Question

我的查询真的不那么复杂。它的运行时间接近250毫秒，这很慢。我使用EXPLAIN分析了查询并发现了seq扫描。我为此查询中使用的所有列都准备了适当的索引。所以我不确定从哪里开始。

这就是我所拥有的：

cl_production=# EXPLAIN SELECT count(DISTINCT events.id) AS count_distinct_events_id FROM "events" INNER JOIN "events_tickets" ON "events_tickets".event_id = "events".id INNER JOIN "tickets" ON "tickets".id = "events_tickets".ticket_id WHERE ((events.occurs_at > '2011-08-20 07:00:00.000000') AND (tickets.company_id = 175));
                                                       QUERY PLAN                                                       
------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=15735.79..15735.80 rows=1 width=4)
   ->  Hash Join  (cost=10540.01..15651.89 rows=33562 width=4)
         Hash Cond: (events_tickets.event_id = events.id)
         ->  Hash Join  (cost=3510.07..7516.61 rows=33562 width=4)
               Hash Cond: (events_tickets.ticket_id = tickets.id)
               ->  Seq Scan on events_tickets  (cost=0.00..1803.80 rows=124980 width=8)
               ->  Hash  (cost=3058.63..3058.63 rows=27475 width=4)
                     ->  Bitmap Heap Scan on tickets  (cost=521.19..3058.63 rows=27475 width=4)
                           Recheck Cond: (company_id = 175)
                           ->  Bitmap Index Scan on index_tickets_on_company_id  (cost=0.00..514.33 rows=27475 width=0)
                                 Index Cond: (company_id = 175)
         ->  Hash  (cost=5963.87..5963.87 rows=64965 width=4)
               ->  Index Scan using index_events_on_occurs_at on events  (cost=0.00..5963.87 rows=64965 width=4)
                     Index Cond: (occurs_at > '2011-08-20 07:00:00'::timestamp without time zone)

如上所述，以下是我的架构文件中的索引：

add_index "events_tickets", ["event_id", "ticket_id"], :name => "index_events_tickets_on_event_id_and_ticket_id", :unique => true
add_index "events_tickets", ["event_id"], :name => "index_events_tickets_on_event_id"
add_index "events_tickets", ["ticket_id"], :name => "index_events_tickets_on_ticket_id"
add_index "events", ["occurs_at"], :name => "index_events_on_occurs_at"
add_index "tickets", ["company_id"], :name => "index_tickets_on_company_id"

我假设序列扫描是什么杀死了这个查询。我在那张桌子上有很完整的索引。所以我迷路了。任何帮助将不胜感激。

感谢。

Answer 1

您应该进行EXPLAIN ANALYZE以获取计划中每个节点的实际时间，而不仅仅是成本估算。

也许这种使用半连接而不是计数不同聚合的查询形式会有所帮助：

SELECT COUNT(*)
FROM events
WHERE EXISTS (SELECT 1
              FROM events_tickets
                   JOIN tickets ON tickets.id = events_tickets.ticket_id
              WHERE tickets.company_id = 175
                    AND events_tickets.event_id = events.id)
      AND events.occurs_at > '2011-08-20 07:00:00'::timestamp

帮助改进查询。试图使用EXPLAIN

1 个答案: