华山论剑之 PostgreSQL sequence (下篇)

rename 对 sequence 的影响

关联列与 sequence 后,即 sequence 属于该列后,drop 表或列时会自动 drop 相关 sequence。

但如果对表或列 rename 后,甚至 rename sequence后,会发生什么呢?

我们来做一下实验。

创建测试表 tb_test_sequence_rename

alvindb=>
CREATE TABLE tb_test_sequence_rename (
    test_id BIGSERIAL PRIMARY KEY,
    create_time TIMESTAMP DEFAULT clock_timestamp()
);
CREATE TABLE

查看表与 sequence:

alvindb=> \d tb_test_sequence_rename
                                            Table "alvin.tb_test_sequence_rename"
   Column    |            Type             | Collation | Nullable |                         Default                          
-------------+-----------------------------+-----------+----------+----------------------------------------------------------
 test_id     | bigint                      |           | not null | nextval('tb_test_sequence_rename_test_id_seq'::regclass)
 create_time | timestamp without time zone |           |          | clock_timestamp()
Indexes:
    "tb_test_sequence_rename_pkey" PRIMARY KEY, btree (test_id)

alvindb=> \d tb_test_sequence_rename_test_id_seq
             Sequence "alvin.tb_test_sequence_rename_test_id_seq"
  Type  | Start | Minimum |       Maximum       | Increment | Cycles? | Cache 
--------+-------+---------+---------------------+-----------+---------+-------
 bigint |     1 |       1 | 9223372036854775807 |         1 | no      |     1
Owned by: alvin.tb_test_sequence_rename.test_id

alvindb=> 

对表进行 rename :

alvindb=> ALTER TABLE tb_test_sequence_rename RENAME TO tb_test_sequence_rename2;
ALTER TABLE

通过如下结果,我们可以看到, rename 表后 'Owned by' 也会随之自动变化

alvindb=> \d tb_test_sequence_rename2
                                           Table "alvin.tb_test_sequence_rename2"
   Column    |            Type             | Collation | Nullable |                         Default                          
-------------+-----------------------------+-----------+----------+----------------------------------------------------------
 test_id     | bigint                      |           | not null | nextval('tb_test_sequence_rename_test_id_seq'::regclass)
 create_time | timestamp without time zone |           |          | clock_timestamp()
Indexes:
    "tb_test_sequence_rename_pkey" PRIMARY KEY, btree (test_id)

alvindb=> \d tb_test_sequence_rename_test_id_seq
             Sequence "alvin.tb_test_sequence_rename_test_id_seq"
  Type  | Start | Minimum |       Maximum       | Increment | Cycles? | Cache 
--------+-------+---------+---------------------+-----------+---------+-------
 bigint |     1 |       1 | 9223372036854775807 |         1 | no      |     1
Owned by: alvin.tb_test_sequence_rename2.test_id

alvindb=> 

我们再 rename 列试一下,

alvindb=> ALTER TABLE tb_test_sequence_rename2 RENAME test_id TO test_id2;
ALTER TABLE

通过如下结果,我们可以看到, rename 列后 'Owned by' 也会随之自动变化

alvindb=> \d tb_test_sequence_rename2
                                           Table "alvin.tb_test_sequence_rename2"
   Column    |            Type             | Collation | Nullable |                         Default                          
-------------+-----------------------------+-----------+----------+----------------------------------------------------------
 test_id2    | bigint                      |           | not null | nextval('tb_test_sequence_rename_test_id_seq'::regclass)
 create_time | timestamp without time zone |           |          | clock_timestamp()
Indexes:
    "tb_test_sequence_rename_pkey" PRIMARY KEY, btree (test_id2)

alvindb=> \d tb_test_sequence_rename_test_id_seq
             Sequence "alvin.tb_test_sequence_rename_test_id_seq"
  Type  | Start | Minimum |       Maximum       | Increment | Cycles? | Cache 
--------+-------+---------+---------------------+-----------+---------+-------
 bigint |     1 |       1 | 9223372036854775807 |         1 | no      |     1
Owned by: alvin.tb_test_sequence_rename2.test_id2

alvindb=> 

我们来 rename 一下 sequence,

alvindb=> ALTER SEQUENCE tb_test_sequence_rename_test_id_seq RENAME TO tb_test_sequence_rename_test_id_seq2;
ALTER SEQUENCE

