1

我有一个关于 PostgreSQL 性能差异的问题。

我正在 MAC OS X 系统上开发一个 django Web 应用程序，该 Web 应用程序必须部署在 FreeBSD 服务器上。在系统的一个页面上有一个 ajax 控制的数据输入字段。在此字段中，您可以输入城市名称，当您输入两个或更多字符时，系统开始在数据库中查找城市，并在字符串开头显示符合这两个（或更多）字符的城市下拉列表.

在我将它部署在 FreeBSD 服务器上之前，所有这些似乎都运行良好。第一次部署没问题，但第二次部署显示出巨大的性能差异。

这是测试结果列表：

system1 proc: Intel Core 2 Duo 3.06 GHz, mem: 8GB. :
OS: OS X 10.6.8, 10.8.0 Darwin Kernel Version 10.8.0: Tue Jun  7 16:33:36 PDT 2011; 
root:xnu-1504.15.3~1/RELEASE_I386 i386
DB: PostgreSQL 8.4.5 on i386-apple-darwin10.5.0, compiled by GCC i686-apple-darwin10-gcc-4.2.1 (GCC) 4.2.1 (Apple Inc. build 5664), 64-bit

Tests system1:
query   parameter       time in ms
01      'de%'           909
02      'de%'           886
03      'den%'          132
04      'den %'         115
05      'den h%'        115
06      'den ha%'       117
07      'den haa%'      95
08      'den haag%'     100

host: system1
guest: parallels virtual machine, proc: 2 cpu, mem: 1 GB.
system2 
OS: 8.2-RELEASE FreeBSD 8.2-RELEASE #0: Thu Feb 17 02:41:51 UTC 2011     root@mason.cse.buffalo.edu:/usr/obj/usr/src/sys/GENERIC  amd64
DB: PostgreSQL 8.4.7 on amd64-portbld-freebsd8.2, compiled by GCC cc (GCC) 4.2.1 20070719  [FreeBSD], 64-bit

Tests system2:
query   parameter       time in ms
01      'de%'           1178
02      'de%'           857
03      'den%'          298
04      'den %'         233
05      'den h%'        134
06      'den ha%'       132
07      'den haa%'      132
08      'den haag%'     136

host: system1
guest: parallels virtual machine, proc: 2 cpu, mem: 1 GB.
system3
OS: 8.3-RELEASE FreeBSD 8.3-RELEASE #0: Mon Apr  9 21:23:18 UTC 2012     root@mason.cse.buffalo.edu:/usr/obj/usr/src/sys/GENERIC  amd64
DB: PostgreSQL 8.4.11 on amd64-portbld-freebsd8.3, compiled by GCC cc (GCC) 4.2.2 20070831 prerelease [FreeBSD], 64-bit

Tests system3:
query   parameter       time in ms
01      'de%'           7096
02      'de%'           7012
03      'den%'          6228
04      'den %'         6237
05      'den h%'        6145
06      'den ha%'       5640
07      'den haa%'      5512
08      'den haag%'     5561

这些参数模仿在 ajax 应用程序中输入数据的方式。这些结果来自通过 pgAdmin3 直接查询数据库，而不是通过应用程序。我没有把查询放在这里，因为我认为这不应该是相关的。数据库是相同的，并且在所有三个数据库实例上都使用相同的查询。

现在我可以理解system1：OS X系统，裸硬件和system2：运行FreeBSD的虚拟机之间的性能差异。

我不明白的是 system2 和 system3 之间的巨大性能差异，它们都是虚拟机在同一主机下运行。测试是在每个 vm 单独运行的情况下完成的。

有谁知道为什么会发生这种情况？

对于所有系统，我首先进行了 VACUUM ANALYZE。

system1: PostgreSQL 8.4.5 on i386-apple-darwin10.5.0
"Limit  (cost=3894.65..3894.70 rows=20 width=36) (actual time=1445.667..1445.671 rows=20 loops=1)"
"  ->  Sort  (cost=3894.65..3903.99 rows=3736 width=36) (actual time=1445.666..1445.666 rows=20 loops=1)"
"        Sort Key: public.geopoint_language.name_language, 



