Retro Eye care Haitian Deep Dark Default

Seata Transaction

Background

Apache ShardingSphere provides BASE transactions that integrate the Seata implementation. All references to Seata integration in this article refer to Seata AT mode.

Prerequisites

ShardingSphere’s Seata integration is only available in apache/incubator-seata:v2.1.0 or higher. For Seata Client corresponding to the org.apache.seata:seata-all Maven module, this limitation applies to both HotSpot VM and GraalVM Native Image. Introduce Maven dependencies and exclude the outdated Maven dependency of org.antlr:antlr4-runtime:4.8 in org.apache.seata:seata-all.

<project>
    <dependencies>
      <dependency>
         <groupId>org.apache.shardingsphere</groupId>
         <artifactId>shardingsphere-jdbc</artifactId>
         <version>${shardingsphere.version}</version>
      </dependency>
      <dependency>
         <groupId>org.apache.shardingsphere</groupId>
         <artifactId>shardingsphere-transaction-base-seata-at</artifactId>
         <version>${shardingsphere.version}</version>
      </dependency>
      <dependency>
        <groupId>org.apache.seata</groupId>
        <artifactId>seata-all</artifactId>
        <version>2.1.0</version>
        <exclusions>
            <exclusion>
               <groupId>org.antlr</groupId>
               <artifactId>antlr4-runtime</artifactId>
            </exclusion>
        </exclusions>
      </dependency>
    </dependencies>
</project>

Affected by Calcite, commons-lang:commons-lang and org.apache.commons:commons-pool2 used by ShardingSphere JDBC have dependency conflicts with Seata Client. Users need to consider whether to resolve dependency conflicts based on actual scenarios.

When using ShardingSphere’s Seata integration module, the database instance connected to ShardingSphere should implement both ShardingSphere’s dialect parsing support and Seata AT mode’s dialect parsing support. Such databases include but are not limited to mysql, gvenzl/oracle-free, gvenzl/oracle-xe, postgres, mcr.microsoft.com/mssql/server and other Docker Images.

Procedure

  1. Start Seata Server
  2. Create the log table
  3. Add the Seata configuration

Sample

Start Seata Server

Follow the steps in one of the links below to download and start Seata Server.

The proper way to start Seata Server is to instantiate it through the Docker Image of apache/seata-server in Docker Hub.

Create undo_log table

Create the undo_log table in each real database instance involved in ShardingSphere. The SQL content is based on the corresponding database in https://github.com/apache/incubator-seata/tree/v2.1.0/script/client/at/db . The following content takes MySQL as an example.

CREATE TABLE IF NOT EXISTS `undo_log`
(
    `branch_id`     BIGINT       NOT NULL COMMENT 'branch transaction id',
    `xid`           VARCHAR(128) NOT NULL COMMENT 'global transaction id',
    `context`       VARCHAR(128) NOT NULL COMMENT 'undo_log context,such as serialization',
    `rollback_info` LONGBLOB     NOT NULL COMMENT 'rollback info',
    `log_status`    INT(11)      NOT NULL COMMENT '0:normal status,1:defense status',
    `log_created`   DATETIME(6)  NOT NULL COMMENT 'create datetime',
    `log_modified`  DATETIME(6)  NOT NULL COMMENT 'modify datetime',
    UNIQUE KEY `ux_undo_log` (`xid`, `branch_id`)
    ) ENGINE = InnoDB AUTO_INCREMENT = 1 DEFAULT CHARSET = utf8mb4 COMMENT ='AT transaction mode undo table';
ALTER TABLE `undo_log` ADD INDEX `ix_log_created` (`log_created`);

Modify configuration

Write the following content in the YAML configuration file of ShardingSphere of your own project, refer to Distributed Transaction.

If Java API is used when initializing ShardingSphere JDBC DataSource, refer to Distributed Transaction.

transaction:
  defaultType: BASE
  providerType: Seata

Add the seata.conf file to the root directory of the classpath. The configuration file format refers to the JavaDoc of org.apache.seata.config.FileConfiguration.

