13. Spring Data Cloud Spanner

You can the use Spring Boot Starter for Spring Data Spanner to autoconfigure Google Cloud Spanner in your Spring application. It contains all the necessary setup that makes it easy to authenticate with your Google Cloud project. The following configuration options are available:

Spring Data Repositories can be configured via the @EnableSpannerRepositories annotation on your main @Configuration class. With our Spring Boot Starter for Spring Data Cloud Spanner, @EnableSpannerRepositories is automatically added. It is not required to add it to any other class, unless there is a need to override finer grain configuration parameters provided by @EnableSpannerRepositories.

Our Spring Boot autoconfiguration creates the following beans available in the Spring application context:

Simple constructors are supported on POJOs. The constructor arguments can be a subset of the persistent properties. Every constructor argument needs to have the same name and type as a persistent property on the entity and the constructor should set the property from the given argument. Arguments that are not directly set to properties are not supported.

@Table(name = "traders")
public class Trader {
	@PrimaryKey
	@Column(name = "trader_id")
	String traderId;

	String firstName;

	String lastName;

	@NotMapped
	Double temporaryNumber;

	public Trader(String traderId, String firstName) {
	    this.traderId = traderId;
	    this.firstName = firstName;
	}
}

The @Table annotation can provide the name of the Spanner table that stores instances of the annotated class, one per row. This annotation is optional, and if not given, the name of the table is inferred from the class name with the first character uncapitalized.

@Table(name = "trades_#{tableNameSuffix}")
public class Trade {
	// ...
}

For a simple table, you may only have a primary key consisting of a single column. Even in that case, the @PrimaryKey annotation is required. @PrimaryKey identifies the one or more ID properties corresponding to the primary key.

Spanner has first class support for composite primary keys of multiple columns. You have to annotate all of your POJO’s fields that the primary key consists of with @PrimaryKey as below:

@Table(name = "trades")
public class Trade {
	@PrimaryKey(keyOrder = 2)
	@Column(name = "trade_id")
	private String tradeId;

	@PrimaryKey(keyOrder = 1)
	@Column(name = "trader_id")
	private String traderId;

	private String action;

	private Double price;

	private Double shares;

	private String symbol;
}

The keyOrder parameter of @PrimaryKey identifies the properties corresponding to the primary key columns in order, starting with 1 and increasing consecutively. Order is important and must reflect the order defined in the Spanner schema. In our example the DDL to create the table and its primary key is as follows:

CREATE TABLE trades (
    trader_id STRING(MAX),
    trade_id STRING(MAX),
    action STRING(15),
    symbol STRING(10),
    price FLOAT64,
    shares FLOAT64
) PRIMARY KEY (trader_id, trade_id)

Spanner does not have automatic ID generation. For most use-cases, sequential IDs should be used with caution to avoid creating data hotspots in the system. Read Spanner Primary Keys documentation for a better understanding of primary keys and recommended practices.

All accessible properties on POJOs are automatically recognized as a Spanner column. Column naming is generated by the PropertyNameFieldNamingStrategy by default defined on the SpannerMappingContext bean. The @Column annotation optionally provides a different column name than that of the property.

If an object of type B is embedded as a property of A, then the columns of B will be saved in the same Cloud Spanner table as those of A. If B has primary key columns, those columns will be included in the primary key of A. B can also have embedded properties. Embedding allows reuse of columns between multiple entities, and can be useful for implementing parent-child situations, because Cloud Spanner requires child tables to include the key columns of their parents.

For example:

class X {
  @PrimaryKey
  String grandParentId;

  long age;
}

class A {
  @PrimaryKey
  @Embedded
  X grandParent;

  @PrimaryKey(keyOrder = 2)
  String parentId;

  String value;
}

@Table(name = "items")
class B {
  @PrimaryKey
  @Embedded
  A parent;

  @PrimaryKey(keyOrder = 2)
  String id;

  @Column(name = "child_value")
  String value;
}

Entities of B can be stored in a table defined as:

CREATE TABLE items (
    grandParentId STRING(MAX),
    parentId STRING(MAX),
    id STRING(MAX),
    value STRING(MAX),
    child_value STRING(MAX),
    age INT64
) PRIMARY KEY (grandParentId, parentId, id)

Note that embedded properties' column names must all be unique.

Currently there is no support to map relationships between objects. I.e., currently we do not have ways to establish parent-children relationships directly via annotations. This feature is actively being worked on.

Spring Data Spanner supports the following types for regular fields:

Spanner supports ARRAY types for columns. ARRAY columns are mapped to List fields in POJOS.

Example:

List<Double> curve;

Spring Data Spanner supports the following inner types:

Cloud Spanner queries can construct STRUCT values that appear as columns in the result. Cloud Spanner requires STRUCT values appear in ARRAYs at the root level: SELECT ARRAY(SELECT STRUCT(1 as val1, 2 as val2)) as pair FROM Users.

