下推更多的计算
PolarDB-X会尽可能将更多的计算下推到存储层MySQL。下推计算能够减少数据传输,减少网络层和PolarDB-X层的开销,提升SQL语句的执行效率。PolarDB-X支持下推几乎所有算子,包括:
- 过滤条件,如WHERE或HAVING中的条件。
- 聚合算子,如COUNT,GROUP BY等,会分成两阶段进行聚合计算。
- 排序算子,如ORDER BY。
- JOIN和子查询,两边JOIN Key分片方式必须一样,或其中一边为广播表。
如下示例讲解如何将更多的计算下推到MySQL来加速执行
> EXPLAIN select * from customer, nation where c_nationkey = n_nationkey and n_regionkey = 3; Project(c_custkey="c_custkey", c_name="c_name", c_address="c_address", c_nationkey="c_nationkey", c_phone="c_phone", c_acctbal="c_acctbal", c_mktsegment="c_mktsegment", c_comment="c_comment", n_nationkey="n_nationkey", n_name="n_name", n_regionkey="n_regionkey", n_comment="n_comment") BKAJoin(condition="c_nationkey = n_nationkey", type="inner") Gather(concurrent=true) LogicalView(tables="nation", shardCount=2, sql="SELECT * FROM `nation` AS `nation` WHERE (`n_regionkey` = ?)") Gather(concurrent=true) LogicalView(tables="customer_[0-7]", shardCount=8, sql="SELECT * FROM `customer` AS `customer` WHERE (`c_nationkey` IN ('?'))")
若执行计划中出现了BKAJOIN,BKAJOIN每次从左表获取一批数据,就会拼成一个IN查询取出右表相关联的行,并在最后执行JOIN操作。由于左表数据量很大,需要取很多次才能完成查询,执行很慢。
无法下推JOIN的原因是:当前情况下,nation是按主键n_nationkey切分的,而本查询的JOIN Key是c_custkey,二者不同,所以下推失败。
考虑到nation (国家)表数据量并不大、且几乎没有修改操作,可以将其重建成如下广播表:
--- 修改后 --- CREATE TABLE `nation` ( `n_nationkey` int(11) NOT NULL, `n_name` varchar(25) NOT NULL, `n_regionkey` int(11) NOT NULL, `n_comment` varchar(152) DEFAULT NULL, PRIMARY KEY (`n_nationkey`) ) BROADCAST; --- 声明为广播表
修改后,可以看到执行计划中不再出现JOIN,几乎所有计算都被下推到存储层MySQL执行了(LogicalView中),而上层仅仅是将结果收集并返回给用户(Gather算子),执行性能大大增强。
> EXPLAIN select * from customer, nation where c_nationkey = n_nationkey and n_regionkey = 3; Gather(concurrent=true) LogicalView(tables="customer_[0-7],nation", shardCount=8, sql="SELECT * FROM `customer` AS `customer` INNER JOIN `nation` AS `nation` ON ((`nation`.`n_regionkey` = ?) AND (`customer`.`c_nationkey` = `nation`.`n_nationkey`))")
更多关于下推的原理和优化,请参见查询改写与下推。
增加索引
PolarDB-X支持全局二级索引
以下以慢SQL示例来讲解如何通过GSI下推更多算子
> EXPLAIN select o_orderkey, c_custkey, c_name from orders, customer where o_custkey = c_custkey and o_orderdate = '2019-11-11' and o_totalprice > 100; Project(o_orderkey="o_orderkey", c_custkey="c_custkey", c_name="c_name") HashJoin(condition="o_custkey = c_custkey", type="inner") Gather(concurrent=true) LogicalView(tables="customer_[0-7]", shardCount=8, sql="SELECT `c_custkey`, `c_name` FROM `customer` AS `customer`") Gather(concurrent=true) LogicalView(tables="orders_[0-7]", shardCount=8, sql="SELECT `o_orderkey`, `o_custkey` FROM `orders` AS `orders` WHERE ((`o_orderdate` = ?) AND (`o_totalprice` > ?))")
