文章主要用来记录，我在分析线上环境的慢查询时的分析工具。
问题描述
标签
programming
mysql
字数
3062 字
阅读时间
18 分钟
有一个两张表 join 的查询耗时 40s 甚至 60s。花费了很多时间才做了优化。
explain 语句
explain analyze 语句
optimizer_trace 分析 explain analyze
使用 mysql histogram 来验证 optimizer_trace 的分析结果
根据 histogram 的结果，可以尝试强制指定 join 时的驱动表。 因此 straight_join 可以解决问题
实验环境
mysql 版本 8.0.37 docker image OS macos 15 OrbStack 分配了 8 G RAM CPU M3 Pro
我们聚一个例子，需要两张表，如下
第一张 users 表
sql
users 表插入数据 500_000 行，其中 deleted_at is not null 的数据很少（不到 10 条）
第二张 posts 表
sql
posts 表插入数据 1500_000 行
查询如下
sql
SELECT
posts.*
FROM
posts
INNER JOIN users ON users.id = posts.user_id
AND users.deleted_at IS NULL
ORDER BY
posts.id DESC
LIMIT 100 OFFSET 0;explain 结果
sql
mysql> EXPLAIN SELECT posts.* FROM posts INNER JOIN users ON users.id = posts.user_id AND users.deleted_at IS NULL ORDER BY posts.id DESC LIMIT 100 OFFSET 0\G
*************************** 1. row ***************************
id: 1
select_type: SIMPLE
table: users
partitions: NULL
type: ALL
possible_keys: PRIMARY
key: NULL
key_len: NULL
ref: NULL
rows: 497082
filtered: 10.00
Extra: Using where; Using temporary; Using filesort
*************************** 2. row ***************************
id: 1
select_type: SIMPLE
table: posts
partitions: NULL
type: ref
possible_keys: user_id
key: user_id
key_len: 8
ref: sql_tests.users.id
rows: 17
filtered: 100.00
Extra: NULL
2 rows in set, 1 warning (0.01 sec)explain analyze 结果
sql
mysql> EXPLAIN ANALYZE SELECT posts.* FROM posts INNER JOIN users ON users.id = posts.user_id AND users.deleted_at IS NULL ORDER BY posts.id DESC LIMIT 100 OFFSET 0\G
*************************** 1. row ***************************
EXPLAIN: -> Limit: 100 row(s) (actual time=48196..48196 rows=100 loops=1)
-> Sort: posts.id DESC, limit input to 100 row(s) per chunk (actual time=48196..48196 rows=100 loops=1)
-> Stream results (cost=357662 rows=877647) (actual time=0.757..47686 rows=15e+6 loops=1)
-> Nested loop inner join (cost=357662 rows=877647) (actual time=0.741..46726 rows=15e+6 loops=1)
-> Filter: (users.deleted_at is null) (cost=50485 rows=49708) (actual time=0.318..133 rows=500000 loops=1)
-> Table scan on users (cost=50485 rows=497082) (actual time=0.315..114 rows=500000 loops=1)
-> Index lookup on posts using user_id (user_id=users.id) (cost=4.41 rows=17.7) (actual time=0.0505..0.0924 rows=30 loops=500000)
1 row in set (48.20 sec)使用 optimizer_trace 进行分析
set session optimizer_trace='enabled=on';再次运行 explain analyze 或者重新执行一次 sql
EXPLAIN ANALYZE SELECT
posts.*
FROM
posts
INNER JOIN users ON users.id = posts.user_id
AND users.deleted_at IS NULL
ORDER BY
posts.id DESC
LIMIT 100 OFFSET 0;关闭 optimizer_trace
set session optimizer_trace='enabled=off';查看 optimizer_trace 的结果
sql
select trace from information_schema.optimizer_trace\G
mysql> select trace from information_schema.optimizer_trace\G
*************************** 1. row ***************************
trace: {
"steps": [
{
"join_preparation": {
"select#": 1,
"steps": [
{
"expanded_query": "/* select#1 */ select `posts`.`id` AS `id`,`posts`.`source` AS `source`,`posts`.`user_id` AS `user_id` from (`posts` join `users` on(((`users`.`id` = `posts`.`user_id`) and (`users`.`deleted_at` is null)))) order by `posts`.`id` desc limit 0,100"
},
{
"transformations_to_nested_joins": {
"transformations": [
"JOIN_condition_to_WHERE",
"parenthesis_removal"
],
