Postgresql11分区索引

pg11分区索引使用总结

在 PostgreSQL 10 中，分区上的索引需要基于各个分区手动创建，而不能基于分区的父表创建索引。PostgreSQL 11 可以基于分区表创建索引。分区表上的索引并不会创建一个物理上的索引，而是为每个分区上的索引创建一个模板.

分区自动索引

如果在分区表上创建了一个索引，PostgreSQL自动为每个分区创建具有相同属性的索引，其主表本身并没有自身索引。

自动创建的索引，名称按照 “{partition name}_{column name}idx” 的模式定义。多个字段的复合索引使用下划线（）连接字段名称。如果索引名称已经存在，在名称的最后添加一个数字。如果名称过长，使用缩写。

CREATE TABLE measurement (
    city_id         int not null,
    logdate         date not null,
    peaktemp        int,
    unitsales       int
) PARTITION BY RANGE (logdate);

--建子分区measurement_y2018
CREATE TABLE measurement_y2018 PARTITION OF measurement FOR VALUES FROM ('2018-01-01') TO ('2019-01-01');
--建子分区measurement_y2019
CREATE TABLE measurement_y2019 PARTITION OF measurement FOR VALUES FROM ('2019-01-01') TO ('2020-01-01');

--在主表非分区列建索引
CREATE INDEX idx_measurement_peaktemp ON measurement(peaktemp);
CREATE INDEX idx_measurement_peaktemp_city_id ON measurement(city_id);
--在主表分区表分区列建索引
CREATE INDEX idx_measurement_peaktemp_1 ON measurement(logdate,peaktemp);
 --在子分区表分区列建索引
CREATE INDEX idx_measurement_peaktemp_y2018 ON measurement_y2018(logdate);
 --在子分区表非分区列建索引
CREATE INDEX idx_measurement_peaktemp_y2018_2 ON measurement_y2018(peaktemp);
  --在子分区表非分区列建复合索引
CREATE INDEX idx_measurement_peaktemp_y2018_1 ON measurement_y2018(logdate,peaktemp);

查询数据库索引如下：

postgres=# select * from pg_indexes where tablename like '%measurement%' order by tablename;
 schemaname |     tablename     |               indexname                | tablespace |                                                    indexdef

------------+-------------------+----------------------------------------+------------+----------------------------------------------------------------
-------------------------------------------------
 public     | measurement_y2018 | idx_measurement_peaktemp_y2018         |            | CREATE INDEX idx_measurement_peaktemp_y2018 ON public.measureme
nt_y2018 USING btree (logdate)
 public     | measurement_y2018 | idx_measurement_peaktemp_y2018_2       |            | CREATE INDEX idx_measurement_peaktemp_y2018_2 ON public.measure
ment_y2018 USING btree (peaktemp)
 public     | measurement_y2018 | idx_measurement_peaktemp_y2018_1       |            | CREATE INDEX idx_measurement_peaktemp_y2018_1 ON public.measure
ment_y2018 USING btree (logdate, peaktemp)
 public     | measurement_y2018 | measurement_y2018_logdate_peaktemp_idx |            | CREATE INDEX measurement_y2018_logdate_peaktemp_idx ON public.m
easurement_y2018 USING btree (logdate, peaktemp)
 public     | measurement_y2018 | measurement_y2018_peaktemp_idx         |            | CREATE INDEX measurement_y2018_peaktemp_idx ON public.measureme
nt_y2018 USING btree (peaktemp)
 public     | measurement_y2019 | measurement_y2019_logdate_peaktemp_idx |            | CREATE INDEX measurement_y2019_logdate_peaktemp_idx ON public.m
easurement_y2019 USING btree (logdate, peaktemp)
 public     | measurement_y2019 | measurement_y2019_peaktemp_idx         |            | CREATE INDEX measurement_y2019_peaktemp_idx ON public.measureme
nt_y2019 USING btree (peaktemp)
(7 rows)

以_idx结尾的为由主表为分区表自动创建的索引

随后新增的分区或者通过 ATTACH PARTITION 挂载的分区都会自动创建相应的索引。

CREATE TABLE measurement_y2020 (LIKE measurement);
ALTER TABLE measurement ATTACH PARTITION measurement_y2020
  FOR VALUES FROM ('2020-01-01') TO ('2021-01-01');

postgres=# \d measurement_y2020
           Table "public.measurement_y2020"
  Column   |  Type   | Collation | Nullable | Default
-----------+---------+-----------+----------+---------
 city_id   | integer |           | not null |
 logdate   | date    |           | not null |
 peaktemp  | integer |           |          |
 unitsales | integer |           |          |
Partition of: measurement FOR VALUES FROM ('2020-01-01') TO ('2021-01-01')
Indexes:
    "measurement_y2020_logdate_peaktemp_idx" btree (logdate, peaktemp)
    "measurement_y2020_peaktemp_idx" btree (peaktemp)

自动创建的索引不能单独删除，可以通过分区表统一删除。

postgres=# DROP INDEX measurement_y2020_peaktemp_idx;
ERROR:  cannot drop index measurement_y2020_peaktemp_idx because index idx_measurement_peaktemp requires it
HINT:  You can drop index idx_measurement_peaktemp instead.

