本節內容來源於PGer的一個問題:
Q:
由於多版本的存在，那麼全表掃描是不是需要更長的時間了呢？

A:
關於全表掃描,不妨考慮2種極端的情況:
1.insert資料(事務已提交,下同),沒有執行update/delete,沒有dead tuple,全表掃描效率沒有影響;
2.insert資料,執行了大量的update/delete,同時禁用了autovacuum也沒有手工執行vacuum,那麼存在大量的dead tuple,效能上一是需要更多的IO操作,二是需要執行額外的CPU判斷(對於所有的tuple都要執行可見性判斷).
其判斷邏輯如下:
```
((Xmin == my-transaction &&　　　　　　　inserted by the current transaction
　Cmin < my-command &&　　　　　　　　　　before this command, and
　(Xmax is null ||　　　　　　　　　　　　the row has not been deleted, or
　　(Xmax == my-transaction &&　　　　　　it was deleted by the current transaction
　　　Cmax >= my-command)))　　　　　　　but not before this command,
||　　　　　　　　　　　　　　　　　　　　　or
　(Xmin is committed &&　　　　　　　　　　the row was inserted by a committed transaction, and
　　(Xmax is null ||　　　　　　　　　　　　the row has not been deleted, or
　　　(Xmax == my-transaction &&　　　　　the row is being deleted by this transaction
　　　　Cmax >= my-command) ||　　　　　　but it’s not deleted "yet", or
　　　　(Xmax != my-transaction &&　　　　the row was deleted by another transaction
　　　　　Xmax is not committed))))　　　　that has not been committed
```
簡單做個實驗,建立一張表t_fts,
1.插入資料,大小為s1,執行全表掃描,時間為m秒;
2.update所有行,大小為s2,執行全表掃描,時間為n秒.
理論上來說,n應為m的s2/s1倍左右(相對於IO時間,如果tuple數不多,CPU時間可以忽略不計).
建立資料表,插入資料:
```
testdb=# drop table if exists t_fts;
DROP TABLE
testdb=# create table t_fts(id int,c1 varchar(200),c2 varchar(200));
CREATE TABLE
testdb=# 
testdb=# insert into t_fts select x,lpad('c1'||x,200,'x'),lpad('c1'||x,200,'x') from generate_series(1,2000000) as x;
INSERT 0 2000000
testdb=# select pg_size_pretty(pg_table_size('t_fts'));
 pg_size_pretty 
----------------
 868 MB
(1 row)
```
禁用autovacuum,執行查詢:
```
testdb=# alter system set autovacuum=off;
ALTER SYSTEM
testdb=# show  autovacuum;
 autovacuum 
------------
 off
(1 row)

testdb=# explain analyze verbose select * from t_fts;
                                                         QUERY PLAN                                                         
----------------------------------------------------------------------------------------------------------------------------
 Seq Scan on public.t_fts  (cost=0.00..131112.16 rows=2000016 width=412) (actual time=0.048..1086.289 rows=2000000 loops=1)
   Output: id, c1, c2
 Planning Time: 30.762 ms
 Execution Time: 1181.360 ms
(4 rows)
```

執行update:
```
testdb=# update t_fts set c1 = lpad('c1'||(id+1),200,id+1||''),c2 =  lpad('c1'||(id+1),200,id+1||'');
UPDATE 2000000
testdb=# select pg_size_pretty(pg_table_size('t_fts'));
 pg_size_pretty 
----------------
 1737 MB
(1 row)

testdb=# explain analyze verbose select * from t_fts;
                                                          QUERY PLAN                                                         
  
-----------------------------------------------------------------------------------------------------------------------------
--
 Seq Scan on public.t_fts  (cost=0.00..262223.14 rows=4000014 width=412) (actual time=3168.414..6117.780 rows=2000000 loops=1
)
   Output: id, c1, c2
 Planning Time: 5.493 ms
 Execution Time: 6205.705 ms
(4 rows)

testdb=# explain analyze verbose select * from t_fts;
                                                          QUERY PLAN                                                         
 
-----------------------------------------------------------------------------------------------------------------------------
-
 Seq Scan on public.t_fts  (cost=0.00..262223.14 rows=4000014 width=412) (actual time=776.660..2311.270 rows=2000000 loops=1)
   Output: id, c1, c2
 Planning Time: 0.426 ms
 Execution Time: 2391.895 ms
(4 rows)

testdb=# explain analyze verbose select * from t_fts;
                                                          QUERY PLAN                                                         
 
-----------------------------------------------------------------------------------------------------------------------------
-
 Seq Scan on public.t_fts  (cost=0.00..262223.14 rows=4000014 width=412) (actual time=728.758..2293.157 rows=2000000 loops=1)
   Output: id, c1, c2
 Planning Time: 0.481 ms
 Execution Time: 2373.241 ms
(4 rows)

```

**參考資料**
[Concurrency Control](http://www.interdb.jp/pg/pgsql05.html)
[Understanding System Columns in PostgreSQL](https://severalnines.com/blog/understanding-system-columns-postgresql)

PostgreSQL DBA(55) - MVCC#8(對全表掃描的影響)

相關文章