Postgresql 哪个查询更快？

我得到两个答案：。但我仍然不知道哪一个是有效的。下面是对这两个方面的解释：

解释

my

查询：

 Hash Anti Join  (cost=136.09..138.49 rows=1 width=556)
   Hash Cond: (t1.id = t2.parent_id)
   CTE tree
     ->  Recursive Union  (cost=0.14..132.81 rows=101 width=556)
           ->  Index Scan using resource_type_idx_parent_resource_type_id on resource_type  (cost=0
                 Index Cond: (parent_resource_type_id IS NULL)
           ->  Hash Join  (cost=0.33..12.26 rows=10 width=556)
                 Hash Cond: (rt.parent_resource_type_id = tree.id)
                 ->  Seq Scan on resource_type rt  (cost=0.00..11.30 rows=130 width=524)
                 ->  Hash  (cost=0.20..0.20 rows=10 width=36)
                       ->  WorkTable Scan on tree  (cost=0.00..0.20 rows=10 width=36)
   ->  CTE Scan on tree t1  (cost=0.00..2.02 rows=101 width=556)
   ->  Hash  (cost=2.02..2.02 rows=101 width=4)
         ->  CTE Scan on tree t2  (cost=0.00..2.02 rows=101 width=4)

Laurenz-Albe

回答：

 Nested Loop Anti Join  (cost=132.81..388.39 rows=100 width=556)
   Join Filter: (t2.deep_name ~~ (t1.deep_name || '.%'::text))
   CTE tree
     ->  Recursive Union  (cost=0.14..132.81 rows=101 width=556)
           ->  Index Scan using resource_type_idx_parent_resource_type_id on resource_type  (cost=0
                 Index Cond: (parent_resource_type_id IS NULL)
           ->  Hash Join  (cost=0.33..12.26 rows=10 width=556)
                 Hash Cond: (rt.parent_resource_type_id = tree.id)
                 ->  Seq Scan on resource_type rt  (cost=0.00..11.30 rows=130 width=524)
                 ->  Hash  (cost=0.20..0.20 rows=10 width=36)
                       ->  WorkTable Scan on tree  (cost=0.00..0.20 rows=10 width=36)
   ->  CTE Scan on tree t1  (cost=0.00..2.02 rows=101 width=556)
   ->  CTE Scan on tree t2  (cost=0.00..2.02 rows=101 width=32)

区别在于：

--- CTE Scan on tree  (cost=132.81..134.83 rows=101 w (Selection)
+++ (clipboard)
@@ -1,5 +1,5 @@
- Hash Anti Join  (cost=136.09..138.49 rows=1 width=556)
-   Hash Cond: (t1.id = t2.parent_id)
+ Nested Loop Anti Join  (cost=132.81..388.39 rows=100 width=556)
+   Join Filter: (t2.deep_name ~~ (t1.deep_name || '.%'::text))
    CTE tree
      ->  Recursive Union  (cost=0.14..132.81 rows=101 width=556)
            ->  Index Scan using resource_type_idx_parent_resource_type_id on resource_type  (cost=0
@@ -10,5 +10,4 @@
                  ->  Hash  (cost=0.20..0.20 rows=10 width=36)
                        ->  WorkTable Scan on tree  (cost=0.00..0.20 rows=10 width=36)
    ->  CTE Scan on tree t1  (cost=0.00..2.02 rows=101 width=556)
-   ->  Hash  (cost=2.02..2.02 rows=101 width=4)
-         ->  CTE Scan on tree t2  (cost=0.00..2.02 rows=101 width=4)
+   ->  CTE Scan on tree t2  (cost=0.00..2.02 rows=101 width=32)

---

哪一个更快？如何从解释结果中知道这一点？

在这两种方法上运行

explain（ANALYZE，BUFFERS）

。差异（如果有）并不显著。CTE只产生100行，并且（反）自连接可以完全在内存中执行。缝合缓冲区。顺便说一句：CTE和~~运算符都可能导致非平凡数据量的降级。

EXPLAIN ANALYZE

是您的朋友。在我看来，第一个版本的Hash-Anti-join略优于嵌套循环。