openGauss每日一练第19天|统计信息、执行计划、垃圾收集操作

这节课我们学习收集统计信息、打印执行计划、垃圾收集和checkpoint，详细内容如下：

课程学习

1.准备数据

Create schema tpcds;
CREATE TABLE tpcds.customer_address
(
ca_address_sk integer NOT NULL ,
ca_address_id character(16),
ca_street_number character(10) ,
ca_street_name character varying(60) ,
ca_street_type character(15) ,
ca_suite_number character(10) ,
ca_city character varying(60) ,
ca_county character varying(30) ,
ca_state character(2) ,
ca_zip character(10) ,
ca_country character varying(20) ,
ca_gmt_offset numeric(5,2) ,
ca_location_type character(20)
);
insert into tpcds.customer_address values
(1, 'AAAAAAAABAAAAAAA', '18', 'Jackson', 'Parkway', 'Suite 280', 'Fairfield', 'Maricopa County', 'AZ', '86192' ,'United States', -7.00, 'condo'),
(2, 'AAAAAAAACAAAAAAA', '362', 'Washington 6th', 'RD', 'Suite 80', 'Fairview', 'Taos County', 'NM', '85709', 'United States', -7.00, 'condo'),
(3, 'AAAAAAAADAAAAAAA', '585', 'Dogwood Washington', 'Circle', 'Suite Q', 'Pleasant Valley', 'York County', 'PA', '12477', 'United States', -5.00, 'single family');

–使用序列的generate_series(1,N)函数对表插入数据

insert into tpcds.customer_address values(generate_series(10, 10000));

这一步插入进去的数据只有第一个sk字段，其他都是空的。

2.收集统计信息

–查看系统表中表的统计信息

select relname, relpages, reltuples from pg_class where relname = 'customer_address';

     relname      | relpages | reltuples 
------------------+----------+-----------
 customer_address |        0 |         0
(1 row)

—使用ANALYZE VERBOSE语句更新统计信息，并输出表的相关信息

analyze VERBOSE tpcds.customer_address;

INFO:  analyzing "tpcds.customer_address"(gaussdb pid=1)
INFO:  ANALYZE INFO : "customer_address": scanned 55 of 55 pages, containing 9994 live rows and 0 dead rows; 9994 rows in sample, 9994 estimated total rows(gaussdb pid=1)
ANALYZE

–查看系统表中表的统计信息

select relname, relpages, reltuples from pg_class where relname = 'customer_address';

     relname      | relpages | reltuples 
------------------+----------+-----------
 customer_address |       55 |      9994
(1 row)

这就说明了pg_class里面的统计信息是lazy load的，如果不analyze就一直是过期（或空）的数据信息。

3.打印执行计划

–使用默认的打印格式

SET explain_perf_mode=normal;

–显示表简单查询的执行计划

EXPLAIN SELECT * FROM tpcds.customer_address;

                              QUERY PLAN                               
-----------------------------------------------------------------------
 Seq Scan on customer_address  (cost=0.00..154.94 rows=9994 width=151)
(1 row)

–以JSON格式输出的执行计划（explain_perf_mode为normal时）

EXPLAIN(FORMAT JSON) SELECT * FROM tpcds.customer_address;

                 QUERY PLAN                 
--------------------------------------------
 [                                         +
   {                                       +
     "Plan": {                             +
       "Node Type": "Seq Scan",            +
       "Relation Name": "customer_address",+
       "Alias": "customer_address",        +
       "Startup Cost": 0.00,               +
       "Total Cost": 154.94,               +
       "Plan Rows": 9994,                  +
       "Plan Width": 151                   +
     }                                     +
   }                                       +
 ]
(1 row)

–禁止开销估计的执行计划

EXPLAIN(COSTS FALSE)SELECT * FROM tpcds.customer_address;

           QUERY PLAN          
------------------------------
 Seq Scan on customer_address
(1 row)

–带有聚集函数查询的执行计划

EXPLAIN SELECT SUM(ca_address_sk) FROM tpcds.customer_address WHERE ca_address_sk<100;

                               QUERY PLAN                                
-------------------------------------------------------------------------
 Aggregate  (cost=180.16..180.17 rows=1 width=12)
   ->  Seq Scan on customer_address  (cost=0.00..179.93 rows=94 width=4)
         Filter: (ca_address_sk < 100)
(3 rows)

–有索引条件的执行计划

create index customer_address_idx on tpcds.customer_address(ca_address_sk);
EXPLAIN SELECT * FROM tpcds.customer_address WHERE ca_address_sk<100;