"                                                  ->  Seq Scan on geopoint  (cost=0.00..618.94 rows=625 width=11) (actual time=0.018..6.277 rows=728 loops=1)"
"                                                        Filter: (((admin2_code)::text = ''::text) AND (country_id = 154))"
"Total runtime: 1446.377 ms"


system2: PostgreSQL 8.4.7 on amd64-portbld-freebsd8.2
"Limit  (cost=3794.26..3794.31 rows=20 width=36) (actual time=64848.595..64848.903 rows=20 loops=1)"
"  ->  Sort  (cost=3794.26..3800.41 rows=2460 width=36) (actual time=64848.583..64848.679 rows=20 loops=1)"
"        Sort Key: public.geopoint_language.name_language, adm2_state_language.name_language, adm1_state_language.name_language"
"        Sort Method:  top-N heapsort  Memory: 27kB"
"        ->  HashAggregate  (cost=3704.20..3728.80 rows=2460 width=36) (actual time=64843.710..64846.020 rows=417 loops=1)"
"              ->  Append  (cost=1055.08..3667.30 rows=2460 width=36) (actual time=1201.210..64839.849 rows=417 loops=1)"
"                    ->  Hash Semi Join  (cost=1055.08..1784.76 rows=163 width=48) (actual time=1201.199..64707.965 rows=362 loops=1)"
"                          Hash Cond: (adm1_state_language.language_id = public.language.id)"
"                          ->  Merge Join  (cost=1049.36..1776.47 rows=287 width=64) (actual time=1200.852..64703.138 rows=362 loops=1)"
"                                Merge Cond: (public.state.pstate_id = public.state.id)"
"                                ->  Nested Loop  (cost=843.55..3856.49 rows=470 width=72) (actual time=1199.848..64693.793 rows=362 loops=1)"
"                                      ->  Nested Loop  (cost=843.55..2641.98 rows=4159 width=58) (actual time=39.390..64356.535 rows=13768 loops=1)"
"                                            Join Filter: ((public.state.admin2_code)::text = (public.geopoint.admin2_code)::text)"
"                                            ->  Nested Loop Semi Join  (cost=217.62..758.39 rows=8 width=51) (actual time=38.388..146.171 rows=862 loops=1)"
"                                                  ->  Nested Loop  (cost=217.62..754.28 rows=15 width=47) (actual time=38.361..124.606 rows=862 loops=1)"
"                                                        ->  Nested Loop Semi Join  (cost=217.62..616.50 rows=20 width=34) (actual time=38.332..87.939 rows=918 loops=1)"
"                                                              ->  Merge Join  (cost=217.62..606.79 rows=35 width=30) (actual time=38.300..58.248 rows=1066 loops=1)"
"                                                                    Merge Cond: (adm1_state_language.state_id = public.state.pstate_id)"
"                                                                    ->  Index Scan using idx_state_language_1 on state_language adm1_state_language  (cost=0.00..392.25 rows=7120 width=21) (actual time=0.019..16.666 rows=2826 loops=1)"
"                                                                          Index Cond: (preferred = true)"
"                                                                    ->  Sort  (cost=196.31..196.48 rows=68 width=9) (actual time=6.551..12.945 rows=1065 loops=1)"
"                                                                          Sort Key: public.state.pstate_id"
"                                                                          Sort Method:  quicksort  Memory: 45kB"
"                                                                          ->  Seq Scan on state  (cost=0.00..194.24 rows=68 width=9) (actual time=0.019..3.928 rows=431 loops=1)"
"                                                                                Filter: (((admin2_code)::text <> ''::text) AND (country_id = 154) AND ((admin3_code)::text = ''::text) AND ((admin4_code)::text = ''::text))"
"                                                              ->  Index Scan using pk_language_all on language  (cost=0.00..0.27 rows=1 width=4) (actual time=0.011..0.011 rows=1 loops=1066)"
"                                                                    Index Cond: (public.language.id = adm1_state_language.language_id)"
"                                                                    Filter: public.language.enabled"
"                                                        ->  Index Scan using idx_state_language_1 on state_language adm2_state_language  (cost=0.00..6.87 rows=1 width=21) (actual time=0.011..0.018 rows=1 loops=918)"
"                                                              Index Cond: ((adm2_state_language.state_id = public.state.id) AND (adm2_state_language.language_id = adm1_state_language.language_id) AND (adm2_state_language.preferred = true))"
"                                                  ->  Index Scan using pk_language_all on language  (cost=0.00..0.27 rows=1 width=4) (actual time=0.008..0.008 rows=1 loops=862)"
"                                                        Index Cond: (public.language.id = adm1_state_language.language_id)"
"                                                        Filter: public.language.enabled"
"                                            ->  Materialize  (cost=625.93..695.80 rows=6987 width=12) (actual time=0.005..36.752 rows=6884 loops=862)"
"                                                  ->  Seq Scan on geopoint  (cost=0.00..618.94 rows=6987 width=12) (actual time=0.015..45.945 rows=6884 loops=1)"
"                                                        Filter: (((admin2_code)::text <> ''::text) AND (country_id = 154))"
"                                      ->  Index Scan using idx_geopoint_language_3 on geopoint_language  (cost=0.00..0.28 rows=1 width=18) (actual time=0.013..0.013 rows=0 loops=13768)"
"                                            Index Cond: ((public.geopoint_language.geopoint_id = public.geopoint.id) AND (public.geopoint_language.language_id = adm1_state_language.language_id))"
"                                            Filter: (public.geopoint_language.endonym AND ((public.geopoint_language.name_language)::text ~~* 'de%%'::text))"
"                                ->  Sort  (cost=205.72..206.42 rows=282 width=4) (actual time=0.984..3.008 rows=342 loops=1)"
"                                      Sort Key: public.state.id"
"                                      Sort Method:  quicksort  Memory: 25kB"
"                                      ->  Seq Scan on state  (cost=0.00..194.24 rows=282 width=4) (actual time=0.427..0.850 rows=12 loops=1)"
"                                            Filter: ((country_id = 154) AND ((admin2_code)::text = ''::text) AND ((admin3_code)::text = ''::text) AND ((admin4_code)::text = ''::text))"
"                          ->  Hash  (cost=5.51..5.51 rows=17 width=4) (actual time=0.235..0.235 rows=17 loops=1)"
"                                ->  Seq Scan on language  (cost=0.00..5.51 rows=17 width=4) (actual time=0.015..0.127 rows=17 loops=1)"
"                                      Filter: enabled"
"                    ->  Subquery Scan "*SELECT* 2"  (cost=1844.22..1880.91 rows=2297 width=35) (actual time=125.544..127.175 rows=55 loops=1)"
"                          ->  Merge Join  (cost=1844.22..1857.94 rows=2297 width=35) (actual time=125.533..126.563 rows=55 loops=1)"
"                                Merge Cond: ((adm1_state_language.language_id = public.geopoint_language.language_id) AND ((public.state.admin1_code)::text = (public.geopoint.admin1_code)::text))"
"                                ->  Sort  (cost=392.00..392.29 rows=117 width=26) (actual time=82.529..82.647 rows=24 loops=1)"
"                                      Sort Key: adm1_state_language.language_id, public.state.admin1_code"
