How to perform a major version upgrade for RDS for PostgreSQL - Amazon Relational Database Service

How to perform a major version upgrade for RDS for PostgreSQL

We recommend the following process when performing a major version upgrade on an Amazon RDS for PostgreSQL database:

  1. Have a version-compatible parameter group ready – If you are using a custom parameter group, you have two options. You can specify a default parameter group for the new DB engine version. Or you can create your own custom parameter group for the new DB engine version. For more information, see Parameter groups for Amazon RDS and Working with DB cluster parameter groups for Multi-AZ DB clusters.

  2. Check for unsupported database classes – Check that your database's instance class is compatible with the PostgreSQL version you are upgrading to. For more information, see Supported DB engines for DB instance classes.

  3. Check for unsupported usage:

    • Prepared transactions – Commit or roll back all open prepared transactions before attempting an upgrade.

      You can use the following query to verify that there are no open prepared transactions on your database. 

      SELECT count(*) FROM pg_catalog.pg_prepared_xacts;

    • Reg* data types – Remove all uses of the reg* data types before attempting an upgrade. Except for regtype and regclass, you can't upgrade the reg* data types. The pg_upgrade utility can't persist this data type, which is used by Amazon RDS to do the upgrade.

      To verify that there are no uses of unsupported reg* data types, use the following query for each database. 

      
SELECT count(*) FROM pg_catalog.pg_class c, pg_catalog.pg_namespace n, pg_catalog.pg_attribute a
  WHERE c.oid = a.attrelid
      AND NOT a.attisdropped
      AND a.atttypid IN ('pg_catalog.regproc'::pg_catalog.regtype,
                         'pg_catalog.regprocedure'::pg_catalog.regtype,
                         'pg_catalog.regoper'::pg_catalog.regtype,
                         'pg_catalog.regoperator'::pg_catalog.regtype,
                         'pg_catalog.regconfig'::pg_catalog.regtype,
                         'pg_catalog.regdictionary'::pg_catalog.regtype)
      AND c.relnamespace = n.oid
      AND n.nspname NOT IN ('pg_catalog', 'information_schema');

  4. Check for invalid databases:

    • Ensure there are no invalid databases. The datconnlimit column in the pg_database catalog includes a value of -2 to mark databases as invalid that were interrupted during a DROP DATABASE operation.

      Use the following query to check for invalid databases:

      SELECT datname FROM pg_database WHERE datconnlimit = - 2;

    • The previous query returns invalid database names. You can use DROP DATABASE invalid_db_name; to drop invalid databases. You can also use the following command to drop invalid databases:

      SELECT 'DROP DATABASE ' || quote_ident(datname) || ';' FROM pg_database WHERE datconnlimit = -2 \gexec

    For more information about invalid databases, see Understanding the behavior of autovacuum with invalid databases.

  5. Handle logical replication slots – An upgrade can't occur if the database has any logical replication slots. Logical replication slots are typically used for AWS DMS migration and for replicating tables from the database to data lakes, BI tools, and other targets. Before upgrading, make sure that you know the purpose of any logical replication slots that are in use, and confirm that it's okay to delete them. If the logical replication slots are still being used, you shouldn't delete them, and you can't proceed with the upgrade.

    If the logical replication slots aren't needed, you can delete them using the following SQL:

    SELECT * FROM pg_replication_slots WHERE slot_type NOT LIKE 'physical';
SELECT pg_drop_replication_slot(slot_name);

    Logical replication setups that use the pglogical extension also need to have slots dropped for a successful major version upgrade. For information about how to identify and drop slots created using the pglogical extension, see Managing logical replication slots for RDS for PostgreSQL.

  6. Handle read replicas – An upgrade of a Single-AZ DB instance or Multi-AZ DB instance deployment also upgrades the in-Region read replicas along with the primary DB instance. Amazon RDS doesn't upgrade Multi-AZ DB cluster read replicas.

    You can't upgrade read replicas separately. If you could, it could lead to situations where the primary and replica databases have different PostgreSQL major versions. However, read replica upgrades might increase downtime on the primary DB instance. To prevent a read replica upgrade, promote the replica to a standalone instance or delete it before starting the upgrade process.

    The upgrade process recreates the read replica's parameter group based on the read replica's current parameter group. You can apply a custom parameter group to a read replica only after the upgrade completes by modifying the read replica. For more information about read replicas, see Working with read replicas for Amazon RDS for PostgreSQL.