"expanded_query": "/* select#1 */ select `posts`.`id` AS `id`,`posts`.`source` AS `source`,`posts`.`user_id` AS `user_id` from `posts` join `users` where ((`users`.`id` = `posts`.`user_id`) and (`users`.`deleted_at` is null)) order by `posts`.`id` desc limit 0,100"
}
}
]
}
},
{
"join_optimization": {
"select#": 1,
"steps": [
{
"condition_processing": {
"condition": "WHERE",
"original_condition": "((`users`.`id` = `posts`.`user_id`) and (`users`.`deleted_at` is null))",
"steps": [
{
"transformation": "equality_propagation",
"resulting_condition": "((`users`.`deleted_at` is null) and multiple equal(`users`.`id`, `posts`.`user_id`))"
},
{
"transformation": "constant_propagation",
"resulting_condition": "((`users`.`deleted_at` is null) and multiple equal(`users`.`id`, `posts`.`user_id`))"
},
{
"transformation": "trivial_condition_removal",
"resulting_condition": "((`users`.`deleted_at` is null) and multiple equal(`users`.`id`, `posts`.`user_id`))"
}
]
}
},
{
"substitute_generated_columns": {
}
},
{
"table_dependencies": [
{
"table": "`posts`",
"row_may_be_null": false,
"map_bit": 0,
"depends_on_map_bits": [
]
},
{
"table": "`users`",
"row_may_be_null": false,
"map_bit": 1,
"depends_on_map_bits": [
]
}
]
},
{
"ref_optimizer_key_uses": [
{
"table": "`posts`",
"field": "user_id",
"equals": "`users`.`id`",
"null_rejecting": true
},
{
"table": "`users`",
"field": "id",
"equals": "`posts`.`user_id`",
"null_rejecting": true
}
]
},
{
"rows_estimation": [
{
"table": "`posts`",
"table_scan": {
"rows": 14612375,
"cost": 10224
}
},
{
"table": "`users`",
"table_scan": {
"rows": 497082,
"cost": 777.25
}
}
]
},
{
"considered_execution_plans": [
{
"plan_prefix": [
],
"table": "`users`",
"best_access_path": {
"considered_access_paths": [
{
"access_type": "ref",
"index": "PRIMARY",
"usable": false,
"chosen": false
},
{
"rows_to_scan": 497082,
"filtering_effect": [
],
"final_filtering_effect": 0.1,
"access_type": "scan",
"resulting_rows": 49708.2,
"cost": 50485.4,
"chosen": true
}
]
},
"condition_filtering_pct": 100,
"rows_for_plan": 49708.2,
"cost_for_plan": 50485.4,
"rest_of_plan": [
{
"plan_prefix": [
"`users`"
],
"table": "`posts`",
"best_access_path": {
"considered_access_paths": [
{
"access_type": "ref",
"index": "user_id",
"rows": 17.656,
"cost": 307176,
"chosen": true
},
{
"rows_to_scan": 14612375,
"filtering_effect": [
],
"final_filtering_effect": 1,
"access_type": "scan",
"using_join_cache": true,
"buffers_needed": 3,
"resulting_rows": 1.46124e+07,
"cost": 7.26355e+10,
"chosen": false
}
]
},
"condition_filtering_pct": 100,
"rows_for_plan": 877647,
"cost_for_plan": 357662,
"sort_cost": 877647,
"new_cost_for_plan": 1.23531e+06,
"chosen": true
}
]
},
{
"plan_prefix": [
],
"table": "`posts`",
"best_access_path": {
"considered_access_paths": [
{
"access_type": "ref",
"index": "user_id",
"usable": false,
"chosen": false
},
{
"rows_to_scan": 14612375,
"filtering_effect": [
],
"final_filtering_effect": 1,
"access_type": "scan",
"resulting_rows": 1.46124e+07,
"cost": 1.47146e+06,
"chosen": true
}
]
},
"condition_filtering_pct": 100,
"rows_for_plan": 1.46124e+07,
"cost_for_plan": 1.47146e+06,
"pruned_by_cost": true
}
]
},
{
"attaching_conditions_to_tables": {
"original_condition": "((`posts`.`user_id` = `users`.`id`) and (`users`.`deleted_at` is null))",
"attached_conditions_computation": [
],
"attached_conditions_summary": [
{
"table": "`users`",
"attached": "(`users`.`deleted_at` is null)"
},
{
"table": "`posts`",
