sharding-jdbc源码解析之sql单表或绑定表路由

时间 2019-11-06

标签 sharding jdbc 源码解析 sql 绑定路由栏目 Java 繁體版

原文原文链接

说在前面sql

本文转自“天河聊技术”微信公众号数据库

sql路由这里的内容比较多，包含单表路由或者绑定表路由、多库多表路由、笛卡尔积路由，分三部分来介绍，今天先介绍单表或绑定表路由。微信

sql路由源码解析app

com.dangdang.ddframe.rdb.sharding.routing.PreparedStatementRoutingEngine、com.dangdang.ddframe.rdb.sharding.routing.StatementRoutingEngine两个sql路由引擎类，预编译的用的比较多，咱们以预编译的Statement的引擎类来跟踪下sharding-jdbc是对sql怎么进行路由的。ide

上层sql执行器接收到逻辑sql后再进行sql路由的时候会建立预编译statement对象的路由器，所以会调用其构造器性能

/**
 * 预解析的SQL路由器.
 * 
 * @author zhangliang
 */
public final class PreparedStatementRoutingEngine {

//    逻辑sql
    private final String logicSQL;

//sql路由器
    private final SQLRouter sqlRouter;

//    sql语句对象
    private SQLStatement sqlStatement;
    
    public PreparedStatementRoutingEngine(final String logicSQL, final ShardingContext shardingContext) {
        this.logicSQL = logicSQL;
        sqlRouter = SQLRouterFactory.createSQLRouter(shardingContext);
    }

/**
 * 路由引擎工厂.
 * 
 * @author zhangiang
 */
@NoArgsConstructor(access = AccessLevel.PRIVATE)
public final class SQLRouterFactory {
    
    /**
     * 建立SQL路由器.
     * 
     * @param shardingContext 数据源运行期上下文
     * @return SQL路由器
     */
//    这里是静态工厂方法实现
    public static SQLRouter createSQLRouter(final ShardingContext shardingContext) {
        return HintManagerHolder.isDatabaseShardingOnly() ? new DatabaseHintSQLRouter(shardingContext) : new ParsingSQLRouter(shardingContext);
    }
}

接下来会建立ParsingSQLRouter对象ui

**
 * 须要解析的SQL路由器.
 * 
 * @author zhangiang
 */
public final class ParsingSQLRouter implements SQLRouter {

//    分库分表配置对象
    private final ShardingRule shardingRule;

//    支持的数据库类型
    private final DatabaseType databaseType;

//    是否要展现sql
    private final boolean showSQL;
    
    private final List<Number> generatedKeys;

//    上面这些属性值都是存储在分片上下文中
    public ParsingSQLRouter(final ShardingContext shardingContext) {
        shardingRule = shardingContext.getShardingRule();
        databaseType = shardingContext.getDatabaseType();
        showSQL = shardingContext.isShowSQL();
        generatedKeys = new LinkedList<>();
    }

这个方法是sql路由的入口方法this

/**
 * SQL路由.
 * 当第一次路由时进行SQL解析,以后的路由复用第一次的解析结果.
 * 
 * @param parameters SQL中的参数
 * @return 路由结果
 */
public SQLRouteResult route(final List<Object> parameters) {//sql路由业务方法
    if (null == sqlStatement) {
        sqlStatement = sqlRouter.parse(logicSQL, parameters.size());
    }
    return sqlRouter.route(logicSQL, parameters, sqlStatement);
}

进入到这个parse方法对象

sqlStatement = sqlRouter.parse(logicSQL, parameters.size());

@Override
    public SQLStatement parse(final String logicSQL, final int parametersSize) {
//        建立sql解析引擎
        SQLParsingEngine parsingEngine = new SQLParsingEngine(databaseType, logicSQL, shardingRule);
//        开启度量上下文
        Context context = MetricsContext.start("Parse SQL");
        
//        sql解析器解析得到sql语句对象
        SQLStatement result = parsingEngine.parse();
        if (result instanceof InsertStatement) {
            ((InsertStatement) result).appendGenerateKeyToken(shardingRule, parametersSize);
        }
        MetricsContext.stop(context);
        return result;
    }

进入下面的sql路由方法，返回路由结果ip

return sqlRouter.route(logicSQL, parameters, sqlStatement);

private RoutingResult route(final List<Object> parameters, final SQLStatement sqlStatement) {
        Collection<String> tableNames = sqlStatement.getTables().getTableNames();
        RoutingEngine routingEngine;
//        若是表集合是1，或者是绑定表路由就走简单路由规则
        if (1 == tableNames.size() || shardingRule.isAllBindingTables(tableNames)) {
            routingEngine = new SimpleRoutingEngine(shardingRule, parameters, tableNames.iterator().next(), sqlStatement);//单表路由
        } else {
            // TODO 可配置是否执行笛卡尔积
            routingEngine = new ComplexRoutingEngine(shardingRule, parameters, tableNames, sqlStatement);
        }
        return routingEngine.route();//tianhe TODO 笛卡尔积
    }