通过如下结果,我们可以看到, rename sequence 后 'Owned by' 也会随之自动变化,并且 Default 中的 sequence 也会随之变化

alvindb=> \d tb_test_sequence_rename2
                                            Table "alvin.tb_test_sequence_rename2"
   Column    |            Type             | Collation | Nullable |                          Default                          
-------------+-----------------------------+-----------+----------+-----------------------------------------------------------
 test_id2    | bigint                      |           | not null | nextval('tb_test_sequence_rename_test_id_seq2'::regclass)
 create_time | timestamp without time zone |           |          | clock_timestamp()
Indexes:
    "tb_test_sequence_rename_pkey" PRIMARY KEY, btree (test_id2)

alvindb=> \d tb_test_sequence_rename_test_id_seq2
            Sequence "alvin.tb_test_sequence_rename_test_id_seq2"
  Type  | Start | Minimum |       Maximum       | Increment | Cycles? | Cache 
--------+-------+---------+---------------------+-----------+---------+-------
 bigint |     1 |       1 | 9223372036854775807 |         1 | no      |     1
Owned by: alvin.tb_test_sequence_rename2.test_id2

alvindb=> 

通过以上三个 rename 实验,可以发现,正常 rename 不会对 sequence 的使用产生影响

无论是 rename 表名,列名,还是 sequence 的名字,如我们所期望,PostgreSQL 都会智能地作出相应的修改。

复制表或迁移表时 sequence 的相关操作

以下表为例,

alvindb=> \d tb_test_bigserial
                                            Table "alvin.tb_test_bigserial"
   Column    |            Type             | Collation | Nullable |                      Default                       
-------------+-----------------------------+-----------+----------+----------------------------------------------------
 test_id     | bigint                      |           | not null | nextval('tb_test_bigserial_test_id_seq'::regclass)
 create_time | timestamp without time zone |           |          | clock_timestamp()
Indexes:
    "tb_test_bigserial_pkey" PRIMARY KEY, btree (test_id)
alvindb=> \d tb_test_bigserial_test_id_seq
                Sequence "alvin.tb_test_bigserial_test_id_seq"
  Type  | Start | Minimum |       Maximum       | Increment | Cycles? | Cache 
--------+-------+---------+---------------------+-----------+---------+-------
 bigint |     1 |       1 | 9223372036854775807 |         1 | no      |     1
Owned by: alvin.tb_test_bigserial.test_id

在一些情况下,为避免 DDL 或大量 DML 对表 tb_test_bigserial 的影响,我们可以通过 RENAME 表的方式实现:

  1. 根据表 tb_test_bigserial 复制出相同表结构的表 tb_test_bigserial_new
  2. tb_test_bigserial_new 进行 DDL 或 大量 DML
  3. tb_test_bigserial_new RENAME 回 tb_test_bigserial

根据表 tb_test_bigserial 复制出相同表结构的表 tb_test_bigserial_new

alvindb=> CREATE TABLE tb_test_bigserial_new (LIKE tb_test_bigserial INCLUDING ALL);
CREATE TABLE
alvindb=> \d tb_test_bigserial_new
                                          Table "alvin.tb_test_bigserial_new"
   Column    |            Type             | Collation | Nullable |                      Default                       
-------------+-----------------------------+-----------+----------+----------------------------------------------------
 test_id     | bigint                      |           | not null | nextval('tb_test_bigserial_test_id_seq'::regclass)
 create_time | timestamp without time zone |           |          | clock_timestamp()
Indexes:
    "tb_test_bigserial_new_pkey" PRIMARY KEY, btree (test_id)

tb_test_bigserial_new 进行 DDL 或 大量 DML 后,将 tb_test_bigserial_new RENAME 回 tb_test_bigserial

alvindb=> BEGIN;
BEGIN
alvindb=> ALTER TABLE tb_test_bigserial RENAME TO tb_test_bigserial_old;
ALTER TABLE
alvindb=> ALTER TABLE tb_test_bigserial_new RENAME TO tb_test_bigserial;
ALTER TABLE
alvindb=> END;
COMMIT