"attached": "(`posts`.`user_id` = `users`.`id`)"
}
]
}
},
{
"optimizing_distinct_group_by_order_by": {
"simplifying_order_by": {
"original_clause": "`posts`.`id` desc",
"items": [
{
"item": "`posts`.`id`"
}
],
"resulting_clause_is_simple": false,
"resulting_clause": "`posts`.`id` desc"
}
}
},
{
"finalizing_table_conditions": [
{
"table": "`users`",
"original_table_condition": "(`users`.`deleted_at` is null)",
"final_table_condition ": "(`users`.`deleted_at` is null)"
},
{
"table": "`posts`",
"original_table_condition": "(`posts`.`user_id` = `users`.`id`)",
"final_table_condition ": null
}
]
},
{
"refine_plan": [
{
"table": "`users`"
},
{
"table": "`posts`"
}
]
},
{
"considering_tmp_tables": [
{
"adding_tmp_table_in_plan_at_position": 2,
"write_method": "write_all_rows"
},
{
"adding_sort_to_table": ""
}
]
}
]
}
},
{
"join_execution": {
"select#": 1,
"steps": [
{
"sorting_table": "<temporary>",
"filesort_information": [
{
"direction": "desc",
"expression": "`posts`.`id`"
}
],
"filesort_priority_queue_optimization": {
"limit": 100,
"chosen": true
},
"filesort_execution": [
],
"filesort_summary": {
"memory_available": 262144,
"key_size": 8,
"row_size": 535,
"max_rows_per_buffer": 101,
"num_rows_estimate": 877647,
"num_rows_found": 14999897,
"num_initial_chunks_spilled_to_disk": 0,
"peak_memory_used": 54843,
"sort_algorithm": "std::stable_sort",
"unpacked_addon_fields": "using_priority_queue",
"sort_mode": "<fixed_sort_key, additional_fields>"
}
}
]
}
}
]
}
1 row in set (0.01 sec)这里我们重点关注这一部分，可以发现，优化器认为扫描 users 表之后，只会得到 5w 行数据。而实际 explain analyze 得到的是 50w 行记录。
text
"table": "`users`",
"best_access_path": {
"considered_access_paths": [
{
"access_type": "ref",
"index": "PRIMARY",
"usable": false,
"chosen": false
},
{
"rows_to_scan": 497082,
"filtering_effect": [
],
"final_filtering_effect": 0.1,
"access_type": "scan",
"resulting_rows": 49708.2,
"cost": 50485.4,
"chosen": true
}
]
},根据上述结果，怀疑是优化器拿到的 users.deleted_at 的数据不准，因此做出了误判。
验证：
使用 mysql histogram
mysql> ANALYZE TABLE users UPDATE HISTOGRAM ON deleted_at WITH 1024 BUCKETS;
+-----------------+-----------+----------+-------------------------------------------------------+
| Table | Op | Msg_type | Msg_text |
+-----------------+-----------+----------+-------------------------------------------------------+
| sql_tests.users | histogram | status | Histogram statistics created for column 'deleted_at'. |
+-----------------+-----------+----------+-------------------------------------------------------+
1 row in set (2.81 sec)重复使用 optimizer_trace 得到新的 optimizer_trace 的结果
vmysql> set session optimizer_trace='enabled=on';
Query OK, 0 rows affected (0.01 sec)
mysql> EXPLAIN ANALYZE SELECT posts.* FROM posts INNER JOIN users ON users.id = posts.user_id AND users.deleted_at IS NULL ORDER BY posts.id DESC LIMIT 100 OFFSET 0\G
*************************** 1. row ***************************
EXPLAIN: -> Limit: 100 row(s) (cost=3.65e+6 rows=100) (actual time=0.183..2.55 rows=100 loops=1)
-> Nested loop inner join (cost=3.65e+6 rows=100) (actual time=0.181..2.55 rows=100 loops=1)
-> Index scan on posts using PRIMARY (reverse) (cost=0.07 rows=100) (actual time=0.11..0.142 rows=100 loops=1)
-> Filter: (users.deleted_at is null) (cost=0.25 rows=1) (actual time=0.0238..0.0239 rows=1 loops=100)
-> Single-row index lookup on users using PRIMARY (id=posts.user_id) (cost=0.25 rows=1) (actual time=0.0237..0.0237 rows=1 loops=100)
