Creating and Managing Indexes

Context

Indexes accelerate data access but increase the processing time of insertion, update, and deletion operations. Therefore, before creating an index, consider whether it is necessary and select the columns where indexes are to be created. You can determine whether to create an index for a table by analyzing the service processing and data use of applications, as well as columns that are frequently used as search criteria or need to be collated.

Indexes are created based on columns in database tables. Therefore, you must correctly identify which columns require indexes. You are advised to create indexes for any of the following columns:

• Columns that are often searched and queried. This speeds up searches.
• Columns that function as primary keys. This enforces the uniqueness of the columns and the data collation structures in organized tables.
• Columns that are often joined. This increases the join efficiency.
• Columns that are often searched by range. The index helps collate data, and therefore the specified ranges are contiguous.
• Columns that often need to be collated. The index helps collate data, reducing the time for a collation query.
• Columns where the WHERE clause is executed frequently. This speeds up condition judgment.
• Columns that often appear after the keywords ORDER BY, GROUP BY, and DISTINCT.
  

  • After an index is created, the system automatically determines when to reference it. If the system determines that indexing is faster than sequential scan, the index will be used.
  • After an index is successfully created, it must be synchronized with the associated table to ensure new data can be accurately located, which increases the data operation load. Therefore, delete unnecessary indexes periodically.
  • Partitioned table indexes are classified into local indexes and global indexes. A local index corresponds to a specific partition, and a global index corresponds to the entire partitioned table.

Procedure

To create a partitioned table, see Creating and Managing Partitioned Tables.

• Create an index.
  • Create the partitioned table index tpcds_web_returns_p2_index1 without specifying the partition name.

    1	gaussdb=# CREATE INDEX tpcds_web_returns_p2_index1 ON tpcds.web_returns_p2 (ca_address_id) LOCAL;

    If the following information is displayed, the creation is successful:

    1	CREATE INDEX

  • Create the partitioned table index tpcds_web_returns_p2_index2 with the partition name specified.

    1 2 3 4 5 6 7 8 9 10 11

    gaussdb=# CREATE INDEX tpcds_web_returns_p2_index2 ON tpcds.web_returns_p2 (ca_address_sk) LOCAL ( PARTITION web_returns_p2_P1_index, PARTITION web_returns_p2_P2_index TABLESPACE example3, PARTITION web_returns_p2_P3_index TABLESPACE example4, PARTITION web_returns_p2_P4_index, PARTITION web_returns_p2_P5_index, PARTITION web_returns_p2_P6_index, PARTITION web_returns_p2_P7_index, PARTITION web_returns_p2_P8_index ) TABLESPACE example2;

    If the following information is displayed, the creation is successful:

    1	CREATE INDEX

• Change the tablespace of an index partition.
  • Change the tablespace of index partition web_returns_p2_P2_index to example1.

    1	gaussdb=# ALTER INDEX tpcds.tpcds_web_returns_p2_index2 MOVE PARTITION web_returns_p2_P2_index TABLESPACE example1;

    If the following information is displayed, the modification is successful:

    1	ALTER INDEX

  • Change the tablespace of index partition web_returns_p2_P3_index to example2.

    1	gaussdb=# ALTER INDEX tpcds.tpcds_web_returns_p2_index2 MOVE PARTITION web_returns_p2_P3_index TABLESPACE example2;

    If the following information is displayed, the modification is successful:

    1	ALTER INDEX

• Rename an index partition.
  Rename the index partition web_returns_p2_P8_index to web_returns_p2_P8_index_new.

  1	gaussdb=# ALTER INDEX tpcds.tpcds_web_returns_p2_index2 RENAME PARTITION web_returns_p2_P8_index TO web_returns_p2_P8_index_new;

  If the following information is displayed, the rename operation is successful:

  1	ALTER INDEX

• Query indexes.
  • Query all indexes defined by the system and users.

    1	gaussdb=# SELECT RELNAME FROM PG_CLASS WHERE RELKIND='i';

  • Query information about a specified index.

    1	gaussdb=# \di+ tpcds.tpcds_web_returns_p2_index2

• Delete indexes.

  1 2	gaussdb=# DROP INDEX tpcds.tpcds_web_returns_p2_index1; gaussdb=# DROP INDEX tpcds.tpcds_web_returns_p2_index2;

  If the following information is displayed, the deletion is successful:

  1	DROP INDEX

GaussDB supports four methods for creating indexes. For details, see Table 1.

• After an index is created, the system automatically determines when to reference it. If the system determines that indexing is faster than sequential scan, the index will be used.

• After an index is successfully created, it must be synchronized with the associated table to ensure new data can be accurately located, which increases the data operation load. Therefore, delete unnecessary indexes periodically.

• Create an ordinary table.

  1 2	gaussdb=# CREATE TABLE tpcds.customer_address_bak AS TABLE tpcds.customer_address; INSERT 0 0

• Create an ordinary index.
  For the tpcds.customer_address_bak table, you need to perform the following operations frequently:

  1	gaussdb=# SELECT ca_address_sk FROM tpcds.customer_address_bak WHERE ca_address_sk=14888;

  Generally, the database system needs to scan the tpcds.customer_address_bak table row by row to find all matched tuples. If the size of the tpcds.customer_address_bak table is large but only a few (possibly zero or one) of the WHERE conditions are met, the performance of this sequential scan is low. If the database system uses an index to maintain the ca_address_sk attribute, the database system only needs to search a few tree layers for the matched tuples. This greatly improves data query performance. Furthermore, indexes can improve the update and deletion operation performance in the database.

  Create an index.

  1 2	gaussdb=# CREATE INDEX index_wr_returned_date_sk ON tpcds.customer_address_bak (ca_address_sk); CREATE INDEX

• Create a unique index.

  Create a unique index on the SM_SHIP_MODE_SK column in the tpcds.ship_mode_t1 table.

  gaussdb=# CREATE UNIQUE INDEX ds_ship_mode_t1_index1 ON tpcds.ship_mode_t1(SM_SHIP_MODE_SK);

• Create a composite index.
  Frequently query records with ca_address_sk being 5050 and ca_street_number smaller than 1000 in the tpcds.customer_address_bak table.

  1	gaussdb=# SELECT ca_address_sk,ca_address_id FROM tpcds.customer_address_bak WHERE ca_address_sk = 5050 AND ca_street_number < 1000;

  Define a composite index on ca_address_sk and ca_street_number columns.

  1 2	gaussdb=# CREATE INDEX more_column_index ON tpcds.customer_address_bak(ca_address_sk ,ca_street_number ); CREATE INDEX

• Create a partial index.

  If you only want to find records with ca_address_sk being 5050, you can create a partial index to facilitate your query.

  1 2	gaussdb=# CREATE INDEX part_index ON tpcds.customer_address_bak(ca_address_sk) WHERE ca_address_sk = 5050; CREATE INDEX

• Create an expression index.
  Frequently query records with ca_street_number smaller than 1000.

  1	gaussdb=# SELECT * FROM tpcds.customer_address_bak WHERE trunc(ca_street_number) < 1000;

  The following expression index can be created for this query task:

  1 2	gaussdb=# CREATE INDEX para_index ON tpcds.customer_address_bak (trunc(ca_street_number)); CREATE INDEX

• Delete the tpcds.customer_address_bak table.

  1 2	gaussdb=# DROP TABLE tpcds.customer_address_bak; DROP TABLE