GaussDB(DWS)运维 -- sequence常见运维操作

【DWS的序列概述】

GaussDB(DWS)是一个share nothing架构的产品，seqeunce编号需要通过GTM统一分配。集群规模越大、节点数越多、sequence调用越频繁，sequnce对GTM的压力就越大。对于seqeunce的使用一般有如下建议

1. 建议业务中不要大量使用sequence，建议不要超过50个
2. 如果业务允许，建议使用uuid替换sequence
3. 使用sequence的时候，需要定义cache值不小于100

【序列相关信息查询】

建议定义如下视图进行查询

CREATE OR REPLACE FUNCTION public.dfm_get_seqence_info
(
    OUT namespace text,  -- 序列的schema
    OUT sequencename text, -- 序列名称
    OUT min_value bigint,  -- 序列最小值
    OUT max_value bigint,  -- 序列最大值
    OUT start_value bigint,  -- 起始值
    OUT increment_by bigint,  -- 增加步长
    OUT cache_value bigint,  -- cache值
    OUT is_cycled boolean -- 是否可以循环使用
)
 RETURNS SETOF record
 LANGUAGE plpgsql
 NOT FENCED NOT SHIPPABLE
AS $function$
DECLARE
    rowObject record;
    rowDetail record;
    queryObjectStr text;
    queryDetailStr text;
BEGIN
    --Get all the node names
    queryObjectStr := 'SELECT c.oid, relname, n.nspname FROM pg_class c INNER JOIN pg_namespace n on c.relnamespace = n.oid WHERE c.relkind = ''S''';
    FOR rowObject IN EXECUTE(queryObjectStr) LOOP
        queryDetailStr := 'SELECT start_value, increment_by, max_value, min_value, cache_value, is_cycled FROM ' || quote_ident(rowObject.nspname) || '.' || quote_ident(rowObject.relname);
        FOR rowDetail IN EXECUTE(queryDetailStr) LOOP
            namespace = rowObject.nspname;
            sequencename = rowObject.relname;
            start_value = rowDetail.start_value;
            increment_by = rowDetail.increment_by;
            max_value = rowDetail.max_value;
            min_value = rowDetail.min_value;
            cache_value = rowDetail.cache_value;
            is_cycled = rowDetail.is_cycled;
            RETURN next;
        END LOOP;
    END LOOP;
    RETURN;
END; $function$
;

CREATE VIEW public.dfm_get_seqence_info AS SELECT * FROM public.dfm_get_seqence_info();

示例

postgres=# CREATE SEQUENCE public.test_seq;
CREATE SEQUENCE
Time: 13.917 ms
postgres=# CREATE TABLE public.test_serial(a bigserial, b int) DISTRIBUTE BY ROUNDROBIN;
NOTICE:  CREATE TABLE will create implicit sequence "test_serial_a_seq" for serial column "test_serial.a"
SQLSTATE: 00000
LOCATION:  CreateSeqOwnedByTable, parse_utilcmd.cpp:144
CREATE TABLE
Time: 16.572 ms
postgres=# CREATE SEQUENCE public.test_seq_1;
CREATE SEQUENCE
Time: 13.412 ms
postgres=# CREATE TABLE public.test_serial_default(a bigint default nextval('public.test_seq_1'::regclass), b int) DISTRIBUTE BY ROUNDROBIN;
CREATE TABLE
Time: 12.006 ms
postgres=# SELECT * FROM public.dfm_get_seqence_info;
 namespace |   sequencename    | min_value |      max_value      | start_value | increment_by | cache_value | is_cycled
-----------+-------------------+-----------+---------------------+-------------+--------------+-------------+-----------
 public    | test_seq          |         1 | 9223372036854775807 |           1 |            1 |           1 | f
 public    | test_serial_a_seq |         1 | 9223372036854775807 |           1 |            1 |           1 | f
 public    | test_seq_1        |         1 | 9223372036854775807 |           1 |            1 |           1 | f
(3 rows)

