温馨提示×

温馨提示×

您好,登录后才能下订单哦!

密码登录×
登录注册×
其他方式登录
点击 登录注册 即表示同意《亿速云用户服务条款》

sharding-jdbc路由的原理及应用

发布时间:2021-06-22 17:48:31 来源:亿速云 阅读:809 作者:chen 栏目:大数据

这篇文章主要介绍“sharding-jdbc路由的原理及应用”,在日常操作中,相信很多人在sharding-jdbc路由的原理及应用问题上存在疑惑,小编查阅了各式资料,整理出简单好用的操作方法,希望对大家解答”sharding-jdbc路由的原理及应用”的疑惑有所帮助!接下来,请跟着小编一起来学习吧!

路由引擎主要分为两大类:

  • 分片路由(直接路由、标准路由、笛卡尔积路由)

  • 广播路由(全库表路由、全库路由、全实例路由、单播路由、阻断路由)

具体路由类型含义参考官网路由引擎

https://shardingsphere.apache.org/document/current/cn/features/sharding/principle/route/

主要分析查询路由

1.路由ParsingSQLRouter#route入口

@RequiredArgsConstructor
public final class ParsingSQLRouter implements ShardingRouter {
    @Override
    public SQLRouteResult route(final SQLStatement sqlStatement, final List<Object> parameters) {
        //优化,处理条件占位符参数与真实数据、分页、group by etc.
        OptimizedStatement optimizedStatement = OptimizeEngineFactory.newInstance(shardingRule, shardingMetaData.getTable(), sqlStatement, parameters).optimize();
        boolean needMergeShardingValues = isNeedMergeShardingValues(sqlStatement);
        if (optimizedStatement instanceof ShardingWhereOptimizedStatement && needMergeShardingValues) {
            checkSubqueryShardingValues(sqlStatement, ((ShardingWhereOptimizedStatement) optimizedStatement).getShardingConditions());
            mergeShardingConditions(((ShardingWhereOptimizedStatement) optimizedStatement).getShardingConditions());
        }
        //路由入口
        RoutingResult routingResult = RoutingEngineFactory.newInstance(shardingRule, shardingMetaData.getDataSource(), optimizedStatement).route();
        if (needMergeShardingValues) {
            Preconditions.checkState(1 == routingResult.getRoutingUnits().size(), "Must have one sharding with subquery.");
        }
        if (optimizedStatement instanceof ShardingInsertOptimizedStatement) {
            setGeneratedValues((ShardingInsertOptimizedStatement) optimizedStatement);
        }
        SQLRouteResult result = new SQLRouteResult(optimizedStatement);
        result.setRoutingResult(routingResult);
        return result;
    }
    ... ...
}

2.路由工厂并路由RoutingEngineFactory#route

@NoArgsConstructor(access = AccessLevel.PRIVATE)
public final class RoutingEngineFactory {
    