建立简单路由引擎
routingEngine = new SimpleRoutingEngine(shardingRule, parameters, tableNames.iterator().next(), sqlStatement);//单表路由

/**
 * 简单路由引擎.
 * 
 * @author zhangliang
 */
@RequiredArgsConstructor
public final class SimpleRoutingEngine implements RoutingEngine {
//    分库分表配置对象
    private final ShardingRule shardingRule;
//    sql参数
    private final List<Object> parameters;

//    逻辑表名
    private final String logicTableName;

//    sql语句对象
    private final SQLStatement sqlStatement;

不是单表路由，就走多库多表路由引擎，建立多库多表路由对象

routingEngine = new ComplexRoutingEngine(shardingRule, parameters, tableNames, sqlStatement);

/**
 * 混合多库表路由引擎.
 * 
 * @author gaohongtao
 * @author zhangliang
 */
@RequiredArgsConstructor
@Slf4j
public final class ComplexRoutingEngine implements RoutingEngine {
    
    private final ShardingRule shardingRule;
    
    private final List<Object> parameters;
    
    private final Collection<String> logicTables;
    
    private final SQLStatement sqlStatement;

return routingEngine.route();//tianhe TODO 笛卡尔积

这里是路由逻辑，这里有三种实现，一种是单表或者绑定表路由，一种是多库多表路由，一种是笛卡尔积路由

单表或者绑定表路由

@Override
    public RoutingResult route() {
//        根据逻辑表名得到表规则配置对象
        TableRule tableRule = shardingRule.getTableRule(logicTableName);
//        根据表规则配置对象得到数据源集合
        Collection<String> routedDataSources = routeDataSources(tableRule);
        Map<String, Collection<String>> routedMap = new LinkedHashMap<>(routedDataSources.size());
        for (String each : routedDataSources) {
            routedMap.put(each, routeTables(tableRule, each));
        }
        return generateRoutingResult(tableRule, routedMap);
    }

  根据逻辑表名得到表规则配置对象
        TableRule tableRule = shardingRule.getTableRule(logicTableName);

/**
     * 根据逻辑表名称查找分片规则.
     *
     * @param logicTableName 逻辑表名称
     * @return 该逻辑表的分片规则
     */
    public TableRule getTableRule(final String logicTableName) {
//        根据逻辑表返回表规则配置对象
        Optional<TableRule> tableRule = tryFindTableRule(logicTableName);
        if (tableRule.isPresent()) {
            return tableRule.get();
        }
//        若是默认数据源不为空就根据默认数据源建立表配置规则对象
        if (dataSourceRule.getDefaultDataSource().isPresent()) {
            return createTableRuleWithDefaultDataSource(logicTableName, dataSourceRule);
        }
        throw new ShardingJdbcException("Cannot find table rule and default data source with logic table: '%s'", logicTableName);
    }

//        若是默认数据源不为空就根据默认数据源建立表配置规则对象
        if (dataSourceRule.getDefaultDataSource().isPresent()) {
            return createTableRuleWithDefaultDataSource(logicTableName, dataSourceRule);
        }

//    根据默认数据源建立部分库数据分片策略，数据表不分表分片策略对象，并建立表配置规则对象进行装载
    private TableRule createTableRuleWithDefaultDataSource(final String logicTableName, final DataSourceRule defaultDataSourceRule) {
        Map<String, DataSource> defaultDataSourceMap = new HashMap<>(1);
        defaultDataSourceMap.put(defaultDataSourceRule.getDefaultDataSourceName(), defaultDataSourceRule.getDefaultDataSource().get());
        return TableRule.builder(logicTableName)
                .dataSourceRule(new DataSourceRule(defaultDataSourceMap))
                .databaseShardingStrategy(new DatabaseShardingStrategy("", new NoneDatabaseShardingAlgorithm()))
                .tableShardingStrategy(new TableShardingStrategy("", new NoneTableShardingAlgorithm())).build();
    }

返回到这里

@Override
    public RoutingResult route() {
//        根据逻辑表名得到表规则配置对象
        TableRule tableRule = shardingRule.getTableRule(logicTableName);
//        根据表规则配置对象得到数据源集合
        Collection<String> routedDataSources = routeDataSources(tableRule);
        Map<String, Collection<String>> routedMap = new LinkedHashMap<>(routedDataSources.size());
        for (String each : routedDataSources) {
            routedMap.put(each, routeTables(tableRule, each));
        }
        return generateRoutingResult(tableRule, routedMap);
    }