1 row in set (0.02 sec)
mysql> set session optimizer_trace='enabled=off';
Query OK, 0 rows affected (0.00 sec)耗时大幅缩减
得到的 optimizer_trace 的结果
sql
mysql> select trace from information_schema.optimizer_trace\G
*************************** 1. row ***************************
trace: {
"steps": [
{
"join_preparation": {
"select#": 1,
"steps": [
{
"expanded_query": "/* select#1 */ select `posts`.`id` AS `id`,`posts`.`source` AS `source`,`posts`.`user_id` AS `user_id` from (`posts` join `users` on(((`users`.`id` = `posts`.`user_id`) and (`users`.`deleted_at` is null)))) order by `posts`.`id` desc limit 0,100"
},
{
"transformations_to_nested_joins": {
"transformations": [
"JOIN_condition_to_WHERE",
"parenthesis_removal"
],
"expanded_query": "/* select#1 */ select `posts`.`id` AS `id`,`posts`.`source` AS `source`,`posts`.`user_id` AS `user_id` from `posts` join `users` where ((`users`.`id` = `posts`.`user_id`) and (`users`.`deleted_at` is null)) order by `posts`.`id` desc limit 0,100"
}
}
]
}
},
{
"join_optimization": {
"select#": 1,
"steps": [
{
"condition_processing": {
"condition": "WHERE",
"original_condition": "((`users`.`id` = `posts`.`user_id`) and (`users`.`deleted_at` is null))",
"steps": [
{
"transformation": "equality_propagation",
"resulting_condition": "((`users`.`deleted_at` is null) and multiple equal(`users`.`id`, `posts`.`user_id`))"
},
{
"transformation": "constant_propagation",
"resulting_condition": "((`users`.`deleted_at` is null) and multiple equal(`users`.`id`, `posts`.`user_id`))"
},
{
"transformation": "trivial_condition_removal",
"resulting_condition": "((`users`.`deleted_at` is null) and multiple equal(`users`.`id`, `posts`.`user_id`))"
}
]
}
},
{
"substitute_generated_columns": {
}
},
{
"table_dependencies": [
{
"table": "`posts`",
"row_may_be_null": false,
"map_bit": 0,
"depends_on_map_bits": [
]
},
{
"table": "`users`",
"row_may_be_null": false,
"map_bit": 1,
"depends_on_map_bits": [
]
}
]
},
{
"ref_optimizer_key_uses": [
{
"table": "`posts`",
"field": "user_id",
"equals": "`users`.`id`",
"null_rejecting": true
},
{
"table": "`users`",
"field": "id",
"equals": "`posts`.`user_id`",
"null_rejecting": true
}
]
},
{
"rows_estimation": [
{
"table": "`posts`",
"table_scan": {
"rows": 14612375,
"cost": 10224
}
},
{
"table": "`users`",
"table_scan": {
"rows": 473065,
"cost": 777.25
}
}
]
},
{
"considered_execution_plans": [
{
"plan_prefix": [
],
"table": "`users`",
"best_access_path": {
"considered_access_paths": [
{
"access_type": "ref",
"index": "PRIMARY",
"usable": false,
"chosen": false
},
{
"rows_to_scan": 473065,
"filtering_effect": [
{
"condition": "(`users`.`deleted_at` is null)",
"histogram_selectivity": 0.99998
}
],
"final_filtering_effect": 0.99998,
"access_type": "scan",
"resulting_rows": 473056,
"cost": 48083.8,
"chosen": true
}
]
},
"condition_filtering_pct": 100,
"rows_for_plan": 473056,
"cost_for_plan": 48083.8,
"rest_of_plan": [
{
"plan_prefix": [
"`users`"
],
"table": "`posts`",
"best_access_path": {
"considered_access_paths": [
{
"access_type": "ref",
"index": "user_id",
"rows": 17.656,
"cost": 2.92329e+06,
"chosen": true
},
{
"rows_to_scan": 14612375,
"filtering_effect": [
],
"final_filtering_effect": 1,
"access_type": "scan",
"using_join_cache": true,
"buffers_needed": 24,
"resulting_rows": 1.46124e+07,
"cost": 6.91247e+11,
"chosen": false
}
]
},
"condition_filtering_pct": 100,
