다. COPY를 사용하는 것이 Spanner 데이터베이스에 데이터를 빠르게 삽입하는 가장 효율적인 방법입니다.

COPY 작업은 기본적으로 원자적입니다. Spanner의 원자적 작업은 커밋 크기 한도에 의해 제한됩니다. 자세한 내용은 CRUD 한도를 참조하세요.

다음 예는 비원자 COPY 작업을 실행하는 방법을 보여줍니다. 이를 통해 COPY 작업이 커밋 크기 제한을 초과할 수 있습니다.

psql

#!/bin/bash

# Get the source directory of this script.
directory=${BASH_SOURCE%/*}/

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

# Copy data to Spanner from a tab-separated text file using the COPY command.
psql -c "COPY singers (singer_id, first_name, last_name) FROM STDIN" \
  < "${directory}singers_data.txt"
psql -c "COPY albums FROM STDIN" \
  < "${directory}albums_data.txt"

echo "Copied singers and albums"

자바

import java.io.IOException;
import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.SQLException;
import org.postgresql.PGConnection;
import org.postgresql.copy.CopyManager;

class WriteDataWithCopy {

  static void writeDataWithCopy(String host, int port, String database)
      throws SQLException, IOException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      // Unwrap the PostgreSQL JDBC connection interface to get access to
      // a CopyManager.
      PGConnection pgConnection = connection.unwrap(PGConnection.class);
      CopyManager copyManager = pgConnection.getCopyAPI();

      // Enable 'partitioned_non_atomic' mode. This ensures that the COPY operation
      // will succeed even if it exceeds Spanner's mutation limit per transaction.
      connection
          .createStatement()
          .execute("set spanner.autocommit_dml_mode='partitioned_non_atomic'");
      long numSingers =
          copyManager.copyIn(
              "COPY singers (singer_id, first_name, last_name) FROM STDIN",
              WriteDataWithCopy.class.getResourceAsStream("singers_data.txt"));
      System.out.printf("Copied %d singers\n", numSingers);

      long numAlbums =
          copyManager.copyIn(
              "COPY albums FROM STDIN",
              WriteDataWithCopy.class.getResourceAsStream("albums_data.txt"));
      System.out.printf("Copied %d albums\n", numAlbums);
    }
  }
}

Go

import (
	"context"
	"fmt"
	"os"

	"github.com/jackc/pgx/v5"
)

func WriteDataWithCopy(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	// Enable 'partitioned_non_atomic' mode. This ensures that the COPY operation
	// will succeed even if it exceeds Spanner's mutation limit per transaction.
	conn.Exec(ctx, "set spanner.autocommit_dml_mode='partitioned_non_atomic'")

	file, err := os.Open("samples/singers_data.txt")
	if err != nil {
		return err
	}
	tag, err := conn.PgConn().CopyFrom(ctx, file,
		"copy singers (singer_id, first_name, last_name) from stdin")
	if err != nil {
		return err
	}
	fmt.Printf("Copied %v singers\n", tag.RowsAffected())

	file, err = os.Open("samples/albums_data.txt")
	if err != nil {
		return err
	}
	tag, err = conn.PgConn().CopyFrom(ctx, file,
		"copy albums from stdin")
	if err != nil {
		return err
	}
	fmt.Printf("Copied %v albums\n", tag.RowsAffected())

	return nil
}

Node.js

import { Client } from 'pg';
import { pipeline } from 'node:stream/promises'
import fs from 'node:fs'
import { from as copyFrom } from 'pg-copy-streams'
import path from "path";

async function writeDataWithCopy(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  // Enable 'partitioned_non_atomic' mode. This ensures that the COPY operation
  // will succeed even if it exceeds Spanner's mutation limit per transaction.
  await connection.query("set spanner.autocommit_dml_mode='partitioned_non_atomic'");
  // Copy data from a csv file to Spanner using the COPY command.
  // Note that even though the command says 'from stdin', the actual input comes from a file.
  const copySingersStream = copyFrom('copy singers (singer_id, first_name, last_name) from stdin');
  const ingestSingersStream = connection.query(copySingersStream);
  const sourceSingersStream = fs.createReadStream(path.join(__dirname, 'singers_data.txt'));
  await pipeline(sourceSingersStream, ingestSingersStream);
  console.log(`Copied ${copySingersStream.rowCount} singers`);

  const copyAlbumsStream = copyFrom('copy albums from stdin');
  const ingestAlbumsStream = connection.query(copyAlbumsStream);
  const sourceAlbumsStream = fs.createReadStream(path.join(__dirname, 'albums_data.txt'));
  await pipeline(sourceAlbumsStream, ingestAlbumsStream);
  console.log(`Copied ${copyAlbumsStream.rowCount} albums`);

  // Close the connection.
  await connection.end();
}

Python

import os
import string
import psycopg


def write_data_with_copy(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:

        script_dir = os.path.dirname(os.path.abspath(__file__))
        singers_file_path = os.path.join(script_dir, "singers_data.txt")
        albums_file_path = os.path.join(script_dir, "albums_data.txt")

        conn.autocommit = True
        block_size = 1024
        with conn.cursor() as cur:
            with open(singers_file_path, "r") as f:
                with cur.copy("COPY singers (singer_id, first_name, last_name) "
                              "FROM STDIN") as copy:
                    while data := f.read(block_size):
                        copy.write(data)
            print("Copied %d singers" % cur.rowcount)

            with open(albums_file_path, "r") as f:
                with cur.copy("COPY albums "
                              "FROM STDIN") as copy:
                    while data := f.read(block_size):
                        copy.write(data)
            print("Copied %d albums" % cur.rowcount)

C#

using Npgsql;

namespace dotnet_snippets;

public static class WriteDataWithCopySample
{
    public static void WriteDataWithCopy(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        // Enable 'partitioned_non_atomic' mode. This ensures that the COPY operation
        // will succeed even if it exceeds Spanner's mutation limit per transaction.
        using var cmd = new NpgsqlCommand("set spanner.autocommit_dml_mode='partitioned_non_atomic'", connection);
        cmd.ExecuteNonQuery();

        var singerCount = 0;
        using var singerReader = new StreamReader("singers_data.txt");
        using (var singerWriter = connection.BeginTextImport("COPY singers (singer_id, first_name, last_name) FROM STDIN"))
        {
            while (singerReader.ReadLine() is { } line)
            {
                singerWriter.WriteLine(line);
                singerCount++;
            }
        }
        Console.WriteLine($"Copied {singerCount} singers");

        var albumCount = 0;
        using var albumReader = new StreamReader("albums_data.txt");
        using (var albumWriter = connection.BeginTextImport("COPY albums FROM STDIN"))
        {
            while (albumReader.ReadLine() is { } line)
            {
                albumWriter.WriteLine(line);
                albumCount++;
            }
        }
        Console.WriteLine($"Copied {albumCount} albums");
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./write_data_with_copy.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar write example-db

Go

go run sample_runner.go write example-db

Node.js

npm start write example-db

Python

python write_data_with_copy.py example-db

C#

dotnet run write example-db

다음과 같이 표시됩니다.

Copied 5 singers
Copied 5 albums

SQL을 사용하여 데이터 쿼리

Spanner는 데이터 읽기용 SQL 인터페이스를 지원하며, 개발자는 Google Cloud CLI를 사용하여 명령줄에서 액세스하거나 PostgreSQL 드라이버를 사용하여 프로그래매틱 방식으로 액세스할 수 있습니다.

명령줄에서

다음 SQL 문을 실행하여 Albums 테이블에서 모든 열의 값을 읽습니다.

gcloud spanner databases execute-sql example-db --instance=test-instance \
    --sql='SELECT singer_id, album_id, album_title FROM albums'

결과가 다음과 같이 표시됩니다.

SingerId AlbumId AlbumTitle
1        1       Total Junk
1        2       Go, Go, Go
2        1       Green
2        2       Forever Hold Your Peace
2        3       Terrified

PostgreSQL 드라이버 사용

명령줄에서 SQL 문을 실행하는 방법 외에 PostgreSQL 드라이버를 사용하여 프로그래매틱 방식으로 SQL 문을 실행하는 방법도 있습니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

psql -c "SELECT singer_id, album_id, album_title
         FROM albums"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.ResultSet;
import java.sql.SQLException;

class QueryData {
  static void queryData(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      try (ResultSet resultSet =
          connection
              .createStatement()
              .executeQuery("SELECT singer_id, album_id, album_title FROM albums")) {
        while (resultSet.next()) {
          System.out.printf(
              "%d %d %s\n",
              resultSet.getLong("singer_id"),
              resultSet.getLong("album_id"),
              resultSet.getString("album_title"));
        }
      }
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func QueryData(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	rows, err := conn.Query(ctx, "SELECT singer_id, album_id, album_title "+
		"FROM albums")
	defer rows.Close()
	if err != nil {
		return err
	}
	for rows.Next() {
		var singerId, albumId int64
		var title string
		err = rows.Scan(&singerId, &albumId, &title)
		if err != nil {
			return err
		}
		fmt.Printf("%v %v %v\n", singerId, albumId, title)
	}

	return rows.Err()
}

Node.js

import { Client } from 'pg';

async function queryData(host: string, port: number, database: string): Promise<void> {
  // Connect to Spanner through PGAdapter.
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  const result = await connection.query("SELECT singer_id, album_id, album_title " +
      "FROM albums");
  for (const row of result.rows) {
    console.log(`${row["singer_id"]} ${row["album_id"]} ${row["album_title"]}`);
  }

