Spring Cloud Sleuth基于DataSource的链路跟踪

• Spring Cloud Sleuth基于DataSource的链路跟踪
• • 现状
  • p6spy和datasource-proxy
  • 原理
  • 集成步骤

最新的spring-cloud-sleuth(3.1.0-M1)已经支持基于datasource/jdbc的链路跟踪了，相关文档请参考：sleuth-jdbc-integration

并且spring-cloud-sleuth提供了两套解决方案自由选择：

• 基于JDBC的 p6spy
• 基于DataSource的 datasource-proxy

两种解决方案各有优劣，这里只讨论基于Datasource的datasource-proxy。

虽然最新的spring-cloud-sleuth已经有了数据源链路跟踪的集成方案，但是截止到2021-09-17，中央仓库的spring-cloud-sleuth版本还停留在3.0.4，这个版本并没有支持数据源链路跟踪，并且实际情况中还在使用spring-cloud-sleuth 2.x甚至更低版本，如何在低版中集成数据源链路跟踪呢？

这里基于新版的spring-cloud-sleuth源码和datasource-proxy与spring boot的集成(spring-cloud-sleuth参考了spring-boot-data-source-decorator)，用最简单的方式支持数据源的链路跟踪——不管使用的什么数据库，不管使用了什么JDBC，只要是使用了spring-datasource，都可以方便集成。

为什么不使用p6spy呢？

和mysql链路跟踪(brave-instrumentation-mysql)类似，p6spy严重依赖JDBC实现，虽然p6spy支持了很多种常见的数据库Driver，但是总感觉不放心。

反正一定会使用spring-datasource，不如基于这一层抽象来做扩展，完全忽略数据库Driver的差异，这样即使切换数据库Driver，也不会有影响，多数据源的情况下，如果设置多个Driver也无需担心不支持的问题。

即使不使用spring-datasource，那么也一定会有另外一个抽象来代替spring datasource，如果替换了spring datasource，也只需基于新的抽象实现一套链路跟踪逻辑即可。

所以最终选择了datasource-proxy。

datasource-proxy的原理这里仅简述一下，具体请参考datasource-proxy

其原理可以概括为：

1. 基于JDK动态代理
2. 代理java.sql.*/javax.sql.*相关类，在Datasource/Connection/Statement执行getConnection()/createStatment()/executeQuery()/executeUpdate()等方法前后埋点；
3. 埋点位置和操作抽象成为listener，实现listener并在生成代理类的时候配置listener，就可以完成日志打印、链路跟踪等功能；
4. spring boot集成时，实现一个listener，在获取连接和执行sql前后，基于spring-cloud-sleuth埋点，完成链路跟踪；
5. 在DataSource初始化完成后(BeanPostProcessor.postProcessAfterInitialization())，替换DataSource为data-source-proxy代理类。

这里仅讨论基于spring-boot/spring-cloud 2.x 的集成。

1. 引入spring-cloud-sleuth和datasource-proxy：

   
     org.springframework.cloud
     spring-cloud-starter-sleuth
     2.2.10-RELEASE
   
   
   
     net.ttddyy
     datasource-proxy
     1.7
   
   

2. 然后定义一些配置：

    # 是否打开datasource链路监控，默认打开
    spring.sleuth.datasource.enabled=true
    # 支持那些阶段的链路监控，包含连接、查询和获取结果，默认全部
    spring.sleuth.datasource.include=CONNECTION,QUERY,FETCH
    # 默认数据库服务名称，localEndpoint.serviceName，默认default-datasource
    spring.sleuth.datasource.default-name=default-datasource
    # 多数据源的情况下，可以指定不停数据源的服务名称，key为DataSource在spring容器中的bean名称
    # 如果未指定，将使用默认数据库服务名称
    spring.sleuth.datasource.name-mapping.orderDataDource=order-mysql
    spring.sleuth.datasource.name-mapping.productDataDource=product-oracle
   

