PostgreSQL主鍵索引膨脹的重建方法

普通的索引膨脹處理比較簡單，主鍵的索引膨脹也不復雜，只是在新舊索引交替時有一些小處理。本試驗在primary key上通過CONCURRENTLY建立第二索引來解決索引膨脹問題，適用9.3、9.4，其他版本使用前請實際測試。

創建測試表

                                    Table "swrd.mytbl"
 Column |            Type             |                     Modifiers
--------+-----------------------------+----------------------------------------------------
 id     | integer                     | not null default nextval('mytbl_id_seq'::regclass)
 val    | timestamp without time zone | default now()
Indexes:
    "mytbl_pkey" PRIMARY KEY, btree (id)

生成測試數據

生成測試數據步驟略，這里為了清楚看到測試的情況，生成10000000條。

swrd=# SELECT COUNT(*) FROM mytbl;
  count
----------
 10000000
(1 row)

創建第二索引

在id上創建第二索引，記得使用CONCURRENTLY參數

swrd=# CREATE UNIQUE INDEX CONCURRENTLY ON mytbl USING btree(id);
CREATE INDEX

可以看到id字段上同時有兩個索引mytbl_pkey和mytbl_id_idx

swrd=# SELECT schemaname,relname,indexrelname,pg_relation_size(indexrelid) AS index_size,idx_scan,idx_tup_read,idx_tup_fetch FROM pg_stat_user_indexes WHERE indexrelname IN (SELECT indexname FROM pg_indexes WHERE schemaname = 'swrd' AND tablename = 'mytbl');
 schemaname | relname | indexrelname | index_size | idx_scan | idx_tup_read | idx_tup_fetch
------------+---------+--------------+------------+----------+--------------+---------------
 swrd       | mytbl   | mytbl_pkey   |  224632832 |        0 |            0 |             0
 swrd       | mytbl   | mytbl_id_idx |  224641024 |        0 |            0 |             0
(2 rows)

替換索引

開啟事務刪除主鍵索引同時將第二索引更新為主鍵的約束

swrd=# begin;
BEGIN
swrd=# ALTER TABLE mytbl DROP CONSTRAINT mytbl_pkey;
ALTER TABLE
swrd=# ALTER TABLE mytbl ADD CONSTRAINT mytbl_id_idx PRIMARY KEY USING INDEX mytbl_id_idx;
ALTER TABLE
swrd=# END;
COMMIT

檢查測試表的索引，可見現在只有第二索引了

swrd=# SELECT schemaname,relname,indexrelname,pg_relation_size(indexrelid) AS index_size,idx_scan,idx_tup_read,idx_tup_fetch FROM pg_stat_user_indexes WHERE indexrelname IN (SELECT indexname FROM pg_indexes WHERE schemaname = 'swrd' AND tablename = 'mytbl');
 schemaname | relname | indexrelname | index_size | idx_scan | idx_tup_read | idx_tup_fetch
------------+---------+--------------+------------+----------+--------------+---------------
 swrd       | mytbl   | mytbl_id_idx |  224641024 |        0 |            0 |             0
(1 row)

檢查表的定義

檢查表定義，可以看到與最初建表時是一樣的

swrd=# \d+ mytbl
                                                        Table "swrd.mytbl"
 Column |            Type             |                     Modifiers                      | Storage | Stats target | Description
--------+-----------------------------+----------------------------------------------------+---------+--------------+-------------
 id     | integer                     | not null default nextval('mytbl_id_seq'::regclass) | plain   |              |
 val    | timestamp without time zone | default now()                                      | plain   |              |
Indexes:
    "mytbl_id_idx" PRIMARY KEY, btree (id)

參考：PostgreSQL主鍵索引膨脹的重建方法