本文介紹了執行計劃管理的工作流程和執行計劃調優實例。
對于每一條SQL，優化器都會生成相應執行計劃。很多情況下，應用請求的SQL都是重復的（僅參數不同），參數化之后的SQL完全相同。這時，可以按照參數化之后的SQL構造一個緩存，將除了參數以外的各種信息（比如執行計劃）緩存起來，稱為執行計劃緩存（Plan Cache）。
對于較復雜的查詢（例如涉及到多個表的Join），為了使其執行計劃能保持相對穩定，不因為版本升級等原因發生變化，PolarDB-X會為每個SQL記錄一組執行計劃，該執行計劃會被持久化地保存，即使版本升級也會保留，此過程稱為執行計劃管理（Plan Management）。
工作流程
當PolarDB-X收到一條查詢SQL時，會經歷以下流程：
對查詢SQL進行參數化處理，將所有參數替換為占位符
?。以參數化的SQL作為Key，查找執行計劃緩存中是否存在此SQL的緩存，如果沒有，則調用優化器進行優化。
如果該SQL是簡單查詢，則直接執行，跳過執行計劃管理相關步驟。
如果該SQL是復雜查詢，則使用基線（Baseline）中固化的執行計劃，如果有多個，則選擇代價最低的那個。
執行計劃緩存
PolarDB-X默認開啟執行計劃緩存功能。EXPLAIN結果中的HitCache表示當前SQL是否命中執行計劃緩存。開啟執行計劃緩存后，PolarDB-X會對SQL做參數化處理，參數化會將SQL中的常量用占位符?替換，并構建出相應的參數列表。在執行計劃中也可以看到LogicalView算子的SQL中含有?。
執行計劃管理
對于復雜SQL，經過執行計劃緩存之后，還會經過執行計劃管理流程。
執行計劃緩存和執行計劃管理都是采用參數化后的SQL作為Key來查找執行計劃，執行計劃緩存中會緩存所有SQL的執行計劃，而執行計劃管理僅對復雜查詢SQL進行處理。由于受到具體參數的影響，SQL模版和最優的執行計劃并非一一對應的。
在執行計劃管理中，每一條SQL對應一個基線，每個基線中包含一個或多個執行計劃。實際使用中，會根據當時的參數選擇其中代價最小的執行計劃來執行。當執行計劃緩存中的執行計劃進入執行計劃管理時，SPM會操作一個流程判斷該執行計劃是否是已知的，是已知的話，是否代價是最小的；不是已知的話，是否需要執行以判斷該執行計劃的優化程度。
運維命令
PolarDB-X提供了豐富的指令集用于管理執行計劃，語法如下：
BASELINE (LOAD|PERSIST|CLEAR|VALIDATE|LIST|DELETE) [Signed Integer,Signed Integer....]
BASELINE (ADD|FIX) SQL (HINT Select Statemtnt)BASELINE (ADD|FIX) SQL <HINT> <Select Statement>：將SQL以HINT修復過后的執行計劃記錄固定下來。
BASELINE LOAD：將系統表中指定的基線信息刷新到內存并使其生效。
BASELINE LOAD_PLAN：將系統表中指定的執行計劃信息刷新到內存并使其生效。
BASELINE LIST：列出當前所有的基線信息。
BASELINE PERSIST：將指定的基線落盤。
BASELINE PERSIST_PLAN：將指定的執行計劃落盤。
BASELINE CLEAR：清理內存中某個基線。
BASELINE CLEAR_PLAN：清理內存中某個執行計劃。
BASELINE DELETE：刪除磁盤中某個基線。
BASELINE DELETE_PLAN：刪除磁盤中某個執行計劃。
執行計劃調優實踐
數據發生變化或PolarDB-X優化器引擎升級后，針對同一條SQL，有可能會出現更好的執行計劃。SPM在自動演化時會將查詢計劃枚舉過程中自動發現的更優執行計劃加入到SQL的基線中。除此以外，您也可以通過SPM的指令主動優化執行計劃。
執行
EXPLAIN命令，發現一條SQL的執行計劃中使用的是Hash Join，并且在Baseline List的基線中，該SQL僅有這一個執行計劃。EXPLAIN SELECT * FROM lineitem JOIN part ON l_partkey=p_partkey WHERE p_name LIKE '%green%';返回執行計劃信息如下：
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | LOGICAL PLAN | +----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | Gather(parallel=true) | | ParallelHashJoin(condition="l_partkey = p_partkey", type="inner") | | LogicalView(tables="[00-03].lineitem", shardCount=4, sql="SELECT `l_orderkey`, `l_partkey`, `l_suppkey`, `l_linenumber`, `l_quantity`, `l_extendedprice`, `l_discount`, `l_tax`, `l_returnflag`, `l_linestatus`, `l_shipdate`, `l_commitdate`, `l_receiptdate`, `l_shipinstruct`, `l_shipmode`, `l_comment` FROM `lineitem` AS `lineitem`", parallel=true) | | LogicalView(tables="[00-03].part", shardCount=4, sql="SELECT `p_partkey`, `p_name`, `p_mfgr`, `p_brand`, `p_type`, `p_size`, `p_container`, `p_retailprice`, `p_comment` FROM `part` AS `part` WHERE (`p_name` LIKE ?)", parallel=true) | | HitCache:true | | | | | +----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ 7 rows in set (0.06 sec)執行以下命令，查看基線：
BASELINE LIST;返回信息如下：