3. 定义配置类：

    @ConfigurationProperties(prefix = "spring.sleuth.datasource")
    @Data
   public class SleuthDatasourceProperties {
      
      private boolean enabled = true;
    
      
      private List include = Arrays.asList(TraceType.CONNECTION, TraceType.QUERY, TraceType.FETCH);
    
      
      private String defaultName = "default-datasource";
    
      
      private Map nameMapping = new HashMap<>(4);
    
      
      public enum TraceType {
        
        CONNECTION,
        
        QUERY,
        
        FETCH
      }
    }
   

4. 编写链路监听策略类，定义方法埋点前后逻辑：

   public class TracingListenerStrategy {
   
       private static final String SPAN_SQL_QUERY_TAG_NAME = "query-sql";
   
       private static final String SPAN_ROW_COUNT_TAG_NAME = "row-count";
   
       private static final String SPAN_CONNECTION_POSTFIX = "/connection";
   
       private static final String SPAN_QUERY_POSTFIX = "/query";
   
       private static final String SPAN_FETCH_POSTFIX = "/fetch";
   
       private static final String JDBC_PREFIX = "jdbc:/";
   
       private final Map openConnections = new ConcurrentHashMap<>();
   
       private final Tracer tracer;
   
       private final List traceTypes;
   
       TracingListenerStrategy(Tracer tracer, List traceTypes) {
           this.tracer = tracer;
           this.traceTypes = traceTypes;
       }
   
       void beforeGetConnection(C connectionKey, String dataSourceName) {
           SpanWithScope spanWithScope = null;
           if (traceTypes.contains(KocaTraceProperties.Database.TraceType.CONNECTION)) {
               Span connectionSpan = tracer.nextSpan().name(JDBC_PREFIX + dataSourceName + SPAN_CONNECTION_POSTFIX);
               connectionSpan.remoteServiceName(dataSourceName);
               connectionSpan.kind(Span.Kind.CLIENT);
               connectionSpan.start();
               spanWithScope = new SpanWithScope(connectionSpan, tracer.withSpanInScope(connectionSpan));
           }
           ConnectionInfo connectionInfo = new ConnectionInfo(spanWithScope);
           openConnections.put(connectionKey, connectionInfo);
       }
   
       void afterGetConnection(C connectionKey, Throwable t) {
           if (t != null) {
               ConnectionInfo connectionInfo = openConnections.remove(connectionKey);
               connectionInfo.getSpan().ifPresent(connectionSpan -> {
                   connectionSpan.getSpan().error(t);
                   connectionSpan.finish();
               });
           }
       }
   
       void beforeQuery(C connectionKey, S statementKey, String dataSourceName) {
           SpanWithScope spanWithScope = null;
           if (traceTypes.contains(KocaTraceProperties.Database.TraceType.QUERY)) {
               Span statementSpan = tracer.nextSpan().name(JDBC_PREFIX + dataSourceName + SPAN_QUERY_POSTFIX);
               statementSpan.remoteServiceName(dataSourceName);
               statementSpan.kind(Span.Kind.CLIENT);
               statementSpan.start();
               spanWithScope = new SpanWithScope(statementSpan, tracer.withSpanInScope(statementSpan));
           }
           StatementInfo statementInfo = new StatementInfo(spanWithScope);
           ConnectionInfo connectionInfo = openConnections.get(connectionKey);
           if (connectionInfo == null) {
               // Connection may be closed after statement preparation, but before statement execution.
               return;
           }
           connectionInfo.getNestedStatements().put(statementKey, statementInfo);
       }
   
       void addQueryRowCount(C connectionKey, S statementKey, int rowCount) {
           ConnectionInfo connectionInfo = openConnections.get(connectionKey);
           if (connectionInfo == null) {
               // Connection is already closed
               return;
           }
           StatementInfo statementInfo = connectionInfo.getNestedStatements().get(statementKey);
           statementInfo.getSpan().ifPresent(statementSpan -> {
               statementSpan.getSpan().tag(SPAN_ROW_COUNT_TAG_NAME, String.valueOf(rowCount));
           });
       }
   
       void afterQuery(C connectionKey, S statementKey, String sql, Throwable t) {
           ConnectionInfo connectionInfo = openConnections.get(connectionKey);
           if (connectionInfo == null) {
               // Connection may be closed after statement preparation, but before statement execution.
               return;
           }
           StatementInfo statementInfo = connectionInfo.getNestedStatements().get(statementKey);
           statementInfo.getSpan().ifPresent(statementSpan -> {
               statementSpan.getSpan().tag(SPAN_SQL_QUERY_TAG_NAME, sql);
               if (t != null) {
                   statementSpan.getSpan().error(t);
               }
               statementSpan.finish();
           });
       }
   