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def query_data(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        conn.autocommit = True
        with conn.cursor() as cur:
            cur.execute("SELECT singer_id, album_id, album_title "
                        "FROM albums")
            for album in cur:
                print(album)

C#

using Npgsql;

namespace dotnet_snippets;

public static class QueryDataSample
{
    public static void QueryData(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        using var cmd = new NpgsqlCommand("SELECT singer_id, album_id, album_title FROM albums", connection);
        using var reader = cmd.ExecuteReader();
        while (reader.Read())
        {
            Console.WriteLine($"{reader.GetInt64(0)} {reader.GetInt64(1)} {reader.GetString(2)}");
        }
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./query_data.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar query example-db

Go

go run sample_runner.go query example-db

Node.js

npm start query example-db

Python

python query_data.py example-db

C#

dotnet run query example-db

결과:

1 1 Total Junk
1 2 Go, Go, Go
2 1 Green
2 2 Forever Hold Your Peace
2 3 Terrified

SQL 매개변수를 사용하여 쿼리

애플리케이션에 자주 실행되는 쿼리가 있는 경우 이를 매개변수화하여 성능을 개선할 수 있습니다. 매개변수화된 결과 쿼리를 캐시하고 다시 사용할 수 있으므로 컴파일 비용이 절감됩니다. 자세한 내용은 쿼리 매개변수를 사용하여 자주 실행되는 쿼리 속도 향상을 참조하세요.

다음 예시에서는 WHERE 절의 매개변수를 사용하여 LastName의 특정 값이 포함된 레코드를 쿼리합니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

# Create a prepared statement to use a query parameter.
# Using a prepared statement for executing the same SQL string multiple
# times increases the execution speed of the statement.
psql -c "PREPARE select_singer AS
         SELECT singer_id, first_name, last_name
         FROM singers
         WHERE last_name = \$1" \
     -c "EXECUTE select_singer ('Garcia')"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.PreparedStatement;
import java.sql.ResultSet;
import java.sql.SQLException;

class QueryDataWithParameter {
  static void queryDataWithParameter(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      try (PreparedStatement statement =
          connection.prepareStatement(
              "SELECT singer_id, first_name, last_name "
                  + "FROM singers "
                  + "WHERE last_name = ?")) {
        statement.setString(1, "Garcia");
        try (ResultSet resultSet = statement.executeQuery()) {
          while (resultSet.next()) {
            System.out.printf(
                "%d %s %s\n",
                resultSet.getLong("singer_id"),
                resultSet.getString("first_name"),
                resultSet.getString("last_name"));
          }
        }
      }
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func QueryDataWithParameter(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	rows, err := conn.Query(ctx,
		"SELECT singer_id, first_name, last_name "+
			"FROM singers "+
			"WHERE last_name = $1", "Garcia")
	defer rows.Close()
	if err != nil {
		return err
	}
	for rows.Next() {
		var singerId int64
		var firstName, lastName string
		err = rows.Scan(&singerId, &firstName, &lastName)
		if err != nil {
			return err
		}
		fmt.Printf("%v %v %v\n", singerId, firstName, lastName)
	}

	return rows.Err()
}

Node.js

import { Client } from 'pg';

async function queryWithParameter(host: string, port: number, database: string): Promise<void> {
  // Connect to Spanner through PGAdapter.
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  const result = await connection.query(
      "SELECT singer_id, first_name, last_name " +
      "FROM singers " +
      "WHERE last_name = $1", ["Garcia"]);
  for (const row of result.rows) {
    console.log(`${row["singer_id"]} ${row["first_name"]} ${row["last_name"]}`);
  }

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def query_data_with_parameter(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        conn.autocommit = True
        with conn.cursor() as cur:
            cur.execute("SELECT singer_id, first_name, last_name "
                        "FROM singers "
                        "WHERE last_name = %s", ("Garcia",))
            for singer in cur:
                print(singer)

C#

using Npgsql;

namespace dotnet_snippets;

public static class QueryDataWithParameterSample
{
    public static void QueryDataWithParameter(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        using var cmd = new NpgsqlCommand("SELECT singer_id, first_name, last_name "
                                          + "FROM singers "
                                          + "WHERE last_name = $1", connection);
        cmd.Parameters.Add(new NpgsqlParameter { Value = "Garcia" });
        using var reader = cmd.ExecuteReader();
        while (reader.Read())
        {
            Console.WriteLine($"{reader["singer_id"]} {reader["first_name"]} {reader["last_name"]}");
        }
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./query_data_with_parameter.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar querywithparameter example-db

Go

go run sample_runner.go querywithparameter example-db

Node.js

npm start querywithparameter example-db

Python

python query_data_with_parameter.py example-db

C#

dotnet run querywithparameter example-db

결과:

12 Melissa Garcia

데이터베이스 스키마 업데이트

Albums 테이블에 MarketingBudget이라는 새 열을 추가해야 한다고 가정합니다. 기존 테이블에 새 열을 추가하려면 데이터베이스 스키마를 업데이트해야 합니다. Spanner는 데이터베이스에서 트래픽이 계속 처리되는 동안 데이터베이스의 스키마 업데이트를 지원합니다. 스키마 업데이트 시 데이터베이스를 오프라인으로 전환할 필요가 없고 전체 테이블 또는 열을 잠그지 않습니다. 스키마 업데이트 중에도 데이터베이스에 계속 데이터를 쓸 수 있습니다. 스키마 업데이트에서 지원되는 스키마 업데이트와 스키마 변경 성능에 대해 자세히 알아보세요.

열 추가

명령줄에서 Google Cloud CLI를 사용하거나 PostgreSQL 드라이버를 프로그래매틱 방식으로 사용하여 열을 추가할 수 있습니다.

명령줄에서

다음과 같은 ALTER TABLE 명령어를 사용하여 테이블에 새 열을 추가합니다.

gcloud spanner databases ddl update example-db --instance=test-instance \
    --ddl='ALTER TABLE albums ADD COLUMN marketing_budget BIGINT'

다음과 같이 표시됩니다.

Schema updating...done.

PostgreSQL 드라이버 사용

PostgreSQL 드라이버를 사용하여 DDL 문을 실행하여 스키마를 수정합니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

psql -c "ALTER TABLE albums ADD COLUMN marketing_budget bigint"
echo "Added marketing_budget column"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.SQLException;

class AddColumn {
  static void addColumn(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      connection.createStatement().execute("alter table albums add column marketing_budget bigint");
      System.out.println("Added marketing_budget column");
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func AddColumn(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	_, err = conn.Exec(ctx,
		"ALTER TABLE albums "+
			"ADD COLUMN marketing_budget bigint")
	if err != nil {
		return err
	}
	fmt.Println("Added marketing_budget column")

	return nil
}

Node.js

import { Client } from 'pg';

async function addColumn(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  await connection.query(
      "ALTER TABLE albums " +
      "ADD COLUMN marketing_budget bigint");
  console.log("Added marketing_budget column");

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def add_column(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        # DDL can only be executed when autocommit=True.
        conn.autocommit = True
        with conn.cursor() as cur:
            cur.execute("ALTER TABLE albums "
                        "ADD COLUMN marketing_budget bigint")
            print("Added marketing_budget column")

C#

using Npgsql;

namespace dotnet_snippets;

public static class AddColumnSample
{
    public static void AddColumn(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        using var cmd = connection.CreateCommand();
        cmd.CommandText = "alter table albums add column marketing_budget bigint";
        cmd.ExecuteNonQuery();
        Console.WriteLine("Added marketing_budget column");
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./add_column.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar addmarketingbudget example-db

Go

go run sample_runner.go addmarketingbudget example-db

Node.js

npm start addmarketingbudget example-db

Python

python add_column.py example-db

C#

dotnet run addmarketingbudget example-db

다음과 같이 표시됩니다.

Added marketing_budget column

DDL 일괄 실행

여러 스키마 수정을 일괄 실행하는 것이 좋습니다. PostgreSQL 드라이버의 기본 제공 일괄 처리 기능을 사용하거나, 모든 DDL 문을 세미콜론으로 구분된 하나의 SQL 문자열로 제출하거나, START BATCH DDLRUN BATCH 문을 사용하여 여러 DDL 문을 하나의 일괄 처리로 실행할 수 있습니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

# Use a single SQL command to batch multiple statements together.
# Executing multiple DDL statements as one batch is more efficient
# than executing each statement individually.
# Separate the statements with semicolons.
psql << SQL

CREATE TABLE venues (
  venue_id    bigint not null primary key,
  name        varchar(1024),
  description jsonb
);

CREATE TABLE concerts (
  concert_id bigint not null primary key ,
  venue_id   bigint not null,
  singer_id  bigint not null,
  start_time timestamptz,
  end_time   timestamptz,
  constraint fk_concerts_venues foreign key
    (venue_id) references venues (venue_id),
  constraint fk_concerts_singers foreign key
    (singer_id) references singers (singer_id)
);

