ORDER BY 导致性能下降

请帮助优化请求。我不明白为什么使用 ORDER BY 实际时间会增加几倍至 150 毫秒，而没有 ORDER BY 15-5 毫秒。也许需要某种索引，但我不知道是哪一个

explain analyze
with recursive src as(   
select  
    d.station_id1 border_form, 
    d.station_id2 border_to, 
    d.country_key_station_id1 code1, 
    d.country_key_station_id2 code2,
    '/' || d.country_key_station_id1 || '/' || d.country_key_station_id2 || '/' as route,
    '' as route_distance,
    d.distance
from 
    test_distance.distances_border d 
where 
    d.country_key_station_id1 = 20
union all
select
    s.border_form,
    d1.station_id2 as border_to,
    s.code1, 
    d1.country_key_station_id2 as code2,    
    s.route || d1.country_key_station_id2 || '/' as route,
    s.route_distance || dc.distance || '+' as route_distance,
    s.distance + dc.distance as distance            
from 
    test_distance.distances_border d1 
join src s on d1.country_key_station_id1 = s.code2
    and not s.route like '%/' || d1.country_key_station_id2 || '/%'
join test_distance.distances_border_in_country dc on s.border_to = 
dc.station_id1 
    and d1.station_id1 = dc.station_id2         
)
select 
border_form, 
border_to, 
code1, 
code2,
    dist_start,
dist_transit,
dist_end,   
route,
distance + dist_start + dist_end as distance 
from (
select 
    border_form, 
    border_to, 
    code1, 
    code2,
    test_distance.get_distance('910000' , border_form, code1) as dist_start,
    route_distance as dist_transit,
    test_distance.get_distance(border_to , '720903', code2) as dist_end,    
    route,
    distance 
from src s 
where code2 = 29
order by distance
) r

如果按距离排序完全放在最后，那么实际时间会变得更长，从 180 毫秒开始，如果有人给我一个想法，我将不胜感激

查询计划

Subquery Scan on r  (cost=830.05..830.07 rows=1 width=160) (actual time=73.354..73.357 rows=22 loops=1)
CTE src
->  Recursive Union  (cost=0.00..825.38 rows=185 width=125) (actual time=0.030..7.470 rows=1046 loops=1)
      ->  Seq Scan on distances_border d  (cost=0.00..4.70 rows=35 width=26) (actual time=0.024..0.135 rows=35 loops=1)
            Filter: (country_key_station_id1 = 20)
            Rows Removed by Filter: 125
      ->  Hash Join  (cost=32.86..81.70 rows=15 width=125) (actual time=0.948..1.189 rows=169 loops=6)
            Hash Cond: (((dc.station_id1)::text = (s_1.border_to)::text) AND ((dc.station_id2)::text = (d1.station_id1)::text))
            ->  Seq Scan on distances_border_in_country dc  (cost=0.00..27.60 rows=1660 width=18) (actual time=0.002..0.065 rows=1660 loops=5)
            ->  Hash  (cost=28.68..28.68 rows=279 width=166) (actual time=0.882..0.882 rows=508 loops=6)
                  Buckets: 1024  Batches: 1  Memory Usage: 8kB
                  ->  Hash Join  (cost=11.38..28.68 rows=279 width=166) (actual time=0.273..0.749 rows=508 loops=6)
                        Hash Cond: (d1.country_key_station_id1 = s_1.code2)
                        Join Filter: (s_1.route !~~ (('%/'::text || (d1.country_key_station_id2)::text) || '/%'::text))
                        Rows Removed by Join Filter: 1475
                        ->  Seq Scan on distances_border d1  (cost=0.00..3.60 rows=160 width=22) (actual time=0.002..0.010 rows=160 loops=6)
                        ->  Hash  (cost=7.00..7.00 rows=350 width=152) (actual time=0.046..0.046 rows=174 loops=6)
                              Buckets: 1024  Batches: 1  Memory Usage: 23kB
                              ->  WorkTable Scan on src s_1  (cost=0.00..7.00 rows=350 width=152) (actual time=0.001..0.017 rows=174 loops=6)
   ->  Sort  (cost=4.67..4.68 rows=1 width=152) (actual time=73.351..73.352 rows=22 loops=1)
    Sort Key: s.distance
    Sort Method: quicksort  Memory: 28kB
    ->  CTE Scan on src s  (cost=0.00..4.66 rows=1 width=152) (actual time=7.986..73.318 rows=22 loops=1)
          Filter: (code2 = 29)
          Rows Removed by Filter: 1024
    Planning time: 1.990 ms
    Execution time: 73.654 ms

我看到好像一直在排序，但是我不知道如何理解写的内容:(