解释分析比postgres中的实际查询慢

Question

我有以下查询

select * from activity_feed where user_id in (select following_id from user_follow where follower_id=:user_id)
union 
select * from activity_feed where project_id in (select project_id from user_project_follow where user_id=:user_id)
order by id desc limit 30

根据 postico

，运行时间约为 14 毫秒

但是当我对此查询执行 explain analyze 时，计划时间为 0.5 毫秒，执行时间约为 800 毫秒（这是我实际期望的）。这是因为没有 explain analyze 的查询正在返回缓存结果吗？我仍然得到不到 20 毫秒的结果。使用其他值。

哪一个更能体现我在制作中的表现？我还意识到这是一个相当低效的查询，我似乎无法找出可以提高效率的索引。我可能不得不不使用 union

编辑：执行计划

Limit  (cost=1380.94..1380.96 rows=10 width=148) (actual time=771.111..771.405 rows=10 loops=1)
  ->  Sort  (cost=1380.94..1385.64 rows=1881 width=148) (actual time=771.097..771.160 rows=10 loops=1)
        Sort Key: activity_feed."timestamp" DESC
        Sort Method: top-N heapsort  Memory: 27kB
        ->  HashAggregate  (cost=1321.48..1340.29 rows=1881 width=148) (actual time=714.888..743.273 rows=4462 loops=1)
              Group Key: activity_feed.id, activity_feed."timestamp", activity_feed.user_id, activity_feed.verb, activity_feed.object_type, activity_feed.object_id, activity_feed.project_id, activity_feed.privacy_level, activity_feed.local_time, activity_feed.local_date
              ->  Append  (cost=5.12..1274.46 rows=1881 width=148) (actual time=0.998..682.466 rows=4487 loops=1)
                    ->  Hash Join  (cost=5.12..610.43 rows=1350 width=70) (actual time=0.982..326.089 rows=3013 loops=1)
                          Hash Cond: (activity_feed.user_id = user_follow.following_id)
                          ->  Seq Scan on activity_feed  (cost=0.00..541.15 rows=24215 width=70) (actual time=0.016..150.535 rows=24215 loops=1)
                          ->  Hash  (cost=4.78..4.78 rows=28 width=8) (actual time=0.911..0.922 rows=29 loops=1)
                                Buckets: 1024  Batches: 1  Memory Usage: 10kB
                                ->  Index Only Scan using unique_user_follow_pair on user_follow  (cost=0.29..4.78 rows=28 width=8) (actual time=0.022..0.334 rows=29 loops=1)
                                      Index Cond: (follower_id = '17420532762804570'::bigint)
                                      Heap Fetches: 0
                    ->  Hash Join  (cost=30.50..635.81 rows=531 width=70) (actual time=0.351..301.945 rows=1474 loops=1)
                          Hash Cond: (activity_feed_1.project_id = user_project_follow.project_id)
                          ->  Seq Scan on activity_feed activity_feed_1  (cost=0.00..541.15 rows=24215 width=70) (actual time=0.027..143.896 rows=24215 loops=1)
                          ->  Hash  (cost=30.36..30.36 rows=11 width=8) (actual time=0.171..0.182 rows=11 loops=1)
                                Buckets: 1024  Batches: 1  Memory Usage: 9kB
                                ->  Index Only Scan using idx_user_project_follow_temp on user_project_follow  (cost=0.28..30.36 rows=11 width=8) (actual time=0.020..0.102 rows=11 loops=1)
                                      Index Cond: (user_id = '17420532762804570'::bigint)
                                      Heap Fetches: 11
Planning Time: 0.571 ms
Execution Time: 771.774 ms

提前感谢您的帮助！

Answer 1

像您在这里显示的那样非常慢的时钟访问（当 TIMING 默认为 ON 时慢近 100 倍！）通常表示旧硬件或旧内核 IME。如果您对性能非常讲究，则无法信任 EXPLAIN (ANALYZE) 来获取良好数据可能会非常令人沮丧，因此您应该考虑升级硬件或 OS.

解释分析比postgres中的实际查询慢

Explain analyze slower than actual query in postgres

sql

postgresql

indexing

query-optimization

sql-execution-plan