PostgreSQL10当sequence 带有cache时，正常关闭PG Cluster时对sequence的影响

postgres=# select * from testa;
 id |   info    
----+-----------
  1 | abc
  2 | abcd
 11 | test
  3 | asdfghjkl
  4 | asdf4
  5 | asdf5
  6 | asdf6
  7 | asdf7
  8 | asdf8
  9 | asdf9
 10 | asdf10
 11 | asdf11
  9 | asdf9
(13 rows)


postgres=# 
postgres=# select sequencename,sequenceowner,cache_size,last_value from pg_sequences;
 sequencename | sequenceowner | cache_size | last_value 
--------------+---------------+------------+------------
 testa_id_seq | postgres      |        100 |        100
(1 row)


postgres=# \q
[postgres@pgserver ~]$ pg_ctl stop -m fast  --->>>>注意这是fast关闭，也就是“干净的”关闭PG Cluster。
waiting for server to shut down.... done
server stopped
[postgres@pgserver ~]$ pg_ctl start--->>>>启动PG Cluster
waiting for server to start....2018-04-23 09:32:08.316 EDT [19339] LOG:  listening on IPv4 address "0.0.0.0",port 5432
2018-04-23 09:32:08.317 EDT [19339] LOG:  listening on IPv6 address "::",port 5432
2018-04-23 09:32:08.327 EDT [19339] LOG:  listening on Unix socket "/tmp/.s.PGSQL.5432"
2018-04-23 09:32:08.403 EDT [19340] LOG:  database system was shut down at 2018-04-23 09:32:02 EDT
2018-04-23 09:32:08.412 EDT [19339] LOG:  database system is ready to accept connections
 done
server started
[postgres@pgserver ~]$ psql
psql (10beta2)
Type "help" for help.


postgres=# select sequencename,last_value from pg_sequences;
 sequencename | sequenceowner | cache_size | last_value 
--------------+---------------+------------+------------
 testa_id_seq | postgres      |        100 |        100 ---->>>此时与“pg_ctl stop -m fast”之前一样。
(1 row)


postgres=# insert into testa(info)  values('asdf???');
INSERT 0 1
postgres=# select * from testa;
 id  |   info    
-----+-----------
   1 | abc
   2 | abcd
  11 | test
   3 | asdfghjkl
   4 | asdf4
   5 | asdf5
   6 | asdf6
   7 | asdf7
   8 | asdf8
   9 | asdf9
  10 | asdf10
  11 | asdf11
   9 | asdf9
 101 | asdf??? --->>>注意插入的id值是101，并不是预想的12。
(14 rows)


postgres=# select sequencename,last_value from pg_sequences;
 sequencename | sequenceowner | cache_size | last_value 
--------------+---------------+------------+------------
 testa_id_seq | postgres      |        100 |        200---->>>此处从100变为200。
(1 row)


postgres=#

Oracle 带cache的 sequence 在正常关闭和异常关闭时的表现：
http://www.xifenfei.com/2015/04/%E6%AD%A3%E5%B8%B8%E5%85%B3%E9%97%AD%E6%95%B0%E6%8D%AE%E5%BA%93sequence-cache%E4%B8%8D%E4%B8%BA0-sequence%E4%B8%8D%E8%B7%B3%E8%B7%83.html


结论：PG Cluster正常关闭时，没用真正用过的cache并不回写到数据字典中，这样的后果就是在PG Cluster起来之后，该sequence被插入的第一个值就是pg_sequences.last_value+1

估计PG对sequence的设计是这样的：只要sequence有cache了，那么PG就视为这些cache已经实际上被使用了，虽然可能在实际上并没有真正使用。参见：https://www.postgresql.org/docs/10/static/view-pg-sequences.html