PostgreSQL的慢查询，订购LIMIT

Question

• 列表项

mydbab=# \d+ table1;

    Table "dbmydb.table1" 
    Column  | Type | Modifiers | Scol5rage | Stats target | Description 
----------------+---------+-----------+------------+--------------+------------- 
pk    | bigint | not null | plain  |    | 
col1   | bigint | not null | plain  |    | 
col2   | citext | not null | extended |    | 
col3   | citext |   | extended |    |   
col4   | citext | not null | extended |    | 
col5   | citext |   | extended |    | 
col6   | citext |   | extended |    | 
bcol7   | citext |   | extended |    | 
col8   | citext |   | extended |    | 
col9   | citext |   | extended |    | 
col10   | bigint | not null | plain  |    | 
col11   | citext |   | extended |    | 
col12   | bigint |   | plain  |    |  
col13   | integer | default 1 | plain  |    | 
col14   | integer | default 0 | plain  |    | 
col15   | bigint |   | plain  |    | 
col16   | integer | default 0 | plain  |    | 
col17   | integer | default 0 | plain  |    |  
col18   | integer | default 0 | plain  |    | 
col19   | citext |   | extended |    | 
col20   | citext |   | extended |    | 
col21   | citext |   | extended |    | 

指标：

"table1_pk" PRIMARY KEY, btree (pk) 
"table1_idx3" btree (col10) 
"table1_idx4" btree ("col4") 
"table1_idx5" btree (col11) 
"table1_idx6" btree (col15) 
"table1_fk_idx" btree (col1) 

检查约束：

"table1_col6_c" CHECK (length(col6::text) <= 5000) 
"table1_col4_c" CHECK (length("col4"::text) <= 253) 
"table1_col5_c” CHECK (length(col5::text) <= 5000) 
"table1_col3_c" CHECK (length(col3::text) <= 253) 
"table1_col11_c" CHECK (length(col11::text) <= 253) 
"table1_col20_c” CHECK (length(col20::text) <= 200) 
"table1_col21_c” CHECK (length(col21::text) <= 500) 
"table1_col7_c” CHECK (length(col7::text) <= 500) 
"table1_col9_c" CHECK (length(col9::text) <= 100) 
"table1_col5_c" CHECK (length("col5"::text) <= 5000) 
"table1_col19_c” CHECK (length(col19::text) <= 100) 

外键约束：

"table1_fk" FOREIGN KEY (col1) REFERENCES table2(col1) ON DELETE CASCADE 

被引用：

TABLE “table3” CONSTRAINT “table3_fk2" FOREIGN KEY (pk) REFERENCES table1(pk) ON DELETE CASCADE 
TABLE “table4” CONSTRAINT “table4_fk2" FOREIGN KEY (pk) REFERENCES table1(pk) ON DELETE CASCADE 
TABLE “table5” CONSTRAINT “table5_fk1" FOREIGN KEY (pk) REFERENCES table1(pk) 
TABLE “table6” CONSTRAINT “table6_fk" FOREIGN KEY (pk) REFERENCES table1(pk) ON DELETE CASCADE 
TABLE “table7” CONSTRAINT “table7_fk1" FOREIGN KEY (pk) REFERENCES table1(pk) ON DELETE CASCADE 

选项：autovacuum_vacuum_scale_factor=0.05

查询计划

explain SELECT * FROM table1 WHERE ((table1.col1 = 814000000002054) AND (((table1.pk >= 238000000000000) AND (table1.pk <= 238999999999999)) OR ((table1.pk >= 0) AND (table1.pk <= 999999999999)))) ORDER BY table1.col15 DESC LIMIT 7 

(query gets timed out for explain analyze) 

Limit (cost=0.56..607.64 rows=7 width=603) 

Index Scan Backward using table1_idx6 on table1 (cost=0.56..12622561.13 rows=145548 width=603) Filter: ((col1 = 814000000002054::bigint) AND (((pk >= 238000000000000::bigint) AND (pk <= 238999999999999::bigint)) OR ((pk >= 0) AND (pk <= 999999999999::bigint)))) 

问题： 表有50M记录。 此查询需要15分钟以上才能运行！我们已经校准到自动真空到0.05。自动分析是PG的默认设置。 甚至将查询更改为ORDER BY table1.col15 ASC LIMIT 7，即使查询大约在同一时间。 插入/更新操作只需要ms。

历史：当数据较少时结果速度更快 - 查询在MySQL中没有问题，几周前迁移到Pg。

Worker_mem是956MB。

select version();

PostgreSQL 9.4.0 on x86_64-unknown-linux-gnu, compiled by gcc (GCC) 4.1.2 20080704 (Red Hat 4.1.2-54), 64-bit 
(1 row) 

Answer 1

尝试创建一个多列索引：

CREATE INDEX ON table1(col15, col1, pk);