这时,查看新表结构:

alvindb=> \d tb_test_bigserial
                                            Table "alvin.tb_test_bigserial"
   Column    |            Type             | Collation | Nullable |                      Default                       
-------------+-----------------------------+-----------+----------+----------------------------------------------------
 test_id     | bigint                      |           | not null | nextval('tb_test_bigserial_test_id_seq'::regclass)
 create_time | timestamp without time zone |           |          | clock_timestamp()
Indexes:
    "tb_test_bigserial_new_pkey" PRIMARY KEY, btree (test_id)

此处,我们只关注 sequence。上述的索引的名字可以根据需要决定是否需要 RENAME 回原来的名字。

查看 sequence,

alvindb=> \d tb_test_bigserial_test_id_seq
                Sequence "alvin.tb_test_bigserial_test_id_seq"
  Type  | Start | Minimum |       Maximum       | Increment | Cycles? | Cache 
--------+-------+---------+---------------------+-----------+---------+-------
 bigint |     1 |       1 | 9223372036854775807 |         1 | no      |     1
Owned by: alvin.tb_test_bigserial_old.test_id

从以上 'Owned by' 可以看出,sequence tb_test_bigserial_test_id_seq 还是归旧表 tb_test_bigserial_old 的列所有。

从上文“rename 对 sequence 的影响”我们知道,这是正常的。

此时 DROP 旧表,会提示新表 tb_test_bigserial 还在依赖 sequence tb_test_bigserial_test_id_seq

alvindb=> DROP TABLE tb_test_bigserial_old;
ERROR:  cannot drop table tb_test_bigserial_old because other objects depend on it
DETAIL:  default value for column test_id of table tb_test_bigserial depends on sequence tb_test_bigserial_test_id_seq
HINT:  Use DROP ... CASCADE to drop the dependent objects too.
alvindb=> 

以下 RENAME 表时关键的一步:

alvindb=> ALTER SEQUENCE tb_test_bigserial_test_id_seq OWNED BY tb_test_bigserial.test_id;
ALTER SEQUENCE

通过上述 SQL,sequence tb_test_bigserial_test_id_seq 就归新表的列所有了。

在日常操作中,我们有可能忘记修改 sequence 的所属关系。以致后来 DROP 老表时加了 CASCADE,将 sequence 也一起 DROP 掉了,从而引发问题。

此时 DROP 表就不报错了:

alvindb=> DROP TABLE tb_test_bigserial_old;
DROP TABLE

以下是 RENAME 后所期望的结果(注意 sequence 的 'Owned by'):


alvindb=> \d tb_test_bigserial
                                            Table "alvin.tb_test_bigserial"
   Column    |            Type             | Collation | Nullable |                      Default                       
-------------+-----------------------------+-----------+----------+----------------------------------------------------
 test_id     | bigint                      |           | not null | nextval('tb_test_bigserial_test_id_seq'::regclass)
 create_time | timestamp without time zone |           |          | clock_timestamp()
Indexes:
    "tb_test_bigserial_new_pkey" PRIMARY KEY, btree (test_id)

alvindb=> \d tb_test_bigserial_test_id_seq
                Sequence "alvin.tb_test_bigserial_test_id_seq"
  Type  | Start | Minimum |       Maximum       | Increment | Cycles? | Cache 
--------+-------+---------+---------------------+-----------+---------+-------
 bigint |     1 |       1 | 9223372036854775807 |         1 | no      |     1
Owned by: alvin.tb_test_bigserial.test_id

sequence cache

从上面可以看到,sequence 默认的 cache 是1。

在高并发的系统中,为了提高性能,可以增大 cache,如 10,即一次将 sequence 的10个数值 load 到内存中。这样,只需要访问 sequence 1 次,而不是10次。

下面,将 sequence tb_test_bigserial_test_id_seq 的 cache 改成 10:

alvindb=> SELECT pg_backend_pid();
 pg_backend_pid 
----------------
          16649
(1 row)
alvindb=> ALTER SEQUENCE tb_test_bigserial_test_id_seq CACHE 10;
ALTER SEQUENCE
alvindb=> \d tb_test_bigserial_test_id_seq
                Sequence "alvin.tb_test_bigserial_test_id_seq"
  Type  | Start | Minimum |       Maximum       | Increment | Cycles? | Cache 
--------+-------+---------+---------------------+-----------+---------+-------
 bigint |     1 |       1 | 9223372036854775807 |         1 | no      |    10
Owned by: alvin.tb_test_bigserial.test_id

