Skip to content

Commit

Permalink
MDEV-9750: Quick memory exhaustion with 'extended_keys=on' ...
Browse files Browse the repository at this point in the history
(Variant #5, full patch, for 10.5)

Do not produce SEL_ARG graphs that would yield huge numbers of ranges.
Introduce a concept of SEL_ARG graph's "weight". If we are about to
produce a graph whose "weight" exceeds the limit, remove the parts
of SEL_ARG graph that represent the biggest key parts. Do so until
the graph's is within the limit.

Includes
- debug code to verify SEL_ARG graph weight
- A user-visible @@optimizer_max_sel_arg_weight to control the optimization
- Logging the optimization into the optimizer trace.
  • Loading branch information
spetrunia committed Jan 29, 2021
1 parent a2eb974 commit c367203
Show file tree
Hide file tree
Showing 11 changed files with 811 additions and 158 deletions.
4 changes: 4 additions & 0 deletions mysql-test/main/mysqld--help.result
Original file line number Diff line number Diff line change
Expand Up @@ -681,6 +681,9 @@ The following specify which files/extra groups are read (specified before remain
max_connections*5 or max_connections + table_cache*2
(whichever is larger) number of file descriptors
(Automatically configured unless set explicitly)
--optimizer-max-sel-arg-weight=#
The maximum weight of the SEL_ARG graph. Set to 0 for no
limit
--optimizer-prune-level=#
Controls the heuristic(s) applied during query
optimization to prune less-promising partial plans from
Expand Down Expand Up @@ -1637,6 +1640,7 @@ old-alter-table DEFAULT
old-mode
old-passwords FALSE
old-style-user-limits FALSE
optimizer-max-sel-arg-weight 32000
optimizer-prune-level 1
optimizer-search-depth 62
optimizer-selectivity-sampling-limit 100
Expand Down
179 changes: 179 additions & 0 deletions mysql-test/main/range_notembedded.result
Original file line number Diff line number Diff line change
Expand Up @@ -35,3 +35,182 @@ json_detailed(JSON_EXTRACT(trace, '$**.ranges'))
]
set optimizer_trace=@tmp_21958;
drop table t2;
#
# MDEV-9750: Quick memory exhaustion with 'extended_keys=on'...
#
create table t1 (
kp1 int,
kp2 int,
kp3 int,
kp4 int,
key key1(kp1, kp2, kp3,kp4)
);
insert into t1 values (1,1,1,1),(2,2,2,2),(3,3,3,3);
analyze table t1;
Table Op Msg_type Msg_text
test.t1 analyze status Engine-independent statistics collected
test.t1 analyze status OK
show variables like 'optimizer_max_sel_arg_weight';
Variable_name Value
optimizer_max_sel_arg_weight 32000
set @tmp_9750=@@optimizer_trace;
set optimizer_trace=1;
explain select * from t1 where
kp1 in (1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20) and
kp2 in (1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20) and
kp3 in (1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20) and
kp4 in (1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20)
;
id select_type table type possible_keys key key_len ref rows Extra
1 SIMPLE t1 index key1 key1 20 NULL 3 Using where; Using index
set @json= (select json_detailed(JSON_EXTRACT(trace, '$**.range_scan_alternatives'))
from information_schema.optimizer_trace);
# This will show 3-component ranges.
# The ranges were produced, but the optimizer has cut away kp4
# to keep the number of ranges at manageable level:
select left(@json, 500);
left(@json, 500)
[