       void beforeResultSetNext(C connectionKey, S statementKey, R resultSetKey, String dataSourceName) {
           if (!traceTypes.contains(KocaTraceProperties.Database.TraceType.FETCH)) {
               return;
           }
           ConnectionInfo connectionInfo = openConnections.get(connectionKey);
           // ConnectionInfo may be null if Connection was closed before ResultSet
           if (connectionInfo == null) {
               return;
           }
           if (connectionInfo.getNestedResultSetSpans().containsKey(resultSetKey)) {
               // ResultSet span is already created
               return;
           }
           Span resultSetSpan = tracer.nextSpan().name(JDBC_PREFIX + dataSourceName + SPAN_FETCH_POSTFIX);
           resultSetSpan.remoteServiceName(dataSourceName);
           resultSetSpan.kind(Span.Kind.CLIENT);
           resultSetSpan.start();
           SpanWithScope spanWithScope = new SpanWithScope(resultSetSpan, tracer.withSpanInScope(resultSetSpan));
           connectionInfo.getNestedResultSetSpans().put(resultSetKey, spanWithScope);
   
           StatementInfo statementInfo = connectionInfo.getNestedStatements().get(statementKey);
           // StatementInfo may be null when Statement is proxied and instance returned from ResultSet is different from
           // instance returned in query method
           // in this case if Statement is closed before ResultSet span won't be finished immediately, but when 
           // Connection is closed
           if (statementInfo != null) {
               statementInfo.getNestedResultSetSpans().put(resultSetKey, spanWithScope);
           }
       }
   
       void afterResultSetClose(C connectionKey, R resultSetKey, int rowCount, Throwable t) {
           ConnectionInfo connectionInfo = openConnections.get(connectionKey);
           // ConnectionInfo may be null if Connection was closed before ResultSet
           if (connectionInfo == null) {
               return;
           }
           SpanWithScope resultSetSpan = connectionInfo.getNestedResultSetSpans().remove(resultSetKey);
           // ResultSet span may be null if Statement or ResultSet were already closed
           if (resultSetSpan == null) {
               return;
           }
   
           if (rowCount != -1) {
               resultSetSpan.getSpan().tag(SPAN_ROW_COUNT_TAG_NAME, String.valueOf(rowCount));
           }
           if (t != null) {
               resultSetSpan.getSpan().error(t);
           }
           resultSetSpan.finish();
       }
   
       void afterStatementClose(C connectionKey, S statementKey) {
           ConnectionInfo connectionInfo = openConnections.get(connectionKey);
           // ConnectionInfo may be null if Connection was closed before Statement
           if (connectionInfo == null) {
               return;
           }
           StatementInfo statementInfo = connectionInfo.getNestedStatements().remove(statementKey);
           if (statementInfo != null) {
               statementInfo.getNestedResultSetSpans().forEach((resultSetKey, span) -> {
                   connectionInfo.getNestedResultSetSpans().remove(resultSetKey);
                   span.finish();
               });
               statementInfo.getNestedResultSetSpans().clear();
           }
       }
   
       void afterCommit(C connectionKey, Throwable t) {
           ConnectionInfo connectionInfo = openConnections.get(connectionKey);
           if (connectionInfo == null) {
               // Connection is already closed
               return;
           }
           connectionInfo.getSpan().ifPresent(connectionSpan -> {
               if (t != null) {
                   connectionSpan.getSpan().error(t);
               }
               connectionSpan.getSpan().annotate("commit");
           });
       }
   
       void afterRollback(C connectionKey, Throwable t) {
           ConnectionInfo connectionInfo = openConnections.get(connectionKey);
           if (connectionInfo == null) {
               // Connection is already closed
               return;
           }
           connectionInfo.getSpan().ifPresent(connectionSpan -> {
               if (t != null) {
                   connectionSpan.getSpan().error(t);
               } else {
                   connectionSpan.getSpan().tag("error", "Transaction rolled back");
               }
               connectionSpan.getSpan().annotate("rollback");
           });
       }
   
       void afterConnectionClose(C connectionKey, Throwable t) {
           ConnectionInfo connectionInfo = openConnections.remove(connectionKey);
           if (connectionInfo == null) {
               // connection is already closed
               return;
           }
           connectionInfo.getNestedResultSetSpans().values().forEach(SpanWithScope::finish);
           connectionInfo.getNestedStatements().values()
               .forEach(statementInfo -> statementInfo.getSpan().ifPresent(SpanWithScope::finish));
           connectionInfo.getSpan().ifPresent(connectionSpan -> {
               if (t != null) {
                   connectionSpan.getSpan().error(t);
               }
               connectionSpan.finish();
           });
       }
   
       private final class ConnectionInfo {
   
           private final SpanWithScope span;
   
           private final Map nestedStatements = new ConcurrentHashMap<>();
   
           private final Map nestedResultSetSpans = new ConcurrentHashMap<>();
   
           private ConnectionInfo(SpanWithScope span) {
               this.span = span;
           }
   
