postgresql数据库表膨胀之pg_repack安装及使用

pg_repack是一个可以在线重建表和索引的扩展。它会在数据库中建立一个和需要清理的目标表一样的临时表，将目标表中的数据COPY到临时表，并在临时表上建立与目标表一样的索引，然后通过重命名的方式用临时表替换目标表。

环境：

1）redhat-release：CentOS Linux release 7.6.1810 (Core)

2）database version：postgresql 14.6

一、部署pg_repack

--解压编译

[postgres@tencent soft]$ unzip pg_repack-master.zip

[postgres@tencent soft]$ cd pg_repack-master/

[postgres@tencent pg_repack-master]$ ll

total 36

drwxrwxr-x 3 postgres postgres 4096 Jun 15 04:53 bin

-rw-rw-r-- 1 postgres postgres 1662 Jun 15 04:53 COPYRIGHT

drwxrwxr-x 2 postgres postgres 4096 Jun 15 04:53 doc

drwxrwxr-x 3 postgres postgres 4096 Jun 15 04:53 lib

-rw-rw-r-- 1 postgres postgres 1616 Jun 15 04:53 Makefile

-rw-rw-r-- 1 postgres postgres 1381 Jun 15 04:53 META.json

drwxrwxr-x 2 postgres postgres 4096 Jun 15 04:53 msvc

-rw-rw-r-- 1 postgres postgres 2191 Jun 15 04:53 README.rst

drwxrwxr-x 4 postgres postgres 4096 Jun 15 04:53 regress

[postgres@tencent pg_repack-master]$ make &&make install

--所在数据库安装pg_repack扩展

[postgres@tencent pg_repack-master]$ psql -d testdb test

psql (14.6)

Type "help" for help.

testdb=# select * from pg_available_extensions where name like 'pg_repack*';

name | default_version | installed_version | comment

------+-----------------+-------------------+---------

(0 rows)

testdb=# create extension pg_repack;

CREATE EXTENSION

testdb=# \dx

List of installed extensions

Name | Version | Schema | Description

-------+-------+----------+---------------------------------------

btree_gist | 1.6 | test | support for indexing common datatypes in GiST

pageinspect | 1.9 | public | inspect the contents of database pages at a low level

pg_freespacemap | 1.2 | test | examine the free space map (FSM)

pg_repack | 1.4.8 | test | Reorganize tables in PostgreSQL databases with minimal locks

pg_visibility | 1.2 | test | examine the visibility map (VM) and page-level visibility info

pgstattuple | 1.5 | test | show tuple-level statistics

plpgsql | 1.0 | pg_catalog | PL/pgSQL procedural language

(7 rows)

--查看pg_repack包含的函数

testdb=# \dxS+ pg_repack

Objects in extension "pg_repack"

Object description

------------------------------------------------------------------

function repack.conflicted_triggers(oid)

function repack.disable_autovacuum(regclass)

function repack.get_alter_col_storage(oid)

function repack.get_assign(oid,text)

function repack.get_columns_for_create_as(oid)

function repack.get_compare_pkey(oid,text)

function repack.get_create_index_type(oid,name)

function repack.get_create_trigger(oid,oid)

function repack.get_drop_columns(oid,text)

function repack.get_enable_trigger(oid)

function repack.get_index_columns(oid,text)

function repack.get_order_by(oid,oid)

function repack.get_storage_param(oid)

function repack.get_table_and_inheritors(regclass)

function repack.oid2text(oid)

function repack.repack_apply(cstring,cstring,cstring,cstring,cstring,integer)

function repack.repack_drop(oid,integer)

function repack.repack_indexdef(oid,oid,name,boolean)

function repack.repack_index_swap(oid)

function repack.repack_swap(oid)

function repack.repack_trigger()

function repack.version()

function repack.version_sql()

schema repack

view repack.primary_keys

view repack.tables

(26 rows)

二、安装pgstattuple模块

（提供多种函数来获得元组层的统计信息）

testdb=# create extension pgstattuple;

CREATE EXTENSION

三、创建测试环境

3.1 创建表（表必须有主键或者唯一约束）

testdb=# create table repack_test(id int primary key ,name varchar(10));

CREATE TABLE

testdb=# \d repack_test

Table "test.repack_test"

Column | Type | Collation | Nullable | Default

--------+-----------------------+-----------+----------+---------

id | integer | | not null |

name | character varying(10) | | |

Indexes:

"repack_test_pkey" PRIMARY KEY, btree (id)

3.2 插入数据

testdb=# insert into repack_test select generate_series(1,50000000),'a';

INSERT 0 50000000

--用pg_stattuple查看表情况

testdb=# select * from pgstattuple('repack_test');

------------+-------------+------------+---------------+------------------+----------------+--------------------+------------+--------------

1812389888 | 50000000 | 1500000000 | 82.76 | 0 | 0 | 0 | 6195196 | 0.34

(1 row)

--查看表大小

testdb=# select pg_size_pretty(pg_relation_size('repack_test'));

pg_size_pretty

----------------

1728 MB

(1 row)

3.3 修改表数据

--update 表repack_test

testdb=# update repack_test set name ='b' where id<2500000;

UPDATE 2499999

--用pg_stattuple查看表情况

testdb=# select * from pgstattuple('repack_test');

------------+-------------+------------+---------------+------------------+----------------+--------------------+------------+--------------

1903009792 | 50000000 | 1500000000 | 78.82 | 2499999 | 74999970 | 3.94 | 6505400 | 0.34

(1 row)

可以发现产生了2499999条死亡元组。

--查看表大小

testdb=# select pg_size_pretty(pg_relation_size('repack_test'));

pg_size_pretty

----------------

1815 MB

(1 row)

四、执行pg_repack

--执行pg_repack

[postgres@tencent bin]$ ./pg_repack -p 4519 -d testdb -Utest --no-order --table test.repack_test

INFO: repacking table "test.repack_test"

# test.repack_test 表示test模式下repack_test表，默认自动查找postgres数据库下表repack_test，执行过程中会报没有发现表，所以需要指定schema。

另，此处执行没有输入密码，因为配置了.pgpass文件

--用pg_stattuple查看表情况

testdb=# select * from pgstattuple('repack_test');

------------+-------------+------------+---------------+------------------+----------------+--------------------+------------+--------------

1812389888 | 50000000 | 1500000000 | 82.76 | 0 | 0 | 0 | 6195196 | 0.34

(1 row)

当前死亡元组数归零。

--查看表大小

testdb=# select pg_size_pretty(pg_relation_size('repack_test'));

pg_size_pretty

----------------

1728 MB

(1 row)

表数据恢复原始大小。

五、pg_repack主要模式

repack数据库：

postgres@node1 contrib]$ pg_repack -p 5432 -d postgres --no-order --jobs 4 --elevel=info

repack模式：

pg_repack -p 5432 -d postgres --schema=public --no-order --jobs 4 --elevel=info

repack表和索引：

pg_repack -p 5432 -d postgres --no-order --table public.repack_test --elevel=info

repack所有索引：

pg_repack -p 5432 -d postgres --no-order --only-indexes --table public.repack_test --elevel=info

repack指定索引：

pg_repack -p 5432 -d postgres --index public.repack_test_pkey --elevel=info