                                           QUERY PLAN                                           
------------------------------------------------------------------------------------------------
 [Bypass]
 Index Scan using customer_address_idx on customer_address  (cost=0.00..9.90 rows=94 width=151)
   Index Cond: (ca_address_sk < 100)
(3 rows)

4.垃圾收集

–VACUUM回收表或B-Tree索引中已经删除的行所占据的存储空间

update tpcds.customer_address set ca_address_sk = ca_address_sk + 1 where ca_address_sk <100;

VACUUM (VERBOSE, ANALYZE) tpcds.customer_address;

INFO:  vacuuming "tpcds.customer_address"(gaussdb pid=1)
INFO:  index "customer_address_idx" now contains 10087 row versions in 31 pages(gaussdb pid=1)
DETAIL:  0 index row versions were removed.
0 index pages have been deleted, 0 are currently reusable.
CPU 0.00s/0.00u sec elapsed 0.00 sec.
INFO:  "customer_address": found 0 removable, 10087 nonremovable row versions in 55 out of 55 pages(gaussdb pid=1)
DETAIL:  93 dead row versions cannot be removed yet.
There were 0 unused item pointers.
0 pages are entirely empty.
CPU 0.00s/0.00u sec elapsed 0.00 sec.
INFO:  analyzing "tpcds.customer_address"(gaussdb pid=1)
INFO:  ANALYZE INFO : "customer_address": scanned 55 of 55 pages, containing 9994 live rows and 93 dead rows; 9994 rows in sample, 9994 estimated total rows(gaussdb pid=1)
VACUUM

5.事务日志检查点

–检查点（CHECKPOINT）是一个事务日志中的点，所有数据文件都在该点被更新以反映日志中的信息，所有数据文件都将被刷新到磁盘

CHECKPOINT;

6.清理数据

drop schema tpcds cascade;

课后作业

1.创建分区表，并用generate_series(1,N)函数对表插入数据

Create schema tpcds;
CREATE TABLE tpcds.products
(
sk int NOT NULL ,
id char(16),
description char(100)
);

insert into tpcds.products values
(1, 'a', 'aaa'),
(2, 'b', 'bbb'),
(3, 'c', 'ccc');

insert into tpcds.products values(generate_series(10, 10000));

2.收集表统计信息

omm=# analyze verbose tpcds.products;
INFO:  analyzing "tpcds.products"(gaussdb pid=1)
INFO:  ANALYZE INFO : "products": scanned 45 of 45 pages, containing 9994 live rows and 0 dead rows; 9994 rows in sample, 9994 estimated total rows(gaussdb pid=1)
ANALYZE
omm=# select relname, relpages, reltuples from pg_class where relname = 'products';
 relname  | relpages | reltuples 
----------+----------+-----------
 products |       45 |      9994
(1 row)

3.显示简单查询的执行计划；建立索引并显示有索引条件的执行计划

explain select * from tpcds.products;

                           QUERY PLAN                           
---------------------------------------------------------------
 Seq Scan on products  (cost=0.00..144.94 rows=9994 width=122)
(1 row)

create index products_sk_idx on tpcds.products(sk);
explain select * from tpcds.products WHERE sk<100;

                                    QUERY PLAN                                     
-----------------------------------------------------------------------------------
 [Bypass]
 Index Scan using products_sk_idx on products  (cost=0.00..9.90 rows=94 width=122)
   Index Cond: (sk < 100)
(3 rows)

4.更新表数据，并做垃圾收集

update tpcds.products set sk = sk + 1 where sk <100;

vacuum (verbose, analyze) tpcds.products;

INFO:  vacuuming "tpcds.products"(gaussdb pid=1)
INFO:  index "products_sk_idx" now contains 10087 row versions in 31 pages(gaussdb pid=1)
DETAIL:  0 index row versions were removed.
0 index pages have been deleted, 0 are currently reusable.
CPU 0.00s/0.00u sec elapsed 0.00 sec.
INFO:  "products": found 0 removable, 10087 nonremovable row versions in 45 out of 45 pages(gaussdb pid=1)
DETAIL:  93 dead row versions cannot be removed yet.
There were 0 unused item pointers.
0 pages are entirely empty.
CPU 0.00s/0.00u sec elapsed 0.00 sec.
INFO:  analyzing "tpcds.products"(gaussdb pid=1)
INFO:  ANALYZE INFO : "products": scanned 45 of 45 pages, containing 9994 live rows and 93 dead rows; 9994 rows in sample, 9994 estimated total rows(gaussdb pid=1)
VACUUM

5.清理数据

checkpoint;

drop schema tpcds cascade;