阿飛Javaer，轉(zhuǎn)載請注明原創(chuàng)出處，謝謝！
路由條件
ParsingSQLRouter.java中決定是簡單路由還是復(fù)雜路由的條件如下；
private RoutingResult route(final List<Object> parameters, final SQLStatement sqlStatement) {
Collection<String> tableNames = sqlStatement.getTables().getTableNames();
RoutingEngine routingEngine;
if (1 == tableNames.size()
|| shardingRule.isAllBindingTables(tableNames)
|| shardingRule.isAllInDefaultDataSource(tableNames)) {
routingEngine = new SimpleRoutingEngine(shardingRule, parameters, tableNames.iterator().next(), sqlStatement);
} else {
// TODO config for cartesian set
routingEngine = new ComplexRoutingEngine(shardingRule, parameters, tableNames, sqlStatement);
}
return routingEngine.route();
}
- 是否只有一張表--tableNames.size()
說明：這個"一張表"并不是指SQL中只有一張表，而是有分庫分表規(guī)則的表數(shù)量，例如下面這段構(gòu)造ShardingRule的源碼，tableRules()有兩個表，所以tableNames.size()的值為2；如果(Arrays.asList(orderTableRule))即只有1個表，那么tableNames.size()的值為1；
ShardingRule.builder()
.dataSourceRule(dataSourceRule)
.tableRules(Arrays.asList(orderTableRule, userTableRule))
.databaseShardingStrategy(*** ***).tableShardingStrategy(*** ***) .build();
- 是否都是綁定表--shardingRule.isAllBindingTables(tableNames)
說明：isAllBindingTables(tableNames)判斷tableNames是否都屬于綁定表，例如下面這段構(gòu)造ShardingRule的源碼，.bindingTableRules()里的參數(shù)就是綁定表集合，這里是t_order和t_order_item都是綁定表，那么：
SELECT od.user_id, od.order_id, oi.item_id, od.status FROM t_order od join t_order_item oi on od.order_id=oi.order_id這個SQL只有t_order和t_order_item兩個表且都是綁定表，那么shardingRule.isAllBindingTables(tableNames)為true；
ShardingRule.builder()
.dataSourceRule(dataSourceRule)
.tableRules(Arrays.asList(orderTableRule, orderItemTableRule, userTableRule))
.bindingTableRules(Collections.singletonList(new BindingTableRule(Arrays.asList(orderTableRule, orderItemTableRule))))
. *** ***;
- 是否都在默認(rèn)數(shù)據(jù)源中--shardingRule.isAllInDefaultDataSource(tableNames)
說明：sharding-jdbc判斷邏輯源碼如下，即只要在表規(guī)則集合中能夠匹配到邏輯表，就認(rèn)為不屬于默認(rèn)數(shù)據(jù)源中（默認(rèn)數(shù)據(jù)源不分庫分表），例如
ShardingRule.builder().dataSourceRule(dataSourceRule).tableRules(Arrays.asList(orderTableRule, orderItemTableRule, userTableRule))，根據(jù)tableRules參數(shù)可知，主要SQL中有t_user，t_order，t_order_item三個表的任意一個表，那么shardingRule.isAllInDefaultDataSource(tableNames)都為false；
public boolean isAllInDefaultDataSource(final Collection<String> logicTables) {
for (String each : logicTables) {
if (tryFindTableRule(each).isPresent()) {
return false;
}
}
return !logicTables.isEmpty();
}
public Optional<TableRule> tryFindTableRule(final String logicTableName) {
for (TableRule each : tableRules) {
if (each.getLogicTable().equalsIgnoreCase(logicTableName)) {
return Optional.of(each);
}
}
return Optional.absent();
}
構(gòu)造復(fù)雜路由
綜上分析，如果三個條件都不滿足就走復(fù)雜路由ComplexRoutingEngine，構(gòu)造這種場景：
t_order和t_order_item分庫分表且綁定表關(guān)系，加入一個新的分庫分表t_user；ShardingRule如下：
ShardingRule shardingRule = ShardingRule.builder()
.dataSourceRule(dataSourceRule)
.tableRules(Arrays.asList(orderTableRule, orderItemTableRule, userTableRule))
.bindingTableRules(Collections.singletonList(new BindingTableRule(Arrays.asList(orderTableRule, orderItemTableRule))))
.databaseShardingStrategy(new DatabaseShardingStrategy("user_id", new ModuloDatabaseShardingAlgorithm()))
.tableShardingStrategy(new TableShardingStrategy("order_id", new ModuloTableShardingAlgorithm()))
.build();
執(zhí)行的SQL為：
SELECT od.user_id, od.order_id, oi.item_id, od.status
FROM `t_user` tu
join t_order od on tu.user_id=od.user_id
join t_order_item oi on od.order_id=oi.order_id
where tu.`status`='VALID' and tu.user_id=?