//        根据表规则配置对象得到数据源集合
        Collection<String> routedDataSources = routeDataSources(tableRule);

根据分片列获取分片值

getShardingValues(strategy.getShardingColumns());

//        根据真实的数据源名称和分片值计算静态分片
        Collection<String> result = strategy.doStaticSharding(tableRule.getActualDatasourceNames(), shardingValues);

/**
     * 计算静态分片.
     *
     * @param availableTargetNames 全部的可用分片资源集合
     * @param shardingValues 分片值集合
     * @return 分库后指向的数据源名称集合
     */
    public Collection<String> doStaticSharding(final Collection<String> availableTargetNames, final Collection<ShardingValue<?>> shardingValues) {
        Collection<String> result = new TreeSet<>(String.CASE_INSENSITIVE_ORDER);
//        若是没有解析到传入的数据源分片值，要走全库路由
        if (shardingValues.isEmpty()) {
            result.addAll(availableTargetNames);
        } else {
//            若是传入分片值，根据分片值去获取具体的数据源
            result.addAll(doSharding(shardingValues, availableTargetNames));
        }
        return result;
    }

注意上面的数据库路由的默认实现，若是不传入数据库分片值会走全库路由的，数据量大的话是会影响性能的，因此建议必需要传入分片值，阿里的TDDL这里的实现是直接报错的。

//            若是传入分片值，根据分片值去获取具体的数据源
            result.addAll(doSharding(shardingValues, availableTargetNames));

private Collection<String> doSharding(final Collection<ShardingValue<?>> shardingValues, final Collection<String> availableTargetNames) {
//        若是没分片
        if (shardingAlgorithm instanceof NoneKeyShardingAlgorithm) {
            return Collections.singletonList(((NoneKeyShardingAlgorithm) shardingAlgorithm).doSharding(availableTargetNames, shardingValues.iterator().next()));
        }
//        若是按一个分片值分片
        if (shardingAlgorithm instanceof SingleKeyShardingAlgorithm) {
            SingleKeyShardingAlgorithm<?> singleKeyShardingAlgorithm = (SingleKeyShardingAlgorithm<?>) shardingAlgorithm;
            ShardingValue shardingValue = shardingValues.iterator().next();
            switch (shardingValue.getType()) {
                case SINGLE:
//                    = 元算符分片
                    return Collections.singletonList(singleKeyShardingAlgorithm.doEqualSharding(availableTargetNames, shardingValue));
                case LIST:
//                    in运算符分片
                    return singleKeyShardingAlgorithm.doInSharding(availableTargetNames, shardingValue);
                case RANGE:
//                    between运算符分片
                    return singleKeyShardingAlgorithm.doBetweenSharding(availableTargetNames, shardingValue);
                default:
//                    如今只支持这三种运算符分片
                    throw new UnsupportedOperationException(shardingValue.getType().getClass().getName());
            }
        }
//        若是是多个分片值
        if (shardingAlgorithm instanceof MultipleKeysShardingAlgorithm) {
            return ((MultipleKeysShardingAlgorithm) shardingAlgorithm).doSharding(availableTargetNames, shardingValues);
        }
//        其余方式的分片不支持
        throw new UnsupportedOperationException(shardingAlgorithm.getClass().getName());
    }

返回到这里

@Override
    public RoutingResult route() {
//        根据逻辑表名得到表规则配置对象
        TableRule tableRule = shardingRule.getTableRule(logicTableName);
//        根据表规则配置对象得到数据源集合
        Collection<String> routedDataSources = routeDataSources(tableRule);
        Map<String, Collection<String>> routedMap = new LinkedHashMap<>(routedDataSources.size());
        for (String each : routedDataSources) {
            routedMap.put(each, routeTables(tableRule, each));
        }
        return generateRoutingResult(tableRule, routedMap);
    }

根据数据源和表配置规则组装路由map

routedMap.put(each, routeTables(tableRule, each));

下面这个方法是获取路由的表的集合

private Collection<String> routeTables(final TableRule tableRule, final String routedDataSource) {
//        获取表分片策略
        TableShardingStrategy strategy = shardingRule.getTableShardingStrategy(tableRule);
//        获取分片值
        List<ShardingValue<?>> shardingValues = HintManagerHolder.isUseShardingHint() ? getTableShardingValuesFromHint(strategy.getShardingColumns())
                : getShardingValues(strategy.getShardingColumns());//doDynamicSharding
//        若是是动态分片走动态分片，若是是静态分片走静态分片
        Collection<String> result = tableRule.isDynamic() ? strategy.doDynamicSharding(shardingValues) : strategy.doStaticSharding(tableRule.getActualTableNames(routedDataSource), shardingValues);
        Preconditions.checkState(!result.isEmpty(), "no table route info");
        return result;
    }