[

{
"index": "key1",
"ranges":
[
"(1,1,1) <= (kp1,kp2,kp3) <= (1,1,1)",
"(1,1,2) <= (kp1,kp2,kp3) <= (1,1,2)",
"(1,1,3) <= (kp1,kp2,kp3) <= (1,1,3)",
"(1,1,4) <= (kp1,kp2,kp3) <= (1,1,4)",
"(1,1,5) <= (kp1,kp2,kp3) <= (1,1,5)",
"(1,1,6) <= (kp1,kp2,kp3) <= (1,1,6)",
"(1,1,7) <= (kp1,kp2,kp3) <= (1,1,7)",
"
## Repeat the above with low max_weight:
set @tmp9750_weight=@@optimizer_max_sel_arg_weight;
set optimizer_max_sel_arg_weight=20;
explain select * from t1 where
kp1 in (1,2,3,4,5,6,7,8,9,10) and
kp2 in (1,2,3,4,5,6,7,8,9,10) and
kp3 in (1,2,3,4,5,6,7,8,9,10) and
kp4 in (1,2,3,4,5,6,7,8,9,10)
;
id select_type table type possible_keys key key_len ref rows Extra
1 SIMPLE t1 index key1 key1 20 NULL 3 Using where; Using index
set @trace= (select trace from information_schema.optimizer_trace);
set @json= json_detailed(json_extract(@trace, '$**.range_scan_alternatives'));
select left(@json, 500);
left(@json, 500)
[

[

{
"index": "key1",
"ranges":
[
"(1) <= (kp1) <= (1)",
"(2) <= (kp1) <= (2)",
"(3) <= (kp1) <= (3)",
"(4) <= (kp1) <= (4)",
"(5) <= (kp1) <= (5)",
"(6) <= (kp1) <= (6)",
"(7) <= (kp1) <= (7)",
"(8) <= (kp1) <= (8)",
"(9) <= (kp1) <= (9)",
"(10) <= (kp1) <= (10)"

set @json= json_detailed(json_extract(@trace, '$**.setup_range_conditions'));
select left(@json, 2500);
left(@json, 2500)
[

[

{
"sel_arg_weight_heuristic":
{
"key1_field": "kp1",
"key2_field": "kp2",
"key1_weight": 10,
"key2_weight": 10
}
},

{
"sel_arg_weight_heuristic":
{
"key1_field": "kp1",
"key2_field": "kp3",
"key1_weight": 10,
"key2_weight": 10
}
},

{
"sel_arg_weight_heuristic":
{
"key1_field": "kp1",
"key2_field": "kp4",
"key1_weight": 10,
"key2_weight": 10
}
}
]
]
## Repeat the above with a bit higher max_weight:
set @tmp9750_weight=@@optimizer_max_sel_arg_weight;
set optimizer_max_sel_arg_weight=120;
explain select * from t1 where
kp1 in (1,2,3,4,5,6,7,8,9,10) and
kp2 in (1,2,3,4,5,6,7,8,9,10) and
kp3 in (1,2,3,4,5,6,7,8,9,10) and
kp4 in (1,2,3,4,5,6,7,8,9,10)
;
id select_type table type possible_keys key key_len ref rows Extra
1 SIMPLE t1 index key1 key1 20 NULL 3 Using where; Using index
set @json= (select json_detailed(JSON_EXTRACT(trace, '$**.range_scan_alternatives'))
from information_schema.optimizer_trace);
select left(@json, 1500);
left(@json, 1500)
[