在 session 1(pid 16649)中插入数据

alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> SELECT * FROM tb_test_bigserial ORDER BY 2 DESC;
 test_id |        create_time         
---------+----------------------------
       2 | 2021-05-01 15:21:00.116979
       1 | 2021-05-01 15:20:56.405149
(2 rows)

alvindb=> SELECT * FROM tb_test_bigserial_test_id_seq;
 last_value | log_cnt | is_called 
------------+---------+-----------
         10 |      32 | t
(1 row)
alvindb=> SELECT lastval();
 lastval 
---------
       2
(1 row)

从以上结果可以看到,此 session 中一次取了10个数据,last_value 是10,is_called 是 true,即新的 session 中从11开始。

lastval() 是2,即当前 session 中,上一次 nextval 返回的是2。需要注意,lastval() 并没有参数,即它返回最近任意 sequence 的 nextval 值。

如下,

alvindb=> SELECT nextval('tb_test_sequence_rename_test_id_seq2');
 nextval 
---------
       1
(1 row)

alvindb=> SELECT lastval();
 lastval 
---------
       1
(1 row)

在 session 2(pid 14287)中插入数据:

alvindb=> SELECT pg_backend_pid();
 pg_backend_pid 
----------------
          14287
(1 row)
alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> SELECT * FROM tb_test_bigserial ORDER BY 2 DESC;
 test_id |        create_time         
---------+----------------------------
      12 | 2021-05-01 15:22:02.582036
      11 | 2021-05-01 15:22:01.313919
       2 | 2021-05-01 15:21:00.116979
       1 | 2021-05-01 15:20:56.405149
(4 rows)
alvindb=> SELECT * FROM tb_test_bigserial_test_id_seq;
 last_value | log_cnt | is_called 
------------+---------+-----------
         20 |      32 | t
(1 row)
alvindb=> SELECT lastval();
 lastval 
---------
      12
(1 row)

从以上结果可以看到,虽然在 session 1中,sequence tb_test_bigserial_test_id_seq 只取了两次值,但 session 2中是从11开始取值的。

因为 cache 是10,所以1-10是预留给了 session 1。11-20是预留给了 session 2。

在 session 2中,last_value 是20,lastval()是12。

如果现在再看 session 1呢?

alvindb=> SELECT pg_backend_pid();
 pg_backend_pid 
----------------
          16649
(1 row)

alvindb=> SELECT now();
              now              
-------------------------------
 2021-05-01 15:23:54.674943+08
(1 row)

alvindb=> SELECT * FROM tb_test_bigserial_test_id_seq;
 last_value | log_cnt | is_called 
------------+---------+-----------
         20 |      32 | t
(1 row)

alvindb=> SELECT lastval();
 lastval 
---------
       2
(1 row)

现在看到 session 1中的 last_value 变成了 20,而 lastval() 还是2。

对比 session 1 和 session 2,可以看出,last_value 是与 session 无关的。而 lastval() 是与 session 有关,但与是哪个 sequence 无关。

现在,在 session 3 中验证一下,

alvindb=> SELECT pg_backend_pid();
 pg_backend_pid 
----------------
          15809
(1 row)

alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> SELECT * FROM tb_test_bigserial ORDER BY 2 DESC;
 test_id |        create_time         
---------+----------------------------
      22 | 2021-05-01 15:25:08.643196
      21 | 2021-05-01 15:25:04.005421
      12 | 2021-05-01 15:22:02.582036
      11 | 2021-05-01 15:22:01.313919
       2 | 2021-05-01 15:21:00.116979
       1 | 2021-05-01 15:20:56.405149
(6 rows)
alvindb=> SELECT now();
              now              
-------------------------------
 2021-05-01 15:25:24.499995+08
(1 row)

alvindb=> SELECT * FROM tb_test_bigserial_test_id_seq;
 last_value | log_cnt | is_called 
------------+---------+-----------
         30 |      32 | t
(1 row)

alvindb=> SELECT lastval();
 lastval 
---------
      22
(1 row)

可以看出,以上推论是正确的。session 3 中 sequence 是从21开始的,插入两行数据后,last_value 变成 30,lastval() 在当前 session 中是22。

现在有一个问题,后插入的数据的 test_id 一定比先插入的大吗?能否根据 test_id 排序来确定是哪一行数据先插入的呢?