    /**
     * Create new instance of routing engine.
     * 
     * @param shardingRule sharding rule
     * @param shardingDataSourceMetaData sharding data source meta data
     * @param optimizedStatement optimized statement
     * @return new instance of routing engine
     */
    public static RoutingEngine newInstance(final ShardingRule shardingRule, final ShardingDataSourceMetaData shardingDataSourceMetaData, final OptimizedStatement optimizedStatement) {
        SQLStatement sqlStatement = optimizedStatement.getSQLStatement();
        Collection<String> tableNames = sqlStatement.getTables().getTableNames();
        //全库路由
        if (sqlStatement instanceof TCLStatement) {
            return new DatabaseBroadcastRoutingEngine(shardingRule);
        }
        //全库表路由
        if (sqlStatement instanceof DDLStatement) {
            return new TableBroadcastRoutingEngine(shardingRule, optimizedStatement);
        }
        //阻断路由
        if (sqlStatement instanceof DALStatement) {
            return getDALRoutingEngine(shardingRule, sqlStatement, tableNames);
        }
        //全实例路由
        if (sqlStatement instanceof DCLStatement) {
            return getDCLRoutingEngine(shardingRule, optimizedStatement, shardingDataSourceMetaData);
        }
        //默认库路由
        if (shardingRule.isAllInDefaultDataSource(tableNames)) {
            return new DefaultDatabaseRoutingEngine(shardingRule, tableNames);
        }
        //全库路由
        if (shardingRule.isAllBroadcastTables(tableNames)) {
            return sqlStatement instanceof SelectStatement ? new UnicastRoutingEngine(shardingRule, tableNames) : new DatabaseBroadcastRoutingEngine(shardingRule);
        }
        //单播路由
        if (optimizedStatement instanceof ShardingWhereOptimizedStatement && ((ShardingWhereOptimizedStatement) optimizedStatement).getShardingConditions().isAlwaysFalse() || tableNames.isEmpty()) {
            return new UnicastRoutingEngine(shardingRule, tableNames);
        }
        Preconditions.checkState(optimizedStatement instanceof ShardingWhereOptimizedStatement);
        //分片路由
        return getShardingRoutingEngine(shardingRule, (ShardingWhereOptimizedStatement) optimizedStatement, tableNames);
    }
    ... ...
    private static RoutingEngine getShardingRoutingEngine(final ShardingRule shardingRule, final ShardingWhereOptimizedStatement optimizedStatement, final Collection<String> tableNames) {
        ///根据解析出来逻辑表获取分片表,如:SELECT i.* FROM t_order o, t_order_item i WHERE o.order_id = i.order_id and o.order_id = ?
        //则shardingTableNames为t_order,t_order_item
        Collection<String> shardingTableNames = shardingRule.getShardingLogicTableNames(tableNames);
        //满足以下两个条件走标准路由,否则走复合路由
        //1、是否只有一张分片表
        //2、绑定的逻辑表(配置绑定表)是否包含所有分片表
        if (1 == shardingTableNames.size() || shardingRule.isAllBindingTables(shardingTableNames)) {
            //标准路由,获取第一张表路由即可,参考官网案例说明
            return new StandardRoutingEngine(shardingRule, shardingTableNames.iterator().next(), optimizedStatement);
        }
        // TODO config for cartesian set
        //复合路由
        return new ComplexRoutingEngine(shardingRule, tableNames, optimizedStatement);
    }
}

3.标准路由StandardRoutingEngine#route

标准路由场景

  • t_order和t_order_item是分库分表并且是绑定表;如第二步55行判断shardingTableNames.size()=1或者shardingTableNames都是绑定表时,这时会走标准路由

@RequiredArgsConstructor
public final class StandardRoutingEngine implements RoutingEngine {

    //分库分表规则
    private final ShardingRule shardingRule;
    //逻辑表 t_order
    private final String logicTableName;
    //sql解析并优化后的结果
    private final ShardingWhereOptimizedStatement optimizedStatement;
    
    @Override
    public RoutingResult route() {
        //insert、update、delete判断表是否是单表
        if (isDMLForModify(optimizedStatement.getSQLStatement()) && !optimizedStatement.getSQLStatement().getTables().isSingleTable()) {
            throw new SQLParsingException("Cannot support Multiple-Table for '%s'.", optimizedStatement.getSQLStatement());
        }
        //路由数据节点、封装路由结果
        return generateRoutingResult(getDataNodes(shardingRule.getTableRule(logicTableName)));
    }
   
    ... ... 
       
    private RoutingResult generateRoutingResult(final Collection<DataNode> routedDataNodes) {
        RoutingResult result = new RoutingResult();
        //根据数据节点封装路由单元、表单元
        for (DataNode each : routedDataNodes) {
            //路由单元 demo_ds_0
            RoutingUnit routingUnit = new RoutingUnit(each.getDataSourceName());
            //表单元 逻辑表:真实表 t_order:t_order_0
            routingUnit.getTableUnits().add(new TableUnit(logicTableName, each.getTableName()));
            result.getRoutingUnits().add(routingUnit);
        }
        return result;
    }
    
    private Collection<DataNode> getDataNodes(final TableRule tableRule) {
        //判断database、table分片策略同时是Hint(直接路由)
        if (shardingRule.isRoutingByHint(tableRule)) {
            return routeByHint(tableRule);
        }
        //database、table分片策略都不是Hint
        if (isRoutingByShardingConditions(tableRule)) {
            //根据分片条件、策略路由到对应的database、table,同对分片键判断
            return routeByShardingConditions(tableRule);
        }
        //database或table分片策略有一个是Hint
        return routeByMixedConditions(tableRule);
    }
    ... ...
}