[

{
"index": "key1",
"ranges":
[
"(1,1) <= (kp1,kp2) <= (1,1)",
"(1,2) <= (kp1,kp2) <= (1,2)",
"(1,3) <= (kp1,kp2) <= (1,3)",
"(1,4) <= (kp1,kp2) <= (1,4)",
"(1,5) <= (kp1,kp2) <= (1,5)",
"(1,6) <= (kp1,kp2) <= (1,6)",
"(1,7) <= (kp1,kp2) <= (1,7)",
"(1,8) <= (kp1,kp2) <= (1,8)",
"(1,9) <= (kp1,kp2) <= (1,9)",
"(1,10) <= (kp1,kp2) <= (1,10)",
"(2,1) <= (kp1,kp2) <= (2,1)",
"(2,2) <= (kp1,kp2) <= (2,2)",
"(2,3) <= (kp1,kp2) <= (2,3)",
"(2,4) <= (kp1,kp2) <= (2,4)",
"(2,5) <= (kp1,kp2) <= (2,5)",
"(2,6) <= (kp1,kp2) <= (2,6)",
"(2,7) <= (kp1,kp2) <= (2,7)",
"(2,8) <= (kp1,kp2) <= (2,8)",
"(2,9) <= (kp1,kp2) <= (2,9)",
"(2,10) <= (kp1,kp2) <= (2,10)",
"(3,1) <= (kp1,kp2) <= (3,1)",
"(3,2) <= (kp1,kp2) <= (3,2)",
"(3,3) <= (kp1,kp2) <= (3,3)",
"(3,4) <= (kp1,kp2) <= (3,4)",
"(3,5) <= (kp1,kp2) <= (3,5)",
"(3,6) <= (kp1,kp2) <= (3,6)",
"(3,7) <= (kp1,kp2) <= (3,7)",
"(3,8) <= (kp1,kp2) <= (3,8)",
"(3,9) <= (kp1,kp2) <= (3,9)",
"(3,10) <= (kp1,kp2
set optimizer_max_sel_arg_weight= @tmp9750_weight;
set optimizer_trace=@tmp_9750;
drop table t1;
66 changes: 66 additions & 0 deletions mysql-test/main/range_notembedded.test
Original file line number Diff line number Diff line change
Expand Up @@ -31,3 +31,69 @@ from information_schema.optimizer_trace;
set optimizer_trace=@tmp_21958;

drop table t2;
--echo #
--echo # MDEV-9750: Quick memory exhaustion with 'extended_keys=on'...
--echo #

create table t1 (
kp1 int,
kp2 int,
kp3 int,
kp4 int,
key key1(kp1, kp2, kp3,kp4)
);

insert into t1 values (1,1,1,1),(2,2,2,2),(3,3,3,3);
analyze table t1;

show variables like 'optimizer_max_sel_arg_weight';

# 20 * 20 * 20 *20 = 400*400 = 160,000 ranges
set @tmp_9750=@@optimizer_trace;
set optimizer_trace=1;
explain select * from t1 where
kp1 in (1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20) and
kp2 in (1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20) and
kp3 in (1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20) and
kp4 in (1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20)
;

set @json= (select json_detailed(JSON_EXTRACT(trace, '$**.range_scan_alternatives'))
from information_schema.optimizer_trace);
--echo # This will show 3-component ranges.
--echo # The ranges were produced, but the optimizer has cut away kp4
--echo # to keep the number of ranges at manageable level:
select left(@json, 500);

--echo ## Repeat the above with low max_weight:
set @tmp9750_weight=@@optimizer_max_sel_arg_weight;
set optimizer_max_sel_arg_weight=20;
explain select * from t1 where
kp1 in (1,2,3,4,5,6,7,8,9,10) and
kp2 in (1,2,3,4,5,6,7,8,9,10) and
kp3 in (1,2,3,4,5,6,7,8,9,10) and
kp4 in (1,2,3,4,5,6,7,8,9,10)
;
set @trace= (select trace from information_schema.optimizer_trace);
set @json= json_detailed(json_extract(@trace, '$**.range_scan_alternatives'));
select left(@json, 500);

set @json= json_detailed(json_extract(@trace, '$**.setup_range_conditions'));
select left(@json, 2500);

--echo ## Repeat the above with a bit higher max_weight:
set @tmp9750_weight=@@optimizer_max_sel_arg_weight;
set optimizer_max_sel_arg_weight=120;
explain select * from t1 where
kp1 in (1,2,3,4,5,6,7,8,9,10) and
kp2 in (1,2,3,4,5,6,7,8,9,10) and
kp3 in (1,2,3,4,5,6,7,8,9,10) and
kp4 in (1,2,3,4,5,6,7,8,9,10)
;
set @json= (select json_detailed(JSON_EXTRACT(trace, '$**.range_scan_alternatives'))
from information_schema.optimizer_trace);
select left(@json, 1500);

set optimizer_max_sel_arg_weight= @tmp9750_weight;
set optimizer_trace=@tmp_9750;
drop table t1;
Loading

0 comments on commit c367203

Please sign in to comment.