下面,我们回到 session 2 继续插入数据,

alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> SELECT * FROM tb_test_bigserial ORDER BY 2 DESC;
 test_id |        create_time         
---------+----------------------------
      14 | 2021-05-01 15:28:16.500833
      13 | 2021-05-01 15:28:04.276826
      22 | 2021-05-01 15:25:08.643196
      21 | 2021-05-01 15:25:04.005421
      12 | 2021-05-01 15:22:02.582036
      11 | 2021-05-01 15:22:01.313919
       2 | 2021-05-01 15:21:00.116979
       1 | 2021-05-01 15:20:56.405149
(8 rows)

可以看出,在 session 2中,sequence 继续从 13 开始。这是因为 session 2中一次性 cache 的 10 个数值(11-20)还没有用完。

所以,如果 cache 设置大于1,就不能以 sequence 的大小判断其插入的顺序。

可以用 create_time 来排序。但需要注意,在事务中,now() 返回的时间是一样的,而 clock_timestamp() 返回的则是实际的时间,

这也是在创建表 tb_test_bigserial 时采用 clock_timestamp() 而不是 now() 的原因。

现在如果调用 setval 会发生什么呢?

我们在 session 1中调用 setval,

alvindb=> SELECT pg_backend_pid();
 pg_backend_pid 
----------------
          16649
(1 row)
alvindb=> SELECT setval('tb_test_bigserial_test_id_seq', 51);
 setval 
--------
     51
(1 row)
alvindb=> SELECT * FROM tb_test_bigserial_test_id_seq;
 last_value | log_cnt | is_called 
------------+---------+-----------
         51 |       0 | t
(1 row)
alvindb=> SELECT lastval();
 lastval 
---------
      51
(1 row)

上述 SQL 已将 last_value 设为 51,即下一个返回 52。

现在我们在 session 2中继续插入,

alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> SELECT * FROM tb_test_bigserial ORDER BY 2 DESC;
 test_id |        create_time         
---------+----------------------------
      18 | 2021-05-01 15:28:22.692747
      17 | 2021-05-01 15:28:22.371691
      16 | 2021-05-01 15:28:22.021806
      15 | 2021-05-01 15:28:20.917822
      14 | 2021-05-01 15:28:16.500833
      13 | 2021-05-01 15:28:04.276826
      22 | 2021-05-01 15:25:08.643196
      21 | 2021-05-01 15:25:04.005421
      12 | 2021-05-01 15:22:02.582036
      11 | 2021-05-01 15:22:01.313919
       2 | 2021-05-01 15:21:00.116979
       1 | 2021-05-01 15:20:56.405149
(12 rows)
alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> SELECT * FROM tb_test_bigserial ORDER BY 2 DESC;
 test_id |        create_time         
---------+----------------------------
      19 | 2021-05-01 15:28:26.485763
      18 | 2021-05-01 15:28:22.692747
      17 | 2021-05-01 15:28:22.371691
      16 | 2021-05-01 15:28:22.021806
      15 | 2021-05-01 15:28:20.917822
      14 | 2021-05-01 15:28:16.500833
      13 | 2021-05-01 15:28:04.276826
      22 | 2021-05-01 15:25:08.643196
      21 | 2021-05-01 15:25:04.005421
      12 | 2021-05-01 15:22:02.582036
      11 | 2021-05-01 15:22:01.313919
       2 | 2021-05-01 15:21:00.116979
       1 | 2021-05-01 15:20:56.405149
(13 rows)

alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> SELECT * FROM tb_test_bigserial ORDER BY 2 DESC;
 test_id |        create_time         
---------+----------------------------
      20 | 2021-05-01 15:28:29.747721
      19 | 2021-05-01 15:28:26.485763
      18 | 2021-05-01 15:28:22.692747
      17 | 2021-05-01 15:28:22.371691
      16 | 2021-05-01 15:28:22.021806
      15 | 2021-05-01 15:28:20.917822
      14 | 2021-05-01 15:28:16.500833
      13 | 2021-05-01 15:28:04.276826
      22 | 2021-05-01 15:25:08.643196
      21 | 2021-05-01 15:25:04.005421
      12 | 2021-05-01 15:22:02.582036
      11 | 2021-05-01 15:22:01.313919
       2 | 2021-05-01 15:21:00.116979
       1 | 2021-05-01 15:20:56.405149
(14 rows)

alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> SELECT * FROM tb_test_bigserial ORDER BY 2 DESC;
 test_id |        create_time         
---------+----------------------------
      52 | 2021-05-01 15:28:32.084771
      20 | 2021-05-01 15:28:29.747721
      19 | 2021-05-01 15:28:26.485763
      18 | 2021-05-01 15:28:22.692747
      17 | 2021-05-01 15:28:22.371691
      16 | 2021-05-01 15:28:22.021806
      15 | 2021-05-01 15:28:20.917822
      14 | 2021-05-01 15:28:16.500833
      13 | 2021-05-01 15:28:04.276826
      22 | 2021-05-01 15:25:08.643196
      21 | 2021-05-01 15:25:04.005421
      12 | 2021-05-01 15:22:02.582036
      11 | 2021-05-01 15:22:01.313919
       2 | 2021-05-01 15:21:00.116979
       1 | 2021-05-01 15:20:56.405149
(15 rows)

alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1
alvindb=> SELECT * FROM tb_test_bigserial ORDER BY 2 DESC;
 test_id |        create_time         
---------+----------------------------
      53 | 2021-05-01 15:28:45.026188
      52 | 2021-05-01 15:28:32.084771
      20 | 2021-05-01 15:28:29.747721
      19 | 2021-05-01 15:28:26.485763
      18 | 2021-05-01 15:28:22.692747
      17 | 2021-05-01 15:28:22.371691
      16 | 2021-05-01 15:28:22.021806
      15 | 2021-05-01 15:28:20.917822
      14 | 2021-05-01 15:28:16.500833
      13 | 2021-05-01 15:28:04.276826
      22 | 2021-05-01 15:25:08.643196
      21 | 2021-05-01 15:25:04.005421
      12 | 2021-05-01 15:22:02.582036
      11 | 2021-05-01 15:22:01.313919
       2 | 2021-05-01 15:21:00.116979
       1 | 2021-05-01 15:20:56.405149
(16 rows)

可以看出,setval 后,session 2中并没有立即生效,而是等 session 2中预分配的10个数值消耗完,才继续从 sequence 又 cache 了10个数值。

另外,可以看出,上述数据中 test_id 列的数据并不是连续的。这也是使用 cache 的原因。

所以,在高并发的系统中,为了提高 sequence 的性能,不仅仅是增大 cache,以上提到的问题都需要注意。

sequence 权限

创建用户 simon。

postgres=# CREATE USER simon WITH PASSWORD 'simon';
CREATE ROLE

现在将表 tb_test_bigserial 的读写权限授予 simon

alvindb=# \c alvindb alvin;
You are now connected to database "alvindb" as user "alvin".
alvindb=> \dp+ tb_test_bigserial
                                   Access privileges
 Schema |       Name        | Type  | Access privileges | Column privileges | Policies 
--------+-------------------+-------+-------------------+-------------------+----------
 alvin  | tb_test_bigserial | table |                   |                   | 
(1 row)

alvindb=> GRANT USAGE ON SCHEMA alvin TO "simon";
GRANT
alvindb=> GRANT SELECT, INSERT, UPDATE, DELETE, TRUNCATE ON alvin.tb_test_bigserial TO "simon";
GRANT
alvindb=> \dp+ tb_test_bigserial
                                    Access privileges
 Schema |       Name        | Type  |  Access privileges  | Column privileges | Policies 
--------+-------------------+-------+---------------------+-------------------+----------
 alvin  | tb_test_bigserial | table | alvin=arwdDxt/alvin+|                   | 
        |                   |       | simon=arwdD/alvin   |                   | 
(1 row)