SQL

echo "Added venues and concerts tables"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.SQLException;
import java.sql.Statement;

class DdlBatch {
  static void ddlBatch(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      try (Statement statement = connection.createStatement()) {
        // Create two new tables in one batch.
        statement.addBatch(
            "CREATE TABLE venues ("
                + "  venue_id    bigint not null primary key,"
                + "  name        varchar(1024),"
                + "  description jsonb"
                + ")");
        statement.addBatch(
            "CREATE TABLE concerts ("
                + "  concert_id bigint not null primary key ,"
                + "  venue_id   bigint not null,"
                + "  singer_id  bigint not null,"
                + "  start_time timestamptz,"
                + "  end_time   timestamptz,"
                + "  constraint fk_concerts_venues foreign key"
                + "    (venue_id) references venues (venue_id),"
                + "  constraint fk_concerts_singers foreign key"
                + "    (singer_id) references singers (singer_id)"
                + ")");
        statement.executeBatch();
      }
      System.out.println("Added venues and concerts tables");
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func DdlBatch(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	// Executing multiple DDL statements as one batch is
	// more efficient than executing each statement
	// individually.
	br := conn.SendBatch(ctx, &pgx.Batch{QueuedQueries: []*pgx.QueuedQuery{
		{SQL: "CREATE TABLE venues (" +
			"  venue_id    bigint not null primary key," +
			"  name        varchar(1024)," +
			"  description jsonb" +
			")"},
		{SQL: "CREATE TABLE concerts (" +
			"  concert_id bigint not null primary key ," +
			"  venue_id   bigint not null," +
			"  singer_id  bigint not null," +
			"  start_time timestamptz," +
			"  end_time   timestamptz," +
			"  constraint fk_concerts_venues foreign key" +
			"    (venue_id) references venues (venue_id)," +
			"  constraint fk_concerts_singers foreign key" +
			"    (singer_id) references singers (singer_id)" +
			")"},
	}})
	if _, err := br.Exec(); err != nil {
		return err
	}
	if err := br.Close(); err != nil {
		return err
	}
	fmt.Println("Added venues and concerts tables")

	return nil
}

Node.js

import { Client } from 'pg';

async function ddlBatch(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  // Executing multiple DDL statements as one batch is
  // more efficient than executing each statement
  // individually.
  await connection.query("start batch ddl");
  await connection.query("CREATE TABLE venues (" +
      "  venue_id    bigint not null primary key," +
      "  name        varchar(1024)," +
      "  description jsonb" +
      ")");
  await connection.query("CREATE TABLE concerts (" +
      "  concert_id bigint not null primary key ," +
      "  venue_id   bigint not null," +
      "  singer_id  bigint not null," +
      "  start_time timestamptz," +
      "  end_time   timestamptz," +
      "  constraint fk_concerts_venues foreign key" +
      "    (venue_id) references venues (venue_id)," +
      "  constraint fk_concerts_singers foreign key" +
      "    (singer_id) references singers (singer_id)" +
      ")");
  await connection.query("run batch");
  console.log("Added venues and concerts tables");

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def ddl_batch(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        # DDL can only be executed when autocommit=True.
        conn.autocommit = True
        # Use a pipeline to batch multiple statements together.
        # Executing multiple DDL statements as one batch is
        # more efficient than executing each statement
        # individually.
        with conn.pipeline():
            # The following statements are buffered on PGAdapter
            # until the pipeline ends.
            conn.execute("CREATE TABLE venues ("
                         "  venue_id    bigint not null primary key,"
                         "  name        varchar(1024),"
                         "  description jsonb"
                         ")")
            conn.execute("CREATE TABLE concerts ("
                         "  concert_id bigint not null primary key ,"
                         "  venue_id   bigint not null,"
                         "  singer_id  bigint not null,"
                         "  start_time timestamptz,"
                         "  end_time   timestamptz,"
                         "  constraint fk_concerts_venues foreign key"
                         "    (venue_id) references venues (venue_id),"
                         "  constraint fk_concerts_singers foreign key"
                         "    (singer_id) references singers (singer_id)"
                         ")")
        print("Added venues and concerts tables")

C#

using Npgsql;

namespace dotnet_snippets;

public static class DdlBatchSample
{
    public static void DdlBatch(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        // Create two new tables in one batch.
        var batch = connection.CreateBatch();
        batch.BatchCommands.Add(new NpgsqlBatchCommand(
            "CREATE TABLE venues ("
            + "  venue_id    bigint not null primary key,"
            + "  name        varchar(1024),"
            + "  description jsonb"
            + ")"));
        batch.BatchCommands.Add(new NpgsqlBatchCommand(
            "CREATE TABLE concerts ("
            + "  concert_id bigint not null primary key ,"
            + "  venue_id   bigint not null,"
            + "  singer_id  bigint not null,"
            + "  start_time timestamptz,"
            + "  end_time   timestamptz,"
            + "  constraint fk_concerts_venues foreign key"
            + "    (venue_id) references venues (venue_id),"
            + "  constraint fk_concerts_singers foreign key"
            + "    (singer_id) references singers (singer_id)"
            + ")"));
        batch.ExecuteNonQuery();
        Console.WriteLine("Added venues and concerts tables");
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./ddl_batch.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar ddlbatch example-db

Go

go run sample_runner.go ddlbatch example-db

Node.js

npm start ddlbatch example-db

Python

python ddl_batch.py example-db

C#

dotnet run ddlbatch example-db

다음과 같이 표시됩니다.

Added venues and concerts tables

새 열에 데이터 쓰기

다음 코드는 새 열에 데이터를 씁니다. 이 코드는 MarketingBudgetAlbums(1, 1)로 키가 지정된 행에서는 100000으로, Albums(2, 2)로 키가 지정된 행에서는 500000으로 설정합니다.

PGAdapter는 PostgreSQL COPY 명령어를 변형으로 변환합니다. COPY 명령어는 기본적으로 Insert 변형으로 변환됩니다. set spanner.copy_upsert=true를 실행하여 COPY 명령어를 InsertOrUpdate 변형으로 변환합니다. 이렇게 하면 Spanner의 기존 데이터를 업데이트할 수 있습니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

# Instruct PGAdapter to use insert-or-update for COPY statements.
# This enables us to use COPY to update data.
psql -c "set spanner.copy_upsert=true" \
     -c "COPY albums (singer_id, album_id, marketing_budget) FROM STDIN
         WITH (DELIMITER ';')" \
<< DATA
1;1;100000
2;2;500000
DATA

echo "Copied albums using upsert"

자바

import java.io.IOException;
import java.io.StringReader;
import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.SQLException;
import org.postgresql.PGConnection;
import org.postgresql.copy.CopyManager;

class UpdateDataWithCopy {

  static void updateDataWithCopy(String host, int port, String database)
      throws SQLException, IOException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      // Unwrap the PostgreSQL JDBC connection interface to get access to
      // a CopyManager.
      PGConnection pgConnection = connection.unwrap(PGConnection.class);
      CopyManager copyManager = pgConnection.getCopyAPI();

      // Enable 'partitioned_non_atomic' mode. This ensures that the COPY operation
      // will succeed even if it exceeds Spanner's mutation limit per transaction.
      connection
          .createStatement()
          .execute("set spanner.autocommit_dml_mode='partitioned_non_atomic'");

      // Instruct PGAdapter to use insert-or-update for COPY statements.
      // This enables us to use COPY to update existing data.
      connection.createStatement().execute("set spanner.copy_upsert=true");

      // COPY uses mutations to insert or update existing data in Spanner.
      long numAlbums =
          copyManager.copyIn(
              "COPY albums (singer_id, album_id, marketing_budget) FROM STDIN",
              new StringReader("1\t1\t100000\n" + "2\t2\t500000\n"));
      System.out.printf("Updated %d albums\n", numAlbums);
    }
  }
}

Go

import (
	"context"
	"fmt"
	"io"

	"github.com/jackc/pgx/v5"
)

func UpdateDataWithCopy(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	// Enable non-atomic mode. This makes the COPY operation non-atomic,
	// and allows it to exceed the Spanner mutation limit.
	if _, err := conn.Exec(ctx,
		"set spanner.autocommit_dml_mode='partitioned_non_atomic"); err != nil {
		return err
	}
	// Instruct PGAdapter to use insert-or-update for COPY statements.
	// This enables us to use COPY to update data.
	if _, err := conn.Exec(ctx, "set spanner.copy_upsert=true"); err != nil {
		return err
	}

	// Create a pipe that can be used to write the data manually that we want to copy.
	reader, writer := io.Pipe()
	// Write the data to the pipe using a separate goroutine. This allows us to stream the data
	// to the COPY operation row-by-row.
	go func() error {
		for _, record := range []string{"1\t1\t100000\n", "2\t2\t500000\n"} {
			if _, err := writer.Write([]byte(record)); err != nil {
				return err
			}
		}
		if err := writer.Close(); err != nil {
			return err
		}
		return nil
	}()
	tag, err := conn.PgConn().CopyFrom(ctx, reader, "COPY albums (singer_id, album_id, marketing_budget) FROM STDIN")
	if err != nil {
		return err
	}
	fmt.Printf("Updated %v albums\n", tag.RowsAffected())

	return nil
}

Node.js

import { Client } from 'pg';
import { pipeline } from 'node:stream/promises'
import { from as copyFrom } from 'pg-copy-streams'
import {Readable} from "stream";

async function updateDataWithCopy(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  // Enable 'partitioned_non_atomic' mode. This ensures that the COPY operation
  // will succeed even if it exceeds Spanner's mutation limit per transaction.
  await connection.query("set spanner.autocommit_dml_mode='partitioned_non_atomic'");

  // Instruct PGAdapter to use insert-or-update for COPY statements.
  // This enables us to use COPY to update existing data.
  await connection.query("set spanner.copy_upsert=true");