分区表唯一约束

对于 PostgreSQL 10，只能基于分区创建唯一约束（PRIMARY KEY 和 UNIQUE KEY），而不能针对分区的父表创建唯一约束。PostgreSQL 11 支持分区表上的唯一约束。

CREATE TABLE rtable(c1 INT, c2 VARCHAR(10)) PARTITION BY RANGE(c1);
ALTER TABLE rtable ADD CONSTRAINT pk_rtable PRIMARY KEY(c1);
\d rtable
                      Table "public.rtable"
 Column |         Type          | Collation | Nullable | Default 
--------+-----------------------+-----------+----------+---------
 c1     | integer               |           | not null | 
 c2     | character varying(10) |           |          | 
Partition key: RANGE (c1)
Indexes:
    "pk_rtable" PRIMARY KEY, btree (c1)
Number of partitions: 0

添加分区或者加载（ATTACH）分区时自动创建相应的主键：

CREATE TABLE rtable100 PARTITION OF rtable FOR VALUES FROM (1) TO (100);
\d rtable100
                    Table "public.rtable100"
 Column |         Type          | Collation | Nullable | Default 
--------+-----------------------+-----------+----------+---------
 c1     | integer               |           | not null | 
 c2     | character varying(10) |           |          | 
Partition of: rtable FOR VALUES FROM (1) TO (100)
Indexes:
    "rtable100_pkey" PRIMARY KEY, btree (c1)

如果在分区表上创建了唯一约束，无法再创建基于外部表（FOREIGN TABLE）的分区。因为无法为外部表创建唯一约束。

CREATE FOREIGN TABLE rtable200 PARTITION OF rtable FOR VALUES FROM (101) TO (200) SERVER remote1;
ERROR:  cannot create index on foreign table "rtable200"

主键约束或唯一约束必须包含分区字段。这样才能确保整个分区表内的唯一性，因为每个分区上的唯一约束只维护自身的唯一性。

CREATE TABLE rtable1(c1 INT, c2 VARCHAR(10)) PARTITION BY RANGE(c1);
ALTER TABLE rtable1 ADD CONSTRAINT pk_table1 PRIMARY KEY(c2);
ERROR:  insufficient columns in PRIMARY KEY constraint definition
DETAIL:  PRIMARY KEY constraint on table "rtable1" lacks column "c1" which is part of the partition key.

分区表+索引查询分析

查询条件无分区键，全表扫描,各个子分区都会查询一遍

postgres=# explain select * from measurement where city_id =1;
                                            QUERY PLAN
--------------------------------------------------------------------------------------------------
 Append  (cost=4.22..44.43 rows=27 width=16)
   ->  Bitmap Heap Scan on measurement_y2018  (cost=4.22..14.76 rows=9 width=16)
         Recheck Cond: (city_id = 1)
         ->  Bitmap Index Scan on measurement_y2018_city_id_idx  (cost=0.00..4.22 rows=9 width=0)
               Index Cond: (city_id = 1)
   ->  Bitmap Heap Scan on measurement_y2019  (cost=4.22..14.76 rows=9 width=16)
         Recheck Cond: (city_id = 1)
         ->  Bitmap Index Scan on measurement_y2019_city_id_idx  (cost=0.00..4.22 rows=9 width=0)
               Index Cond: (city_id = 1)
   ->  Bitmap Heap Scan on measurement_y2020  (cost=4.22..14.76 rows=9 width=16)
         Recheck Cond: (city_id = 1)
         ->  Bitmap Index Scan on measurement_y2020_city_id_idx  (cost=0.00..4.22 rows=9 width=0)
               Index Cond: (city_id = 1)
(13 rows)

查询条件包括分区键(logdate)，先定位具体的分区子表，再在子表进行索引查询

postgres=# explain select * from measurement where city_id =1 and  logdate ='2019-03-01';
                                                   QUERY PLAN
-----------------------------------------------------------------------------------------------------------------
 Append  (cost=8.69..12.71 rows=1 width=16)
   ->  Bitmap Heap Scan on measurement_y2019  (cost=8.69..12.71 rows=1 width=16)
         Recheck Cond: ((city_id = 1) AND (logdate = '2019-03-01'::date))
         ->  BitmapAnd  (cost=8.69..8.69 rows=1 width=0)
               ->  Bitmap Index Scan on measurement_y2019_city_id_idx  (cost=0.00..4.22 rows=9 width=0)
                     Index Cond: (city_id = 1)
               ->  Bitmap Index Scan on measurement_y2019_logdate_peaktemp_idx  (cost=0.00..4.22 rows=9 width=0)
                     Index Cond: (logdate = '2019-03-01'::date)