+-------------+--------------------------------------------------------------------------------+------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------+----------+ | BASELINE_ID | PARAMETERIZED_SQL | PLAN_ID | EXTERNALIZED_PLAN | FIXED | ACCEPTED | +-------------+--------------------------------------------------------------------------------+------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------+----------+ | -399023558 | SELECT * FROM lineitem JOIN part ON l_partkey = p_partkey WHERE p_name LIKE ? | -935671684 | Gather(parallel=true) ParallelHashJoin(condition="l_partkey = p_partkey", type="inner") LogicalView(tables="[00-03].lineitem", shardCount=4, sql="SELECT `l_orderkey`, `l_partkey`, `l_suppkey`, `l_linenumber`, `l_quantity`, `l_extendedprice`, `l_discount`, `l_tax`, `l_returnflag`, `l_linestatus`, `l_shipdate`, `l_commitdate`, `l_receiptdate`, `l_shipinstruct`, `l_shipmode`, `l_comment` FROM `lineitem` AS `lineitem`", parallel=true) LogicalView(tables="[00-03].part", shardCount=4, sql="SELECT `p_partkey`, `p_name`, `p_mfgr`, `p_brand`, `p_type`, `p_size`, `p_container`, `p_retailprice`, `p_comment` FROM `part` AS `part` WHERE (`p_name` LIKE ?)", parallel=true) | 0 | 1 | +-------------+--------------------------------------------------------------------------------+------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------+----------+ 1 row in set (0.02 sec)假如這個SQL在某些條件下采用BKA Join（Lookup Join）會有更好的性能，那么首先需要想辦法利用HINT引導PolarDB-X生成符合預期的執行計劃。BKA Join的HINT為
/*+TDDL:BKA_JOIN(lineitem, part)*/。通過EXPLAIN [HINT] [SQL]觀察出來的執行計劃是否符合預期：EXPLAIN /*+TDDL:bka_join(lineitem, part)*/ SELECT * FROM lineitem JOIN part ON l_partkey=p_partkey WHERE p_name LIKE '%green%';返回的執行計劃信息如下：
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | LOGICAL PLAN | +----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | Gather(parallel=true) | | ParallelBKAJoin(condition="l_partkey = p_partkey", type="inner") | | LogicalView(tables="[00-03].lineitem", shardCount=4, sql="SELECT `l_orderkey`, `l_partkey`, `l_suppkey`, `l_linenumber`, `l_quantity`, `l_extendedprice`, `l_discount`, `l_tax`, `l_returnflag`, `l_linestatus`, `l_shipdate`, `l_commitdate`, `l_receiptdate`, `l_shipinstruct`, `l_shipmode`, `l_comment` FROM `lineitem` AS `lineitem`", parallel=true) | | Gather(concurrent=true) | | LogicalView(tables="[00-03].part", shardCount=4, sql="SELECT `p_partkey`, `p_name`, `p_mfgr`, `p_brand`, `p_type`, `p_size`, `p_container`, `p_retailprice`, `p_comment` FROM `part` AS `part` WHERE (`p_name` LIKE ?)") | | HitCache:false | | | | | +----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ 8 rows in set (0.14 sec)此時由于Hint的干預，Join的算法已修正為BKA Join。但是這并不會對基線造成變動，如果想以后每次遇到這條SQL都使用上面的計劃，還需要將其加入到基線中。
采用執行計劃管理的
Baseline Add指令為該SQL增加一個執行計劃。此時會同時有兩套執行計劃存在于該SQL的基線中，查詢計劃枚舉優化器會根據代價選擇一個執行計劃執行。BASELINE ADD SQL /*+TDDL:bka_join(lineitem, part)*/ SELECT * FROM lineitem JOIN part ON l_partkey=p_partkey WHERE p_name LIKE '%green%';返回信息如下：
+-------------+--------+ | BASELINE_ID | STATUS | +-------------+--------+ | -399023558 | OK | +-------------+--------+ 1 row in set (0.09 sec)執行以下命令，查看基線：
