我有一个下面的查询，它会每分钟命中一次数据库
SELECT n.notificationid ,n.notificationtype
FROM notification n
INNER JOIN usernotification un ON n.notificationid = un.notificationid
LEFT JOIN vhctotal vt ON vt.vhcid = n.vhcid
WHERE CASE WHEN n.notificationtype = 2 THEN n.datechecked = '' ELSE to_timestamp(n.datechecked, 'DD/MM/YYYY hh24:mi:ss') > NOW() - INTERVAL '30 days' END
AND date_trunc('day', n.vhcdate::TIMESTAMP) = CURRENT_DATE
AND un.commentstatus = 0
AND un.userid = 30427
OR (
un.commentstatus = 1
AND un.updateddatetime > current_timestamp - interval '5 minutes'
)
上面的查询需要很长时间才能执行。当我检查那些表索引没有正确完成。
我想为这些表创建索引。
我是否必须为WHERE子句中使用的列创建索引？
它会提供更好的性能吗？请建议。
编辑：
QUERY PLAN
Nested Loop Left Join (cost=25.96..2136.39 rows=1 width=237) (actual time=49.911..49.911 rows=0 loops=1)
Join Filter: (vt.vhcid = n.vhcid)
-> Hash Join (cost=25.96..2089.81 rows=1 width=163) (actual time=49.908..49.908 rows=0 loops=1)
Hash Cond: (un.notificationid = n.notificationid)
Join Filter: ((CASE WHEN ((n.notificationtype)::text = '2'::text) THEN ((n.datechecked)::text = ''::text) ELSE (to_timestamp((n.datechecked)::text, 'DD/MM/YYYY hh24:mi:ss'::text) > (now() - '30 days'::interval)) END AND (date_trunc('day'::text, (n.vhcdate)::timestamp without time zone) = ('now'::cstring)::date) AND (un.commentstatus = 0) AND (un.userid = 30427)) OR ((un.commentstatus = 1) AND (un.updateddatetime > (now() - '00:05:00'::interval))))
Rows Removed by Join Filter: 26406
-> Seq Scan on usernotification un (cost=0.00..528.36 rows=26136 width=36) (actual time=0.009..3.937 rows=26406 loops=1)
-> Hash (cost=18.76..18.76 rows=576 width=127) (actual time=0.420..0.420 rows=577 loops=1)
Buckets: 1024 Batches: 1 Memory Usage: 96kB
-> Seq Scan on notification n (cost=0.00..18.76 rows=576 width=127) (actual time=0.004..0.105 rows=577 loops=1)
-> Seq Scan on vhctotal vt (cost=0.00..33.48 rows=1048 width=74) (never executed)
Total runtime: 50.033 ms
版本：postgres-9.2.14