構(gòu)造的這個場景：tableNames.size()=3（三張表t_user，t_order，t_order_item都有分庫分表規(guī)則，所以值為3），shardingRule.isAllBindingTables(tableNames)為false（t_user表不屬于綁定表范圍）；shardingRule.isAllInDefaultDataSource(tableNames)為false（三張表都不屬于默認(rèn)數(shù)據(jù)源中的表）；所以這個SQL會走復(fù)雜路由的邏輯；
ComplexRoutingEngine
復(fù)雜路由引擎的核心邏輯就是拆分成多個簡單路由，然后求笛卡爾積，復(fù)雜路由核心源碼如下：
@RequiredArgsConstructor
@Slf4j
public final class ComplexRoutingEngine implements RoutingEngine {
// 分庫分表規(guī)則
private final ShardingRule shardingRule;
// SQL請求參數(shù)，豬油一個user_id的值為10
private final List<Object> parameters;
// 邏輯表集合：t_order，t_order_item，t_user，三個邏輯表
private final Collection<String> logicTables;
// SQL解析結(jié)果
private final SQLStatement sqlStatement;
// 復(fù)雜路由的核心邏輯
@Override
public RoutingResult route() {
Collection<RoutingResult> result = new ArrayList<>(logicTables.size());
Collection<String> bindingTableNames = new TreeSet<>(String.CASE_INSENSITIVE_ORDER);
// 遍歷邏輯表集合
for (String each : logicTables) {
Optional<TableRule> tableRule = shardingRule.tryFindTableRule(each);
// 如果遍歷的表配置了分庫分表規(guī)則
if (tableRule.isPresent()) {
// 如果綁定關(guān)系表已經(jīng)處理過，那么不需要再處理，例如t_order處理過，由于t_order_item與其是綁定關(guān)系，那么不需要再處理
if (!bindingTableNames.contains(each)) {
// 根據(jù)當(dāng)前遍歷的邏輯表構(gòu)造一個簡單路由規(guī)則
result.add(new SimpleRoutingEngine(shardingRule, parameters, tableRule.get().getLogicTable(), sqlStatement).route());
}
// 根據(jù)當(dāng)前邏輯表，查找其對應(yīng)的所有綁定表，例如根據(jù)t_order就能夠查詢出t_order和t_order_item；假如配置了.bindingTableRules(***t_point, t_point_detail***)，那么，根據(jù)t_point能查詢出t_point和t_point_detail，其目的是N個綁定表只需要路由一個綁定表即可，因為綁定表之間的路由關(guān)系完全一致。
Optional<BindingTableRule> bindingTableRule = shardingRule.findBindingTableRule(each);
if (bindingTableRule.isPresent()) {
bindingTableNames.addAll(Lists.transform(bindingTableRule.get().getTableRules(), new Function<TableRule, String>() {
@Override
public String apply(final TableRule input) {
return input.getLogicTable();
}
}));
}
}
}
log.trace("mixed tables sharding result: {}", result);
// 如果是復(fù)雜路由，但是路由結(jié)果為空，那么拋出異常
if (result.isEmpty()) {
throw new ShardingJdbcException("Cannot find table rule and default data source with logic tables: '%s'", logicTables);
}
// 如果結(jié)果的size為1，那么直接返回即可
if (1 == result.size()) {
return result.iterator().next();
}
// 對剛剛的路由結(jié)果集合計算笛卡爾積，就是最終復(fù)雜的路由結(jié)果
return new CartesianRoutingEngine(result).route();
}
}
由上面源碼分析可知，會分別對t_user和t_order構(gòu)造簡單路由（t_order_item和t_order是綁定關(guān)系，二者取其一即可）；
- t_user只分庫不分表（因為構(gòu)造TableRule時邏輯表和實際表一致），且請求參數(shù)為user_id=10，所以t_user這個邏輯表的簡單路由結(jié)果為：數(shù)據(jù)源ds_jdbc_0，實際表t_user；
- t_order分庫分表，且請求參數(shù)user_id被解析為t_user的條件（笛卡爾積路由引擎會處理），所以t_order的簡單路由結(jié)果為：數(shù)據(jù)源ds_jdbc_0和ds_jdbc_1，實際表t_order_0和t_order_1；
debug的result如下：
result detail
CartesianRoutingEngine