           Optional getSpan() {
               return Optional.ofNullable(span);
           }
   
           Map getNestedStatements() {
               return nestedStatements;
           }
   
           Map getNestedResultSetSpans() {
               return nestedResultSetSpans;
           }
       }
   
       private final class StatementInfo {
   
           private final SpanWithScope span;
   
           private final Map nestedResultSetSpans = new ConcurrentHashMap<>();
   
           private StatementInfo(SpanWithScope span) {
               this.span = span;
           }
   
           Optional getSpan() {
               return Optional.ofNullable(span);
           }
   
           Map getNestedResultSetSpans() {
               return nestedResultSetSpans;
           }
       }
   
       private static final class SpanWithScope {
   
           private final Span span;
   
           private final Tracer.SpanInScope spanInScope;
   
           private SpanWithScope(Span span, Tracer.SpanInScope spanInScope) {
               this.span = span;
               this.spanInScope = spanInScope;
           }
   
           Span getSpan() {
               return span;
           }
   
           void finish() {
               spanInScope.close();
               span.finish();
           }
       }
   
   }
   

   由于数据库SQL执行是异步的，有CONNECTION、STATEMENT、FETCH等阶段，JDK动态代理拦截各个方法时，并不知道这个方法属于那一次数据库调用，但是同一次数据库调用的数据库连接信息是一样的，因此需要在before CONNECTION阶段将Span信息暂存在Map中，Map的Key即是数据库连接信息，after CONNECTION阶段取出数据库连接对应的Span并处理，STATEMENT和FETCH阶段与CONNECTION类似。

5. 实现链路监听：

   public class TracingQueryExecutionListener implements QueryExecutionListener, MethodExecutionListener, Ordered {
   
       private static final String EXECUTE_UPDATE = "executeUpdate";
   
       private static final String GET_ConNECTION = "getConnection";
   
       private static final String NEXT = "next";
   
       private static final String COMMIT = "commit";
   
       private static final String ROLLBACK = "rollback";
   
       private static final String CLOSE = "close";
   
       private final TracingListenerStrategy strategy;
   
       TracingQueryExecutionListener(Tracer tracer, List traceTypes) {
           this.strategy = new TracingListenerStrategy<>(tracer, traceTypes);
       }
   
       @Override
       public void beforeQuery(ExecutionInfo execInfo, List queryInfoList) {
           strategy.beforeQuery(execInfo.getConnectionId(), execInfo.getStatement(), execInfo.getDataSourceName());
       }
   
       @Override
       public void afterQuery(ExecutionInfo execInfo, List queryInfoList) {
           if (EXECUTE_UPDATE.equals(execInfo.getMethod().getName()) && execInfo.getThrowable() == null) {
               strategy.addQueryRowCount(execInfo.getConnectionId(), execInfo.getStatement(), (int) execInfo.getResult());
           }
           String sql = queryInfoList.stream().map(QueryInfo::getQuery).collect(Collectors.joining("n"));
           strategy.afterQuery(execInfo.getConnectionId(), execInfo.getStatement(), sql, execInfo.getThrowable());
       }
   
       @Override
       public void beforeMethod(MethodExecutionContext executionContext) {
           Object target = executionContext.getTarget();
           String methodName = executionContext.getMethod().getName();
           String dataSourceName = executionContext.getProxyConfig().getDataSourceName();
           String connectionId = executionContext.getConnectionInfo().getConnectionId();
           if (target instanceof DataSource) {
               if (GET_CONNECTION.equals(methodName)) {
                   strategy.beforeGetConnection(connectionId, dataSourceName);
               }
           } else if (target instanceof ResultSet) {
               ResultSet resultSet = (ResultSet) target;
               if (NEXT.equals(methodName)) {
                   try {
                       strategy.beforeResultSetNext(connectionId, resultSet.getStatement(), resultSet, dataSourceName);
                   } catch (SQLException ignore) {
                       // ignore
                   }
               }
           }
       }
   