4.复合路由CartesianRoutingEngine#route

复杂路由场景

  • t_order和t_order_item是分库分表并且是绑定表;新增一个t_user分库分表,这时第二步的55行,shardingTableNames.size()=3,且t_user未配置成绑定表,这时会走复合路由

@RequiredArgsConstructor
public final class ComplexRoutingEngine implements RoutingEngine {
    //分库分表规则
    private final ShardingRule shardingRule;
    //逻辑表t_order、t_order_item
    private final Collection<String> logicTables;
    //sql解析并优化后的结果
    private final ShardingWhereOptimizedStatement optimizedStatement;
    
    @Override
    public RoutingResult route() {
        Collection<RoutingResult> result = new ArrayList<>(logicTables.size());
        Collection<String> bindingTableNames = new TreeSet<>(String.CASE_INSENSITIVE_ORDER);
        //遍历逻辑表
        for (String each : logicTables) {
            Optional<TableRule> tableRule = shardingRule.findTableRule(each);
            //表是否配置了分库分表规则
            if (tableRule.isPresent()) {
                // 如果绑定关系表已经处理过,那么不需要再处理,如t_order处理过,由于t_order_item与其是绑定关系,那么不需要再处理;
                if (!bindingTableNames.contains(each)) {
                    //构建标准路由并路由
                    result.add(new StandardRoutingEngine(shardingRule, tableRule.get().getLogicTable(), optimizedStatement).route());
                }
                //根据逻辑表查找对应的所有绑定表,如根据t_order就能查询到t_order、t_order_item,因为t_order和t_order_item是绑定表
                Optional<BindingTableRule> bindingTableRule = shardingRule.findBindingTableRule(each);
                if (bindingTableRule.isPresent()) {
                    //添加绑定表
                    bindingTableNames.addAll(Lists.transform(bindingTableRule.get().getTableRules(), new Function<TableRule, String>() {
                        
                        @Override
                        public String apply(final TableRule input) {
                            return input.getLogicTable();
                        }
                    }));
                }
            }
        }
        if (result.isEmpty()) {
            throw new ShardingException("Cannot find table rule and default data source with logic tables: '%s'", logicTables);
        }
        if (1 == result.size()) {
            return result.iterator().next();
        }
        //笛卡尔积路由
        return new CartesianRoutingEngine(result).route();
    }
}

5.笛卡尔积路由

笛卡尔积路由场景

  • 笛卡尔路由是最复杂的情况,它无法根据绑定表的关系定位分片规则,因此非绑定表之间的关联查询需要拆解为笛卡尔积组合执行。如上面事例新增的t_user分库分表,t_user和t_order未配置绑定表关系,如下SQL就需要走笛卡尔积路由

  • SELECT * FROM t_user u JOIN t_order o ON u.user_id = o.user_id AND u.user_id in(1 , 2)

@RequiredArgsConstructor
public final class CartesianRoutingEngine implements RoutingEngine {
    
    private final Collection<RoutingResult> routingResults;
    
    @Override
    public RoutingResult route() {
        RoutingResult result = new RoutingResult();
        //获取数据源对应的逻辑表集合
        for (Entry<String, Set<String>> entry : getDataSourceLogicTablesMap().entrySet()) {
            //通过数据源名称和逻辑表的名称获取实际的表组,即[["t_user_0","t_user_1"],["t_order_0", "t_order_1]]
            List<Set<String>> actualTableGroups = getActualTableGroups(entry.getKey(), entry.getValue());
            //把逻辑表名封装成TableUnit表单元
            //TableUnit包含数据源名称、逻辑表名、实际表名(通过这三个属性可以确定最终访问的表)
            List<Set<TableUnit>> routingTableGroups = toRoutingTableGroups(entry.getKey(), actualTableGroups);
            //封装RoutingUnit路由单元
            //cartesianProduct计算笛卡尔积
            result.getRoutingUnits().addAll(getRoutingUnits(entry.getKey(), Sets.cartesianProduct(routingTableGroups)));
        }
        return result;
    }
    