BASELINE LIST;返回信息如下：
+-------------+--------------------------------------------------------------------------------+-------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------+----------+ | BASELINE_ID | PARAMETERIZED_SQL | PLAN_ID | EXTERNALIZED_PLAN | FIXED | ACCEPTED | +-------------+--------------------------------------------------------------------------------+-------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------+----------+ | -399023558 | SELECT * FROM lineitem JOIN part ON l_partkey = p_partkey WHERE p_name LIKE ? | -1024543942 | Gather(parallel=true) ParallelBKAJoin(condition="l_partkey = p_partkey", type="inner") LogicalView(tables="[00-03].lineitem", shardCount=4, sql="SELECT `l_orderkey`, `l_partkey`, `l_suppkey`, `l_linenumber`, `l_quantity`, `l_extendedprice`, `l_discount`, `l_tax`, `l_returnflag`, `l_linestatus`, `l_shipdate`, `l_commitdate`, `l_receiptdate`, `l_shipinstruct`, `l_shipmode`, `l_comment` FROM `lineitem` AS `lineitem`", parallel=true) Gather(concurrent=true) LogicalView(tables="[00-03].part", shardCount=4, sql="SELECT `p_partkey`, `p_name`, `p_mfgr`, `p_brand`, `p_type`, `p_size`, `p_container`, `p_retailprice`, `p_comment` FROM `part` AS `part` WHERE (`p_name` LIKE ?)") | 0 | 1 | | -399023558 | SELECT * FROM lineitem JOIN part ON l_partkey = p_partkey WHERE p_name LIKE ? | -935671684 | Gather(parallel=true) ParallelHashJoin(condition="l_partkey = p_partkey", type="inner") LogicalView(tables="[00-03].lineitem", shardCount=4, sql="SELECT `l_orderkey`, `l_partkey`, `l_suppkey`, `l_linenumber`, `l_quantity`, `l_extendedprice`, `l_discount`, `l_tax`, `l_returnflag`, `l_linestatus`, `l_shipdate`, `l_commitdate`, `l_receiptdate`, `l_shipinstruct`, `l_shipmode`, `l_comment` FROM `lineitem` AS `lineitem`", parallel=true) LogicalView(tables="[00-03].part", shardCount=4, sql="SELECT `p_partkey`, `p_name`, `p_mfgr`, `p_brand`, `p_type`, `p_size`, `p_container`, `p_retailprice`, `p_comment` FROM `part` AS `part` WHERE (`p_name` LIKE ?)", parallel=true) | 0 | 1 | +-------------+--------------------------------------------------------------------------------+-------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------+----------+ 2 rows in set (0.03 sec)通過以上
Baseline List指令展示出來的結果，可以看到基于BKA_JOIN的執行計劃已增加到該SQL的基線中。此時對這條SQL執行EXPLAIN命令，發現隨SQL中p_name LIKE ?條件變化，PolarDB-X會選擇不同的執行計劃。如果需要PolarDB-X固定使用上述的執行計劃（而非在兩個中挑選一個），可以采用
Baseline Fix指令強制PolarDB-X使用指定的執行計劃。BASELINE FIX SQL /*+TDDL:bka_join(lineitem, part)*/ SELECT * FROM lineitem JOIN part ON l_partkey=p_partkey WHERE p_name LIKE '%green%';返回信息如下：