"rows_for_plan": 8.35226e+06,
"cost_for_plan": 2.97137e+06,
"sort_cost": 8.35226e+06,
"new_cost_for_plan": 1.13236e+07,
"chosen": true
}
]
},
{
"plan_prefix": [
],
"table": "`posts`",
"best_access_path": {
"considered_access_paths": [
{
"access_type": "ref",
"index": "user_id",
"usable": false,
"chosen": false
},
{
"rows_to_scan": 14612375,
"filtering_effect": [
],
"final_filtering_effect": 1,
"access_type": "scan",
"resulting_rows": 1.46124e+07,
"cost": 1.47146e+06,
"chosen": true
}
]
},
"condition_filtering_pct": 100,
"rows_for_plan": 1.46124e+07,
"cost_for_plan": 1.47146e+06,
"rest_of_plan": [
{
"plan_prefix": [
"`posts`"
],
"table": "`users`",
"best_access_path": {
"considered_access_paths": [
{
"access_type": "eq_ref",
"index": "PRIMARY",
"rows": 1,
"cost": 5.11433e+06,
"chosen": true,
"cause": "clustered_pk_chosen_by_heuristics"
},
{
"rows_to_scan": 473065,
"filtering_effect": [
{
"condition": "(`users`.`deleted_at` is null)",
"histogram_selectivity": 0.99998
}
],
"final_filtering_effect": 0.99998,
"access_type": "scan",
"using_join_cache": true,
"buffers_needed": 29376,
"resulting_rows": 473056,
"cost": 6.91269e+11,
"chosen": false
}
]
},
"condition_filtering_pct": 99.998,
"rows_for_plan": 1.46121e+07,
"cost_for_plan": 6.58579e+06,
"chosen": true
}
]
}
]
},
{
"attaching_conditions_to_tables": {
"original_condition": "((`users`.`id` = `posts`.`user_id`) and (`users`.`deleted_at` is null))",
"attached_conditions_computation": [
],
"attached_conditions_summary": [
{
"table": "`posts`",
"attached": null
},
{
"table": "`users`",
"attached": "((`users`.`id` = `posts`.`user_id`) and (`users`.`deleted_at` is null))"
}
]
}
},
{
"optimizing_distinct_group_by_order_by": {
"simplifying_order_by": {
"original_clause": "`posts`.`id` desc",
"items": [
{
"item": "`posts`.`id`"
}
],
"resulting_clause_is_simple": true,
"resulting_clause": "`posts`.`id` desc"
}
}
},
{
"reconsidering_access_paths_for_index_ordering": {
"clause": "ORDER BY",
"steps": [
],
"index_order_summary": {
"table": "`posts`",
"index_provides_order": true,
"order_direction": "desc",
"index": "PRIMARY",
"plan_changed": true,
"access_type": "index"
}
}
},
{
"finalizing_table_conditions": [
{
"table": "`users`",
"original_table_condition": "((`users`.`id` = `posts`.`user_id`) and (`users`.`deleted_at` is null))",
"final_table_condition ": "(`users`.`deleted_at` is null)"
}
]
},
{
"refine_plan": [
{
"table": "`posts`"
},
{
"table": "`users`"
}
]
},
{
"considering_tmp_tables": [
]
}
]
}
},
{
"join_execution": {
"select#": 1,
"steps": [
]
}
}
]
}
1 row in set (0.01 sec)再次关注扫描 users 表的预估可以发现，先扫描 users 表会得到大概 50w 行记录。而且这里也用到了 histogram 提供的统计数据。
text
"table": "`users`",
"best_access_path": {
"considered_access_paths": [
{
"access_type": "ref",
"index": "PRIMARY",
"usable": false,
"chosen": false
},
{
"rows_to_scan": 473065,
"filtering_effect": [
{
"condition": "(`users`.`deleted_at` is null)",
"histogram_selectivity": 0.99998
}
],
"final_filtering_effect": 0.99998,
"access_type": "scan",
"resulting_rows": 473056,
"cost": 48083.8,
"chosen": true
}
]
},因此可以得知，如果这里可以使用 posts 表作为驱动表，或许可以优化整体的查询效率。 使用 STRAIGHT_JOIN 强制 posts 表作为驱动表，执行上述查询 首先删除 histogram 数据
sql
mysql> ANALYZE TABLE users DROP HISTOGRAM ON deleted_at;
+-----------------+-----------+----------+-------------------------------------------------------+
| Table | Op | Msg_type | Msg_text |