    private Map<String, Set<String>> getDataSourceLogicTablesMap() {
        //获取数据源的交集,如t_user逻辑表路由到数据源demo_ds_0,而t_order表路由到数据源ds_demo_0和demo_ds_1,数据源交集就是demo_ds_0
        //事例SELECT * FROM t_user_0 u JOIN t_order_0 o ON u.user_id = o.user_id WHERE u.user_id in(1, 2); t_user和t_order不是绑定表关系
        //笛卡尔路由是最复杂的情况,它无法根据绑定表的关系定位分片规则,因此非绑定表之间的关联查询需要拆解为笛卡尔积组合执行
        Collection<String> intersectionDataSources = getIntersectionDataSources();
        Map<String, Set<String>> result = new HashMap<>(routingResults.size());
        //遍历标准路由后的结果集
        for (RoutingResult each : routingResults) {
            //通过数据源的名称获取数据源和逻辑表之间的映射关系
            for (Entry<String, Set<String>> entry : each.getDataSourceLogicTablesMap(intersectionDataSources).entrySet()) {
                if (result.containsKey(entry.getKey())) {
                    result.get(entry.getKey()).addAll(entry.getValue());
                } else {
                    result.put(entry.getKey(), entry.getValue());
                }
            }
        }
        // 返回数据源-逻辑表集合组成的Map,这里就是{"demo_ds_0":["t_user", "t_order"]}
        return result;
    }
    
    private Collection<String> getIntersectionDataSources() {
        Collection<String> result = new HashSet<>();
        for (RoutingResult each : routingResults) {
            if (result.isEmpty()) {
                result.addAll(each.getDataSourceNames());
            }

            //交集
            result.retainAll(each.getDataSourceNames());
        }
        return result;
    }
    ... ...
}

笛卡尔积结果如下:

sharding-jdbc路由的原理及应用

6.直接路由

直接路由场景

  • 满足直接路由的条件相对苛刻,它需要通过Hint(使用HintAPI直接指定路由至库表)方式分片,并且是只分库不分表的前提下,则可以避免SQL解析和之后的结果归并

  • 假如路由算法为value % 2,当一个逻辑库t_order对应2个真实库t_order_0和t_order_1时,路由后SQL将在t_order_1上执行。下方是使用API的代码样例:

    String sql = "SELECT * FROM t_order";
    try (
    //获取Hint实例
    HintManager hintManager = HintManager.getInstance();
    Connection conn = dataSource.getConnection();
    PreparedStatement pstmt = conn.prepareStatement(sql)) {
        //设置数据源分片个数
        hintManager.setDatabaseShardingValue(3);
        try (ResultSet rs = pstmt.executeQuery()) {
            while (rs.next()) {
                //...
            }
        }
    }
@RequiredArgsConstructor
public final class DatabaseHintRoutingEngine implements RoutingEngine {
    //数据源集群
    private final Collection<String> dataSourceNames;
    //Hint数据分片策略
    private final HintShardingStrategy databaseShardingStrategy;
    
    @Override
    public RoutingResult route() {
         //获取当前线程数据源分片
        Collection<Comparable<?>> shardingValues = HintManager.getDatabaseShardingValues();
        Preconditions.checkState(!shardingValues.isEmpty());
        Collection<String> routingDataSources;
        //根据分片策略路由
        routingDataSources = databaseShardingStrategy.doSharding(dataSourceNames, Collections.<RouteValue>singletonList(new ListRouteValue<>("", "", shardingValues)));
        Preconditions.checkState(!routingDataSources.isEmpty(), "no database route info");
        RoutingResult result = new RoutingResult();
        //封装路由单元
        for (String each : routingDataSources) {
            result.getRoutingUnits().add(new RoutingUnit(each));
        }
        return result;
    }
}

到此,关于“sharding-jdbc路由的原理及应用”的学习就结束了,希望能够解决大家的疑惑。理论与实践的搭配能更好的帮助大家学习,快去试试吧!若想继续学习更多相关知识,请继续关注亿速云网站,小编会继续努力为大家带来更多实用的文章!

向AI问一下细节

免责声明:本站发布的内容(图片、视频和文字)以原创、转载和分享为主,文章观点不代表本网站立场,如果涉及侵权请联系站长邮箱:is@yisu.com进行举报,并提供相关证据,一经查实,将立刻删除涉嫌侵权内容。

AI