mirror of
https://github.com/MariaDB/server.git
synced 2025-10-13 11:19:13 +02:00
b3e70bde38This is done by no longer merging inner table dep with outer nested_join ones. This expands the search space of outer joins and introduces suboptimal join order choices already existing in inner joins (e.g. MDEV-36331, MDEV-37346) to outer joins, caused by heuristic pruning. To mitigate this issue, we add a switch that turns on heuristic pruning always (existing behaviour), and make is so that when the switch is off (default) joins with limited number of tables or limited search space* do not perform heuristic pruning. This improves join order optimization for these joins, where otherwise heuristic pruning may remove good joins - see added test cases in the test main.greedy_optimizer. *: about the same as 10 tables with default search depth (7) when optimizer_search_depth=0. If we completely disable heuristic pruning, or use the upper bound 10!, the test main.greedy_optimizer will grind to almost hanging for optimizer_search_depth=0 when more than 20 tables are involved. If we only compare the number of tables with 10, then it will prune when search space is low due to a low search depth when we can afford not to prune.
139 lines
3.7 KiB
Text
139 lines
3.7 KiB
Text
create table t1 (a int, b int, c int);
|
|
insert into t1 select seq,seq/2, seq/4 from seq_1_to_100;
|
|
create table t2 (a int, b int, c int);
|
|
insert into t2 select seq, seq/2, seq/4 from seq_1_to_200;
|
|
analyze table t1,t2 persistent for all;
|
|
Table Op Msg_type Msg_text
|
|
test.t1 analyze status Engine-independent statistics collected
|
|
test.t1 analyze status OK
|
|
test.t2 analyze status Engine-independent statistics collected
|
|
test.t2 analyze status OK
|
|
set optimizer_trace=1;
|
|
set join_cache_level=6;
|
|
set optimizer_switch='hash_join_cardinality=on';
|
|
explain select *
|
|
from t1, t2
|
|
where t1.a=t2.a and t1.a=t2.b and t1.c=t2.c;
|
|
id select_type table type possible_keys key key_len ref rows Extra
|
|
1 SIMPLE t1 ALL NULL NULL NULL NULL 100 Using where
|
|
1 SIMPLE t2 hash_ALL NULL #hash#$hj 15 test.t1.a,test.t1.a,test.t1.c 200 Using where; Using join buffer (flat, BNLH join)
|
|
set @json= (select trace from information_schema.optimizer_trace);
|
|
select json_detailed(json_extract(@json, '$**.hash_join_cardinality')) as JS;
|
|
JS
|
|
[
|
|
{
|
|
"hash_join_columns":
|
|
[
|
|
{
|
|
"field": "a",
|
|
"avg_frequency": 1
|
|
},
|
|
{
|
|
"field": "b",
|
|
"avg_frequency": 2
|
|
},
|
|
{
|
|
"field": "c",
|
|
"avg_frequency": 3.9216
|
|
}
|
|
],
|
|
"rows": 1
|
|
},
|
|
{
|
|
"hash_join_columns":
|
|
[
|
|
{
|
|
"field": "a",
|
|
"avg_frequency": 1
|
|
},
|
|
{
|
|
"field": "a",
|
|
"avg_frequency": 1
|
|
},
|
|
{
|
|
"field": "c",
|
|
"avg_frequency": 3.8462
|
|
}
|
|
],
|
|
"rows": 1
|
|
}
|
|
]
|
|
select json_detailed(json_extract(@json, '$**.rest_of_plan[*].rows_for_plan'))
|
|
as ROWS_FOR_PLAN;
|
|
ROWS_FOR_PLAN
|
|
[
|
|
100,
|
|
200
|
|
]
|
|
explain select *
|
|
from t1, t2 where t1.c=t2.c;
|
|
id select_type table type possible_keys key key_len ref rows Extra
|
|
1 SIMPLE t1 ALL NULL NULL NULL NULL 100 Using where
|
|
1 SIMPLE t2 hash_ALL NULL #hash#$hj 5 test.t1.c 200 Using where; Using join buffer (flat, BNLH join)
|
|
set @json= (select trace from information_schema.optimizer_trace);
|
|
select json_detailed(json_extract(@json, '$**.hash_join_cardinality')) as JS;
|
|
JS
|
|
[
|
|
{
|
|
"hash_join_columns":
|
|
[
|
|
{
|
|
"field": "c",
|
|
"avg_frequency": 3.9216
|
|
}
|
|
],
|
|
"rows": 3.9216
|
|
},
|
|
{
|
|
"hash_join_columns":
|
|
[
|
|
{
|
|
"field": "c",
|
|
"avg_frequency": 3.8462
|
|
}
|
|
],
|
|
"rows": 3.8462
|
|
}
|
|
]
|
|
select json_detailed(json_extract(@json, '$**.rest_of_plan[*].rows_for_plan'))
|
|
as ROWS_FOR_PLAN;
|
|
ROWS_FOR_PLAN
|
|
[
|
|
392.16,
|
|
769.24
|
|
]
|
|
explain select *
|
|
from t1 straight_join t2 where t1.c=t2.c and t2.a<30;
|
|
id select_type table type possible_keys key key_len ref rows Extra
|
|
1 SIMPLE t1 ALL NULL NULL NULL NULL 100 Using where
|
|
1 SIMPLE t2 hash_ALL NULL #hash#$hj 5 test.t1.c 200 Using where; Using join buffer (flat, BNLH join)
|
|
set @json= (select trace from information_schema.optimizer_trace);
|
|
# Note that rows is the same:
|
|
select json_detailed(json_extract(@json, '$**.hash_join_cardinality')) as JS;
|
|
JS
|
|
[
|
|
{
|
|
"hash_join_columns":
|
|
[
|
|
{
|
|
"field": "c",
|
|
"avg_frequency": 3.9216
|
|
}
|
|
],
|
|
"rows": 0.568632
|
|
}
|
|
]
|
|
# Despite available selectivity:
|
|
select json_detailed(json_extract(@json, '$**.selectivity_for_columns')) as JS;
|
|
JS
|
|
[
|
|
[
|
|
{
|
|
"column_name": "a",
|
|
"ranges":
|
|
["NULL < a < 30"],
|
|
"selectivity_from_histogram": 0.145
|
|
}
|
|
]
|
|
]
|
|
drop table t1,t2;
|