"                                      Sort Method:  quicksort  Memory: 26kB"
"                                      ->  Hash Join  (cost=209.42..387.98 rows=117 width=26) (actual time=30.234..82.289 rows=25 loops=1)"
"                                            Hash Cond: (adm1_state_language.language_id = public.language.id)"
"                                            ->  Hash Semi Join  (cost=203.49..380.10 rows=206 width=22) (actual time=29.732..81.508 rows=25 loops=1)"
"                                                  Hash Cond: (adm1_state_language.language_id = public.language.id)"
"                                                  ->  Hash Join  (cost=197.77..371.13 rows=364 width=18) (actual time=29.442..80.926 rows=29 loops=1)"
"                                                        Hash Cond: (adm1_state_language.state_id = public.state.id)"
"                                                        ->  Seq Scan on state_language adm1_state_language  (cost=0.00..143.03 rows=7120 width=21) (actual time=0.008..39.728 rows=7120 loops=1)"
"                                                              Filter: preferred"
"                                                        ->  Hash  (cost=194.24..194.24 rows=282 width=5) (actual time=2.366..2.366 rows=12 loops=1)"
"                                                              ->  Seq Scan on state  (cost=0.00..194.24 rows=282 width=5) (actual time=1.185..2.271 rows=12 loops=1)"
"                                                                    Filter: (((admin2_code)::text = ''::text) AND ((admin3_code)::text = ''::text) AND ((admin4_code)::text = ''::text) AND (country_id = 154))"
"                                                  ->  Hash  (cost=5.51..5.51 rows=17 width=4) (actual time=0.224..0.224 rows=17 loops=1)"
"                                                        ->  Seq Scan on language  (cost=0.00..5.51 rows=17 width=4) (actual time=0.008..0.118 rows=17 loops=1)"
"                                                              Filter: enabled"
"                                            ->  Hash  (cost=5.72..5.72 rows=17 width=4) (actual time=0.445..0.445 rows=17 loops=1)"
"                                                  ->  HashAggregate  (cost=5.55..5.72 rows=17 width=4) (actual time=0.233..0.323 rows=17 loops=1)"
"                                                        ->  Seq Scan on language  (cost=0.00..5.51 rows=17 width=4) (actual time=0.010..0.125 rows=17 loops=1)"
"                                                              Filter: enabled"
"                                ->  Sort  (cost=1452.22..1453.21 rows=396 width=25) (actual time=42.856..43.140 rows=55 loops=1)"
"                                      Sort Key: public.geopoint_language.language_id, public.geopoint.admin1_code"
"                                      Sort Method:  quicksort  Memory: 29kB"
"                                      ->  Hash Join  (cost=626.75..1435.14 rows=396 width=25) (actual time=23.266..42.430 rows=55 loops=1)"
"                                            Hash Cond: (public.geopoint_language.geopoint_id = public.geopoint.id)"
"                                            ->  Seq Scan on geopoint_language  (cost=0.00..802.94 rows=396 width=18) (actual time=3.081..25.949 rows=515 loops=1)"
"                                                  Filter: (endonym AND ((name_language)::text ~~* 'de %%'::text))"
"                                            ->  Hash  (cost=618.94..618.94 rows=625 width=11) (actual time=13.194..13.194 rows=728 loops=1)"
"                                                  ->  Seq Scan on geopoint  (cost=0.00..618.94 rows=625 width=11) (actual time=0.015..8.770 rows=728 loops=1)"
"                                                        Filter: (((admin2_code)::text = ''::text) AND (country_id = 154))"
"Total runtime: 64850.211 ms"