Spring Data Cloud Spanner will attempt to read the column STRUCT values into a property that is an Iterable of an entity type compatible with the schema of the column STRUCT value. For example, the following property can be mapped with the constructed ARRAY<STRUCT> column: List<TwoInts> pair; where the TwoInts type is defined:

class TwoInts {

  int val1;

  int val2;
}

Custom converters can be used extending the type support for user defined types.

For example:

We would like to have a field of type Person on our Trade POJO:

@Table(name = "trades")
public class Trade {
  //...
  Person person;
  //...
}

Where Person is a simple class:

public class Person {

  public String firstName;
  public String lastName;

}

We have to define the two converters:

  public class PersonWriteConverter implements Converter<Person, String> {

    @Override
    public String convert(Person person) {
      return person.firstName + " " + person.lastName;
    }
  }

  public class PersonReadConverter implements Converter<String, Person> {

    @Override
    public Person convert(String s) {
      Person person = new Person();
      person.firstName = s.split(" ")[0];
      person.lastName = s.split(" ")[1];
      return person;
    }
  }

That will be configured in our @Configuration file:

@Configuration
public class ConverterConfiguration {

	@Bean
	public SpannerEntityProcessor spannerEntityProcessor(SpannerMappingContext spannerMappingContext) {
		return new ConverterAwareMappingSpannerEntityProcessor(spannerMappingContext,
				Arrays.asList(new PersonWriteConverter()),
				Arrays.asList(new PersonReadConverter()));
	}
}

If a Converter<Struct, A> is provided, then properties of type List<A> can be used in your entity types.

Spanner has SQL support for running read-only queries. All the query related methods start with query on SpannerTemplate. Using SpannerTemplate you can execute SQL queries that map to POJOs:

List<Trade> trades = this.spannerTemplate.query(Trade.class, Statement.of("SELECT * FROM trades"));

Spanner exposes a Read API for reading single row or multiple rows in a table or in a secondary index.

Using SpannerTemplate you can execute reads, for example:

List<Trade> trades = this.spannerTemplate.readAll(Trade.class);

Main benefit of reads over queries is reading multiple rows of a certain pattern of keys is much easier using the features of the KeySet class.

The write methods of SpannerOperations accept a POJO and writes all of its properties to Spanner. The corresponding Spanner table and entity metadata is obtained from the given object’s actual type.

If a POJO was retrieved from Spanner and its primary key properties values were changed and then written or updated, the operation will occur as if against a row with the new primary key values. The row with the original primary key values will not be affected.

Trade t = new Trade();
this.spannerTemplate.insert(t);

// t was retrieved from a previous operation
this.spannerTemplate.update(t);

// t was retrieved from a previous operation or it's new
this.spannerTemplate.upsert(t);

// t was retrieved from a previous operation or it's new
this.spannerTemplate.update(t, "symbol", "action");

SpannerOperations provides methods to run java.util.Function objects within a single transaction while making available the read and write methods from SpannerOperations.

@Autowired
SpannerOperations mySpannerOperations;

public String doWorkInsideTransaction() {
  return mySpannerOperations.performReadWriteTransaction(
    transActionSpannerOperations -> {
      // work with transActionSpannerOperations here. It is also a SpannerOperations object.

      return "transaction completed";
    }
  );
}

@Autowired
SpannerOperations mySpannerOperations;

public String doWorkInsideTransaction() {
  return mySpannerOperations.performReadOnlyTransaction(
    transActionSpannerOperations -> {
      // work with transActionSpannerOperations here. It is also a SpannerOperations object.

      return "transaction completed";
    }
  );
}

CrudRepository methods work as expected, with one thing Spanner specific: the save and saveAll methods work as update-or-insert.

You can also use PagingAndSortingRepository with Spanner Spring Data. The sorting and pageable findAll methods available from this interface operate on the current state of the Spanner database. As a result, beware that the state of the database (and the results) might change when moving page to page.

The SpannerRepository extends the PagingAndSortingRepository, but adds the read-only and the read-write transaction functionality provided by Spanner. These transactions work very similarly to those of SpannerOperations, but is specific to the repository’s domain type and provides repository functions instead of template functions.

For example, this is a read-write transaction:

@Autowired
SpannerRepository myRepo;

public String doWorkInsideTransaction() {
  return myRepo.performReadOnlyTransaction(
    transactionSpannerRepo -> {
      // work with the single-transaction transactionSpannerRepo here. This is a SpannerRepository object.

      return "transaction completed";
    }
  );
}

public interface TradeRepository extends CrudRepository<Trade, String[]> {
    List<Trade> findByAction(String action);

	int countByAction(String action);

	// Named methods are powerful, but can get unwieldy
	List<Trade> findTop3DistinctByActionAndSymbolIgnoreCaseOrTraderIdOrderBySymbolDesc(
  			String action, String symbol, String traderId);
}

In the example above, the query methods in TradeRepository are generated based on the name of the methods, using the Spring Data Query creation naming convention.