如上分析，求得簡單路由結(jié)果集后，求笛卡爾積就是復(fù)雜路由的最終路由結(jié)果，笛卡爾積路由引擎CartesianRoutingEngine的核心源碼如下：
@RequiredArgsConstructor
@Slf4j
public final class CartesianRoutingEngine implements RoutingEngine {
private final Collection<RoutingResult> routingResults;
@Override
public CartesianRoutingResult route() {
CartesianRoutingResult result = new CartesianRoutingResult();
// getDataSourceLogicTablesMap()的分析參考下面的分析
for (Entry<String, Set<String>> entry : getDataSourceLogicTablesMap().entrySet()) {
// 根據(jù)數(shù)據(jù)源&邏輯表，得到實際表集合，即[["t_user"],["t_order_0","t_order_1"]]
List<Set<String>> actualTableGroups = getActualTableGroups(entry.getKey(), entry.getValue());
// 把邏輯表名封裝，TableUnit的屬性有：數(shù)據(jù)源名稱，邏輯表名，實際表名（這三個屬性才能確定最終訪問的表）
List<Set<TableUnit>> tableUnitGroups = toTableUnitGroups(entry.getKey(), actualTableGroups);
// 計算所有實際表的笛卡爾積
result.merge(entry.getKey(), getCartesianTableReferences(Sets.cartesianProduct(tableUnitGroups)));
}
log.trace("cartesian tables sharding result: {}", result);
return result;
}
// 得到數(shù)據(jù)源-邏輯表集合組成的Map
private Map<String, Set<String>> getDataSourceLogicTablesMap() {
// 這里很關(guān)鍵，是得到數(shù)據(jù)源的交集（上面分析時t_user邏輯表路由到數(shù)據(jù)源ds_jdbc_0，而t_order表路由到數(shù)據(jù)源ds_jdbc_0和ds_jdbc_1，數(shù)據(jù)源交集就是ds_jdbc_0）
Collection<String> intersectionDataSources = getIntersectionDataSources();
Map<String, Set<String>> result = new HashMap<>(routingResults.size());
for (RoutingResult each : routingResults) {
for (Entry<String, Set<String>> entry : each.getTableUnits().getDataSourceLogicTablesMap(intersectionDataSources).entrySet()) {
if (result.containsKey(entry.getKey())) {
result.get(entry.getKey()).addAll(entry.getValue());
} else {
result.put(entry.getKey(), entry.getValue());
}
}
}
// 得到的最終結(jié)果為數(shù)據(jù)源-邏輯表集合組成的Map，這里就是{"ds_jdbc_0":["t_order", "t_user"]}
return result;
}
... ...
}
計算得到的笛卡爾積結(jié)果如下：
image.png
sql.show結(jié)果如下，可以看到重寫后的2條實際SQL：t_user&t_order_0，以及t_user&t_order_1（t_order_item與t_order是綁定表，保持一致即可）：
[INFO ] 2018-05-08 11:13:02,044 --main-- [Sharding-JDBC-SQL] Logic SQL: SELECT od.user_id, od.order_id, oi.item_id, od.status FROM `t_user` tu join t_order od on tu.user_id=od.user_id join t_order_item oi on od.order_id=oi.order_id where tu.`status`='VALID' and tu.user_id=?
... ...
[INFO ] 2018-05-08 11:13:02,059 --main-- [Sharding-JDBC-SQL] Actual SQL: ds_jdbc_0 ::: SELECT od.user_id, od.order_id, oi.item_id, od.status FROM t_user tu join t_order_0 od on tu.user_id=od.user_id join t_order_item_0 oi on od.order_id=oi.order_id where tu.`status`='VALID' and tu.user_id=? ::: [10]
[INFO ] 2018-05-08 11:13:02,059 --main-- [Sharding-JDBC-SQL] Actual SQL: ds_jdbc_0 ::: SELECT od.user_id, od.order_id, oi.item_id, od.status FROM t_user tu join t_order_1 od on tu.user_id=od.user_id join t_order_item_1 oi on od.order_id=oi.order_id where tu.`status`='VALID' and tu.user_id=? ::: [10]