system3: PostgreSQL 8.4.11 on amd64-portbld-freebsd8.3
"Limit  (cost=2715.73..2715.78 rows=20 width=35) (actual time=28363.433..28363.721 rows=20 loops=1)"
"  ->  Sort  (cost=2715.73..2716.02 rows=113 width=35) (actual time=28363.421..28363.516 rows=20 loops=1)"
"        Sort Key: public.geopoint_language.name_language, adm2_state_language.name_language, adm1_state_language.name_language"
"        Sort Method:  top-N heapsort  Memory: 27kB"
"        ->  HashAggregate  (cost=2711.60..2712.73 rows=113 width=35) (actual time=28358.553..28360.849 rows=419 loops=1)"
"              ->  Append  (cost=571.63..2709.90 rows=113 width=35) (actual time=349.502..28354.791 rows=419 loops=1)"
"                    ->  Hash Semi Join  (cost=571.63..884.24 rows=3 width=48) (actual time=349.492..28233.388 rows=362 loops=1)"
"                          Hash Cond: (adm1_state_language.language_id = public.language.id)"
"                          ->  Nested Loop  (cost=565.91..878.47 rows=5 width=64) (actual time=349.172..28228.620 rows=362 loops=1)"
"                                Join Filter: ((public.state.admin2_code)::text = (public.geopoint.admin2_code)::text)"
"                                ->  Merge Join  (cost=565.91..811.16 rows=25 width=61) (actual time=278.147..20849.826 rows=292218 loops=1)"
"                                      Merge Cond: (public.state.pstate_id = adm1_state_language.state_id)"
"                                      Join Filter: (adm1_state_language.language_id = adm2_state_language.language_id)"
"                                      ->  Nested Loop  (cost=233.49..1199.63 rows=277 width=44) (actual time=12.014..10783.473 rows=292218 loops=1)"
"                                            ->  Nested Loop  (cost=233.49..725.31 rows=1 width=26) (actual time=7.617..42.736 rows=862 loops=1)"
"                                                  ->  Merge Join  (cost=233.49..720.62 rows=1 width=13) (actual time=7.585..16.639 rows=431 loops=1)"
"                                                        Merge Cond: (public.state.id = public.state.pstate_id)"
"                                                        ->  Index Scan using pk_state on state  (cost=0.00..523.51 rows=282 width=4) (actual time=1.000..2.268 rows=12 loops=1)"
"                                                              Filter: ((country_id = 154) AND ((admin2_code)::text = ''::text) AND ((admin3_code)::text = ''::text) AND ((admin4_code)::text = ''::text))"
"                                                        ->  Sort  (cost=196.31..196.48 rows=68 width=9) (actual time=6.566..9.020 rows=431 loops=1)"
"                                                              Sort Key: public.state.pstate_id"
"                                                              Sort Method:  quicksort  Memory: 45kB"
"                                                              ->  Seq Scan on state  (cost=0.00..194.24 rows=68 width=9) (actual time=0.015..3.942 rows=431 loops=1)"
"                                                                    Filter: (((admin2_code)::text <> ''::text) AND (country_id = 154) AND ((admin3_code)::text = ''::text) AND ((admin4_code)::text = ''::text))"
"                                                  ->  Index Scan using idx_state_language_1 on state_language adm2_state_language  (cost=0.00..4.68 rows=1 width=21) (actual time=0.012..0.026 rows=2 loops=431)"
"                                                        Index Cond: ((adm2_state_language.state_id = public.state.id) AND (adm2_state_language.preferred = true))"
"                                            ->  Index Scan using idx_geopoint_language_3 on geopoint_language  (cost=0.00..471.76 rows=205 width=18) (actual time=2.051..8.854 rows=339 loops=862)"
"                                                  Index Cond: (public.geopoint_language.language_id = adm2_state_language.language_id)"
"                                                  Filter: (public.geopoint_language.endonym AND ((public.geopoint_language.name_language)::text ~~* 'de%%'::text))"
"                                      ->  Sort  (cost=331.82..337.54 rows=2286 width=29) (actual time=237.631..3511.098 rows=625149 loops=1)"
"                                            Sort Key: adm1_state_language.state_id"
"                                            Sort Method:  quicksort  Memory: 746kB"
"                                            ->  Hash Join  (cost=11.87..204.28 rows=2286 width=29) (actual time=0.996..192.413 rows=7023 loops=1)"
"                                                  Hash Cond: (adm1_state_language.language_id = public.language.id)"
"                                                  ->  Hash Join  (cost=5.94..180.49 rows=4035 width=25) (actual time=0.514..116.195 rows=7023 loops=1)"
"                                                        Hash Cond: (adm1_state_language.language_id = public.language.id)"
"                                                        ->  Seq Scan on state_language adm1_state_language  (cost=0.00..143.03 rows=7120 width=21) (actual time=0.007..38.638 rows=7120 loops=1)"
"                                                              Filter: preferred"
"                                                        ->  Hash  (cost=5.72..5.72 rows=17 width=4) (actual time=0.452..0.452 rows=17 loops=1)"
"                                                              ->  HashAggregate  (cost=5.55..5.72 rows=17 width=4) (actual time=0.260..0.350 rows=17 loops=1)"
"                                                                    ->  Seq Scan on language  (cost=0.00..5.51 rows=17 width=4) (actual time=0.008..0.136 rows=17 loops=1)"