  // Copy data to Spanner using the COPY command.
  const copyStream = copyFrom('COPY albums (singer_id, album_id, marketing_budget) FROM STDIN');
  const ingestStream = connection.query(copyStream);

  // Create a source stream and attach the source to the destination.
  const sourceStream = new Readable();
  const operation = pipeline(sourceStream, ingestStream);
  // Manually push data to the source stream to write data to Spanner.
  sourceStream.push("1\t1\t100000\n");
  sourceStream.push("2\t2\t500000\n");
  // Push a 'null' to indicate the end of the stream.
  sourceStream.push(null);
  // Wait for the copy operation to finish.
  await operation;
  console.log(`Updated ${copyStream.rowCount} albums`);

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def update_data_with_copy(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        conn.autocommit = True
        with conn.cursor() as cur:
            # Instruct PGAdapter to use insert-or-update for COPY statements.
            # This enables us to use COPY to update data.
            cur.execute("set spanner.copy_upsert=true")

            # COPY uses mutations to insert or update existing data in Spanner.
            with cur.copy("COPY albums (singer_id, album_id, marketing_budget) "
                          "FROM STDIN") as copy:
                copy.write_row((1, 1, 100000))
                copy.write_row((2, 2, 500000))
            print("Updated %d albums" % cur.rowcount)

C#

using Npgsql;

namespace dotnet_snippets;

public static class UpdateDataWithCopySample
{
    public static void UpdateDataWithCopy(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        // Enable 'partitioned_non_atomic' mode. This ensures that the COPY operation
        // will succeed even if it exceeds Spanner's mutation limit per transaction.
        using var cmd = connection.CreateCommand();
        cmd.CommandText = "set spanner.autocommit_dml_mode='partitioned_non_atomic'";
        cmd.ExecuteNonQuery();

        // Instruct PGAdapter to use insert-or-update for COPY statements.
        // This enables us to use COPY to update existing data.
        cmd.CommandText = "set spanner.copy_upsert=true";
        cmd.ExecuteNonQuery();

        // COPY uses mutations to insert or update existing data in Spanner.
        using (var albumWriter = connection.BeginTextImport(
                   "COPY albums (singer_id, album_id, marketing_budget) FROM STDIN"))
        {
            albumWriter.WriteLine("1\t1\t100000");
            albumWriter.WriteLine("2\t2\t500000");
        }
        Console.WriteLine($"Updated 2 albums");
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./update_data_with_copy.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar update example-db

Go

go run sample_runner.go update example-db

Node.js

npm start update example-db

Python

python update_data_with_copy.py example-db

C#

dotnet run update example-db

다음과 같이 표시됩니다.

Updated 2 albums

방금 쓴 값을 가져오기 위해 SQL 쿼리를 실행할 수도 있습니다.

다음은 쿼리를 실행하는 코드입니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

psql -c "SELECT singer_id, album_id, marketing_budget
         FROM albums
         ORDER BY singer_id, album_id"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.ResultSet;
import java.sql.SQLException;

class QueryDataWithNewColumn {
  static void queryDataWithNewColumn(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      try (ResultSet resultSet =
          connection
              .createStatement()
              .executeQuery(
                  "SELECT singer_id, album_id, marketing_budget "
                      + "FROM albums "
                      + "ORDER BY singer_id, album_id")) {
        while (resultSet.next()) {
          System.out.printf(
              "%d %d %s\n",
              resultSet.getLong("singer_id"),
              resultSet.getLong("album_id"),
              resultSet.getString("marketing_budget"));
        }
      }
    }
  }
}

Go

import (
	"context"
	"database/sql"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func QueryDataWithNewColumn(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	rows, err := conn.Query(ctx, "SELECT singer_id, album_id, marketing_budget "+
		"FROM albums "+
		"ORDER BY singer_id, album_id")
	defer rows.Close()
	if err != nil {
		return err
	}
	for rows.Next() {
		var singerId, albumId int64
		var marketingBudget sql.NullString
		err = rows.Scan(&singerId, &albumId, &marketingBudget)
		if err != nil {
			return err
		}
		var budget string
		if marketingBudget.Valid {
			budget = marketingBudget.String
		} else {
			budget = "NULL"
		}
		fmt.Printf("%v %v %v\n", singerId, albumId, budget)
	}

	return rows.Err()
}

Node.js

import { Client } from 'pg';

async function queryDataWithNewColumn(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  const result = await connection.query(
      "SELECT singer_id, album_id, marketing_budget "
      + "FROM albums "
      + "ORDER BY singer_id, album_id"
  );
  for (const row of result.rows) {
    console.log(`${row["singer_id"]} ${row["album_id"]} ${row["marketing_budget"]}`);
  }

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def query_data_with_new_column(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        conn.autocommit = True
        with conn.cursor() as cur:
            cur.execute("SELECT singer_id, album_id, marketing_budget "
                        "FROM albums "
                        "ORDER BY singer_id, album_id")
            for album in cur:
                print(album)

C#

using Npgsql;

namespace dotnet_snippets;

public static class QueryDataWithNewColumnSample
{
    public static void QueryWithNewColumnData(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        using var cmd = new NpgsqlCommand("SELECT singer_id, album_id, marketing_budget "
                                          + "FROM albums "
                                          + "ORDER BY singer_id, album_id", connection);
        using var reader = cmd.ExecuteReader();
        while (reader.Read())
        {
            Console.WriteLine($"{reader["singer_id"]} {reader["album_id"]} {reader["marketing_budget"]}");
        }
    }
}

다음 명령어를 사용하여 쿼리를 실행합니다.

psql

PGDATABASE=example-db ./query_data_with_new_column.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar querymarketingbudget example-db

Go

go run sample_runner.go querymarketingbudget example-db

Node.js

npm start querymarketingbudget example-db

Python

python query_data_with_new_column.py example-db

C#

dotnet run querymarketingbudget example-db

다음과 같이 표시됩니다.

1 1 100000
1 2 null
2 1 null
2 2 500000
2 3 null

데이터 업데이트

읽기-쓰기 트랜잭션에서 DML을 사용하여 데이터를 업데이트할 수 있습니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

psql << SQL
  -- Transfer marketing budget from one album to another.
  -- We do it in a transaction to ensure that the transfer is atomic.
  -- Begin a read/write transaction.
  begin;

  -- Increase the marketing budget of album 1 if album 2 has enough budget.
  -- The condition that album 2 has enough budget is guaranteed for the
  -- duration of the transaction, as read/write transactions in Spanner use
  -- external consistency as the default isolation level.
  update albums set
    marketing_budget = marketing_budget + 200000
  where singer_id = 1
    and  album_id = 1
    and exists (
      select album_id
      from albums
      where singer_id = 2
        and  album_id = 2
        and marketing_budget > 200000
      );

  -- Decrease the marketing budget of album 2.      
  update albums set
    marketing_budget = marketing_budget - 200000
  where singer_id = 2
    and  album_id = 2
    and marketing_budget > 200000;

  -- Commit the transaction to make the changes to both marketing budgets
  -- durably stored in the database and visible to other transactions.
  commit;  
SQL

echo "Transferred marketing budget from Album 2 to Album 1"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.PreparedStatement;
import java.sql.ResultSet;
import java.sql.SQLException;

class UpdateDataWithTransaction {

  static void writeWithTransactionUsingDml(String host, int port, String database)
      throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      // Set AutoCommit=false to enable transactions.
      connection.setAutoCommit(false);

      // Transfer marketing budget from one album to another. We do it in a
      // transaction to ensure that the transfer is atomic. There is no need
      // to explicitly start the transaction. The first statement on the
      // connection will start a transaction when AutoCommit=false.
      String selectMarketingBudgetSql =
          "SELECT marketing_budget from albums WHERE singer_id = ? and album_id = ?";
      long album2Budget = 0;
      try (PreparedStatement selectMarketingBudgetStatement =
          connection.prepareStatement(selectMarketingBudgetSql)) {
        // Bind the query parameters to SingerId=2 and AlbumId=2.
        selectMarketingBudgetStatement.setLong(1, 2);
        selectMarketingBudgetStatement.setLong(2, 2);
        try (ResultSet resultSet = selectMarketingBudgetStatement.executeQuery()) {
          while (resultSet.next()) {
            album2Budget = resultSet.getLong("marketing_budget");
          }
        }
        // The transaction will only be committed if this condition still holds
        // at the time of commit. Otherwise, the transaction will be aborted.
        final long transfer = 200000;
        if (album2Budget >= transfer) {
          long album1Budget = 0;
          // Re-use the existing PreparedStatement for selecting the
          // marketing_budget to get the budget for Album 1.
          // Bind the query parameters to SingerId=1 and AlbumId=1.
          selectMarketingBudgetStatement.setLong(1, 1);
          selectMarketingBudgetStatement.setLong(2, 1);
          try (ResultSet resultSet = selectMarketingBudgetStatement.executeQuery()) {
            while (resultSet.next()) {
              album1Budget = resultSet.getLong("marketing_budget");
            }
          }