/**
 * 计算动态分片.
 *
 * @param shardingValues 分片值集合
 * @return 分库后指向的分片资源集合
 */
public Collection<String> doDynamicSharding(final Collection<ShardingValue<?>> shardingValues) {//doDynamicSharding
    Preconditions.checkState(!shardingValues.isEmpty(), "Dynamic table should contain sharding value.");
    Collection<String> availableTargetNames = Collections.emptyList();
    Collection<String> result = new TreeSet<>(String.CASE_INSENSITIVE_ORDER);
    result.addAll(doSharding(shardingValues, availableTargetNames));
    return result;
}

返回到这里

@Override
    public RoutingResult route() {
//        根据逻辑表名得到表规则配置对象
        TableRule tableRule = shardingRule.getTableRule(logicTableName);
//        根据表规则配置对象得到数据源集合
        Collection<String> routedDataSources = routeDataSources(tableRule);
        Map<String, Collection<String>> routedMap = new LinkedHashMap<>(routedDataSources.size());
        for (String each : routedDataSources) {
            routedMap.put(each, routeTables(tableRule, each));
        }
        return generateRoutingResult(tableRule, routedMap);
    }

//        生成路由结果
        return generateRoutingResult(tableRule, routedMap);

private RoutingResult generateRoutingResult(final TableRule tableRule, final Map<String, Collection<String>> routedMap) {
        RoutingResult result = new RoutingResult();
//        遍历roadMap，roadMap里面key值存储的是数据源名称，value值是物理数据表集合
        for (Entry<String, Collection<String>> entry : routedMap.entrySet()) {
//            获取最下数据单元，每一个数据单元是一个DataNode
            Collection<DataNode> dataNodes = tableRule.getActualDataNodes(entry.getKey(), entry.getValue());
            for (DataNode each : dataNodes) {
//                组装数据表单元装载到路由结果中
                result.getTableUnits().getTableUnits().add(new TableUnit(each.getDataSourceName(), logicTableName, each.getTableName()));
            }
        }
        return result;
    }

数据模型

/**
 * SQL路由结果.
 * 
 * @author gaohongtao
 * @author zhangliang
 */
@RequiredArgsConstructor
@Getter
public final class SQLRouteResult {
    
//    sql语句对象
    private final SQLStatement sqlStatement;
    
//    最小sql执行单元集合
    private final Set<SQLExecutionUnit> executionUnits = new LinkedHashSet<>();
    
    private final List<Number> generatedKeys = new LinkedList<>();
}

/**
 * SQL最小执行单元.
 * 
 * @author gaohongtao
 */
@RequiredArgsConstructor
@Getter
@EqualsAndHashCode
@ToString
public final class SQLExecutionUnit {
    
//    具体的数据源
    private final String dataSource;
    
//    具体要执行的物理sql语句
    private final String sql;
}

/**
 * 路由表单元.
 * 
 * @author zhangliang
 */
@RequiredArgsConstructor
@Getter
@EqualsAndHashCode
@ToString
public final class TableUnit {
    
//    数据源名
    private final String dataSourceName;
    
//    逻辑表名
    private final String logicTableName;
    
//    物理表名
    private final String actualTableName;
}

/**
 * 路由表单元集合.
 * 
 * @author zhangliang
 */
@Getter
@ToString
public final class TableUnits {
    
//    路由表单元集合
    private final List<TableUnit> tableUnits = new LinkedList<>();

/**
 *  路由结果.
 * 
 * @author zhangliang
 */
@Getter
public class RoutingResult {
    
//    表路由单元集合
    private final TableUnits tableUnits = new TableUnits();

/**
 * 路由表单元.
 * 
 * @author zhangliang
 */
@RequiredArgsConstructor
@Getter
@EqualsAndHashCode
@ToString
public final class TableUnit {

//    数据源名
    private final String dataSourceName;

//    逻辑表名
    private final String logicTableName;

//    物理表名
    private final String actualTableName;
}

/**
 * 分库分表数据单元.
 * 
 * @author zhangliang
 */
@RequiredArgsConstructor
@Getter
@EqualsAndHashCode
@ToString
public class DataNode {
    
    private static final String DELIMITER = ".";
    
    private final String dataSourceName;//数据库名
    
    private final String tableName;//表名

说到最后

以上介绍，仅供参考。