我有一个PostgreSQL查询，其中有一个With子句（CTE），其中给出了输入值（int数据类型）。对于此查询，输出行数将与With子句中提供的输入值数相同。 with子句中没有联接，主select子句中只有两个表。
该查询将在不到500毫秒的时间内执行，直到输入值的数量为531。如果超过该值，则大约需要15秒。（即使是532个输入值）。
在“解释计划”中，当输入值的数量超过531时，我可以看到主查询的计划从“嵌套循环左连接”更改为“哈希右连接”。
编辑：添加了两种情况的解释分析。
如果输入值的数量为532
Sort (cost=45221.67..45221.68 rows=2 width=322) (actual time=15371.553..15371.576 rows=532 loops=1)
Sort Key: fc.flashnumber DESC
Sort Method: quicksort Memory: 95kB
CTE t
-> GroupAggregate (cost=0.43..4489.54 rows=3586 width=22) (actual time=3.876..21.531 rows=532 loops=1)
Group Key: flashcomment.flashnumber
-> Index Scan using flashcomment_ndx01 on flashcomment (cost=0.43..4435.68 rows=3601 width=22) (actual time=3.861..18.628 rows=713 loops=1)
Index Cond: ((flashnumber)::text = ANY ('{2020022601503,2020022601502,2020022601498,2020022601497,2020022601496,2020022601495,2020022601494,2020022601493,2020022601492,2020022601491,2020022601490,2020022620200219007292020021900630,2020021900629,2020021900628}'::text[]))
Filter: (is_parent = 'N'::bpchar)
-> Nested Loop Left Join (cost=0.43..40732.12 rows=2 width=322) (actual time=3.919..15366.603 rows=532 loops=1)
Join Filter: (upper((fc.stamp_userid)::text) = upper((up.user_id)::text))
Rows Removed by Join Filter: 19483968
-> Nested Loop (cost=0.43..39103.24 rows=1 width=299) (actual time=3.897..34.879 rows=532 loops=1)
-> CTE Scan on t (cost=0.00..71.72 rows=3586 width=90) (actual time=3.879..23.038 rows=532 loops=1)
-> Index Scan using flashcomment_ndx01 on flashcomment fc (cost=0.43..10.87 rows=1 width=299) (actual time=0.012..0.016 rows=1 loops=532)
Index Cond: ((flashnumber)::text = (t.flashnumber)::text)
Filter: ((is_parent = 'N'::bpchar) AND (t.mx_gmttime = stamp_gmttime))
Rows Removed by Filter: 0
-> Seq Scan on user_profile up (cost=0.00..1000.86 rows=35886 width=23) (actual time=0.003..6.865 rows=36625 loops=532)
Planning time: 3.728 ms
Execution time: 15371.683 ms
如果输入值的数量为530
Sort (cost=44660.91..44660.91 rows=2 width=322) (actual time=37.383..37.406 rows=530 loops=1)
Sort Key: fc.flashnumber DESC
Sort Method: quicksort Memory: 95kB
CTE t
-> GroupAggregate (cost=0.43..4473.75 rows=3572 width=22) (actual time=3.922..8.892 rows=530 loops=1)
Group Key: flashcomment.flashnumber
-> Index Scan using flashcomment_ndx01 on flashcomment (cost=0.43..4420.10 rows=3587 width=22) (actual time=3.906..8.644 rows=709 loops=1)
Index Cond: ((flashnumber)::text = ANY ('{2020022601503,2020022601502,2020022601498,2020022601497,2020022601496,2020022601495,2020022601494,2020022601493,2020022601492,2020022601491,2020022601490,2020022601489,2020021900629,2020021900628}'::text[]))
Filter: (is_parent = 'N'::bpchar)
-> Hash Right Join (cost=38960.19..40187.14 rows=2 width=322) (actual time=14.613..34.816 rows=530 loops=1)
Hash Cond: (upper((up.user_id)::text) = upper((fc.stamp_userid)::text))
-> Seq Scan on user_profile up (cost=0.00..1000.86 rows=35886 width=23) (actual time=0.008..6.647 rows=36625 loops=1)
-> Hash (cost=38960.17..38960.17 rows=1 width=299) (actual time=14.586..14.586 rows=530 loops=1)
Buckets: 1024 Batches: 1 Memory Usage: 49kB
-> Nested Loop (cost=0.43..38960.17 rows=1 width=299) (actual time=3.947..14.203 rows=530 loops=1)
-> CTE Scan on t (cost=0.00..71.44 rows=3572 width=90) (actual time=3.925..9.069 rows=530 loops=1)
-> Index Scan using flashcomment_ndx01 on flashcomment fc (cost=0.43..10.88 rows=1 width=299) (actual time=0.009..0.009 rows=1 loops=530)
Index Cond: ((flashnumber)::text = (t.flashnumber)::text)
Filter: ((is_parent = 'N'::bpchar) AND (t.mx_gmttime = stamp_gmttime))
Rows Removed by Filter: 0
Planning time: 3.721 ms
Execution time: 37.531 ms
您可以帮助我提高性能吗？