【序列和表的bind关系查询】

查询sequence和表的bind关系，主要场景有两个

1. 表的字段类型为bigserial或者serial类型，内置生成一个序列
2. 表的字段的默认值为序列的序列号

这两个场景在上述里面都有case

postgres=# SELECT
postgres-#     pg_get_userbyid(c1.relowner) AS seqowner,  -- sequence的用户
postgres-#     n1.nspname AS seqschema,  -- sequence的schema
postgres-#     c1.relname AS seqname, -- -- sequence的名称
postgres-#     (pg_sequence_parameters(c1.oid)).minimum_value AS minvalue, -- sequence的最小值
postgres-#     (pg_sequence_parameters(c1.oid)).maximum_value AS maxvalue, -- sequence的最大值
postgres-#     (pg_sequence_parameters(c1.oid)).increment AS incrementby, -- sequence的步长
postgres-#     pg_get_userbyid(c.relowner) AS tableowner, -- 关联的表的owner
postgres-#     n.nspname AS tableschema,  -- 关联的表的schema
postgres-#     c.relname AS tablename,-- 关联的表的名称
postgres-#     a.attname, -- -- 关联的表的列的名称
postgres-#     d.adsrc as defaultexpression -- 关联的sequence的表达式
postgres-# FROM pg_catalog.pg_class c
postgres-# INNER JOIN pg_catalog.pg_namespace n ON n.oid = c.relnamespace
postgres-# INNER JOIN pg_catalog.pg_attribute a ON a.attrelid = c.oid
postgres-# INNER JOIN pg_catalog.pg_attrdef d ON d.adrelid = c.oid AND d.adnum = a.attnum
postgres-# INNER JOIN pg_catalog.pg_depend p ON p.objid = d.oid
postgres-# INNER JOIN pg_catalog.pg_class c1 ON c1.oid = p.refobjid
postgres-# INNER JOIN pg_catalog.pg_namespace n1 ON n1.oid = c1.relnamespace
postgres-# WHERE c.relkind = 'r'::"char" AND c.oid > 16384
postgres-# AND n.nspname not in ('pg_toast', 'cstore', 'pg_catalog')
postgres-# AND a.attnum > 0
postgres-# AND p.classid='pg_catalog.pg_attrdef'::regclass
postgres-# AND refclassid = 'pg_catalog.pg_class'::regclass
postgres-# AND p.refobjsubid = 0
postgres-# AND c1.relkind = 'S'::"char"
postgres-# ;
 seqowner  | seqschema |      seqname      | minvalue |      maxvalue       | incrementby | tableowner | tableschema |      tablename      | attname |           defaultexpression
-----------+-----------+-------------------+----------+---------------------+-------------+------------+-------------+---------------------+---------+----------------------------------------
 j00565968 | public    | test_serial_a_seq |        1 | 9223372036854775807 |           1 | j00565968  | public      | test_serial         | a       | nextval('test_serial_a_seq'::regclass)
 j00565968 | public    | test_seq_1        |        1 | 9223372036854775807 |           1 | j00565968  | public      | test_serial_default | a       | nextval('test_seq_1'::regclass)
(2 rows)

【替换方案】

实际业务中使用sequence一般有两个场景

• 保证表数据均匀分布。这种场景可以把sequence字段替换为uuid

  postgres=# DROP TABLE public.test_serial;
  DROP TABLE
  Time: 19.871 ms
  postgres=# CREATE TABLE public.test_serial(a text DEFAULT sys_guid(), b int) DISTRIBUTE BY HASH(a);
  CREATE TABLE
  Time: 19.245 ms
  

  
  

  或者使用roundrobin分布替换，然后删除sequence相关列

  postgres=# ALTER TABLE public.test_serial  DISTRIBUTE BY ROUNDROBIN;
  ALTER TABLE
  Time: 218.796 ms
  postgres=# ALTER TABLE public.test_serial DROP COLUMN a;
  ALTER TABLE
  Time: 9.509 ms

• 保证字段数据的唯一性。这种场景可以使用uuid()+主键的方式实现

  postgres=# DROP TABLE public.test_serial;
  DROP TABLE
  Time: 18.786 ms
  postgres=# CREATE TABLE public.test_serial(a text DEFAULT sys_guid(), b int, primary key(a)) DISTRIBUTE BY HASH(a);
  NOTICE:  CREATE TABLE / PRIMARY KEY will create implicit index "test_serial_pkey" for table "test_serial"
  CREATE TABLE
  

  
  

  注：理论上一个集群内的uuid不会重复，但还是建议表上增加主键约束