Complete implementation of WL#1469 "Greedy algorithm to search for an optimal execution plan",

consisting of pos-review fixes and improvements.

mysql-test/r/distinct.result

@@ -388,8 +388,8 @@ SELECT DISTINCTROW email, shipcode FROM t1, t2 WHERE t1.infoID=t2.infoID;
 email	shipcode
 test1@testdomain.com	Z001
 test2@testdomain.com	Z001
-test3@testdomain.com	Z001
 test2@testdomain.com	R002
+test3@testdomain.com	Z001
 SELECT DISTINCTROW email FROM t1 ORDER BY dateentered DESC;
 email
 test1@testdomain.com

mysql-test/r/func_group.result

@@ -183,12 +183,12 @@ insert into t2 values('BBB', 20, 1.0);
 select t1.a1, t1.a2, t2.a1, t2.a2 from t1,t2;
 a1	a2	a1	a2
 10	aaa	AAA	10
-10	NULL	AAA	10
-10	bbb	AAA	10
-20	zzz	AAA	10
 10	aaa	BBB	20
+10	NULL	AAA	10
 10	NULL	BBB	20
+10	bbb	AAA	10
 10	bbb	BBB	20
+20	zzz	AAA	10
 20	zzz	BBB	20
 select max(t1.a1), max(t2.a1) from t1, t2 where t2.a2=9;
 max(t1.a1)	max(t2.a1)

mysql-test/r/select.result

@@ -1393,8 +1393,8 @@ companynr	companynr
 41	40
 explain select distinct t2.companynr,t4.companynr from t2,t4 where t2.companynr=t4.companynr+1;
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
-1	SIMPLE	t2	ALL	NULL	NULL	NULL	NULL	1199	Using temporary
-1	SIMPLE	t4	index	NULL	PRIMARY	1	NULL	12	Using where; Using index
+1	SIMPLE	t4	index	NULL	PRIMARY	1	NULL	12	Using index; Using temporary
+1	SIMPLE	t2	ALL	NULL	NULL	NULL	NULL	1199	Using where
 select t2.fld1,t2.companynr,fld3,period from t3,t2 where t2.fld1 = 38208 and t2.fld1=t3.t2nr and period = 1008 or t2.fld1 = 38008 and t2.fld1 =t3.t2nr and period = 1008;
 fld1	companynr	fld3	period
 038008	37	reporters	1008
@@ -2314,8 +2314,8 @@ left join t4 on id3 = id4 where id2 = 1 or id4 = 1;
 id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
 1	SIMPLE	t3	system	NULL	NULL	NULL	NULL	0	const row not found
 1	SIMPLE	t1	ALL	NULL	NULL	NULL	NULL	2	
-1	SIMPLE	t2	ALL	NULL	NULL	NULL	NULL	1	
-1	SIMPLE	t4	ALL	id4	NULL	NULL	NULL	1	Using where
+1	SIMPLE	t4	ALL	id4	NULL	NULL	NULL	1	
+1	SIMPLE	t2	ALL	NULL	NULL	NULL	NULL	1	Using where
 select * from t1 left join t2 on id1 = id2 left join t3 on id1 = id3
 left join t4 on id3 = id4 where id2 = 1 or id4 = 1;
 id1	id2	id3	id4	id44

mysql-test/t/greedy_optimizer.test

@@ -135,32 +135,39 @@ insert into t7 values (21,2,3,4,5,6);
 # The actual test begins here
 #
 
-# first check the default values for the optimizer paramters
+# Check the default values for the optimizer paramters
+
 select @@plan_search_depth;
 select @@heuristic;
 
--- These are all possible values for the parameters that control the optimizer
--- (total 8 combinations):
+-- This value swithes back to the old implementation of 'find_best()'
+-- set plan_search_depth=63; - old (independent of the heuristic)
+--
+-- These are the values for the parameters that control the greedy optimizer
+-- (total 6 combinations - 3 for plan_search_depth, 2 for heuristic):
 --
--- set plan_search_depth=0;  -- automatic
--- set plan_search_depth=1;  -- min
--- set plan_search_depth=62; -- max - default
--- set plan_search_depth=63; -- old
--- select @@plan_search_depth;
+-- set plan_search_depth=0;  - automatic
+-- set plan_search_depth=1;  - min
+-- set plan_search_depth=62; - max (default)
 --
--- set heuristic=0 -- exhaustive;
--- set heuristic=1 -- heuristic; -- default
--- select @@heuristic;
+-- set heuristic=0 - exhaustive;
+-- set heuristic=1 - heuristic; -- default
+
 