通过用户 simon 插入数据

alvindb=> \c alvindb simon 
You are now connected to database "alvindb" as user "simon".
alvindb=> SET search_path TO alvin;
SET
alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
ERROR:  permission denied for sequence tb_test_bigserial_test_id_seq

此时插入数据提示没有 sequence tb_test_bigserial_test_id_seq 的权限。

现在将 sequence tb_test_bigserial_test_id_seq 也授权一下:

alvindb=> \c alvindb alvin
You are now connected to database "alvindb" as user "alvin".
alvindb=> GRANT ALL ON SEQUENCE tb_test_bigserial_test_id_seq TO "simon";
GRANT
alvindb=> \dp+ tb_test_bigserial
                                    Access privileges
 Schema |       Name        | Type  |  Access privileges  | Column privileges | Policies 
--------+-------------------+-------+---------------------+-------------------+----------
 alvin  | tb_test_bigserial | table | alvin=arwdDxt/alvin+|                   | 
        |                   |       | simon=arwdD/alvin   |                   | 
(1 row)

alvindb=> \dp+ tb_test_bigserial_test_id_seq
                                          Access privileges
 Schema |             Name              |   Type   | Access privileges | Column privileges | Policies 
--------+-------------------------------+----------+-------------------+-------------------+----------
 alvin  | tb_test_bigserial_test_id_seq | sequence | alvin=rwU/alvin  +|                   | 
        |                               |          | simon=rwU/alvin   |                   | 
(1 row)
alvindb=> INSERT INTO tb_test_bigserial(create_time) VALUES (DEFAULT);
INSERT 0 1

将 sequence 的读写权限授予 simon 后,simon 就可以成功向表 tb_test_bigserial 插入数据了。

可以看出,要对其他用户授予读写权限,不仅要授予表的权限,还需要授予对应 sequence 的权限。

总结

sequence 看起来很简单,但如果使用姿势不对,很可能造成 MI (Major Incident)。

对于以下常见的有关 sequence 的问题,需要多留意。

  1. integer vs bigint

    对于交易型的表,log 表,history 表等较大的表,主键 sequence 需要用 bigint。否则可能遇到如下错误:

    ERROR:  nextval: reached maximum value of sequence "tb_test_sequence_test_id_seq" (2147483647)
    

    其实,不止数据库需要考虑这个问题,应用程序中也需要考虑。

    存储主键值的变量等需要用相应的类型,如 Java 中对应的需要使用 long 类型而不是 int 类型。

    开发和测试同学可能都需要排查这个问题,以防止将来生产出现不必要的 MI。

    开发可以直接看代码或查看表的DDL。测试的话,可以通过 setval 来修改当前 sequence 的值进行测试。

    甚至有强制主键 id 必须用 bigint 的数据库规范来避免这个问题。

  2. sequence 'Owned by'

    通过 rename 方式复制表时,需要注意同时修改 sequence 的 'Owned by'。

    否则,新表主键 default 中的 sequence 在 drop 老表时有可能被级联 drop 掉。

  3. sequence nextval

    当涉及到表或 sequence 的迁移,或需要重新创建 sequence,需要注意重新设置 sequence 值。

  4. sequence cache

    当 sequence cache 大于1时,需要注意排序时不要用 sequence 的列,可以考虑根据时间的列排序。

  5. sequence 权限

    在一些情况下,不但要授予相关用户表的权限,也需要注意 sequence 的权限。

公众号

关注 DBA Daily 公众号,第一时间收到文章的更新。
通过一线 DBA 的日常工作,学习实用数据库技术干货!

公众号优质文章推荐

PostgreSQL VACUUM 之深入浅出

华山论剑之 PostgreSQL sequence

[PG Upgrade Series] Extract Epoch Trap

[PG Upgrade Series] Toast Dump Error

GitLab supports only PostgreSQL now

MySQL or PostgreSQL?

PostgreSQL hstore Insight

ReIndex 失败原因调查

PG 数据导入 Hive 乱码问题调查

PostGIS 扩展创建失败原因调查

posted @ 2022-03-09 09:50  DBADaily  阅读(488)  评论(0编辑  收藏  举报