Question

我正在评估postgres从Oracle迁移。以下查询运行速度太慢，请查找解释计划：

explain analyze select DISTINCT EVENT.ID, 
   ORIGIN.ID AS RIGINID,EVENT.PREFERRED_ORIGIN_ID AS PREFERRED_ORIGIN, 
   EVENT.CONTRIBUTOR, ORIGIN.TIME, ORIGIN.LATITUDE, ORIGIN.LONGITUDE, 
   ORIGIN.DEPTH, ORIGIN.EVTYPE, ORIGIN.CATALOG, ORIGIN.AUTHOR OAUTHOR,
   ORIGIN.CONTRIBUTOR OCONTRIBUTOR, 
   MAGNITUDE.ID AS MAGID, MAGNITUDE.MAGNITUDE, MAGNITUDE.TYPE AS MAGTYPE
from event.event 
   left join event.origin on event.id = origin.eventid 
   left join event.magnitude on origin.id = event.magnitude.origin_id 
WHERE EXISTS(
    select origin_id from event.magnitude 
    where magnitude.magnitude>=7.2 
    and origin.id=origin_id) 
order by ORIGIN.TIME desc, MAGNITUDE.MAGNITUDE desc,
    EVENT.ID,EVENT.PREFERRED_ORIGIN_ID,ORIGIN.ID


"Unique  (cost=740549.86..741151.42 rows=15039 width=80) (actual time=17791.557..17799.092 rows=5517 loops=1)"
"  ->  Sort  (cost=740549.86..740587.45 rows=15039 width=80) (actual time=17791.556..17792.220 rows=5517 loops=1)"
"        Sort Key: origin."time", event.magnitude.magnitude, event.id, event.preferred_origin_id, origin.id, event.contributor, origin.latitude, origin.longitude, origin.depth, origin.evtype, origin.catalog, origin.author, origin.contributor, event.magnitude.id, event.magnitude.type"
"        Sort Method:  quicksort  Memory: 968kB"
"        ->  Nested Loop Left Join  (cost=34642.50..739506.42 rows=15039 width=80) (actual time=6.927..17769.788 rows=5517 loops=1)"
"              ->  Hash Semi Join  (cost=34642.50..723750.23 rows=14382 width=62) (actual time=6.912..17744.858 rows=2246 loops=1)"
"                    Hash Cond: (origin.id = event.magnitude.origin_id)"
"                    ->  Merge Left Join  (cost=0.00..641544.72 rows=6133105 width=62) (actual time=0.036..16221.008 rows=6133105 loops=1)"
"                          Merge Cond: (event.id = origin.eventid)"
"                          ->  Index Scan using event_key_index on event  (cost=0.00..163046.53 rows=3272228 width=12) (actual time=0.017..1243.616 rows=3276192 loops=1)"
"                          ->  Index Scan using origin_fk_index on origin  (cost=0.00..393653.81 rows=6133105 width=54) (actual time=0.013..3033.657 rows=6133105 loops=1)"
"                    ->  Hash  (cost=34462.73..34462.73 rows=14382 width=4) (actual time=6.668..6.668 rows=3198 loops=1)"
"                          Buckets: 2048  Batches: 1  Memory Usage: 113kB"
"                          ->  Bitmap Heap Scan on magnitude  (cost=324.65..34462.73 rows=14382 width=4) (actual time=1.682..5.414 rows=3198 loops=1)"
"                                Recheck Cond: (magnitude >= 7.2)"
"                                ->  Bitmap Index Scan on mag_index  (cost=0.00..321.05 rows=14382 width=0) (actual time=1.331..1.331 rows=3198 loops=1)"
"                                      Index Cond: (magnitude >= 7.2)"
"              ->  Index Scan using mag_fkey_index on magnitude  (cost=0.00..1.06 rows=3 width=22) (actual time=0.007..0.009 rows=2 loops=2246)"
"                    Index Cond: (origin.id = event.magnitude.origin_id)"
"Total runtime: 17799.669 ms"

此查询在Oracle中运行1秒钟，而在postgres中需要16秒，差异告诉我，我在某处做错了。这是在具有12G RAM的本地Mac机器上的新安装。

我有：

effective_cache_size=4096MB   
shared_buffer=2048MB  
work_mem=100MB

Answer 1

从DISTINCT子句中删除SELECT。这是多余的，因为无论如何你都选择PRIMARY KEY列。
将LEFT JOIN更改为INNER JOIN。正如其他人所提到的，您的EXISTS子句会消除NULL生成的所有LEFT JOIN输出。
创建两个索引：
```
magnitude (magnitude, origin_id)
magnitude (origin_id, magnitude)
```
（根据您的magnitude >= 7.2条件的选择性，每种方法可能更有效率）

SELECT  EVENT.ID, 
        ORIGIN.ID AS RIGINID,EVENT.PREFERRED_ORIGIN_ID AS PREFERRED_ORIGIN, 
        EVENT.CONTRIBUTOR, ORIGIN.TIME, ORIGIN.LATITUDE, ORIGIN.LONGITUDE, 
        ORIGIN.DEPTH, ORIGIN.EVTYPE, ORIGIN.CATALOG, ORIGIN.AUTHOR OAUTHOR,
        ORIGIN.CONTRIBUTOR OCONTRIBUTOR, 
        MAGNITUDE.ID AS MAGID, MAGNITUDE.MAGNITUDE, MAGNITUDE.TYPE AS MAGTYPE
FROM    event.event
JOIN    event.origin
ON      origin.eventid = event.id
JOIN    event.magnitude
ON      magnitude.origin_id = origin.id
WHERE   origin.id IN
        (
        SELECT  origin_id
        FROM    magnitude
        WHERE   magnitude >= 7.2
        )
ORDER BY
        ORIGIN.TIME desc, MAGNITUDE.MAGNITUDE desc,
        EVENT.ID,EVENT.PREFERRED_ORIGIN_ID,ORIGIN.ID

要检查的示例查询：

WITH    event (id) AS
        (
        VALUES
                (1)
        ),
        origin (id, eventid) AS
        (
        VALUES
                (1, 1),
                (2, 1)
        ),
        magnitude (id, origin_id, magnitude) AS
        (
        VALUES
                (1, 1, 4),
                (2, 1, 8),
                (3, 3, 6)
        )
SELECT  *
FROM    event
LEFT JOIN
        origin
ON      origin.eventid = event.id
LEFT JOIN
        magnitude
ON      magnitude.origin_id = origin.id
WHERE   EXISTS
        (
        SELECT  origin_id
        FROM    magnitude 
        WHERE   magnitude.magnitude >= 7.2 
                AND origin.id = origin_id
        )

4会返回8和origin 1这两个等级。

Answer 2

至少在您发布此问题的其他论坛中已经回答过，将“在event.id = origin.eventid上的左连接event.origin”部分更改为内部联接，因为您的EXISTS查询无论如何，进一步向下限制它。

Answer 3

有一些有用的链接可供学习如何在PostgreSQL wiki解释您自己的执行计划。

Answer 4

根据计划：

排序（成本= 740549.86..740587.45行= 15039宽度= 80）（实际时间= 17791.55 ......

排序运营商是主要的瓶颈。尝试删除order by声明？

Postgres 9.0慢查询

4 个答案: