对于圣诞节,我有一个问题,我真的很想找到如何处理的线索。。
我有一个报告数据库,它由许多小型机器上的卫星数据库中的数据填充。每个卫星数据库每20分钟平均执行一次数据提取。他们使用相同的脚本。然而,它们安装在不同的装置上,分布在全国各地。
现在我有一个SELECT,pentaho reporting使用它在这个report DB中的同一个DB表上执行。虽然一个选择需要毫秒才能执行,但另一个选择需要数小时。它们都在同一个表上执行,在同一个硬件上运行的同一个DB中。
快速一号:
SELECT
res.ticket_id,
res.entry_zone,
res.entry_time,
res.exit_time,
res.parking_time,
res.cost,
co.org_name,
cu.firstname,
cu.surname,
a.name AS article_name,
res.car_id
FROM (SELECT
lh.ticket_id,
z.name AS entry_zone,
lh.park_entered AS entry_time,
lh.park_leaved AS exit_time,
interval_to_hourminsec(lh.park_leaved - lh.park_entered) AS parking_time,
lh.cost,
lh.article_id,
sa.contrib_user_id,
fpl.car_id
FROM longterm_history lh, zones z, sold_articles sa, flexcore_passing_log fpl
WHERE lh.park_leaved BETWEEN '2017-12-18 00:00' AND '2017-12-19 23:59'
AND sa.ticket_id = lh.ticket_id
AND lh.entry_zone = z.zone_number
AND lh.passlog_id = fpl.id
AND lh.park_uuid = 100068
AND z.park_uuid = 100068
AND sa.park_uuid = 100068
AND fpl.park_uuid = 100068
AND lh.entry_zone = 1
) AS res
LEFT OUTER JOIN articles a ON res.article_id = a.article_id AND a.park_uuid = 100068
LEFT OUTER JOIN cont_users cu ON res.contrib_user_id = cu.id AND cu.park_uuid = 100068
LEFT OUTER JOIN cont_orgs co ON cu.org_id = co.id AND co.park_uuid = 100068
ORDER BY res.exit_time ASC
慢速:
SELECT
res.ticket_id,
res.entry_zone,
res.entry_time,
res.exit_time,
res.parking_time,
res.cost,
co.org_name,
cu.firstname,
cu.surname,
a.name AS article_name,
res.car_id
FROM (SELECT
lh.ticket_id,
z.name AS entry_zone,
lh.park_entered AS entry_time,
lh.park_leaved AS exit_time,
interval_to_hourminsec(lh.park_leaved - lh.park_entered) AS parking_time,
lh.cost,
lh.article_id,
sa.contrib_user_id,
fpl.car_id
FROM longterm_history lh, zones z, sold_articles sa, flexcore_passing_log fpl
WHERE lh.park_leaved BETWEEN '2017-12-18 00:00' AND '2017-12-19 23:59'
AND sa.ticket_id = lh.ticket_id
AND lh.entry_zone = z.zone_number
AND lh.passlog_id = fpl.id
AND lh.park_uuid = 100146
AND z.park_uuid = 100146
AND sa.park_uuid = 100146
AND fpl.park_uuid = 100146
AND lh.entry_zone = 1
) AS res
LEFT OUTER JOIN articles a ON res.article_id = a.article_id AND a.park_uuid = 100146
LEFT OUTER JOIN cont_users cu ON res.contrib_user_id = cu.id AND cu.park_uuid = 100146
LEFT OUTER JOIN cont_orgs co ON cu.org_id = co.id AND co.park_uuid = 100146
ORDER BY res.exit_time ASC
如何找出问题所在,是什么导致第二次选择在小时内执行?
我使用的是postgres SQL,服务器版本是9.6.3
通过pentaho数据集成将数据提取到数据库中
编辑:
通过运行两个查询后
EXPLAIN (ANALYZE, BUFFERS)
最大的显著区别在于这一部分:
-> Bitmap Index Scan on longterm_history_park_uuid_idx (cost=0.00..7609.82 rows=352718 width=0) (actual time=492.753..492.753 rows=354537 loops=1)
Index Cond: (park_uuid = 100068)
Buffers: shared read=1238
-> Bitmap Index Scan on longterm_history_park_uuid_idx (cost=0.00..453.11 rows=20890 width=0) (actual time=4.680..4.680 rows=40021 loops=466475)
Index Cond: (park_uuid = 100146)
Buffers: shared hit=65306361 read=139
看来第二次选择速度较慢的原因是
loops=466475
而不是
loops=1
dne按第一选择。但我不知道这意味着什么,也不知道如何修复它。。
编辑2:
我发现tootl可以在线共享计划,以下是链接:
快速查询:
https://explain.depesz.com/s/oYQLB
慢速查询:
https://explain.depesz.com/s/uOtf
快速执行慢速查询,而回迁处于关闭状态:
https://explain.depesz.com/s/4h4F