          // Transfer part of the marketing budget of Album 2 to Album 1.
          album1Budget += transfer;
          album2Budget -= transfer;
          String updateSql =
              "UPDATE albums "
                  + "SET marketing_budget = ? "
                  + "WHERE singer_id = ? and album_id = ?";
          try (PreparedStatement updateStatement = connection.prepareStatement(updateSql)) {
            // Update Album 1.
            int paramIndex = 0;
            updateStatement.setLong(++paramIndex, album1Budget);
            updateStatement.setLong(++paramIndex, 1);
            updateStatement.setLong(++paramIndex, 1);
            // Create a DML batch by calling addBatch
            // on the current PreparedStatement.
            updateStatement.addBatch();

            // Update Album 2 in the same DML batch.
            paramIndex = 0;
            updateStatement.setLong(++paramIndex, album2Budget);
            updateStatement.setLong(++paramIndex, 2);
            updateStatement.setLong(++paramIndex, 2);
            updateStatement.addBatch();

            // Execute both DML statements in one batch.
            updateStatement.executeBatch();
          }
        }
      }
      // Commit the current transaction.
      connection.commit();
      System.out.println("Transferred marketing budget from Album 2 to Album 1");
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func WriteWithTransactionUsingDml(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	// Transfer marketing budget from one album to another. We do it in a
	// transaction to ensure that the transfer is atomic.
	tx, err := conn.Begin(ctx)
	if err != nil {
		return err
	}
	const selectSql = "SELECT marketing_budget " +
		"from albums " +
		"WHERE singer_id = $1 and album_id = $2"
	// Get the marketing_budget of singer 2 / album 2.
	row := tx.QueryRow(ctx, selectSql, 2, 2)
	var budget2 int64
	if err := row.Scan(&budget2); err != nil {
		tx.Rollback(ctx)
		return err
	}
	const transfer = 20000
	// The transaction will only be committed if this condition still holds
	// at the time of commit. Otherwise, the transaction will be aborted.
	if budget2 >= transfer {
		// Get the marketing_budget of singer 1 / album 1.
		row := tx.QueryRow(ctx, selectSql, 1, 1)
		var budget1 int64
		if err := row.Scan(&budget1); err != nil {
			tx.Rollback(ctx)
			return err
		}
		// Transfer part of the marketing budget of Album 2 to Album 1.
		budget1 += transfer
		budget2 -= transfer
		const updateSql = "UPDATE albums " +
			"SET marketing_budget = $1 " +
			"WHERE singer_id = $2 and album_id = $3"
		// Start a DML batch and execute it as part of the current transaction.
		batch := &pgx.Batch{}
		batch.Queue(updateSql, budget1, 1, 1)
		batch.Queue(updateSql, budget2, 2, 2)
		br := tx.SendBatch(ctx, batch)
		_, err = br.Exec()
		if err := br.Close(); err != nil {
			tx.Rollback(ctx)
			return err
		}
	}
	// Commit the current transaction.
	tx.Commit(ctx)
	fmt.Println("Transferred marketing budget from Album 2 to Album 1")

	return nil
}

Node.js

import { Client } from 'pg';

async function writeWithTransactionUsingDml(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  // Transfer marketing budget from one album to another. We do it in a
  // transaction to ensure that the transfer is atomic. node-postgres
  // requires you to explicitly start the transaction by executing 'begin'.
  await connection.query("begin");
  const selectMarketingBudgetSql = "SELECT marketing_budget " +
      "from albums " +
      "WHERE singer_id = $1 and album_id = $2";
  // Get the marketing_budget of singer 2 / album 2.
  const album2BudgetResult = await connection.query(selectMarketingBudgetSql, [2, 2]);
  let album2Budget = album2BudgetResult.rows[0]["marketing_budget"];
  const transfer = 200000;
  // The transaction will only be committed if this condition still holds
  // at the time of commit. Otherwise, the transaction will be aborted.
  if (album2Budget >= transfer) {
    // Get the marketing budget of singer 1 / album 1.
    const album1BudgetResult = await connection.query(selectMarketingBudgetSql, [1, 1]);
    let album1Budget = album1BudgetResult.rows[0]["marketing_budget"];
    // Transfer part of the marketing budget of Album 2 to Album 1.
    album1Budget += transfer;
    album2Budget -= transfer;
    const updateSql = "UPDATE albums " +
        "SET marketing_budget = $1 " +
        "WHERE singer_id = $2 and album_id = $3";
    // Start a DML batch. This batch will become part of the current transaction.
    // TODO: Enable when https://github.com/googleapis/java-spanner/pull/3114 has been merged
    // await connection.query("start batch dml");
    // Update the marketing budget of both albums.
    await connection.query(updateSql, [album1Budget, 1, 1]);
    await connection.query(updateSql, [album2Budget, 2, 2]);
    // TODO: Enable when https://github.com/googleapis/java-spanner/pull/3114 has been merged
    // await connection.query("run batch");
  }
  // Commit the current transaction.
  await connection.query("commit");
  console.log("Transferred marketing budget from Album 2 to Album 1");

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def update_data_with_transaction(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        # Set autocommit=False to use transactions.
        # The first statement that is executed starts the transaction.
        conn.autocommit = False
        with conn.cursor() as cur:
            # Transfer marketing budget from one album to another.
            # We do it in a transaction to ensure that the transfer is atomic.
            # There is no need to explicitly start the transaction. The first
            # statement on the connection will start a transaction when
            # AutoCommit=false.
            select_marketing_budget_sql = ("SELECT marketing_budget "
                                           "from albums "
                                           "WHERE singer_id = %s "
                                           "and album_id = %s")
            # Get the marketing budget of Album #2.
            cur.execute(select_marketing_budget_sql, (2, 2))
            album2_budget = cur.fetchone()[0]
            transfer = 200000
            if album2_budget > transfer:
                # Get the marketing budget of Album #1.
                cur.execute(select_marketing_budget_sql, (1, 1))
                album1_budget = cur.fetchone()[0]
                # Transfer the marketing budgets and write the update back
                # to the database.
                album1_budget += transfer
                album2_budget -= transfer
                update_sql = ("update albums "
                              "set marketing_budget = %s "
                              "where singer_id = %s "
                              "and   album_id = %s")
                # Use a pipeline to execute two DML statements in one batch.
                with conn.pipeline():
                    cur.execute(update_sql, (album1_budget, 1, 1,))
                    cur.execute(update_sql, (album2_budget, 2, 2,))
            else:
                print("Insufficient budget to transfer")
        # Commit the transaction.
        conn.commit()
        print("Transferred marketing budget from Album 2 to Album 1")

C#

using Npgsql;
using System.Data;

namespace dotnet_snippets;

public static class TagsSample
{
    public static void Tags(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        // Start a transaction with isolation level Serializable.
        // Spanner only supports this isolation level. Trying to use a lower
        // isolation level (including the default isolation level READ COMMITTED),
        // will result in an error.
        var transaction = connection.BeginTransaction(IsolationLevel.Serializable);

        // Create a command that uses the current transaction.
        using var cmd = connection.CreateCommand();
        cmd.Transaction = transaction;

        // Set the TRANSACTION_TAG session variable to set a transaction tag
        // for the current transaction.
        cmd.CommandText = "set spanner.transaction_tag='example-tx-tag'";
        cmd.ExecuteNonQuery();

        // Set the STATEMENT_TAG session variable to set the request tag
        // that should be included with the next SQL statement.
        cmd.CommandText = "set spanner.statement_tag='query-marketing-budget'";
        cmd.ExecuteNonQuery();

        // Get the marketing_budget of Album (1,1).
        cmd.CommandText = "select marketing_budget from albums where singer_id=$1 and album_id=$2";
        cmd.Parameters.Add(new NpgsqlParameter { Value = 1L });
        cmd.Parameters.Add(new NpgsqlParameter { Value = 1L });
        var marketingBudget = (long?)cmd.ExecuteScalar();

        // Reduce the marketing budget by 10% if it is more than 1,000.
        if (marketingBudget > 1000L)
        {
            marketingBudget -= (long) (marketingBudget * 0.1);

            // Set the statement tag to use for the update statement.
            cmd.Parameters.Clear();
            cmd.CommandText = "set spanner.statement_tag='reduce-marketing-budget'";
            cmd.ExecuteNonQuery();

            cmd.CommandText = "update albums set marketing_budget=$1 where singer_id=$2 AND album_id=$3";
            cmd.Parameters.Add(new NpgsqlParameter { Value = marketingBudget });
            cmd.Parameters.Add(new NpgsqlParameter { Value = 1L });
            cmd.Parameters.Add(new NpgsqlParameter { Value = 1L });
            cmd.ExecuteNonQuery();
        }

        // Commit the current transaction.
        transaction.Commit();
        Console.WriteLine("Reduced marketing budget");
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./update_data_with_transaction.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar writewithtransactionusingdml example-db

Go

go run sample_runner.go writewithtransactionusingdml example-db

Node.js

npm start writewithtransactionusingdml example-db

Python

python update_data_with_transaction.py example-db

C#

dotnet run writewithtransactionusingdml example-db

다음과 같이 표시됩니다.