There are four properties in seata.conf,

  1. shardingsphere.transaction.seata.at.enable, when this value is true, enable ShardingSphere’s Seata AT integration. The default value is true
  2. shardingsphere.transaction.seata.tx.timeout, global transaction timeout (seconds). The default value is 60
  3. client.application.id, application unique primary key, used to set applicationId of Seata Transaction Manager Client and Seata Resource Manager Client
  4. client.transaction.service.group, transaction group, used to set transactionServiceGroup of Seata Transaction Manager Client and Seata Resource Manager Client. The default value is default

A fully configured seata.conf is as follows,

shardingsphere.transaction.seata.at.enable = true
shardingsphere.transaction.seata.tx.timeout = 60

client {
    application.id = example
    transaction.service.group = default_tx_group
}

A minimally configured seata.conf is as follows. In seata.conf managed by ShardingSphere, the default value of client.transaction.service.group is set to default for historical reasons. Assuming that in the registry.conf of Seata Server and Seata Client used by the user, registry.type and config.type are both file, then for the .conf file configured by config.file.name of registry.conf, the default transaction group name is default_tx_group in apache/incubator-seata:v1.5.1 and later, and my_test_tx_group before apache/incubator-seata:v1.5.1.

client.application.id = example

Modify the registry.conf file of Seata as required.

Usage restrictions

ShardingSphere’s Seata integration does not support isolation levels.

ShardingSphere’s Seata integration places the obtained Seata global transaction into the thread’s local variables. And org.apache.seata.spring.annotation.GlobalTransactionScanner uses Dynamic Proxy to enhance the method. This means that when using ShardingSphere’s Seata integration, users should avoid using the Java API of org.apache.seata:seata-all, unless the user is mixing ShardingSphere’s Seata integration with the TCC mode feature of Seata Client.

For ShardingSphere data source, discuss 6 situations,

  1. Manually obtain the java.sql.Connection instance created from the ShardingSphere data source, and manually calling the setAutoCommit(), commit() and rollback() methods is allowed.

  2. Using the Jakarta EE 8 javax.transaction.Transactional annotation on the function is allowed.

  3. Using Jakarta EE 9/10’s jakarta.transaction.Transactional annotation on functions is allowed.

  4. Using Spring Framework’s org.springframework.transaction.annotation.Transactional annotation on functions is allowed.

  5. Using the org.apache.seata.spring.annotation.GlobalTransactional annotation on the function is not allowed.

  6. Manually create org.apache.seata.tm.api.GlobalTransaction instance from org.apache.seata.tm.api.GlobalTransactionContext, calling the begin(), commit() and rollback() methods of an org.apache.seata.tm.api.GlobalTransaction instance is not allowed.

In actual scenarios where Spring Boot is used, com.alibaba.cloud:spring-cloud-starter-alibaba-seata and org.apache.seata:seata-spring-boot-starter are often transitively imported by other Maven dependencies. To avoid transaction conflicts, users need to set the property seata.enable-auto-data-source-proxy to false in the Spring Boot configuration file. A possible dependency relationship is as follows.

<project>
     <dependencies>
       <dependency>
          <groupId>org.apache.shardingsphere</groupId>
          <artifactId>shardingsphere-jdbc</artifactId>
          <version>${shardingsphere.version}</version>
       </dependency>
       <dependency>
          <groupId>org.apache.shardingsphere</groupId>
          <artifactId>shardingsphere-transaction-base-seata-at</artifactId>
          <version>${shardingsphere.version}</version>
       </dependency>
       <dependency>
          <groupId>org.apache.seata</groupId>
          <artifactId>seata-spring-boot-starter</artifactId>
          <version>2.1.0</version>
          <exclusions>
             <exclusion>
                <groupId>org.antlr</groupId>
                <artifactId>antlr4-runtime</artifactId>
             </exclusion>
          </exclusions>
       </dependency>
     </dependencies>
</project>

The corresponding application.yml under classpath needs to contain the following configuration. In this case, the equivalent configuration of Seata’s registry.conf defined in Spring Boot’s application.yaml is still valid. When downstream projects use the Maven module of org.apache.shardingsphere:shardingsphere-transaction-base-seata-at, it is always encouraged to use registry.conf to configure Seata Client.