 #
-# Compile some simple queries with all combinations of the query
-# optimizer parameters.
+# Compile several queries with all combinations of the query
+# optimizer parameters. Each test query has two variants, where
+# in the second variant the tables in the FROM clause are in
+# inverse order to the tables in the first variant.
+# Due to pre-sorting of tables before compilation, there should
+# be no difference in the plans for each two such query variants.
 #
 
-set heuristic=0;
-select @@heuristic;
+# First, for reference compile the test queries with the 'old' optimization
+# procedure 'find_best'. Notice that 'find_best' does not depend on the
+# choice of heuristic.
 
-set plan_search_depth=0;
+set plan_search_depth=63;
 select @@plan_search_depth;
 
 -- 6-table join, chain
@@ -179,7 +186,13 @@ show status like 'Last_query_cost';
 explain select t1.c11 from t7, t6, t5, t4, t3, t2, t1 where t1.c11 = t2.c21 and t1.c12 = t3.c31 and t1.c13 = t4.c41 and t1.c14 = t5.c51 and t1.c15 = t6.c61 and t1.c16 = t7.c71 and t2.c22 = t3.c32 and t2.c23 = t4.c42 and t2.c24 = t5.c52 and t2.c25 = t6.c62 and t2.c26 = t7.c72 and t3.c33 = t4.c43 and t3.c34 = t5.c53 and t3.c35 = t6.c63 and t3.c36 = t7.c73 and t4.c42 = t5.c54 and t4.c43 = t6.c64 and t4.c44 = t7.c74 and t5.c52 = t6.c65 and t5.c53 = t7.c75 and t6.c62 = t7.c76;
 show status like 'Last_query_cost';
 
-set plan_search_depth=1;
+
+# Test the new optimization procedures
+
+set heuristic=0;
+select @@heuristic;
+
+set plan_search_depth=0;
 select @@plan_search_depth;
 
 -- 6-table join, chain
@@ -198,7 +211,7 @@ show status like 'Last_query_cost';
 explain select t1.c11 from t7, t6, t5, t4, t3, t2, t1 where t1.c11 = t2.c21 and t1.c12 = t3.c31 and t1.c13 = t4.c41 and t1.c14 = t5.c51 and t1.c15 = t6.c61 and t1.c16 = t7.c71 and t2.c22 = t3.c32 and t2.c23 = t4.c42 and t2.c24 = t5.c52 and t2.c25 = t6.c62 and t2.c26 = t7.c72 and t3.c33 = t4.c43 and t3.c34 = t5.c53 and t3.c35 = t6.c63 and t3.c36 = t7.c73 and t4.c42 = t5.c54 and t4.c43 = t6.c64 and t4.c44 = t7.c74 and t5.c52 = t6.c65 and t5.c53 = t7.c75 and t6.c62 = t7.c76;
 show status like 'Last_query_cost';
 
-set plan_search_depth=62;
+set plan_search_depth=1;
 select @@plan_search_depth;
 
 -- 6-table join, chain
@@ -217,7 +230,7 @@ show status like 'Last_query_cost';
 explain select t1.c11 from t7, t6, t5, t4, t3, t2, t1 where t1.c11 = t2.c21 and t1.c12 = t3.c31 and t1.c13 = t4.c41 and t1.c14 = t5.c51 and t1.c15 = t6.c61 and t1.c16 = t7.c71 and t2.c22 = t3.c32 and t2.c23 = t4.c42 and t2.c24 = t5.c52 and t2.c25 = t6.c62 and t2.c26 = t7.c72 and t3.c33 = t4.c43 and t3.c34 = t5.c53 and t3.c35 = t6.c63 and t3.c36 = t7.c73 and t4.c42 = t5.c54 and t4.c43 = t6.c64 and t4.c44 = t7.c74 and t5.c52 = t6.c65 and t5.c53 = t7.c75 and t6.c62 = t7.c76;
 show status like 'Last_query_cost';
 
