PolarDB:Global indexes

Prerequisites

The feature is supported on a PolarDB for PostgreSQL cluster that runs the following engine:

PostgreSQL 14 (revision version 14.6.4.0 or later)

select version();

Limits

• You still can attach or detach partitions in partitioned tables where global indexes are created.

• To create a global index, add the GLOBAL keyword in the statement to create an index. If you do not add this keyword, a local index is created by default.

• You can concurrently create global indexes by adding the CONCURRENTLY keyword.

• You cannot create global indexes on non-partitioned tables or subpartitioned tables.

• Global indexes cannot be expression indexes.

• You cannot create global indexes on the partition key columns of partitioned tables.

Syntax

Create a global index.

CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] name ] ON [ ONLY ] table_name [ USING method ]
    ( { column_name | ( expression ) } [ COLLATE collation ] [ opclass [ ( opclass_parameter = value [, ... ] ) ] ] [ ASC | DESC ] [ NULLS { FIRST | LAST } ] [, ...] )
    [ INCLUDE ( column_name [, ...] ) ]
    [ WITH ( storage_parameter [= value] [, ... ] ) ]
[ GLOBAL/LOCAL ]
    [ TABLESPACE tablespace_name ]
    [ WHERE predicate ]

Examples

Accelerate non-partition key queries

1. Create a partitioned table with a time column as the partition key.

   CREATE TABLE partition_range (
       id INT,
       a INT,
       b INT,
       created_date TIMESTAMP WITHOUT TIME ZONE
   ) PARTITION BY RANGE (created_date);
   
   CREATE TABLE partition_range_part01 PARTITION OF partition_range FOR VALUES FROM (MINVALUE) TO ('2020-01-01 00:00:00');
   CREATE TABLE partition_range_part02 PARTITION OF partition_range FOR VALUES FROM ('2020-01-01 00:00:00') TO ('2020-02-01 00:00:00');
   CREATE TABLE partition_range_part03 PARTITION OF partition_range FOR VALUES FROM ('2020-02-01 00:00:00') TO ('2020-03-01 00:00:00');

2. Query the partition table based on non-partition key conditions.

   EXPLAIN (COSTS OFF) SELECT * FROM partition_range WHERE id = 1;

   The results indicate that all partitions are scanned and that the partition pruning feature cannot be used.

                           QUERY PLAN
   ------------------------------------------------------------
    Append
      ->  Seq Scan on partition_range_part01 partition_range_1
            Filter: (id = 1)
      ->  Seq Scan on partition_range_part02 partition_range_2
            Filter: (id = 1)
      ->  Seq Scan on partition_range_part03 partition_range_3
            Filter: (id = 1)
   (7 rows)

3. Create local indexes on the partitioned table and perform the query again.

   CREATE INDEX partition_range_idx_local ON partition_range(id);
   
   EXPLAIN (COSTS OFF) SELECT * FROM partition_range WHERE id = 1;

   The results indicate that the local indexes of all partitions are scanned because local indexes are created on individual partitions.

                                               QUERY PLAN
   --------------------------------------------------------------------------------------------------
    Append
      ->  Index Scan using partition_range_part01_id_idx on partition_range_part01 partition_range_1
            Index Cond: (id = 1)
      ->  Index Scan using partition_range_part02_id_idx on partition_range_part02 partition_range_2
            Index Cond: (id = 1)
      ->  Index Scan using partition_range_part03_id_idx on partition_range_part03 partition_range_3
            Index Cond: (id = 1)
   (7 rows)

4. Create a global index on the partitioned table by using the GLOBAL keyword and perform the query again.

   CREATE INDEX partition_range_idx_global ON partition_range(id) GLOBAL;
   
   EXPLAIN (COSTS OFF) SELECT * FROM partition_range WHERE id = 1;

   The results indicate that the global index is used to find the partition that contains the data.

                                 QUERY PLAN
   -----------------------------------------------------------------------
    Global Index Scan using partition_range_idx_global on partition_range
      Index Cond: (id = 1)
   (2 rows)

Unique constraint for non-partition keys

The partitioned table in the preceding example is still used. The partition key is created_date, but the unique constraint column is id.

CREATE UNIQUE INDEX partition_range_id_unique_idx ON partition_range(id);
ERROR:  unique constraint on partitioned table must include all partitioning columns
DETAIL:  UNIQUE constraint on table "partition_range" lacks column "created_date" which is part of the partition key.

If you add a unique constraint to a global index, such a restriction is removed.

CREATE UNIQUE INDEX partition_range_id_unique_idx ON partition_range(id) GLOBAL;

Performance test

Use pgbench to generate 80,000 rows of data, and create a partitioned table and a non-partitioned table.

Point query performance on non-partition keys

TPC-B performance on non-partition keys

Conclusions

Global indexes can exponentially improve performance for point queries and DML.