seata:
   enable-auto-data-source-proxy: false

Mixed use with Seata TCC mode features

For the case of setting up ShardingSphere’s Seata integration, In business functions unrelated to ShardingSphere JDBC DataSource, if you need to use Seata Client’s Seata TCC mode-related features in business functions, you can instantiate a non-proxy ordinary TCC interface implementation class, and then use org.apache.integration.tx.api.util.ProxyUtil to create a proxy TCC interface class, and call the functions corresponding to the three stages of the TCC interface implementation class Try, Confirm, and Cancel.

For the org.apache.seata.spring.annotation.GlobalTransactional annotation introduced by the Seata TCC mode or the business functions involved in the Seata TCC mode that need to interact with the database instance, ShardingSphere JDBC DataSource should not be used in the business functions marked by this annotation. Instead, a javax.sql.DataSource instance should be created manually or obtained from a custom Spring Bean.

Transactional propagation across service calls

Transactional propagationn in cross-service call scenarios is not as out-of-the-box as transaction operations within a single microservice. For Seata Server, transactional propagation in cross-service call scenarios requires passing XID to the service provider through service calls and binding it to org.apache.seata.core.context.RootContext. Refer to https://seata.apache.org/docs/user/api/ . This requires discussing two situations,

  1. In the scenario of using ShardingSphere JDBC, transaction scenarios across multiple microservices need to consider using org.apache.seata.core.context.RootContext.getXID() to obtain Seata XID in the context of the starting microservice, and passing it to the end microservice through HTTP or RPC, and processing it in the Filter or Spring WebMVC HandlerInterceptor of the end microservice. Spring WebMVC HandlerInterceptor is only applicable to Spring Boot microservices and is invalid for Quarkus, Micronaut Framework and Helidon.

  2. In the scenario of using ShardingSphere Proxy, multiple microservices operate local transactions against the logical data source of ShardingSphere Proxy. This will be converted into distributed transaction operations on the server side of ShardingSphere Proxy, without considering additional Seata XID.

Introduce a simple scenario to continue discussing the transactional propagation across service calls in the scenario of using ShardingSphere JDBC.

  1. MySQL database instance a-mysql, all databases have created UNDO_LOG table and business table.
  2. MySQL database instance b-mysql, all databases have created UNDO_LOG table and business table.
  3. Seata Server instance a-seata-server using file as configuration center and registration center.
  4. Microservice instance a-service. This microservice creates a ShardingSphere JDBC DataSource that only configures the database instance a-mysql. This ShardingSphere JDBC DataSource configuration uses the Seata AT integration connected to the Seata Server instance a-seata-server, whose Seata Application Id is service-a, whose Seata transaction group is default_tx_group, and the Seata Transaction Coordinator cluster group pointed to by its Virtual Group Mapping is default. This microservice instance a-service exposes a single Restful API GET endpoint as /hello, and the business function aMethod of this Restful API endpoint uses a common local transaction annotation. If this microservice is based on Spring Boot 2,
import org.springframework.transaction.annotation.Transactional;
import org.springframework.web.bind.annotation.GetMapping;
import org.springframework.web.bind.annotation.RestController;

@RestController
public class DemoController {
   @Transactional
   @GetMapping("/hello")
   public String aMethod() {
      // ... Perform an UPDATE operation on the database instance `a-mysql`
      return "Hello World!";
   }
}
  1. Microservice instance b-service. This microservice creates a ShardingSphere JDBC DataSource that only configures the database instance b-mysql. This ShardingSphere JDBC DataSource configuration uses the Seata AT integration connected to the Seata Server instance a-seata-server, whose Seata Application Id is service-b, whose Seata transaction group is default_tx_group, and whose Virtual Group Mapping points to the Seata Transaction Coordinator cluster group as default. The business function bMethod of this microservice instance b-service uses a normal local transaction annotation, and calls the /hello Restful API endpoint of the microservice instance a-service through the HTTP Client in bMethod. If this microservice is based on Spring Boot 2,
import org.springframework.boot.web.client.RestTemplateBuilder;
import org.springframework.stereotype.Service;
import org.springframework.transaction.annotation.Transactional;
import org.springframework.web.client.RestTemplate;