  7. Perform a backup – We recommend that you perform a backup before performing the major version upgrade so that you have a known restore point for your database. If your backup retention period is greater than 0, the upgrade process creates DB snapshots of your database before and after upgrading. To change your backup retention period, see Modifying an Amazon RDS DB instance and Modifying a Multi-AZ DB cluster for Amazon RDS.

    To perform a backup manually, see Creating a DB snapshot for a Single-AZ DB instance for Amazon RDS and Creating a Multi-AZ DB cluster snapshot for Amazon RDS.

  8. Upgrade certain extensions before a major version upgrade – If you plan to skip a major version with the upgrade, you need to update certain extensions before performing the major version upgrade. For example, upgrading from versions 9.5.x or 9.6.x to version 11.x skips a major version. The extensions to update include PostGIS and related extensions for processing spatial data.

    • address_standardizer

    • address_standardizer_data_us

    • postgis_raster

    • postgis_tiger_geocoder

    • postgis_topology

    You cannot directly upgrade to PostgreSQL version 17 if you are using rdkit version 4.6.0 and lower, and PostgreSQL version 16 and lower, due to rdkit incompatibility. Below are the upgrade options:

    • If you are on PostgreSQL version 13 and lower, you need to perform a major version upgrade to version 14.14 and higher 14 versions, 15.9 and higher 15 versions, or 16.5 and higher 16 versions first, and then perform the version upgrade to PostgreSQL 17.

    • If you are on PostgreSQL version 14, 15, or 16, you need to perform a minor version upgrade to 14.14 and higher 14 versions, 15.9 and higher 15 versions, or 16.5 and higher 16 versions, and then upgrade to PostgreSQL version 17.

    Run the following command for each extension that you're using:

    ALTER EXTENSION PostgreSQL-extension UPDATE TO 'new-version';

    For more information, see Upgrading PostgreSQL extensions in RDS for PostgreSQL databases. To learn more about upgrading PostGIS, see Step 6: Upgrade the PostGIS extension.

  9. Drop certain extensions before the major version upgrade – An upgrade that skips a major version to version 11.x doesn't support updating the pgRouting extension. Upgrading from versions 9.4.x, 9.5.x, or 9.6.x to versions 11.x skips a major version. It's safe to drop the pgRouting extension and then reinstall it to a compatible version after the upgrade. For the extension versions you can update to, see Supported PostgreSQL extension versions.

    The tsearch2 and chkpass extensions are no longer supported for PostgreSQL versions 11 or later. If you are upgrading to version 11.x, drop the tsearch2, and chkpass extensions before the upgrade.

  10. Drop unknown data types – Drop unknown data types depending on the target version.

    PostgreSQL version 10 stopped supporting the unknown data type. If a version 9.6 database uses the unknown data type, an upgrade to a version 10 shows an error message such as the following: 

    Database instance is in a state that cannot be upgraded: PreUpgrade checks failed:
The instance could not be upgraded because the 'unknown' data type is used in user tables.
Please remove all usages of the 'unknown' data type and try again."

    To find the unknown data type in your database so you can remove the offending column or change it to a supported data type, use the following SQL:

    SELECT DISTINCT data_type FROM information_schema.columns WHERE data_type ILIKE 'unknown';

  11. Perform an upgrade dry run – We highly recommend testing a major version upgrade on a duplicate of your production database before attempting the upgrade on your production database. You can monitor the execution plans on the duplicate test database for any possible execution plan regressions and to evaluate its performance. To create a duplicate test instance, you can either restore your database from a recent snapshot or do a point-in-time restore of your database to its latest restorable time.

    For more information, see Restoring from a snapshot or Restoring a DB instance to a specified time for Amazon RDS. For Multi-AZ DB clusters, see Restoring from a snapshot to a Multi-AZ DB cluster or Restoring a Multi-AZ DB cluster to a specified time.

    For details on performing the upgrade, see Manually upgrading the engine version.

    In upgrading a version 9.6 database to version 10, be aware that PostgreSQL 10 enables parallel queries by default. You can test the impact of parallelism before the upgrade by changing the max_parallel_workers_per_gather parameter on your test database to 2.

    Note

    The default value for max_parallel_workers_per_gather parameter in the default.postgresql10 DB parameter group is 2.

    For more information, see Parallel Query in the PostgreSQL documentation. To disable parallelism on version 10, set the max_parallel_workers_per_gather parameter to 0.

    During the major version upgrade, the public and template1 databases and the public schema in every database are temporarily renamed. These objects appear in the logs with their original name and a random string appended. The string is appended so that custom settings such as locale and owner are preserved during the major version upgrade. After the upgrade completes, the objects are renamed back to their original names.