Transferred marketing budget from Album 2 to Album 1

트랜잭션 태그 및 요청 태그

트랜잭션 태그 및 요청 태그를 사용하여 Spanner에서 트랜잭션 및 쿼리를 문제 해결할 수 있습니다. SPANNER.TRANSACTION_TAGSPANNER.STATEMENT_TAG 세션 변수를 사용하여 트랜잭션 태그 및 요청 태그를 설정할 수 있습니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

psql << SQL
  -- Start a transaction.
  begin;
  -- Set the TRANSACTION_TAG session variable to set a transaction tag
  -- for the current transaction. This can only be executed at the start
  -- of the transaction.
  set spanner.transaction_TAG='example-tx-tag';

  -- Set the STATEMENT_TAG session variable to set the request tag
  -- that should be included with the next SQL statement.
  set spanner.statement_tag='query-marketing-budget';

  select marketing_budget
  from albums
  where singer_id = 1
    and album_id  = 1;

  -- Reduce the marketing budget by 10% if it is more than 1,000.
  -- Set a statement tag for the update statement.
  set spanner.statement_tag='reduce-marketing-budget';

  update albums
    set marketing_budget = marketing_budget - (marketing_budget * 0.1)::bigint
  where singer_id = 1
    and album_id  = 1
    and marketing_budget > 1000;

  commit;  
SQL

echo "Reduced marketing budget"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.PreparedStatement;
import java.sql.ResultSet;
import java.sql.SQLException;

class Tags {

  static void tags(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      // Set AutoCommit=false to enable transactions.
      connection.setAutoCommit(false);
      // Set the TRANSACTION_TAG session variable to set a transaction tag
      // for the current transaction.
      connection.createStatement().execute("set spanner.transaction_tag='example-tx-tag'");

      // Set the STATEMENT_TAG session variable to set the request tag
      // that should be included with the next SQL statement.
      connection.createStatement().execute("set spanner.statement_tag='query-marketing-budget'");
      long marketingBudget = 0L;
      long singerId = 1L;
      long albumId = 1L;
      try (PreparedStatement statement =
          connection.prepareStatement(
              "select marketing_budget from albums where singer_id=? and album_id=?")) {
        statement.setLong(1, singerId);
        statement.setLong(2, albumId);
        try (ResultSet albumResultSet = statement.executeQuery()) {
          while (albumResultSet.next()) {
            marketingBudget = albumResultSet.getLong(1);
          }
        }
      }
      // Reduce the marketing budget by 10% if it is more than 1,000.
      final long maxMarketingBudget = 1000L;
      final float reduction = 0.1f;
      if (marketingBudget > maxMarketingBudget) {
        marketingBudget -= (long) (marketingBudget * reduction);
        connection.createStatement().execute("set spanner.statement_tag='reduce-marketing-budget'");
        try (PreparedStatement statement =
            connection.prepareStatement(
                "update albums set marketing_budget=? where singer_id=? AND album_id=?")) {
          int paramIndex = 0;
          statement.setLong(++paramIndex, marketingBudget);
          statement.setLong(++paramIndex, singerId);
          statement.setLong(++paramIndex, albumId);
          statement.executeUpdate();
        }
      }

      // Commit the current transaction.
      connection.commit();
      System.out.println("Reduced marketing budget");
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func Tags(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	tx, err := conn.Begin(ctx)
	if err != nil {
		return err
	}

	// Set the TRANSACTION_TAG session variable to set a transaction tag
	// for the current transaction.
	_, _ = tx.Exec(ctx, "set spanner.transaction_tag='example-tx-tag'")

	// Set the STATEMENT_TAG session variable to set the request tag
	// that should be included with the next SQL statement.
	_, _ = tx.Exec(ctx, "set spanner.statement_tag='query-marketing-budget'")

	row := tx.QueryRow(ctx, "select marketing_budget "+
		"from albums "+
		"where singer_id=$1 and album_id=$2", 1, 1)
	var budget int64
	if err := row.Scan(&budget); err != nil {
		tx.Rollback(ctx)
		return err
	}

	// Reduce the marketing budget by 10% if it is more than 1,000.
	if budget > 1000 {
		budget = int64(float64(budget) - float64(budget)*0.1)
		_, _ = tx.Exec(ctx, "set spanner.statement_tag='reduce-marketing-budget'")
		if _, err := tx.Exec(ctx, "update albums set marketing_budget=$1 where singer_id=$2 AND album_id=$3", budget, 1, 1); err != nil {
			tx.Rollback(ctx)
			return err
		}
	}
	// Commit the current transaction.
	tx.Commit(ctx)
	fmt.Println("Reduced marketing budget")

	return nil
}

Node.js

import { Client } from 'pg';

async function tags(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  await connection.query("begin");
  // Set the TRANSACTION_TAG session variable to set a transaction tag
  // for the current transaction.
  await connection.query("set spanner.transaction_tag='example-tx-tag'");
  // Set the STATEMENT_TAG session variable to set the request tag
  // that should be included with the next SQL statement.
  await connection.query("set spanner.statement_tag='query-marketing-budget'");
  const budgetResult = await connection.query(
      "select marketing_budget " +
      "from albums " +
      "where singer_id=$1 and album_id=$2", [1, 1])
  let budget = budgetResult.rows[0]["marketing_budget"];
  // Reduce the marketing budget by 10% if it is more than 1,000.
  if (budget > 1000) {
    budget = budget - budget * 0.1;
    await connection.query("set spanner.statement_tag='reduce-marketing-budget'");
    await connection.query("update albums set marketing_budget=$1 "
        + "where singer_id=$2 AND album_id=$3", [budget, 1, 1]);
  }
  // Commit the current transaction.
  await connection.query("commit");
  console.log("Reduced marketing budget");

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def tags(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        # Set autocommit=False to enable transactions.
        conn.autocommit = False
        with conn.cursor() as cur:
            # Set the TRANSACTION_TAG session variable to set a transaction tag
            # for the current transaction.
            cur.execute("set spanner.transaction_TAG='example-tx-tag'")

            # Set the STATEMENT_TAG session variable to set the request tag
            # that should be included with the next SQL statement.
            cur.execute("set spanner.statement_tag='query-marketing-budget'")

            singer_id = 1
            album_id = 1
            cur.execute("select marketing_budget "
                        "from albums "
                        "where singer_id = %s "
                        "  and album_id  = %s",
                        (singer_id, album_id,))
            marketing_budget = cur.fetchone()[0]

            # Reduce the marketing budget by 10% if it is more than 1,000.
            max_marketing_budget = 1000
            reduction = 0.1
            if marketing_budget > max_marketing_budget:
                # Make sure the marketing_budget remains an int.
                marketing_budget -= int(marketing_budget * reduction)
                # Set a statement tag for the update statement.
                cur.execute(
                    "set spanner.statement_tag='reduce-marketing-budget'")
                cur.execute("update albums set marketing_budget = %s "
                            "where singer_id = %s "
                            "  and album_id  = %s",
                            (marketing_budget, singer_id, album_id,))
            else:
                print("Marketing budget already less than or equal to 1,000")
        # Commit the transaction.
        conn.commit()
        print("Reduced marketing budget")

C#

using Npgsql;
using System.Data;

namespace dotnet_snippets;

public static class TagsSample
{
    public static void Tags(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        // Start a transaction with isolation level Serializable.
        // Spanner only supports this isolation level. Trying to use a lower
        // isolation level (including the default isolation level READ COMMITTED),
        // will result in an error.
        var transaction = connection.BeginTransaction(IsolationLevel.Serializable);

        // Create a command that uses the current transaction.
        using var cmd = connection.CreateCommand();
        cmd.Transaction = transaction;

        // Set the TRANSACTION_TAG session variable to set a transaction tag
        // for the current transaction.
        cmd.CommandText = "set spanner.transaction_tag='example-tx-tag'";
        cmd.ExecuteNonQuery();

        // Set the STATEMENT_TAG session variable to set the request tag
        // that should be included with the next SQL statement.
        cmd.CommandText = "set spanner.statement_tag='query-marketing-budget'";
        cmd.ExecuteNonQuery();

        // Get the marketing_budget of Album (1,1).
        cmd.CommandText = "select marketing_budget from albums where singer_id=$1 and album_id=$2";
        cmd.Parameters.Add(new NpgsqlParameter { Value = 1L });
        cmd.Parameters.Add(new NpgsqlParameter { Value = 1L });
        var marketingBudget = (long?)cmd.ExecuteScalar();

        // Reduce the marketing budget by 10% if it is more than 1,000.
        if (marketingBudget > 1000L)
        {
            marketingBudget -= (long) (marketingBudget * 0.1);

            // Set the statement tag to use for the update statement.
            cmd.Parameters.Clear();
            cmd.CommandText = "set spanner.statement_tag='reduce-marketing-budget'";
            cmd.ExecuteNonQuery();

            cmd.CommandText = "update albums set marketing_budget=$1 where singer_id=$2 AND album_id=$3";
            cmd.Parameters.Add(new NpgsqlParameter { Value = marketingBudget });
            cmd.Parameters.Add(new NpgsqlParameter { Value = 1L });
            cmd.Parameters.Add(new NpgsqlParameter { Value = 1L });
            cmd.ExecuteNonQuery();
        }

        // Commit the current transaction.
        transaction.Commit();
        Console.WriteLine("Reduced marketing budget");
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./tags.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar tags example-db

Go

go run sample_runner.go tags example-db

Node.js

npm start tags example-db

Python

python tags.py example-db

C#

dotnet run tags example-db

읽기 전용 트랜잭션을 사용하여 데이터 검색

같은 타임스탬프에서 읽기를 하나 이상 실행한다고 가정해 봅시다. 읽기 전용 트랜잭션은 트랜잭션 커밋 기록의 일관된 프리픽스를 관찰하므로 애플리케이션이 항상 일관된 데이터를 가져옵니다. 연결을 읽기 전용으로 설정하거나 SET TRANSACTION READ ONLY SQL 문을 사용하여 읽기 전용 트랜잭션을 실행합니다.

