如果我在 select 上使用星号，我的查询会更快。 为什么要这样做。 我正在使用 Postgres

Question

I'm using Postgres 12.6.

My First Query & Explain Result

        select
        *
        FROM   "coupon_coupon"
               LEFT OUTER JOIN "coupon_couponnumber"
                            ON ( "coupon_coupon"."id" =
                                 "coupon_couponnumber"."coupon_id" )
               LEFT OUTER JOIN "coupon_usercoupon"
                            ON ( "coupon_couponnumber"."id" =
                                 "coupon_usercoupon"."coupon_number_id" )

--- explain result
Hash Right Join  (cost=623.52..121842.44 rows=1282129 width=428)
  Hash Cond: (coupon_couponnumber.coupon_id = coupon_coupon.id)
  ->  Merge Left Join  (cost=0.85..117852.86 rows=1282129 width=173)
        Merge Cond: (coupon_couponnumber.id = coupon_usercoupon.coupon_number_id)
        ->  Index Scan using coupon_couponnumber_pkey on coupon_couponnumber  (cost=0.43..50788.71 rows=1282129 width=89)
        ->  Index Scan using coupon_usercoupon_coupon_number_id_idx on coupon_usercoupon  (cost=0.43..49083.82 rows=1185586 width=84)
  ->  Hash  (cost=488.96..488.96 rows=10696 width=255)
        ->  Seq Scan on coupon_coupon  (cost=0.00..488.96 rows=10696 width=255)

My Second Query & Explain Result

select
        coupon_coupon.id
        FROM   "coupon_coupon"
               LEFT OUTER JOIN "coupon_couponnumber"
                            ON ( "coupon_coupon"."id" =
                                 "coupon_couponnumber"."coupon_id" )
               LEFT OUTER JOIN "coupon_usercoupon"
                            ON ( "coupon_couponnumber"."id" =
                                 "coupon_usercoupon"."coupon_number_id" )
--- explain result
Hash Right Join  (cost=48091.56..93461.52 rows=1282129 width=4)
  Hash Cond: (coupon_couponnumber.coupon_id = coupon_coupon.id)
  ->  Hash Right Join  (cost=47468.90..89471.94 rows=1282129 width=4)
        Hash Cond: (coupon_usercoupon.coupon_number_id = coupon_couponnumber.id)
        ->  Seq Scan on coupon_usercoupon  (cost=0.00..24617.86 rows=1185586 width=4)
        ->  Hash  (cost=26433.29..26433.29 rows=1282129 width=8)
              ->  Seq Scan on coupon_couponnumber  (cost=0.00..26433.29 rows=1282129 width=8)
  ->  Hash  (cost=488.96..488.96 rows=10696 width=4)
        ->  Seq Scan on coupon_coupon  (cost=0.00..488.96 rows=10696 width=4)

I found column using asterisk on first query, and the second using specific column. I don't understand why the second query is slower than the first one.

Answer 1

The 1st one using (SELECT *) with a MERGE LEFT JOIN on indexes, while the 2nd using (SELECT AColumn) with HASH RIGHT JOIN .

The cost of a hash join is greater than the merge. It's probably because PG underevaluate the cost of the hash strategy... This is a classical problem when the "planer" (PG optimizer) does not take the lenght of the rows returned properly as a count in the total query cost.