sql - 为什么使用 LIMIT 的横向连接会增加执行时间？

Question

当我运行带有横向连接和内部 LIMIT 的查询时，它使用嵌套循环连接。但是当我删除 LIMIT 时，它使用哈希右连接。为什么？

EXPLAIN ANALYSE
SELECT proxy.*
FROM jobs
LEFT  OUTER JOIN LATERAL (
        SELECT proxy.*
        FROM proxy
        WHERE jobs.id = proxy.job_id
) proxy ON true

Hash Right Join  (cost=2075.47..3029.05 rows=34688 width=12) (actual time=9.951..24.758 rows=35212 loops=1)
Hash Cond: (proxy.job_id = jobs.id)
  ->  Seq Scan on proxy  (cost=0.00..524.15 rows=34015 width=12) (actual time=0.011..2.502 rows=34028 loops=1)
  ->  Hash  (cost=1641.87..1641.87 rows=34688 width=4) (actual time=9.842..9.842 rows=34689 loops=1)
        Buckets: 65536  Batches: 1  Memory Usage: 1732kB
        ->  Index Only Scan using jobs_pkey on jobs  (cost=0.29..1641.87 rows=34688 width=4) (actual time=0.010..4.904 rows=34689 loops=1)
              Heap Fetches: 921

但是当我为查询添加限制时，实际时间从 24 跳到 150:

EXPLAIN ANALYSE
SELECT proxy.*
FROM jobs
LEFT OUTER JOIN LATERAL (
        SELECT proxy.*
        FROM proxy
        WHERE jobs.id = proxy.job_id
        limit 1
) proxy ON true

Nested Loop Left Join  (cost=0.58..290506.19 rows=34688 width=12) (actual time=0.024..155.753 rows=34689 loops=1)
  ->  Index Only Scan using jobs_pkey on jobs  (cost=0.29..1641.87 rows=34688 width=4) (actual time=0.014..3.984 rows=34689 loops=1)
        Heap Fetches: 921
  ->  Limit  (cost=0.29..8.31 rows=1 width=12) (actual time=0.001..0.001 rows=1 loops=34689)
        ->  Index Scan using index_job_proxy_on_job_id on loc_job_source_materials  (cost=0.29..8.31 rows=1 width=12) (actual time=0.001..0.001 rows=1 loops=34689)
              Index Cond: (jobs.id = job_id)

Answer 1

优化器足够聪明，可以将您的第一个查询重写为

SELECT proxy.*
FROM proxy
   RIGHT OUTER JOIN jobs
      ON jobs.id = proxy.job_id;

但是这种优化不能用 LIMIT 子句进行，所以只能进行嵌套循环连接。