执行计划中,orders按照o_orderkey拆分而customer按照c_custkey拆分,由于拆分维度不同JOIN算子不能下推。考虑到2019-11-11当天总价高于100的订单非常多,跨分片JOIN耗时很高,需要在orders表上创建一个GSI来使得JOIN算子可以下推。查询中使用到了orders表的o_orderkey, o_custkey, o_orderdate, o_totalprice四列,其中o_orderkey, o_custkey分别是主表和索引表的拆分键,o_orderdate, o_totalprice作为覆盖列包含在索引中用于避免回表。
> create global index i_o_custkey on orders(`o_custkey`) covering(`o_orderdate`, `o_totalprice`) DBPARTITION BY HASH(`o_custkey`) TBPARTITION BY HASH(`o_custkey`) TBPARTITIONS 4;
增加GSI并通过force index(i_o_custkey)强制使用索引后,跨分片JOIN变为MySQL上的局部JOIN (IndexScan中),并且通过覆盖列避免了回表操作,查询性能得到提升。
> EXPLAIN select o_orderkey, c_custkey, c_name from orders force index(i_o_custkey), customer where o_custkey = c_custkey and o_orderdate = '2019-11-11' and o_totalprice > 100; Gather(concurrent=true) IndexScan(tables="i_o_custkey_[0-7],customer_[0-7]", shardCount=8, sql="SELECT `i_o_custkey`.`o_orderkey`, `customer`.`c_custkey`, `customer`.`c_name` FROM `i_o_custkey` AS `i_o_custkey` INNER JOIN `customer` AS `customer` ON (((`i_o_custkey`.`o_orderdate` = ?) AND (`i_o_custkey`.`o_custkey` = `customer`.`c_custkey`)) AND (`i_o_custkey`.`o_totalprice` > ?))")
更多关于全局二级索引的使用细节,请参见全局二级索引。
执行计划调优
大多数情况下,PolarDB-X的查询优化器可以自动产生最佳的执行计划。但是,少数情况下,可能因为统计信息存在缺失、误差等,导致生成的执行计划不够好,这时,可以通过Hint来干预优化器行为,使之生成更好的执行计划。如下示例将讲解执行计划的调优。
> EXPLAIN select o_orderkey, c_custkey, c_name from orders, customer where o_custkey = c_custkey and o_orderdate = '2019-11-15' and o_totalprice < 10; Project(o_orderkey="o_orderkey", c_custkey="c_custkey", c_name="c_name") HashJoin(condition="o_custkey = c_custkey", type="inner") Gather(concurrent=true) LogicalView(tables="customer_[0-7]", shardCount=8, sql="SELECT `c_custkey`, `c_name` FROM `customer` AS `customer`") Gather(concurrent=true) LogicalView(tables="orders_[0-7]", shardCount=8, sql="SELECT `o_orderkey`, `o_custkey` FROM `orders` AS `orders` WHERE ((`o_orderdate` = ?) AND (`o_totalprice` < ?))")
实际上2019-11-15这一天总价低于10元的订单数量很小,只有几条,这时候用BKAJOIN是比Hash JOIN更好的选择(关于BKAJOIN和Hash JOIN的介绍,请参见JOIN优化和执行
通过如下/*+TDDL:BKA_JOIN(orders, customer)*/ Hint强制优化器使用BKAJOIN(LookupJOIN):
> EXPLAIN /*+TDDL:BKA_JOIN(orders, customer)*/ select o_orderkey, c_custkey, c_name from orders, customer where o_custkey = c_custkey and o_orderdate = '2019-11-15' and o_totalprice < 10; Project(o_orderkey="o_orderkey", c_custkey="c_custkey", c_name="c_name") BKAJoin(condition="o_custkey = c_custkey", type="inner") Gather(concurrent=true) LogicalView(tables="orders_[0-7]", shardCount=8, sql="SELECT `o_orderkey`, `o_custkey` FROM `orders` AS `orders` WHERE ((`o_orderdate` = ?) AND (`o_totalprice` < ?))") Gather(concurrent=true) LogicalView(tables="customer_[0-7]", shardCount=8, sql="SELECT `c_custkey`, `c_name` FROM `customer` AS `customer` WHERE (`c_custkey` IN ('?'))")