    Note

    During the major version upgrade process, you can't do a point-in-time restore of your DB instance or Multi-AZ DB cluster. After Amazon RDS performs the upgrade, it takes an automatic backup of the database. You can perform a point-in-time restore to times before the upgrade began and after the automatic backup of your database has completed.

  12. If an upgrade fails with precheck procedure errors, resolve the issues – During the major version upgrade process, Amazon RDS for PostgreSQL first runs a precheck procedure to identify any issues that might cause the upgrade to fail. The precheck procedure checks all potential incompatible conditions across all databases in the instance.

    If the precheck encounters an issue, it creates a log event indicating the upgrade precheck failed. The precheck process details are in an upgrade log named pg_upgrade_precheck.log for all the databases of a database. Amazon RDS appends a timestamp to the file name. For more information about viewing logs, see Monitoring Amazon RDS log files.

    If a read replica upgrade fails at precheck, replication on the failed read replica is broken and the read replica is put in the terminated state. Delete the read replica and recreate a new read replica based on the upgraded primary DB instance.

    Resolve all of the issues identified in the precheck log and then retry the major version upgrade. The following is an example of a precheck log.

    ------------------------------------------------------------------------
Upgrade could not be run on Wed Apr 4 18:30:52 2018
-------------------------------------------------------------------------
The instance could not be upgraded from 9.6.11 to 10.6 for the following reasons.
Please take appropriate action on databases that have usage incompatible with the requested major engine version upgrade and try the upgrade again.

* There are uncommitted prepared transactions. Please commit or rollback all prepared transactions.* One or more role names start with 'pg_'. Rename all role names that start with 'pg_'.

* The following issues in the database 'my"million$"db' need to be corrected before upgrading:** The ["line","reg*"] data types are used in user tables. Remove all usage of these data types.
** The database name contains characters that are not supported by RDS for PostgreSQL. Rename the database.
** The database has extensions installed that are not supported on the target database version. Drop the following extensions from your database: ["tsearch2"].

* The following issues in the database 'mydb' need to be corrected before upgrading:** The database has views or materialized views that depend on 'pg_stat_activity'. Drop the views.

  13. If a read replica upgrade fails while upgrading the database, resolve the issue – A failed read replica is placed in the incompatible-restore state and replication is terminated on the database. Delete the read replica and recreate a new read replica based on the upgraded primary DB instance.

    Note

    Amazon RDS doesn't upgrade read replicas for Multi-AZ DB clusters. If you perform a major version upgrade on a Multi-AZ DB cluster, then the replication state of its read replicas changes to terminated.

    A read replica upgrade might fail for the following reasons:

    • It was unable to catch up with the primary DB instance even after a wait time.

    • It was in a terminal or incompatible lifecycle state such as storage-full, incompatible-restore, and so on.

    • When the primary DB instance upgrade started, there was a separate minor version upgrade running on the read replica.

    • The read replica used incompatible parameters.

    • The read replica was unable to communicate with the primary DB instance to synchronize the data folder.

  14. Upgrade your production database – When the dry-run major version upgrade is successful, you should be able to upgrade your production database with confidence. For more information, see Manually upgrading the engine version.

  15. Run the ANALYZE operation to refresh the pg_statistic table. You should do this for every database on all your PostgreSQL databases. Optimizer statistics aren't transferred during a major version upgrade, so you need to regenerate all statistics to avoid performance issues. Run the command without any parameters to generate statistics for all regular tables in the current database, as follows:

    ANALYZE VERBOSE;

    The VERBOSE flag is optional, but using it shows you the progress. For more information, see ANALYZE in the PostgreSQL documentation.

    Note

    Run ANALYZE on your system after the upgrade to avoid performance issues.

After the major version upgrade is complete, we recommend the following:

  • A PostgreSQL upgrade doesn't upgrade any PostgreSQL extensions. To upgrade extensions, see Upgrading PostgreSQL extensions in RDS for PostgreSQL databases.

  • Optionally, use Amazon RDS to view two logs that the pg_upgrade utility produces. These are pg_upgrade_internal.log and pg_upgrade_server.log. Amazon RDS appends a timestamp to the file name for these logs. You can view these logs as you can any other log. For more information, see Monitoring Amazon RDS log files.

    You can also upload the upgrade logs to Amazon CloudWatch Logs. For more information, see Publishing PostgreSQL logs to Amazon CloudWatch Logs.

  • To verify that everything works as expected, test your application on the upgraded database with a similar workload. After the upgrade is verified, you can delete this test instance.