@Service
public class DemoService {
   @Transactional
   public void bMethod() {
      RestTemplate restTemplate = new RestTemplateBuilder().build();
      restTemplate.getForEntity("http://a-service/hello", String.class);
      // ... Perform an UPDATE operation on the database instance `b-mysql`
   }
}

For this simple scenario, there is a single Seata Server Cluster, which contains a single Virtual Group as default. This Virtual Group contains a single Seata Server instance as a-seata-server.

Discuss transaction propagation for single service calls. When the business function aMethod of the microservice instance a-service throws an exception, the changes to the MySQL database instance a-mysql in the business function will be rolled back normally.

Discuss transaction propagation for cross-service calls. When the business function bMethod of the microservice instance b-service throws an exception, the changes to the MySQL database instance b-mysql in the business function will be rolled back normally, and the org.apache.seata.core.context.RootContext of the microservice instance a-service is not bound to the Seata XID of the business function bMethod of the microservice instance b-service, so the changes to the MySQL database instance a-mysql in the business function will not be rolled back.

In order to achieve that when the business function bMethod of the microservice instance b-service throws an exception, the changes to the MySQL database instances a-mysql and b-mysql in the business function are rolled back normally, discuss the common processing solutions in different scenarios.

  1. The microservice instances a-service and b-service are both Spring Boot 2 microservices based on Jakarta EE 8. Users can use org.springframework.web.client.RestTemplate in the business function bMethod of the microservice instance b-service to pass the XID to the microservice instance a-service through the service call. The possible transformation logic is as follows.
import org.apache.seata.core.context.RootContext;
import org.springframework.boot.web.client.RestTemplateBuilder;
import org.springframework.stereotype.Service;
import org.springframework.transaction.annotation.Transactional;
import org.springframework.web.client.RestTemplate;

@Service
public class DemoService {
    @Transactional
    public void bMethod() {
        RestTemplate restTemplate = new RestTemplateBuilder().additionalInterceptors((request, body, execution) -> {
                    String xid = RootContext.getXID();
                    if (null != xid) {
                        request.getHeaders().add(RootContext.KEY_XID, xid);
                    }
                    return execution.execute(request, body);
                })
                .build();
        restTemplate.getForEntity("http://a-service/hello", String.class);
        // ... Perform an UPDATE operation on the database instance `b-mysql`
    }
}

At this time, custom org.springframework.web.servlet.config.annotation.WebMvcConfigurer implementations need to be added to the microservice instances a-service and b-service.

import org.apache.seata.integration.http.TransactionPropagationInterceptor;
import org.springframework.context.annotation.Configuration;
import org.springframework.web.servlet.config.annotation.InterceptorRegistry;
import org.springframework.web.servlet.config.annotation.WebMvcConfigurer;

@Configuration
public class CustomWebMvcConfigurer implements WebMvcConfigurer {

    @Override
    public void addInterceptors(InterceptorRegistry registry) {
        registry.addInterceptor(new TransactionPropagationInterceptor());
    }
}

At this time, when the business function bMethod of the microservice instance b-service throws an exception, the changes to the MySQL database instances a-mysql and b-mysql in the business function are rolled back normally.

  1. The microservice instances a-service and b-service are both Spring Boot 3 microservices based on Jakarta EE 9/10. Users can use org.springframework.web.client.RestClient in the business function bMethod of the microservice instance b-service to pass the XID to the microservice instance a-service through a service call. The possible transformation logic is as follows.
import org.apache.seata.core.context.RootContext;
import org.springframework.stereotype.Service;
import org.springframework.transaction.annotation.Transactional;
import org.springframework.web.client.RestClient;