List<Trade> findByAction(String action) would translate to a SELECT * FROM trades WHERE action = ?.

The function List<Trade> findTop3DistinctByActionAndSymbolIgnoreCaseOrTraderIdOrderBySymbolDesc(String action, String symbol, String traderId); will be translated as the equivalent of this SQL query:

SELECT DISTINCT * FROM trades
WHERE ACTION = ? AND LOWER(SYMBOL) = LOWER(?) AND TRADER_ID = ?
ORDER BY SYMBOL DESC
LIMIT 3

Note that the phrase SymbolIgnoreCase is translated to LOWER(SYMBOL) = LOWER(?) indicating a non-case-sensitive matching. The IgnoreCase phrase may only be appended to fields that correspond to columns of type STRING or BYTES. The Spring Data "AllIgnoreCase" phrase appended at the end of the method name is not supported.

The Like or NotLike naming conventions:

List<Trade> findBySymbolLike(String symbolFragment);

The param symbolFragment can contain wildcard characters for string matching such as _ and %.

The Contains and NotContains naming conventions:

List<Trade> findBySymbolContains(String symbolFragment);

The param symbolFragment is a regular expression that is checked for occurrences.

The example above for List<Trade> fetchByActionNamedQuery(String action) does not match the Spring Data Query creation naming convention, so we have to map a parametrized Spanner SQL query to it.

The SQL query for the method can be mapped to repository methods in one of two ways:

The names of the tags of the SQL correspond to the @Param annotated names of the method parameters.

Custom SQL query methods can accept a single Sort or Pageable parameter that is applied on top of any sorting or paging in the SQL:

	@Query("SELECT * FROM trades ORDER BY action DESC")
	List<Trade> sortedTrades(Pageable pageable);

This can be used:

	List<Trade> customSortedTrades = tradeRepository.sortedTrades(PageRequest
  				.of(2, 2, org.springframework.data.domain.Sort.by(Order.asc("id"))));

The results would be sorted by "id" in ascending order.

Trade.fetchByActionNamedQuery=SELECT * FROM trades WHERE trades.action = @tag0

public interface TradeRepository extends CrudRepository<Trade, String[]> {
	// This method uses the query from the properties file instead of one generated based on name.
	List<Trade> fetchByActionNamedQuery(@Param("tag0") String action);
}

public interface TradeRepository extends CrudRepository<Trade, String[]> {
    @Query("SELECT * FROM trades WHERE trades.action = @tag0")
    List<Trade> fetchByActionNamedQuery(@Param("tag0") String action);
}

@Query("SELECT * FROM :com.example.Trade: WHERE trades.action = @tag0")
List<Trade> fetchByActionNamedQuery(String action);

@Query("SELECT * FROM :com.example.Trade: WHERE trades.action = @tag0
  AND price > #{#priceRadius * -1} AND price < #{#priceRadius * 2}")
List<Trade> fetchByActionNamedQuery(String action, Double priceRadius);

Spring Data Spanner supports projections. You can define projection interfaces based on domain types and add query methods that return them in your repository:

public interface TradeProjection {

	String getAction();

	@Value("#{target.symbol + ' ' + target.action}")
	String getSymbolAndAction();
}

public interface TradeRepository extends SpannerRepository<Trade, Key> {

	List<Trade> findByTraderId(String traderId);

	List<TradeProjection> findByAction(String action);

	@Query("SELECT action, symbol FROM trades WHERE action = @action")
	List<TradeProjection> findByQuery(String action);
}

Projections can be provided by name-convention-based query methods as well as by custom SQL queries. If using custom SQL queries, you can further restrict the columns retrieved from Spanner to just those required by the projection to improve performance.

Properties of projection types defined using SpEL use the fixed name target for the underlying domain object. As a result accessing underlying properties take the form target.<property-name>.

When running with Spring Boot, repositories can act as REST services by simply annotating them:

@RepositoryRestResource(collectionResourceRel = "trades", path = "trades")
public interface TradeRepository extends CrudRepository<Trade, String[]> {
}

The @RepositoryRestResource annotation makes this repository available via REST. For example, you can retrieve all Trade objects in the repository by using curl http://<server>:<port>/trades, or any specific trade via curl http://<server>:<port>/trades/<trader_id>,<trade_id>.

The separator between your primary key components, id and trader_id in this case, is a comma by default, but can be configured to any string not found in your key values by extending the SpannerKeyIdConverter class:

@Component
class MySpecialIdConverter extends SpannerKeyIdConverter {

    @Override
    protected String getUrlIdSeparator() {
        return ":";
    }
}

You can also write trades using curl -XPOST -H"Content-Type: application/json" -[email protected] http://<server>:<port>/trades/ where the file test.json holds the JSON representation of a Trade object.

Include this dependency in your pom.xml to enable Spring Data REST Repositories:

<dependency>
  <groupId>org.springframework.boot</groupId>
  <artifactId>spring-boot-starter-data-rest</artifactId>
</dependency>