VACUUM
Function
VACUUM reclaims storage space occupied by tables or B-tree indexes. In normal database operation, rows that have been deleted are not physically removed from their table; they remain present until a VACUUM is done. Therefore, it is necessary to execute VACUUM periodically, especially on frequently-updated tables.
Precautions
- With no table specified, VACUUM processes all the tables that the current user has permission to vacuum in the current database. With a table specified, VACUUM processes only that table.
- To perform VACUUM on a table, you must be the owner of the table, granted with the VACUUM permission on the table, or granted with the gs_role_vacuum_any role. By default, a system administrator has this permission. However, database owners are allowed to VACUUM all tables in their databases, except shared catalogs. (The restriction for shared catalogs means that a true database-wide VACUUM can only be executed by the system administrator). VACUUM skips over any tables that the calling user does not have the permission to vacuum.
- VACUUM cannot be executed inside a transaction block.
- It is recommended that active production databases be vacuumed frequently (at least nightly), in order to remove dead rows. After adding or deleting a large number of rows, it might be a good idea to execute the VACUUM ANALYZE command for the affected table. This will update the system catalogs with the results of all recent changes, and allow the query planner to make better choices in planning queries.
- FULL is recommended only in special scenarios. For example, you wish to physically narrow the table to decrease the occupied disk space after deleting most rows of a table. VACUUM FULL usually shrinks more table size than VACUUM. If the physical space usage does not decrease after you run the command, check whether there are other active transactions (that have started before you delete data transactions and not ended before you run VACUUM FULL). If there are such transactions, run this command again when the transactions quit.
- VACUUM causes a substantial increase in I/O traffic, which might cause poor performance for other active sessions. Therefore, it is sometimes advisable to use the cost-based VACUUM delay feature.
- When VERBOSE is specified, VACUUM prints progress messages to indicate which table is currently being processed. Various statistics about the tables are printed as well.
- When the option list is surrounded by parentheses, the options can be written in any order. If there are no brackets, the options must be given in the order displayed in the syntax.
- VACUUM and VACUUM FULL clear deleted tuples after the delay specified by vacuum_defer_cleanup_age.
- VACUUM ANALYZE executes a VACUUM operation and then an ANALYZE operation for each selected table. This is a handy combination form for routine maintenance scripts.
- Plain VACUUM (without FULL) recycles space and makes it available for reuse. This form of the command can operate in parallel with normal reading and writing of the table, as an exclusive lock is not obtained. VACUUM FULL executes wider processing, including moving rows across blocks to compress tables so they occupy minimum number of disk blocks. This form is much slower and requires an exclusive lock on each table while it is being processed.
- During VACUUM on column-store tables, four internal operations occur: VACUUM on the primary table, VACUUM on the Desc table of the primary table, VACUUM on the delta table of the primary table, and migration of data from the delta table to the primary table. VACUUM does not reclaim the storage space of the delta table. To reclaim it, do VACUUM DELTAMERGE to the column-store table. By default, the VACUUM primary table operation is enabled. You can disable this function by setting the colvacuum_threshold_scale_factor parameter.
- The VACUUM operation on column-store primary tables does not support temporary tables, multi-temperature tables, and time series tables.
- The VACUUM operation on column-store primary tables reclaims space with a delay. To reclaim space immediately, run the vac_fileclear_relation function after VACUUM is executed. Then, an exclusive lock is added to a specified table to reclaim its space.
- Running VACUUM FULL on system catalogs can only be done when the database is offline. Otherwise, table locks, exceptions, and errors may occur.
- If you perform VACUUM FULL when a long-running query accesses a system table, the long-running query may prevent VACUUM FULL from accessing the system table. As a result, the connection times out and an error is reported.
- Running VACUUM FULL on a column-store partitioned table locks the table and its partitions.
- A distributed deadlock may occur when VACUUM FULL and DML statements run concurrently in the following scenarios:
- VACUUM FULL on subpartitions and INSERT/UPDATE/DELETE on the primary table
- VACUUM FULL on a full table and SELECT on a full table or SELECT on subpartitions
- When a VACUUM FULL operation is performed on a table, it triggers table rebuilding. During this rebuilding process, data is dumped into a new data file. Once the rebuilding is complete, the original file is deleted. However, it is important to note that if the table is large, the rebuilding process can consume a large disk space. When the disk space is insufficient, performing the VACUUM FULL operation on large tables may cause the cluster to become read-only.
- In the storage-compute decoupling architecture, a message is displayed indicating that the full database VACCUM, full database VACUUM FULL, and full database VACUUM DELTAMERGE are not supported.
Syntax
- Reclaim space and update statistics. The keyword sequence must be given in the order in which the syntax is displayed.
