postgresql - Postgres 表选择查询太慢

Question

我有一个 gps 跟踪应用程序。它有一个名为 gps_vehicle_data 的表，其中经常存储传入的 gps 数据。我经常查询这个表来处理它，因为它只包含原始数据。最近，我目睹了在表上执行 select 语句的长时间延迟。下面是 EXPLAIN 的结果。我还尝试执行 VACUUM 并粘贴下面的结果。可能是什么原因？

EXPLAIN (ANALYZE, BUFFERS) select * from gps_vehicle_data;

                                                           QUERY PLAN                                                            
---------------------------------------------------------------------------------------------------------------------------------
 Seq Scan on gps_vehicle_data  (cost=0.00..130818.81 rows=1400881 width=1483) (actual time=209.129..62488.822 rows=9635 loops=1)
   Buffers: shared hit=13132 read=103678 dirtied=67 written=25
 Planning time: 0.050 ms
 Execution time: 62500.850 ms

真空输出。

VACUUM (VERBOSE,ANALYSE) gps_vehicle_data;
INFO:  vacuuming "public.gps_vehicle_data"
INFO:  index "gps_vehicle_data_pkey" now contains 1398939 row versions in 10509 pages
DETAIL:  0 index row versions were removed.
0 index pages have been deleted, 0 are currently reusable.
CPU 0.07s/0.09u sec elapsed 9.38 sec.
INFO:  index "gps_vehicle_data_status_idx" now contains 1398939 row versions in 4311 pages
DETAIL:  0 index row versions were removed.
0 index pages have been deleted, 0 are currently reusable.
CPU 0.03s/0.04u sec elapsed 4.50 sec.
INFO:  index "gps_vehicle_data_url_data_idx" now contains 1399004 row versions in 98928 pages
DETAIL:  0 index row versions were removed.
0 index pages have been deleted, 0 are currently reusable.
CPU 0.76s/0.88u sec elapsed 82.74 sec.
INFO:  index "gps_vehicle_data_createdon_idx" now contains 1399007 row versions in 3946 pages
DETAIL:  0 index row versions were removed.
0 index pages have been deleted, 0 are currently reusable.
CPU 0.00s/0.02u sec elapsed 1.92 sec.
INFO:  "gps_vehicle_data": found 0 removable, 1402484 nonremovable row versions in 116884 out of 116884 pages
DETAIL:  1401490 dead row versions cannot be removed yet.
There were 143431 unused item pointers.
Skipped 0 pages due to buffer pins.
0 pages are entirely empty.
CPU 1.70s/2.38u sec elapsed 200.61 sec.
INFO:  vacuuming "pg_toast.pg_toast_17296"
INFO:  index "pg_toast_17296_index" now contains 0 row versions in 1 pages
DETAIL:  0 index row versions were removed.
0 index pages have been deleted, 0 are currently reusable.
CPU 0.00s/0.00u sec elapsed 0.01 sec.
INFO:  "pg_toast_17296": found 0 removable, 0 nonremovable row versions in 0 out of 0 pages
DETAIL:  0 dead row versions cannot be removed yet.
There were 0 unused item pointers.
Skipped 0 pages due to buffer pins.
0 pages are entirely empty.
CPU 0.00s/0.00u sec elapsed 0.01 sec.
INFO:  analyzing "public.gps_vehicle_data"
INFO:  "gps_vehicle_data": scanned 30000 of 116884 pages, containing 335 live rows and 359656 dead rows; 335 rows in sample, 1042851 estimated total rows
VACUUM

Answer 1

您读取了超过 100000 个 block 以获得大约 10000 行，这意味着您的表几乎完全没有任何内容(它正在遭受表膨胀)。

该表一定在某个时候包含了更多的数据，其中大部分已被删除，这导致了膨胀。

正如@a_horse_with_no_name 提到的，您的某些行无法回收，因为有一些旧事务阻止了它们，但是虽然 VACUUM 将释放死行，但它不会重组表以摆脱膨胀。

在这种情况下，正确的解决方案是使用 VACUUM (FULL, ANALYZE) gps_vehicle_data(ANALYZE 只是为了很好的衡量，因为它看起来像您的表统计信息已关闭), 这将重组表。但是请注意，在 VACUUM (FULL) 运行时，对表的所有访问都被阻止。