+-------------+--------+ | BASELINE_ID | STATUS | +-------------+--------+ | -399023558 | OK | +-------------+--------+ 1 row in set (0.07 sec) mysql> baseline list\G *************************** 1. row *************************** BASELINE_ID: -399023558 PARAMETERIZED_SQL: SELECT * FROM lineitem JOIN part ON l_partkey = p_partkey WHERE p_name LIKE ? PLAN_ID: -1024543942 EXTERNALIZED_PLAN: Gather(parallel=true) ParallelBKAJoin(condition="l_partkey = p_partkey", type="inner") LogicalView(tables="[00-03].lineitem", shardCount=4, sql="SELECT `l_orderkey`, `l_partkey`, `l_suppkey`, `l_linenumber`, `l_quantity`, `l_extendedprice`, `l_discount`, `l_tax`, `l_returnflag`, `l_linestatus`, `l_shipdate`, `l_commitdate`, `l_receiptdate`, `l_shipinstruct`, `l_shipmode`, `l_comment` FROM `lineitem` AS `lineitem`", parallel=true) Gather(concurrent=true) LogicalView(tables="[00-03].part", shardCount=4, sql="SELECT `p_partkey`, `p_name`, `p_mfgr`, `p_brand`, `p_type`, `p_size`, `p_container`, `p_retailprice`, `p_comment` FROM `part` AS `part` WHERE (`p_name` LIKE ?)") FIXED: 1 ACCEPTED: 1 *************************** 2. row *************************** BASELINE_ID: -399023558 PARAMETERIZED_SQL: SELECT * FROM lineitem JOIN part ON l_partkey = p_partkey WHERE p_name LIKE ? PLAN_ID: -935671684 EXTERNALIZED_PLAN: Gather(parallel=true) ParallelHashJoin(condition="l_partkey = p_partkey", type="inner") LogicalView(tables="[00-03].lineitem", shardCount=4, sql="SELECT `l_orderkey`, `l_partkey`, `l_suppkey`, `l_linenumber`, `l_quantity`, `l_extendedprice`, `l_discount`, `l_tax`, `l_returnflag`, `l_linestatus`, `l_shipdate`, `l_commitdate`, `l_receiptdate`, `l_shipinstruct`, `l_shipmode`, `l_comment` FROM `lineitem` AS `lineitem`", parallel=true) LogicalView(tables="[00-03].part", shardCount=4, sql="SELECT `p_partkey`, `p_name`, `p_mfgr`, `p_brand`, `p_type`, `p_size`, `p_container`, `p_retailprice`, `p_comment` FROM `part` AS `part` WHERE (`p_name` LIKE ?)", parallel=true) FIXED: 0 ACCEPTED: 1 2 rows in set (0.01 sec)再次執行
EXPLAIN命令，查看執行計劃。EXPLAIN SELECT * FROM lineitem JOIN part ON l_partkey=p_partkey WHERE p_name LIKE '%green%';返回執行計劃信息如下：
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | LOGICAL PLAN | +----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | Gather(parallel=true) | | ParallelBKAJoin(condition="l_partkey = p_partkey", type="inner") | | LogicalView(tables="[00-03].lineitem", shardCount=4, sql="SELECT `l_orderkey`, `l_partkey`, `l_suppkey`, `l_linenumber`, `l_quantity`, `l_extendedprice`, `l_discount`, `l_tax`, `l_returnflag`, `l_linestatus`, `l_shipdate`, `l_commitdate`, `l_receiptdate`, `l_shipinstruct`, `l_shipmode`, `l_comment` FROM `lineitem` AS `lineitem`", parallel=true) | | Gather(concurrent=true) | | LogicalView(tables="[00-03].part", shardCount=4, sql="SELECT `p_partkey`, `p_name`, `p_mfgr`, `p_brand`, `p_type`, `p_size`, `p_container`, `p_retailprice`, `p_comment` FROM `part` AS `part` WHERE (`p_name` LIKE ?)") | | HitCache:true | +----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ 8 rows in set (0.01 sec)Baseline Fix指令執行完后，可以看到BKA Join執行計劃的Fix狀態位已被置為1。此時即使不加HINT，任意條件下對這條SQL執行EXPLAIN命令，都一定會采用這個執行計劃。
文檔內容是否對您有幫助？