1 2
VACUUM [ ( { FULL | FREEZE | VERBOSE | {ANALYZE | ANALYSE }} [,...] ) ] [ table_name [ (column_name [, ...] ) ] ] [ PARTITION ( partition_name ) ];
- Reclaim space, without updating statistics information.
1
VACUUM [ FULL [COMPACT] ] [ FREEZE ] [ VERBOSE ] [ table_name ] [ PARTITION ( partition_name ) ];
- Reclaim space and update statistics information, with a specific order of keywords required.
1 2
VACUUM [ FULL ] [ FREEZE ] [ VERBOSE ] { ANALYZE | ANALYSE } [ VERBOSE ] [ table_name [ (column_name [, ...] ) ] ] [ PARTITION ( partition_name ) ];
- For HDFS and column-store tables, migrate data from the delta table to the primary table. The partition_name parameter is supported only by clusters of version 8.2.1.300 or later.
1
VACUUM DELTAMERGE [ table_name ][partition_name];
- For HDFS tables, delete the empty value partition directory of HDFS table in HDFS storage.
1
VACUUM HDFSDIRECTORY [ table_name ];
Parameter Description
- FULL
Selects "FULL" vacuum, which can reclaim more space, but takes much longer and exclusively locks the table.
FULL options can also contain the COMPACT parameter, which is only used for the HDFS table. Specifying the COMPACT parameter improves VACUUM FULL operation performance.
COMPACT and PARTITION cannot be used at the same time.
Using FULL will cause statistics missing. To collect statistics, add the keyword ANALYZE to VACUUM FULL.
- FREEZE
Is equivalent to executing VACUUM with the vacuum_freeze_min_age parameter set to zero.
- VERBOSE
Prints a detailed vacuum activity report for each table.
- ANALYZE | ANALYSE
Updates statistics used by the planner to determine the most efficient way to execute a query.
- table_name
Indicates the name (optionally schema-qualified) of a specific table to vacuum.
Value range: The name of a specific table to vacuum. Defaults are all tables in the current database.
- column_name
Indicates the name of a specific field to analyze.
Value range: Indicates the name of a specific field to analyze. Defaults are all columns.
- PARTITION
HDFS table does not support PARTITION. COMPACT and PARTITION cannot be used at the same time.
If the PARTITION and COMPACT parameters are used at the same time, the following error message is displayed: COMPACT can not be used with PARTITION.
- partition_name
Indicates the partition name of a specific table to vacuum. Defaults are all partitions.
- DELTAMERGE
For HDFS tables and column-store tables, migrate data in the delta table of HDFS tables or column-store tables to the primary table. If the data volume of the delta table is less than 60,000 rows, the data will not be migrated. Otherwise, the data will be migrated to HDFS, and the delta table will be cleared by TRUNCATE. For a column-store table, this operation always transfers all data in the delta table to the CU.
The following DFX functions are provided to return the data storage in the delta table of a column-store table (for an HDFS table, it can be returned by EXPLAIN ANALYZE):
- pgxc_get_delta_info(TEXT): The input parameter is a column-store table name. The delta table information on each node is collected and displayed, including the number of active tuples, table size, and maximum block ID.
- get_delta_info(TEXT): The input parameter is a column-store table name. The system summarizes the results returned from pgxc_get_delta_info and returns the total number of active tuples, total table size, and maximum block ID in the delta table. When querying delta information about a temporary table, you need to specify the schema of the temporary table. Otherwise, an error is reported, indicating that the table cannot be found.
- HDFSDIRECTORY
Deletes the empty value partition directory of HDFS table in HDFS storage for HDFS table.
Examples
Delete all tables in the current database.
1
|
VACUUM; |
Reclaim the space of partition P2 of the tpcds.web_returns_p1 table without updating statistics.
1
|
VACUUM FULL tpcds.web_returns_p1 PARTITION(P2); |
Reclaim the space of the tpcds.web_returns_p1 table and update statistics.
1
|
VACUUM FULL ANALYZE tpcds.web_returns_p1; |
Delete all tables in the current database and collect statistics about the query optimizer.
1
|
VACUUM ANALYZE; |
Delete only the reason table.
1
|
VACUUM (VERBOSE, ANALYZE) tpcds.reason; |
Perform the DELTAMERGE operation on the column-store table table_delta.
1
|
VACUUM DELTAMERGE tpcds.table_delta; |
Perform the DELTAMERGE operation only on partition p1 of the column-store table table_delta.
1
|
VACUUM DELTAMERGE tpcds.table_delta partition(p1); |
Feedback
Was this page helpful?
Provide feedbackThank you very much for your feedback. We will continue working to improve the documentation.