+-----------------+-----------+----------+-------------------------------------------------------+
| sql_tests.users | histogram | status | Histogram statistics removed for column 'deleted_at'. |
+-----------------+-----------+----------+-------------------------------------------------------+
1 row in set (0.01 sec)使用 STRAIGHT_JOIN 执行，可以大幅降低查询时间
sql
mysql> EXPLAIN ANALYZE SELECT posts.* FROM posts STRAIGHT_JOIN users ON users.id = posts.user_id AND users.deleted_at IS NULL ORDER BY posts.id DESC LIMIT 100 OFFSET 0\G
*************************** 1. row ***************************
EXPLAIN: -> Limit: 100 row(s) (cost=3.65e+6 rows=99.9) (actual time=0.273..1.55 rows=100 loops=1)
-> Nested loop inner join (cost=3.65e+6 rows=99.9) (actual time=0.27..1.54 rows=100 loops=1)
-> Index scan on posts using PRIMARY (reverse) (cost=0.706 rows=999) (actual time=0.208..0.266 rows=100 loops=1)
-> Filter: (users.deleted_at is null) (cost=0.25 rows=0.1) (actual time=0.0124..0.0125 rows=1 loops=100)
-> Single-row index lookup on users using PRIMARY (id=posts.user_id) (cost=0.25 rows=1) (actual time=0.012..0.0121 rows=1 loops=100)
1 row in set (0.02 sec)上述就是整个使用 optimizer_trace 分析的过程。
实验脚本
python
import random
import uuid
import string
import mysql.connector
config = {
'user': 'root',
'password': '',
'host': '127.0.0.1',
'database': 'sql_tests',
'raise_on_warnings': True
}
connection = mysql.connector.connect(**config)
cursor = connection.cursor()
def random_string(length):
return ''.join(random.choices(string.ascii_letters + string.digits, k=length))
def insert_users(count):
sql = '''INSERT INTO users (name, email)
VALUES (%s, %s)'''
cnt = 0
for _ in range(count):
email = str(uuid.uuid4()) + '@example.com'
name = random_string(12)
cursor.execute(sql, (name, email))
cnt += 1
if cnt > 1000:
connection.commit()
cnt = 0
connection.commit()
def insert_logins(count, min_user_id, max_user_id):
sql = '''INSERT INTO posts (source, user_id)
VALUES (%s, %s)'''
cnt = 0
for _ in range(count):
user_id = random.randint(min_user_id, max_user_id)
source = random_string(12)
cursor.execute(sql, (source, user_id))
cnt += 1
if cnt > 1000:
connection.commit()
cnt = 0
connection.commit()
user_count = 500_000
login_count = 15_000_000
insert_users(user_count)
print('insert users successfully')
select_bound_user_id = "select min(id) as min_id, max(id) as max_id from users"
cursor.execute(select_bound_user_id)
res = cursor.fetchone()
min_user_id = res[0]
max_user_id = res[1]
insert_logins(login_count, min_user_id, max_user_id)
print('insert posts successfully')
cursor.close()
connection.close()需要安装
python
pip install mysql-connector-python参考：
- https://dev.mysql.com/doc/refman/8.4/en/join.html#:~:text=official ODBC syntax.-,STRAIGHT_JOIN,-is similar to
- https://timyang.net/data/key-list-pagination/
- https://stackoverflow.com/questions/4481388/why-does-mysql-higher-limit-offset-slow-the-query-down
- https://www.cnblogs.com/weixiaotao/p/10646666.html
- https://dev.mysql.com/doc/refman/8.4/en/analyze-table.html#analyze-table-histogram-statistics-analysis
- https://dev.mysql.com/blog-archive/mysql-8-0-1-accent-and-case-sensitive-collations-for-utf8mb4/
todo
explain 语句
通过 explain 语句可以得到语句可能使用的执行计划，
可以参考: https://dev.mysql.com/doc/refman/8.4/en/explain.html#explain-execution-plan
explain 语句的 json 格式
explain analyze 语句
optimizer_trace
深分页问题
- offset 分页
- cursor 分页
- timestamp 分页
- primary key 分页
- 延迟 join
straight_join
贡献者
zll600