@Service
public class DemoService {
    @Transactional
    public void bMethod() {
        RestClient restClient = RestClient.builder().requestInterceptor((request, body, execution) -> {
                    String xid = RootContext.getXID();
                    if (null != xid) {
                        request.getHeaders().add(RootContext.KEY_XID, xid);
                    }
                    return execution.execute(request, body);
                })
                .build();
        restClient.get().uri("http://a-service/hello").retrieve().body(String.class);
        // ... Perform an UPDATE operation on the database instance `b-mysql`
    }
}

At this time, custom org.springframework.web.servlet.config.annotation.WebMvcConfigurer implementations need to be added to the microservice instances a-service and b-service.

import org.apache.seata.integration.http.JakartaTransactionPropagationInterceptor;
import org.springframework.context.annotation.Configuration;
import org.springframework.web.servlet.config.annotation.InterceptorRegistry;
import org.springframework.web.servlet.config.annotation.WebMvcConfigurer;

@Configuration
public class CustomWebMvcConfigurer implements WebMvcConfigurer {

   @Override
   public void addInterceptors(InterceptorRegistry registry) {
      registry.addInterceptor(new JakartaTransactionPropagationInterceptor());
   }
}

At this time, when the business function bMethod of the microservice instance b-service throws an exception, the changes to the MySQL database instances a-mysql and b-mysql in the business function are rolled back normally.

  1. The microservice instances a-service and b-service are both Spring Boot microservices, but the API gateway middleware used blocks all HTTP requests containing the HTTP Header of TX_XID. The user needs to consider changing the HTTP Header used to pass XID to the microservice instance a-service through service calls, or use the RPC framework to pass XID to the microservice instance a-service through service calls. Refer to https://github.com/apache/incubator-seata/tree/v2.1.0/integration .

  2. The microservice instances a-service and b-service are both microservices such as Quarkus, Micronaut Framework and Helidon. In this case, Spring WebMVC HandlerInterceptor cannot be used. You can refer to the following Spring Boot 3 custom WebMvcConfigurer implementation to implement Filter.

import org.apache.seata.common.util.StringUtils;
import org.apache.seata.core.context.RootContext;
import jakarta.servlet.http.HttpServletRequest;
import jakarta.servlet.http.HttpServletResponse;
import org.springframework.context.annotation.Configuration;
import org.springframework.lang.NonNull;
import org.springframework.web.servlet.HandlerInterceptor;
import org.springframework.web.servlet.config.annotation.InterceptorRegistry;
import org.springframework.web.servlet.config.annotation.WebMvcConfigurer;

@Configuration
public class CustomWebMvcConfigurer implements WebMvcConfigurer {
    @Override
    public void addInterceptors(InterceptorRegistry registry) {
        registry.addInterceptor(new HandlerInterceptor() {
            @Override
            public boolean preHandle(@NonNull HttpServletRequest request, @NonNull HttpServletResponse response, @NonNull Object handler) {
                String rpcXid = request.getHeader(RootContext.KEY_XID);
                String xid = RootContext.getXID();
                if (StringUtils.isBlank(xid) && StringUtils.isNotBlank(rpcXid)) {
                    RootContext.bind(rpcXid);
                }
                return true;
            }

            @Override
            public void afterCompletion(@NonNull HttpServletRequest request, @NonNull HttpServletResponse response, @NonNull Object handler, Exception ex) {
                if (RootContext.inGlobalTransaction()) {
                    String rpcXid = request.getHeader(RootContext.KEY_XID);
                    String xid = RootContext.getXID();
                    if (StringUtils.isNotBlank(xid)) {
                        String unbindXid = RootContext.unbind();
                        if (!StringUtils.equalsIgnoreCase(rpcXid, unbindXid)) {
                            if (StringUtils.isNotBlank(unbindXid)) {
                                RootContext.bind(unbindXid);
                            }
                        }
                    }
                }
            }
        });
    }
}
  1. Both microservice instances a-service and b-service are Spring Boot microservices, but the components used are Spring WebFlux instead of Spring WebMVC. ShardingSphere JDBC cannot handle R2DBC DataSource under the reactive programming API, only JDBC DataSource. Avoid creating ShardingSphere JDBC DataSource in Spring Boot microservices using WebFlux components.