"                                                                          Filter: enabled"
"                                                  ->  Hash  (cost=5.72..5.72 rows=17 width=4) (actual time=0.425..0.425 rows=17 loops=1)"
"                                                        ->  HashAggregate  (cost=5.55..5.72 rows=17 width=4) (actual time=0.231..0.322 rows=17 loops=1)"
"                                                              ->  Seq Scan on language  (cost=0.00..5.51 rows=17 width=4) (actual time=0.010..0.118 rows=17 loops=1)"
"                                                                    Filter: enabled"
"                                ->  Index Scan using pk_city on geopoint  (cost=0.00..2.68 rows=1 width=12) (actual time=0.008..0.011 rows=1 loops=292218)"
"                                      Index Cond: (public.geopoint.id = public.geopoint_language.geopoint_id)"
"                                      Filter: (((public.geopoint.admin2_code)::text <> ''::text) AND (public.geopoint.country_id = 154))"
"                          ->  Hash  (cost=5.51..5.51 rows=17 width=4) (actual time=0.230..0.230 rows=17 loops=1)"
"                                ->  Seq Scan on language  (cost=0.00..5.51 rows=17 width=4) (actual time=0.014..0.124 rows=17 loops=1)"
"                                      Filter: enabled"
"                    ->  Subquery Scan "*SELECT* 2"  (cost=1014.04..1825.63 rows=110 width=35) (actual time=100.899..116.913 rows=57 loops=1)"
"                          ->  Hash Join  (cost=1014.04..1824.53 rows=110 width=35) (actual time=100.889..116.323 rows=57 loops=1)"
"                                Hash Cond: (((public.geopoint.admin1_code)::text = (public.state.admin1_code)::text) AND (public.geopoint_language.language_id = adm1_state_language.language_id))"
"                                ->  Hash Join  (cost=626.75..1432.19 rows=19 width=25) (actual time=19.280..34.017 rows=57 loops=1)"
"                                      Hash Cond: (public.geopoint_language.geopoint_id = public.geopoint.id)"
"                                      ->  Seq Scan on geopoint_language  (cost=0.00..802.94 rows=616 width=18) (actual time=1.942..16.204 rows=720 loops=1)"
"                                            Filter: (endonym AND ((name_language)::text ~~* 'de%%'::text))"
"                                      ->  Hash  (cost=618.94..618.94 rows=625 width=11) (actual time=13.437..13.437 rows=728 loops=1)"
"                                            ->  Seq Scan on geopoint  (cost=0.00..618.94 rows=625 width=11) (actual time=0.012..8.923 rows=728 loops=1)"
"                                                  Filter: (((admin2_code)::text = ''::text) AND (country_id = 154))"
"                                ->  Hash  (cost=385.53..385.53 rows=117 width=26) (actual time=81.568..81.568 rows=25 loops=1)"
"                                      ->  Hash Join  (cost=209.64..385.53 rows=117 width=26) (actual time=29.986..81.410 rows=25 loops=1)"
"                                            Hash Cond: (adm1_state_language.language_id = public.language.id)"
"                                            ->  Hash Join  (cost=203.70..378.69 rows=206 width=22) (actual time=29.457..80.612 rows=25 loops=1)"
"                                                  Hash Cond: (adm1_state_language.language_id = public.language.id)"
"                                                  ->  Hash Join  (cost=197.77..371.13 rows=364 width=18) (actual time=29.011..79.869 rows=29 loops=1)"
"                                                        Hash Cond: (adm1_state_language.state_id = public.state.id)"
"                                                        ->  Seq Scan on state_language adm1_state_language  (cost=0.00..143.03 rows=7120 width=21) (actual time=0.008..38.849 rows=7120 loops=1)"
"                                                              Filter: preferred"
"                                                        ->  Hash  (cost=194.24..194.24 rows=282 width=5) (actual time=2.245..2.245 rows=12 loops=1)"
"                                                              ->  Seq Scan on state  (cost=0.00..194.24 rows=282 width=5) (actual time=1.139..2.167 rows=12 loops=1)"
"                                                                    Filter: (((admin2_code)::text = ''::text) AND ((admin3_code)::text = ''::text) AND ((admin4_code)::text = ''::text) AND (country_id = 154))"
"                                                  ->  Hash  (cost=5.72..5.72 rows=17 width=4) (actual time=0.423..0.423 rows=17 loops=1)"
"                                                        ->  HashAggregate  (cost=5.55..5.72 rows=17 width=4) (actual time=0.226..0.317 rows=17 loops=1)"
"                                                              ->  Seq Scan on language  (cost=0.00..5.51 rows=17 width=4) (actual time=0.008..0.119 rows=17 loops=1)"
"                                                                    Filter: enabled"
"                                            ->  Hash  (cost=5.72..5.72 rows=17 width=4) (actual time=0.449..0.449 rows=17 loops=1)"
"                                                  ->  HashAggregate  (cost=5.55..5.72 rows=17 width=4) (actual time=0.254..0.344 rows=17 loops=1)"
"                                                        ->  Seq Scan on language  (cost=0.00..5.51 rows=17 width=4) (actual time=0.010..0.119 rows=17 loops=1)"
"                                                              Filter: enabled"
"Total runtime: 28366.784 ms"