다음은 같은 읽기 전용 트랜잭션에서 쿼리를 실행하고 읽기를 수행하는 방법을 보여줍니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

psql << SQL
  -- Begin a transaction.
  begin;
  -- Change the current transaction to a read-only transaction.
  -- This statement can only be executed at the start of a transaction.
  set transaction read only;

  -- The following two queries use the same read-only transaction.
  select singer_id, album_id, album_title
  from albums
  order by singer_id, album_id;

  select singer_id, album_id, album_title
  from albums
  order by album_title;

  -- Read-only transactions must also be committed or rolled back to mark
  -- the end of the transaction. There is no semantic difference between
  -- rolling back or committing a read-only transaction.
  commit;  
SQL

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.ResultSet;
import java.sql.SQLException;

class ReadOnlyTransaction {
  static void readOnlyTransaction(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      // Set AutoCommit=false to enable transactions.
      connection.setAutoCommit(false);
      // This SQL statement instructs the JDBC driver to use
      // a read-only transaction.
      connection.createStatement().execute("set transaction read only");

      try (ResultSet resultSet =
          connection
              .createStatement()
              .executeQuery(
                  "SELECT singer_id, album_id, album_title "
                      + "FROM albums "
                      + "ORDER BY singer_id, album_id")) {
        while (resultSet.next()) {
          System.out.printf(
              "%d %d %s\n",
              resultSet.getLong("singer_id"),
              resultSet.getLong("album_id"),
              resultSet.getString("album_title"));
        }
      }
      try (ResultSet resultSet =
          connection
              .createStatement()
              .executeQuery(
                  "SELECT singer_id, album_id, album_title "
                      + "FROM albums "
                      + "ORDER BY album_title")) {
        while (resultSet.next()) {
          System.out.printf(
              "%d %d %s\n",
              resultSet.getLong("singer_id"),
              resultSet.getLong("album_id"),
              resultSet.getString("album_title"));
        }
      }
      // End the read-only transaction by calling commit().
      connection.commit();
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func ReadOnlyTransaction(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	// Start a read-only transaction by supplying additional transaction options.
	tx, err := conn.BeginTx(ctx, pgx.TxOptions{AccessMode: pgx.ReadOnly})

	albumsOrderedById, err := tx.Query(ctx, "SELECT singer_id, album_id, album_title FROM albums ORDER BY singer_id, album_id")
	defer albumsOrderedById.Close()
	if err != nil {
		return err
	}
	for albumsOrderedById.Next() {
		var singerId, albumId int64
		var title string
		err = albumsOrderedById.Scan(&singerId, &albumId, &title)
		if err != nil {
			return err
		}
		fmt.Printf("%v %v %v\n", singerId, albumId, title)
	}

	albumsOrderedTitle, err := tx.Query(ctx, "SELECT singer_id, album_id, album_title FROM albums ORDER BY album_title")
	defer albumsOrderedTitle.Close()
	if err != nil {
		return err
	}
	for albumsOrderedTitle.Next() {
		var singerId, albumId int64
		var title string
		err = albumsOrderedTitle.Scan(&singerId, &albumId, &title)
		if err != nil {
			return err
		}
		fmt.Printf("%v %v %v\n", singerId, albumId, title)
	}

	// End the read-only transaction by calling Commit().
	return tx.Commit(ctx)
}

Node.js

import { Client } from 'pg';

async function readOnlyTransaction(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  // Start a transaction.
  await connection.query("begin");
  // This SQL statement instructs the PGAdapter to make it a read-only transaction.
  await connection.query("set transaction read only");

  const albumsOrderById = await connection.query(
      "SELECT singer_id, album_id, album_title "
      + "FROM albums "
      + "ORDER BY singer_id, album_id");
  for (const row of albumsOrderById.rows) {
    console.log(`${row["singer_id"]} ${row["album_id"]} ${row["album_title"]}`);
  }
  const albumsOrderByTitle = await connection.query(
      "SELECT singer_id, album_id, album_title "
      + "FROM albums "
      + "ORDER BY album_title");
  for (const row of albumsOrderByTitle.rows) {
    console.log(`${row["singer_id"]} ${row["album_id"]} ${row["album_title"]}`);
  }
  // End the read-only transaction by executing commit.
  await connection.query("commit");

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def read_only_transaction(host: string, port: int, database: string):
    with (psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn):
        # Set autocommit=False to enable transactions.
        conn.autocommit = False

        with conn.cursor() as cur:
            # Change the current transaction to a read-only transaction.
            # This statement can only be executed at the start of a transaction.
            cur.execute("set transaction read only")

            # The following two queries use the same read-only transaction.
            cur.execute("select singer_id, album_id, album_title "
                        "from albums "
                        "order by singer_id, album_id")
            for album in cur:
                print(album)

            cur.execute("select singer_id, album_id, album_title "
                        "from albums "
                        "order by album_title")
            for album in cur:
                print(album)

        # Read-only transactions must also be committed or rolled back to mark
        # the end of the transaction. There is no semantic difference between
        # rolling back or committing a read-only transaction.
        conn.commit()

C#

using Npgsql;
using System.Data;

namespace dotnet_snippets;

public static class ReadOnlyTransactionSample
{
    public static void ReadOnlyTransaction(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        // Start a read-only transaction.
        // You must specify Serializable as the isolation level, as the npgsql driver
        // will otherwise automatically set the isolation level to read-committed.
        var transaction = connection.BeginTransaction(IsolationLevel.Serializable);
        using var cmd = connection.CreateCommand();
        cmd.Transaction = transaction;
        // This SQL statement instructs the npgsql driver to use
        // a read-only transaction.
        cmd.CommandText = "set transaction read only";
        cmd.ExecuteNonQuery();

        cmd.CommandText = "SELECT singer_id, album_id, album_title " +
                          "FROM albums " +
                          "ORDER BY singer_id, album_id";
        using (var reader = cmd.ExecuteReader())
        {
            while (reader.Read())
            {
                Console.WriteLine($"{reader["singer_id"]} {reader["album_id"]} {reader["album_title"]}");
            }
        }
        cmd.CommandText = "SELECT singer_id, album_id, album_title "
                          + "FROM albums "
                          + "ORDER BY album_title";
        using (var reader = cmd.ExecuteReader())
        {
            while (reader.Read())
            {
                Console.WriteLine($"{reader["singer_id"]} {reader["album_id"]} {reader["album_title"]}");
            }
        }
        // End the read-only transaction by calling commit().
        transaction.Commit();
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./read_only_transaction.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar readonlytransaction example-db

Go

go run sample_runner.go readonlytransaction example-db

Node.js

npm start readonlytransaction example-db

Python

python read_only_transaction.py example-db

C#

dotnet run readonlytransaction example-db

다음과 비슷한 출력이 표시됩니다.

    1 1 Total Junk
    1 2 Go, Go, Go
    2 1 Green
    2 2 Forever Hold Your Peace
    2 3 Terrified
    2 2 Forever Hold Your Peace
    1 2 Go, Go, Go
    2 1 Green
    2 3 Terrified
    1 1 Total Junk

파티션을 나눈 쿼리 및 Data Boost

partitionQuery API는 쿼리를 더 작은 부분이나 파티션으로 나누고 여러 머신을 사용하여 동시에 파티션을 가져옵니다. 각 파티션은 파티션 토큰으로 식별됩니다. PartitionQuery API는 전체 데이터베이스 내보내기 또는 스캔과 같은 대량 작업에 사용되므로 표준 쿼리 API보다 지연 시간이 깁니다.

Data Boost를 사용하면 프로비저닝된 Spanner 인스턴스의 기존 워크로드에 거의 영향을 주지 않고 분석 쿼리와 데이터 내보내기를 실행할 수 있습니다. Data Boost는 파티션을 나눈 쿼리만 지원합니다.

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

# 'set spanner.data_boost_enabled=true' enables Data Boost for
# all partitioned queries on this connection.

# 'run partitioned query' is a shortcut for partitioning the query
# that follows and executing each of the partitions that is returned
# by Spanner.

psql -c "set spanner.data_boost_enabled=true" \
     -c "run partitioned query
         select singer_id, first_name, last_name
         from singers"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.ResultSet;
import java.sql.SQLException;

class DataBoost {
  static void dataBoost(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      // This enables Data Boost for all partitioned queries on this connection.
      connection.createStatement().execute("set spanner.data_boost_enabled=true");

      // Run a partitioned query. This query will use Data Boost.
      try (ResultSet resultSet =
          connection
              .createStatement()
              .executeQuery(
                  "run partitioned query "
                      + "select singer_id, first_name, last_name "
                      + "from singers")) {
        while (resultSet.next()) {
          System.out.printf(
              "%d %s %s\n",
              resultSet.getLong("singer_id"),
              resultSet.getString("first_name"),
              resultSet.getString("last_name"));
        }
      }
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func DataBoost(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	// This enables Data Boost for all partitioned queries on this connection.
	_, _ = conn.Exec(ctx, "set spanner.data_boost_enabled=true")

	// Run a partitioned query. This query will use Data Boost.
	rows, err := conn.Query(ctx, "run partitioned query select singer_id, first_name, last_name from singers")
	defer rows.Close()
	if err != nil {
		return err
	}
	for rows.Next() {
		var singerId int64
		var firstName, lastName string
		err = rows.Scan(&singerId, &firstName, &lastName)
		if err != nil {
			return err
		}
		fmt.Printf("%v %v %v\n", singerId, firstName, lastName)
	}

	return nil
}

Node.js

import { Client } from 'pg';

async function dataBoost(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  // This enables Data Boost for all partitioned queries on this connection.
  await connection.query("set spanner.data_boost_enabled=true");