       @Override
       public void afterMethod(MethodExecutionContext executionContext) {
           Object target = executionContext.getTarget();
           String methodName = executionContext.getMethod().getName();
           String connectionId = executionContext.getConnectionInfo().getConnectionId();
           Throwable t = executionContext.getThrown();
           if (target instanceof DataSource) {
               if (GET_CONNECTION.equals(methodName)) {
                   strategy.afterGetConnection(connectionId, t);
               }
           } else if (target instanceof Connection) {
               if (COMMIT.equals(methodName)) {
                   strategy.afterCommit(connectionId, t);
               } else if (ROLLBACK.equals(methodName)) {
                   strategy.afterRollback(connectionId, t);
               } else if (CLOSE.equals(methodName)) {
                   strategy.afterConnectionClose(connectionId, t);
               }
           } else if (target instanceof Statement && CLOSE.equals(methodName)) {
               strategy.afterStatementClose(connectionId, (Statement) target);
   
           } else if (target instanceof ResultSet && CLOSE.equals(methodName)) {
               ResultSet resultSet = (ResultSet) target;
               strategy.afterResultSetClose(connectionId, resultSet, -1, t);
   
           }
       }
   
       @Override
       public int getOrder() {
           return Ordered.HIGHEST_PRECEDENCE + 10;
       }
   }
   

6. 添加配置类：

   @Configuration
   @EnableConfigurationProperties(SleuthDatasourceProperties.class)
   @ConditionalOnProperty(
       name = {"spring.sleuth.enabled", "spring.sleuth.datasource.enabled"},
       havingValue = "true", matchIfMissing = true)
   @ConditionalOnBean({DataSource.class, Tracer.class})
   @AutoConfigureAfter({DataSourceAutoConfiguration.class, TraceAutoConfiguration.class})
   public class DataSourceTraceConfig {
   
       @Bean
       public TracingQueryExecutionListener tracingQueryExecutionListener(Tracer tracer,
           KocaTraceProperties kocaTraceProperties) {
           return new TracingQueryExecutionListener(tracer, kocaTraceProperties.getDatabase().getInclude());
       }
   
       @Bean
       public DatasourceProxyBeanPostProcessor datasourceProxyBeanPostProcessor(TracingQueryExecutionListener listener,
           KocaTraceProperties kocaTraceProperties) {
           return new DatasourceProxyBeanPostProcessor(listener, kocaTraceProperties);
       }
   
       static class DatasourceProxyBeanPostProcessor implements BeanPostProcessor {
   
           private TracingQueryExecutionListener listener;
   
           private KocaTraceProperties kocaTraceProperties;
   
           DatasourceProxyBeanPostProcessor(TracingQueryExecutionListener listener,
               KocaTraceProperties kocaTraceProperties) {
               this.listener = listener;
               this.kocaTraceProperties = kocaTraceProperties;
           }
   
           @Override
           public Object postProcessAfterInitialization(Object bean, String beanName) throws BeansException {
               if (bean instanceof DataSource && !(bean instanceof ProxyDataSource) &&
                   !ScopedProxyUtils.isScopedTarget(beanName)) {
                   final ProxyFactory factory = new ProxyFactory(bean);
                   factory.setProxyTargetClass(true);
                   factory.addAdvice(new ProxyDataSourceInterceptor((DataSource) bean, listener, beanName,
                       kocaTraceProperties.getDatabase()));
                   return factory.getProxy();
               }
               return bean;
           }
   
           @Override
           public Object postProcessBeforeInitialization(Object bean, String beanName) throws BeansException {
               return bean;
           }
   
           public static class ProxyDataSourceInterceptor implements MethodInterceptor {
   
               private final DataSource dataSource;
   
               ProxyDataSourceInterceptor(final DataSource dataSource, final TracingQueryExecutionListener listener,
                   String beanName, KocaTraceProperties.Database database) {
                   this.dataSource = ProxyDataSourceBuilder.create(dataSource)
                       .name(database.getNameMapping().getOrDefault(beanName, database.getDefaultName()))
                       .methodListener(listener).listener(listener).build();
               }
   
               @Override
               public Object invoke(final MethodInvocation invocation) throws Throwable {
                   final Method proxyMethod =
                       ReflectionUtils.findMethod(this.dataSource.getClass(), invocation.getMethod().getName());
                   if (proxyMethod != null) {
                       return proxyMethod.invoke(this.dataSource, invocation.getArguments());
                   }
                   return invocation.proceed();
               }
           }
       }
   
   }
   

7. 使用zipkin收集链路并展示效果如下：