system2 和 system3 显示的时间对我来说没有任何意义。

4

1 回答 1

1

我已在 explain.depesz.com: system2 , system3中对您的两个解释计划提出质疑，以便更容易理解和解释它们。

现在这些计划相当大，所以当在两个不同版本的 postgresql 下执行时它们的不同也就不足为奇了。即使是补丁版本做出的更改可能会影响规划器，一个地方的错误修复或小优化也可能会改变另一个地方的决策。

Something that jumps out for me is that the system3 plan has noticeable discrepancies in the estimated vs actual row cardinality of many steps in the execution. For example line 7 and 8 had actual rows 292K, but estimated rows 25 and 277; line 16, 625K vs 2286. In contrast, system2 had no plan lines with such gross mis-estimations, there are a few that missed by one order of magnitude on a fairly modest total count, but nothing missing by several orders of magnitude producing row cardinality in the 100K+ range.

Such differences may indicate that you have incorrect statistics on system3 which cause that pg instance to choose a less efficient plan because it has inaccurate data about the cost of its choices. You may want to try reanalyzing all your tables and possibly increasing the statistics target.

Have you verified that the data is identical on each system? Also check that all the query planning settings are the same on the two systems.

I also wonder why you are even bothering which such an old version of postgres. 8.4 is many years old and there have been numerous substantial improvements in features, performance, and concurrency between 8.4 and 9.2. It seems silly to spend substantial effort tuning a nearly obsolete version of postgres instead of targeting the latest stable release unless you are bound to it for compelling other reasons.

于 2013-03-30T22:27:35.983 回答