-set plan_search_depth=63;
+set plan_search_depth=62;
 select @@plan_search_depth;
 
 -- 6-table join, chain
@@ -297,23 +310,4 @@ show status like 'Last_query_cost';
 explain select t1.c11 from t7, t6, t5, t4, t3, t2, t1 where t1.c11 = t2.c21 and t1.c12 = t3.c31 and t1.c13 = t4.c41 and t1.c14 = t5.c51 and t1.c15 = t6.c61 and t1.c16 = t7.c71 and t2.c22 = t3.c32 and t2.c23 = t4.c42 and t2.c24 = t5.c52 and t2.c25 = t6.c62 and t2.c26 = t7.c72 and t3.c33 = t4.c43 and t3.c34 = t5.c53 and t3.c35 = t6.c63 and t3.c36 = t7.c73 and t4.c42 = t5.c54 and t4.c43 = t6.c64 and t4.c44 = t7.c74 and t5.c52 = t6.c65 and t5.c53 = t7.c75 and t6.c62 = t7.c76;
 show status like 'Last_query_cost';
 
-set plan_search_depth=63;
-select @@plan_search_depth;
-
--- 6-table join, chain
-explain select t1.c11 from t1, t2, t3, t4, t5, t6, t7 where t1.c12 = t2.c21 and t2.c22 = t3.c31 and t3.c32 = t4.c41 and t4.c42 = t5.c51 and t5.c52 = t6.c61 and t6.c62 = t7.c71;
-show status like 'Last_query_cost';
-explain select t1.c11 from t7, t6, t5, t4, t3, t2, t1 where t1.c12 = t2.c21 and t2.c22 = t3.c31 and t3.c32 = t4.c41 and t4.c42 = t5.c51 and t5.c52 = t6.c61 and t6.c62 = t7.c71;
-show status like 'Last_query_cost';
--- 6-table join, star
-explain select t1.c11 from t1, t2, t3, t4, t5, t6, t7 where t1.c11 = t2.c21 and t1.c12 = t3.c31 and t1.c13 = t4.c41 and t1.c14 = t5.c51 and t1.c15 = t6.c61 and t1.c16 = t7.c71;
-show status like 'Last_query_cost';
-explain select t1.c11 from t7, t6, t5, t4, t3, t2, t1 where t1.c11 = t2.c21 and t1.c12 = t3.c31 and t1.c13 = t4.c41 and t1.c14 = t5.c51 and t1.c15 = t6.c61 and t1.c16 = t7.c71;
-show status like 'Last_query_cost';
--- 6-table join, clique
-explain select t1.c11 from t1, t2, t3, t4, t5, t6, t7 where t1.c11 = t2.c21 and t1.c12 = t3.c31 and t1.c13 = t4.c41 and t1.c14 = t5.c51 and t1.c15 = t6.c61 and t1.c16 = t7.c71 and t2.c22 = t3.c32 and t2.c23 = t4.c42 and t2.c24 = t5.c52 and t2.c25 = t6.c62 and t2.c26 = t7.c72 and t3.c33 = t4.c43 and t3.c34 = t5.c53 and t3.c35 = t6.c63 and t3.c36 = t7.c73 and t4.c42 = t5.c54 and t4.c43 = t6.c64 and t4.c44 = t7.c74 and t5.c52 = t6.c65 and t5.c53 = t7.c75 and t6.c62 = t7.c76;
-show status like 'Last_query_cost';
-explain select t1.c11 from t7, t6, t5, t4, t3, t2, t1 where t1.c11 = t2.c21 and t1.c12 = t3.c31 and t1.c13 = t4.c41 and t1.c14 = t5.c51 and t1.c15 = t6.c61 and t1.c16 = t7.c71 and t2.c22 = t3.c32 and t2.c23 = t4.c42 and t2.c24 = t5.c52 and t2.c25 = t6.c62 and t2.c26 = t7.c72 and t3.c33 = t4.c43 and t3.c34 = t5.c53 and t3.c35 = t6.c63 and t3.c36 = t7.c73 and t4.c42 = t5.c54 and t4.c43 = t6.c64 and t4.c44 = t7.c74 and t5.c52 = t6.c65 and t5.c53 = t7.c75 and t6.c62 = t7.c76;
-show status like 'Last_query_cost';
-
 drop table t1,t2,t3,t4,t5,t6,t7;

