lightdb關閉事務自動提交

　　在lightdb/pg中，事務是自動提交的，也就是除非明確通過BEGIN;開啓事務，否則任何語句都是獨立事務。在ltsql客戶端，可以通過\echo :AUTOCOMMIT查看自動提交模式，如下：

[zjh@hs-10-20-30-193 ~]$ ltsql -p9999
ltsql (13.3-22.2)
Type "help" for help.

zjh@postgres=# \echo :AUTOCOMMIT
on
zjh@postgres=# \set AUTOCOMMIT off
zjh@postgres=# \echo :AUTOCOMMIT
off
zjh@postgres=# 

　　因爲ltsql是客戶端工具，不是libpq/ltjdbc實現，所以它不具典型意義。

　　強制自動提交，無參數控制有個缺點，對於一些oracle/mysql不會提交的場景也會自動提交，而我們認爲它不會提交。如explain analyze，如下：

zjh@postgres=*# select count(1) from scott.departments where dept_id<10000;
 count 
-------
    64
(1 row)

zjh@postgres=*# explain analyze update scott.departments set dept_id=dept_id+10000 where dept_id<10000;
                                                       QUERY PLAN                                                        
-------------------------------------------------------------------------------------------------------------------------
 Update on departments  (cost=0.00..54236.67 rows=396542 width=236) (actual time=688.890..688.892 rows=0 loops=1)
   ->  Seq Scan on departments  (cost=0.00..54236.67 rows=396542 width=236) (actual time=0.015..688.719 rows=64 loops=1)
         Filter: (dept_id < 10000)
         Rows Removed by Filter: 6024736
 Planning Time: 0.089 ms
 Execution Time: 689.190 ms
(6 rows)

zjh@postgres=*# select count(1) from scott.departments where dept_id<10000;
 count 
-------
     0
(1 row)

　　explain analyze實際上被提交了，而如果數據被修改後和原來的數據不唯一，就可能無法恢復。這是非常麻煩的。

　　要正確執行更新，需要先BEGIN，然後rollback，如下：

zjh@postgres=# begin;
BEGIN
zjh@postgres=*# select count(1) from scott.departments where dept_id<20000;
 count 
-------
    64
(1 row)

zjh@postgres=*# explain analyze update scott.departments set dept_id=dept_id+10000 where dept_id<20000;
                                                        QUERY PLAN                                                         
---------------------------------------------------------------------------------------------------------------------------
 Update on departments  (cost=0.00..54236.67 rows=396542 width=236) (actual time=694.907..694.908 rows=0 loops=1)
   ->  Seq Scan on departments  (cost=0.00..54236.67 rows=396542 width=236) (actual time=694.636..694.654 rows=64 loops=1)
         Filter: (dept_id < 20000)
         Rows Removed by Filter: 6024736
 Planning Time: 0.064 ms
 Execution Time: 695.004 ms
(6 rows)

zjh@postgres=*# rollback;
ROLLBACK
zjh@postgres=# select count(1) from scott.departments where dept_id<20000;
 count 
-------
    64
(1 row)

　　因爲這很難控制，因此lightdb在內核中基於隱藏字段實現自動更新時間戳，能夠記錄行最後更新時間，這樣能夠做到恢復。如下：

zjh@postgres=# create table test_lut(id int) WITH UPDATE CURRENT_TIMESTAMP;
WARNING:  LightDB DDL check warn! keyword not allowed with [column name]:id
WARNING:  LightDB DDL check warn! no primary key!
WARNING:  LightDB DDL check warn! no gmt_create or no gmt_modified!
CREATE TABLE
zjh@postgres=# insert into test_lut values(1);
INSERT 0 1
zjh@postgres=# select * from test_lut ;
 id 
----
  1
(1 row)

zjh@postgres=# \dS+ test_lut
                                                Table "public.test_lut"
 Column |            Type             | Collation | Nullable |      Default      | Storage | Stats target | Description 
--------+-----------------------------+-----------+----------+-------------------+---------+--------------+-------------
 id     | integer                     |           |          |                   | plain   |              | 
 ltaut  | timestamp without time zone |           |          | CURRENT_TIMESTAMP | plain   |              | 
Access method: heap

zjh@postgres=# select id,ltaut from test_lut ;
 id |           ltaut            
----+----------------------------
  1 | 2022-06-11 20:53:47.295945
(1 row)

zjh@postgres=# update test_lut set id=id+1;
UPDATE 1
zjh@postgres=# select id,ltaut from test_lut ;
 id |           ltaut            
----+----------------------------
  2 | 2022-06-11 20:54:33.041471
(1 row)