  // Run a partitioned query. This query will use Data Boost.
  const singers = await connection.query(
      "run partitioned query "
      + "select singer_id, first_name, last_name "
      + "from singers");
  for (const row of singers.rows) {
    console.log(`${row["singer_id"]} ${row["first_name"]} ${row["last_name"]}`);
  }

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def data_boost(host: string, port: int, database: string):
    with (psycopg.connect("host={host} port={port} dbname={database} "
                          "sslmode=disable".format(host=host,
                                                   port=port,
                                                   database=database)) as conn):
        # Set autocommit=True so each query uses a separate transaction.
        conn.autocommit = True

        with conn.cursor() as cur:
            # This enables Data Boost for all partitioned queries on this
            # connection.
            cur.execute("set spanner.data_boost_enabled=true")

            # Run a partitioned query. This query will use Data Boost.
            cur.execute("run partitioned query "
                        "select singer_id, first_name, last_name "
                        "from singers")
            for singer in cur:
                print(singer)

C#

using Npgsql;

namespace dotnet_snippets;

public static class DataBoostSample
{
    public static void DataBoost(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        using var cmd = connection.CreateCommand();
        // This enables Data Boost for all partitioned queries on this connection.
        cmd.CommandText = "set spanner.data_boost_enabled=true";
        cmd.ExecuteNonQuery();


        // Run a partitioned query. This query will use Data Boost.
        cmd.CommandText = "run partitioned query "
                          + "select singer_id, first_name, last_name "
                          + "from singers";
        using var reader = cmd.ExecuteReader();
        while (reader.Read())
        {
            Console.WriteLine($"{reader["singer_id"]} {reader["first_name"]} {reader["last_name"]}");
        }
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./data_boost.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar databoost example-db

Go

go run sample_runner.go databoost example-db

Node.js

npm start databoost example-db

Python

python data_boost.py example-db

C#

dotnet run databoost example-db

파티션을 나눈 쿼리 실행 및 PGAdapter에서 Data Boost를 사용하는 방법에 대한 자세한 내용은 Data Boost 및 파티션을 나눈 쿼리 문을 참조하세요.

Partitioned DML

Partitioned DML은 다음과 같은 유형의 일괄 업데이트와 삭제를 위해 설계되었습니다.

  • 정기적 클린업 및 가비지 컬렉션:
  • 새 열을 기본값으로 백필:

psql

#!/bin/bash

export PGHOST="${PGHOST:-localhost}"
export PGPORT="${PGPORT:-5432}"
export PGDATABASE="${PGDATABASE:-example-db}"

# Change the DML mode that is used by this connection to Partitioned
# DML. Partitioned DML is designed for bulk updates and deletes.
# See https://cloud.google.com/spanner/docs/dml-partitioned for more
# information.
psql -c "set spanner.autocommit_dml_mode='partitioned_non_atomic'" \
     -c "update albums
         set marketing_budget=0
         where marketing_budget is null"

echo "Updated albums using Partitioned DML"

자바

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.SQLException;

class PartitionedDml {

  static void partitionedDml(String host, int port, String database) throws SQLException {
    String connectionUrl = String.format("jdbc:postgresql://%s:%d/%s", host, port, database);
    try (Connection connection = DriverManager.getConnection(connectionUrl)) {
      // Enable Partitioned DML on this connection.
      connection
          .createStatement()
          .execute("set spanner.autocommit_dml_mode='partitioned_non_atomic'");
      // Back-fill a default value for the MarketingBudget column.
      long lowerBoundUpdateCount =
          connection
              .createStatement()
              .executeUpdate("update albums set marketing_budget=0 where marketing_budget is null");
      System.out.printf("Updated at least %d albums\n", lowerBoundUpdateCount);
    }
  }
}

Go

import (
	"context"
	"fmt"

	"github.com/jackc/pgx/v5"
)

func PartitionedDML(host string, port int, database string) error {
	ctx := context.Background()
	connString := fmt.Sprintf(
		"postgres://uid:pwd@%s:%d/%s?sslmode=disable",
		host, port, database)
	conn, err := pgx.Connect(ctx, connString)
	if err != nil {
		return err
	}
	defer conn.Close(ctx)

	// Enable Partitioned DML on this connection.
	if _, err := conn.Exec(ctx, "set spanner.autocommit_dml_mode='partitioned_non_atomic'"); err != nil {
		return err
	}
	// Back-fill a default value for the MarketingBudget column.
	tag, err := conn.Exec(ctx, "update albums set marketing_budget=0 where marketing_budget is null")
	if err != nil {
		return err
	}
	fmt.Printf("Updated at least %v albums\n", tag.RowsAffected())

	return nil
}

Node.js

import { Client } from 'pg';

async function partitionedDml(host: string, port: number, database: string): Promise<void> {
  const connection = new Client({
    host: host,
    port: port,
    database: database,
  });
  await connection.connect();

  // Enable Partitioned DML on this connection.
  await connection.query("set spanner.autocommit_dml_mode='partitioned_non_atomic'");

  // Back-fill a default value for the MarketingBudget column.
  const lowerBoundUpdateCount = await connection.query(
      "update albums " +
      "set marketing_budget=0 " +
      "where marketing_budget is null");
  console.log(`Updated at least ${lowerBoundUpdateCount.rowCount} albums`);

  // Close the connection.
  await connection.end();
}

Python

import string
import psycopg


def execute_partitioned_dml(host: string, port: int, database: string):
    with psycopg.connect("host={host} port={port} dbname={database} "
                         "sslmode=disable".format(host=host,
                                                  port=port,
                                                  database=database)) as conn:
        conn.autocommit = True
        with conn.cursor() as cur:
            # Change the DML mode that is used by this connection to Partitioned
            # DML. Partitioned DML is designed for bulk updates and deletes.
            # See https://cloud.google.com/spanner/docs/dml-partitioned for more
            # information.
            cur.execute(
                "set spanner.autocommit_dml_mode='partitioned_non_atomic'")

            # The following statement will use Partitioned DML.
            cur.execute("update albums "
                        "set marketing_budget=0 "
                        "where marketing_budget is null")
            print("Updated at least %d albums" % cur.rowcount)

C#

using Npgsql;

namespace dotnet_snippets;

public static class PartitionedDmlSample
{
    public static void PartitionedDml(string host, int port, string database)
    {
        var connectionString = $"Host={host};Port={port};Database={database};SSL Mode=Disable";
        using var connection = new NpgsqlConnection(connectionString);
        connection.Open();

        // Enable Partitioned DML on this connection.
        using var cmd = connection.CreateCommand();
        cmd.CommandText = "set spanner.autocommit_dml_mode='partitioned_non_atomic'";
        cmd.ExecuteNonQuery();

        // Back-fill a default value for the MarketingBudget column.
        cmd.CommandText = "update albums set marketing_budget=0 where marketing_budget is null";
        var lowerBoundUpdateCount = cmd.ExecuteNonQuery();

        Console.WriteLine($"Updated at least {lowerBoundUpdateCount} albums");
    }
}

다음 명령어를 사용하여 샘플을 실행합니다.

psql

PGDATABASE=example-db ./partitioned_dml.sh

자바

java -jar target/pgadapter-snippets/pgadapter-samples.jar partitioneddml example-db

Go

go run sample_runner.go partitioneddml example-db

Node.js

npm start partitioneddml example-db

Python

python partitioned_dml.py example-db

C#

dotnet run datpartitioneddmlboost example-db

삭제

이 튜토리얼에서 사용한 리소스에 대한 추가 비용이 Cloud Billing 계정에 청구되지 않도록 하려면 데이터베이스와 새로 만든 인스턴스를 삭제합니다.

데이터베이스 삭제

인스턴스를 삭제하면 인스턴스 내의 모든 데이터베이스가 자동으로 삭제됩니다. 다음 단계는 인스턴스를 삭제하지 않고 데이터베이스를 삭제하는 방법을 보여줍니다. 인스턴스에 대한 비용은 여전히 발생합니다.

명령줄에서

gcloud spanner databases delete example-db --instance=test-instance

Google Cloud 콘솔 사용

  1. Google Cloud 콘솔에서 Spanner 인스턴스 페이지로 이동합니다.

    인스턴스 페이지로 이동

  2. 인스턴스를 클릭합니다.

  3. 삭제할 데이터베이스를 클릭합니다.

  4. 데이터베이스 세부정보 페이지에서 삭제를 클릭합니다.

  5. 데이터베이스 삭제 여부를 확인하고 삭제를 클릭합니다.

인스턴스 삭제

인스턴스를 삭제하면 해당 인스턴스에서 만든 모든 데이터베이스가 자동으로 삭제됩니다.

명령줄에서

gcloud spanner instances delete test-instance

Google Cloud 콘솔 사용

  1. Google Cloud 콘솔에서 Spanner 인스턴스 페이지로 이동합니다.

    인스턴스 페이지로 이동

  2. 인스턴스를 클릭합니다.

  3. 삭제를 클릭합니다.

  4. 인스턴스 삭제 여부를 확인하고 삭제를 클릭합니다.

다음 단계