1
0
mirror of https://github.com/MariaDB/server.git synced 2025-07-29 05:21:33 +03:00

MDEV-26301 Split optimization refills temporary table too many times

This patch optimizes the number of refills for the lateral derived table
to which a materialized derived table subject to split optimization is
is converted. This optimized number of refills is now considered as the
expected number of refills of the materialized derived table when searching
for the best possible splitting of the table.
This commit is contained in:
Igor Babaev
2023-05-02 23:17:07 -07:00
committed by Oleksandr Byelkin
parent ec79f37718
commit ce7ffe61d8
5 changed files with 910 additions and 32 deletions

View File

@ -227,3 +227,216 @@ where t3.b > 15;
drop table t3, t4;
--echo # End of 10.3 tests
--source include/have_sequence.inc
--echo #
--echo # MDEV-26301: Split optimization refills temporary table too many times
--echo #
# 5 values
create table t1(a int, b int);
insert into t1 select seq,seq from seq_1_to_5;
# 5 value groups of size 2 each
create table t2(a int, b int, key(a));
insert into t2
select A.seq,B.seq from seq_1_to_25 A, seq_1_to_2 B;
# 5 value groups of size 3 each
create table t3(a int, b int, key(a));
insert into t3
select A.seq,B.seq from seq_1_to_5 A, seq_1_to_3 B;
analyze table t1,t2,t3 persistent for all;
explain
select * from
(t1 left join t2 on t2.a=t1.b) left join t3 on t3.a=t1.b;
# Now, create tables for Groups.
create table t10 (
grp_id int,
col1 int,
key(grp_id)
);
# 100 groups of 100 values each
insert into t10
select
A.seq,
B.seq
from
seq_1_to_100 A,
seq_1_to_100 B;
# and X10 multiplier
create table t11 (
col1 int,
col2 int
);
insert into t11
select A.seq, A.seq from seq_1_to_10 A;
analyze table t10,t11 persistent for all;
let $q1=
select * from
(
(t1 left join t2 on t2.a=t1.b)
left join t3 on t3.a=t1.b
) left join (select grp_id, count(*)
from t10 left join t11 on t11.col1=t10.col1
group by grp_id) T on T.grp_id=t1.b;
eval
explain $q1;
--echo # The important part in the below output is:
--echo # "lateral": 1,
--echo # "query_block": {
--echo # "select_id": 2,
--echo # "r_loops": 5, <-- must be 5, not 30.
--source include/analyze-format.inc
eval
analyze format=json $q1;
create table t21 (pk int primary key);
insert into t21 values (1),(2),(3);
create table t22 (pk int primary key);
insert into t22 values (1),(2),(3);
# Same as above but throw in a couple of const tables.
explain
select * from
t21, t22,
(
(t1 left join t2 on t2.a=t1.b)
left join t3 on t3.a=t1.b
) left join (select grp_id, count(*)
from t10 left join t11 on t11.col1=t10.col1
group by grp_id) T on T.grp_id=t1.b
where
t21.pk=1 and t22.pk=2;
explain
select * from
t21,
(
(t1 left join t2 on t2.a=t1.b)
left join t3 on t3.a=t1.b
) left join (select grp_id, count(*)
from
t22 join t10 left join t11 on t11.col1=t10.col1
where
t22.pk=1
group by grp_id) T on T.grp_id=t1.b
where
t21.pk=1;
# And also add a non-const table
create table t5 (
pk int primary key
);
insert into t5 select seq from seq_1_to_1000;
explain
select * from
t21,
(
(((t1 join t5 on t5.pk=t1.b)) left join t2 on t2.a=t1.b)
left join t3 on t3.a=t1.b
) left join (select grp_id, count(*)
from
t22 join t10 left join t11 on t11.col1=t10.col1
where
t22.pk=1
group by grp_id) T on T.grp_id=t1.b
where
t21.pk=1;
drop table t1,t2,t3,t5, t10, t11, t21, t22;
# 5 values
create table t1(a int, b int);
insert into t1 select seq,seq from seq_1_to_5;
# 5 value groups of size 2 each
create table t2(a int, b int, key(a));
insert into t2
select A.seq,B.seq from seq_1_to_25 A, seq_1_to_2 B;
# 5 value groups of size 3 each
create table t3(a int, b int, key(a));
insert into t3
select A.seq,B.seq from seq_1_to_5 A, seq_1_to_3 B;
analyze table t1,t2,t3 persistent for all;
create table t10 (
grp_id int,
col1 int,
key(grp_id)
);
# 100 groups of 100 values each
insert into t10
select
A.seq,
B.seq
from
seq_1_to_100 A,
seq_1_to_100 B;
# and X10 multiplier
create table t11 (
col1 int,
col2 int
);
insert into t11
select A.seq, A.seq from seq_1_to_10 A;
analyze table t10,t11 persistent for all;
let $q=
select *
from
(
(t1 left join t2 on t2.a=t1.b)
left join
t3
on t3.a=t1.b
)
left join
(
select grp_id, count(*)
from t10 left join t11 on t11.col1=t10.col1
group by grp_id
)dt
on dt.grp_id=t1.b;
eval explain $q;
eval $q;
set join_cache_level=4;
eval explain $q;
eval $q;
set join_cache_level=default;
drop index a on t2;
drop index a on t3;
eval explain $q;
eval $q;
drop table t1,t2,t3;
drop table t10, t11;
--echo # End of 10.4 tests