就在你以为一切安全的时候...
eq_range_index_dive_limit
的值是多少?特别是,在 IN
子句中,您是否有更多或更少的项?
这将不包括基准测试,但会深入了解一些内部工作。让我们使用一个工具来查看正在发生的事情——优化器跟踪。
查询:SELECT * FROM canada WHERE id ...
用 3 个值的 OR
,跟踪的一部分如下:
"condition_processing": {
"condition": "WHERE",
"original_condition": "((`canada`.`id` = 296172) or (`canada`.`id` = 295093) or (`canada`.`id` = 293626))",
"steps": [
{
"transformation": "equality_propagation",
"resulting_condition": "(multiple equal(296172, `canada`.`id`) or multiple equal(295093, `canada`.`id`) or multiple equal(293626, `canada`.`id`))"
},
...
"analyzing_range_alternatives": {
"range_scan_alternatives": [
{
"index": "id",
"ranges": [
"293626 <= id <= 293626",
"295093 <= id <= 295093",
"296172 <= id <= 296172"
],
"index_dives_for_eq_ranges": true,
"chosen": true
...
"refine_plan": [
{
"table": "`canada`",
"pushed_index_condition": "((`canada`.`id` = 296172) or (`canada`.`id` = 295093) or (`canada`.`id` = 293626))",
"table_condition_attached": null,
"access_type": "range"
}
]
请注意,ICP被赋予了 "ORs"。这意味着"OR"没有变成"IN",InnoDB将通过ICP执行一堆"="测试。(我不觉得考虑MyISAM值得)。(这是Percona的5.6.22-71.0-log;id是一个次要索引。)
现在来看带有几个值的IN()。
eq_range_index_dive_limit = 10;共有8个值。
"condition_processing": {
"condition": "WHERE",
"original_condition": "(`canada`.`id` in (296172,295093,293626,295573,297148,296127,295588,295810))",
"steps": [
{
"transformation": "equality_propagation",
"resulting_condition": "(`canada`.`id` in (296172,295093,293626,295573,297148,296127,295588,295810))"
},
...
"analyzing_range_alternatives": {
"range_scan_alternatives": [
{
"index": "id",
"ranges": [
"293626 <= id <= 293626",
"295093 <= id <= 295093",
"295573 <= id <= 295573",
"295588 <= id <= 295588",
"295810 <= id <= 295810",
"296127 <= id <= 296127",
"296172 <= id <= 296172",
"297148 <= id <= 297148"
],
"index_dives_for_eq_ranges": true,
"chosen": true
...
"refine_plan": [
{
"table": "`canada`",
"pushed_index_condition": "(`canada`.`id` in (296172,295093,293626,295573,297148,296127,295588,295810))",
"table_condition_attached": null,
"access_type": "range"
}
]
请注意,
IN
似乎没有转换为
OR
。
顺便提一下,注意
常量值已排序。这样做有两个好处:
少跳来跳去,可能会有更好的缓存,更少的I/O来获取所有的值。
如果来自不同连接的两个类似的查询都在事务中,并且它们重叠了列表,那么更有可能出现延迟而不是死锁。
最后,
当IN()带有大量值时。
{
"condition_processing": {
"condition": "WHERE",
"original_condition": "(`canada`.`id` in (293831,292259,292881,293440,292558,295792,292293,292593,294337,295430,295034,297060,293811,295587,294651,295559,293213,295742,292605,296018,294529,296711,293919,294732,294689,295540,293000,296916,294433,297112,293815,292522,296816,293320,293232,295369,291894,293700,291839,293049,292738,294895,294473,294023,294173,293019,291976,294923,294797,296958,294075,293450,296952,297185,295351,295736,296312,294330,292717,294638,294713,297176,295896,295137,296573,292236,294966,296642,296073,295903,293057,294628,292639,293803,294470,295353,297196,291752,296118,296964,296185,295338,295956,296064,295039,297201,297136,295206,295986,292172,294803,294480,294706,296975,296604,294493,293181,292526,293354,292374,292344,293744,294165,295082,296203,291918,295211,294289,294877,293120,295387))",
"steps": [
{
"transformation": "equality_propagation",
"resulting_condition": "(`canada`.`id` in (293831,292259,292881,293440,292558,295792,292293,292593,294337,295430,295034,297060,293811,295587,294651,295559,293213,295742,292605,296018,294529,296711,293919,294732,294689,295540,293000,296916,294433,297112,293815,292522,296816,293320,293232,295369,291894,293700,291839,293049,292738,294895,294473,294023,294173,293019,291976,294923,294797,296958,294075,293450,296952,297185,295351,295736,296312,294330,292717,294638,294713,297176,295896,295137,296573,292236,294966,296642,296073,295903,293057,294628,292639,293803,294470,295353,297196,291752,296118,296964,296185,295338,295956,296064,295039,297201,297136,295206,295986,292172,294803,294480,294706,296975,296604,294493,293181,292526,293354,292374,292344,293744,294165,295082,296203,291918,295211,294289,294877,293120,295387))"
},
...
"analyzing_range_alternatives": {
"range_scan_alternatives": [
{
"index": "id",
"ranges": [
"291752 <= id <= 291752",
"291839 <= id <= 291839",
...
"297196 <= id <= 297196",
"297201 <= id <= 297201"
],
"index_dives_for_eq_ranges": false,
"rows": 111,
"chosen": true
...
"refine_plan": [
{
"table": "`canada`",
"pushed_index_condition": "(`canada`.`id` in (293831,292259,292881,293440,292558,295792,292293,292593,294337,295430,295034,297060,293811,295587,294651,295559,293213,295742,292605,296018,294529,296711,293919,294732,294689,295540,293000,296916,294433,297112,293815,292522,296816,293320,293232,295369,291894,293700,291839,293049,292738,294895,294473,294023,294173,293019,291976,294923,294797,296958,294075,293450,296952,297185,295351,295736,296312,294330,292717,294638,294713,297176,295896,295137,296573,292236,294966,296642,296073,295903,293057,294628,292639,293803,294470,295353,297196,291752,296118,296964,296185,295338,295956,296064,295039,297201,297136,295206,295986,292172,294803,294480,294706,296975,296604,294493,293181,292526,293354,292374,292344,293744,294165,295082,296203,291918,295211,294289,294877,293120,295387))",
"table_condition_attached": null,
"access_type": "range"
}
]
顺便提一下:由于追踪信息过于庞大,我需要这个功能:
@@global.optimizer_trace_max_mem_size = 32222;
BETWEEN 1 AND 4
完全匹配,可能更快)。MySQL/MariaDB 的版本可能很重要。 - Rick James