可以选择执行加如下Hint的查询:
/*+TDDL:BKA_JOIN(orders, customer)*/ select o_orderkey, c_custkey, c_name from orders, customer where o_custkey = c_custkey and o_orderdate = '2019-11-15' and o_totalprice < 10;
以上操作加快了SQL查询速度。为了让Hint发挥作用,可以将应用中的SQL加上Hint,或者更方便的方式是使用执行计划管理(Plan Management)功能对该SQL固定执行计划。具体操作如下:
BASELINE FIX SQL /*+TDDL:BKA_JOIN(orders, customer)*/ select o_orderkey, c_custkey, c_name from orders, customer where o_custkey = c_custkey and o_orderdate = '2019-11-15';
这样一来,对于这条SQL(参数可以不同),PolarDB-X都会采用如上固定的执行计划。更多关于执行计划管理的信息,请参见执行计划管理
并发执行
用户可以通过HINT /*+TDDL:PARALLELISM=4*/ 指定并行度,充分利用多核能力加速计算。比如以下例子:
mysql> explain physical select a.k, count(*) cnt from sbtest1 a, sbtest1 b where a.id = b.k and a.id > 1000 group by k having cnt > 1300 or der by cnt limit 5, 10; +-------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | PLAN | +-------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | ExecutorType: AP_LOCAL | | The Query's MaxConcurrentParallelism: 2 | | Fragment 1 | | Shuffle Output layout: [BIGINT, BIGINT] Output layout: [BIGINT, BIGINT] | | Output partitioning: SINGLE [] Parallelism: 1 | | TopN(sort="cnt ASC", offset=?2, fetch=?3) | | Filter(condition="cnt > ?1") | | HashAgg(group="k", cnt="COUNT()") | | BKAJoin(condition="k = id", type="inner") | | RemoteSource(sourceFragmentIds=[0], type=RecordType(INTEGER_UNSIGNED id, INTEGER_UNSIGNED k)) | | Gather(concurrent=true) | | LogicalView(tables="[000000-000003].sbtest1_[00-15]", shardCount=16, sql="SELECT `k` FROM `sbtest1` AS `sbtest1` WHERE ((`k` > ?) AND (`k` IN (...)))") | | Fragment 0 | | Shuffle Output layout: [BIGINT, BIGINT] Output layout: [BIGINT, BIGINT] | | Output partitioning: SINGLE [] Parallelism: 1 Splits: 16 | | LogicalView(tables="[000000-000003].sbtest1_[00-15]", shardCount=16, sql="SELECT `id`, `k` FROM `sbtest1` AS `sbtest1` WHERE (`id` > ?)") | +-------------------------------------------------------------------------------------------------------------------------------------------------------------------+
默认的并行度并不高,通过强制指定并行度,利用单机或者多机并行模式来加速。
mysql> explain physical /*+TDDL:PARALLELISM=8*/select a.k, count(*) cnt from sbtest1 a, sbtest1 b where a.id = b.k and a.id > 1000 group by k having cnt > 1300 order by cnt limit 5, 10; | +-------------------------------------------------------------------------------------------------------------------------------------------------------------+ | ExecutorMode: AP_LOCAL | | Fragment 0 dependency: [] parallelism: 8 | | BKAJoin(condition="k = id", type="inner") | | Gather(concurrent=true) | | LogicalView(tables="[000000-000003].sbtest1_[00-15]", shardCount=16, sql="SELECT `id`, `k` FROM `sbtest1` AS `sbtest1` WHERE (`id` > ?)") | | Gather(concurrent=true) | | LogicalView(tables="[000000-000003].sbtest1_[00-15]", shardCount=16, sql="SELECT `k` FROM `sbtest1` AS `sbtest1` WHERE ((`k` > ?) AND (`k` IN (...)))") | | Fragment 1 dependency: [] parallelism: 8 | | LocalBuffer | | RemoteSource(sourceFragmentIds=[0], type=RecordType(INTEGER_UNSIGNED id, INTEGER_UNSIGNED k, INTEGER_UNSIGNED k0)) | | Fragment 2 dependency: [0, 1] parallelism: 8 | | Filter(condition="cnt > ?1") | | HashAgg(group="k", cnt="COUNT()") | | RemoteSource(sourceFragmentIds=[1], type=RecordType(INTEGER_UNSIGNED id, INTEGER_UNSIGNED k, INTEGER_UNSIGNED k0)) | | Fragment 3 dependency: [0, 1] parallelism: 1 | | LocalBuffer | | RemoteSource(sourceFragmentIds=[2], type=RecordType(INTEGER_UNSIGNED k, BIGINT cnt)) | | Fragment 4 dependency: [2, 3] parallelism: 1 | | TopN(sort="cnt ASC", offset=?2, fetch=?3) | | RemoteSource(sourceFragmentIds=[3], type=RecordType(INTEGER_UNSIGNED k, BIGINT cnt)) | +---------------------------------------------------------------------------------------------------------