sql/mysql_priv.h

@@ -769,6 +769,8 @@ extern "C" pthread_handler_decl(handle_manager, arg);
 void print_where(COND *cond,const char *info);
 void print_cached_tables(void);
 void TEST_filesort(SORT_FIELD *sortorder,uint s_length);
+void print_plan(JOIN* join, double read_time, double record_count,
+                uint idx, const char *info);
 #endif
 void mysql_print_status(THD *thd);
 /* key.cc */

sql/sql_test.cc

@@ -232,6 +232,102 @@ TEST_join(JOIN *join)
   DBUG_VOID_RETURN;
 }
 
+
+/* 
+  Print the current state during query optimization.
+
+  SYNOPSIS
+    print_plan()
+    join         pointer to the structure providing all context info for
+                 the query
+    read_time    the cost of the best partial plan
+    record_count estimate for the number of records returned by the best
+                 partial plan
+    idx          length of the partial QEP in 'join->positions';
+                 also an index in the array 'join->best_ref';
+    info         comment string to appear above the printout
+
+  DESCRIPTION
+    This function prints to the log file DBUG_FILE the members of 'join' that
+    are used during query optimization (join->positions, join->best_positions,
+    and join->best_ref) and few other related variables (read_time,
+    record_count).
+    Useful to trace query optimizer functions.
+
+  RETURN
+    None
+*/
+
+void
+print_plan(JOIN* join, double read_time, double record_count,
+           uint idx, const char *info)
+{
+  uint i;
+  POSITION pos;
+  JOIN_TAB *join_table;
+  JOIN_TAB **plan_nodes;
+  TABLE*   table;
+
+  if (info == 0)
+    info= "";
+
+  DBUG_LOCK_FILE;
+  if (join->best_read == DBL_MAX)
+  {
+    fprintf(DBUG_FILE,"%s; idx:%u, best: DBL_MAX, current:%g\n",
+            info, idx, read_time);
+  }
+  else
+  {
+    fprintf(DBUG_FILE,"%s; idx: %u, best: %g, current: %g\n",
+            info, idx, join->best_read, read_time);
+  }
+
+  /* Print the tables in JOIN->positions */
+  fputs("     POSITIONS: ", DBUG_FILE);
+  for (i= 0; i < idx ; i++)
+  {
+    pos = join->positions[i];
+    table= pos.table->table;
+    if (table)
+      fputs(table->real_name, DBUG_FILE);
+    fputc(' ', DBUG_FILE);
+  }
+  fputc('\n', DBUG_FILE);
+
+  /*
+    Print the tables in JOIN->best_positions only if at least one complete plan
+    has been found. An indicator for this is the value of 'join->best_read'.
+  */
+  fputs("BEST_POSITIONS: ", DBUG_FILE);
+  if (join->best_read < DBL_MAX)
+  {
+    for (i= 0; i < idx ; i++)
+    {
+      pos= join->best_positions[i];
+      table= pos.table->table;
+      if (table)
+        fputs(table->real_name, DBUG_FILE);
+      fputc(' ', DBUG_FILE);
+    }
+  }
+  fputc('\n', DBUG_FILE);
+
+  /* Print the tables in JOIN->best_ref */
+  fputs("      BEST_REF: ", DBUG_FILE);
+  for (plan_nodes= join->best_ref ; *plan_nodes ; plan_nodes++)
+  {
+    join_table= (*plan_nodes);
+    fputs(join_table->table->real_name, DBUG_FILE);
+    fprintf(DBUG_FILE, "(%u,%u,%u)",
+            join_table->found_records, join_table->records, join_table->read_time);
+    fputc(' ', DBUG_FILE);
+  }
+  fputc('\n', DBUG_FILE);
+
+  DBUG_UNLOCK_FILE;
+}
+
 #endif
 
 typedef struct st_debug_lock