E.20. Postgres Pro Enterprise 13.2.1

Release Date: 2021-04-21

E.20.1. Overview

This release is based on PostgreSQL 13.2 and includes all the new features introduced in PostgreSQL 13, as well as bug fixes implemented in PostgreSQL 13.1 and 13.2 updates. For their detailed description, see PostgreSQL 13, PostgreSQL 13.1, and PostgreSQL 13.2 Release Notes, respectively.

For the list of extension modules and utilities specific to Postgres Pro Enterprise, as well as the main user-visible core changes over vanilla PostgreSQL, see Section 2. As compared to Postgres Pro Enterprise 12.6.1, the following differences are worth mentioning:

  • Implemented the infrastructure for online upgrade to higher minor versions. This is intended to upgrade the installed server to a higher minor version without a server restart, and thus without interrupting the existing client connections. For now, this infrastructure is provided on Linux platforms, except for Astra Linux Smolensk 1.5 and ALT Linux SPT 7.0. To prepare for online upgrade, you should create your database cluster using pg-setup with the initdb --enable-online-upgrade arguments.

  • Improved the multimaster extension and made it open source. Now it uses the Paxos consensus algorithm and two-phase commit protocol with generations to determine the transaction outcome. It allows you to use not only the 2+1 cluster configuration, which was the only recommended and robust multimaster configuration before.

  • Added support for lz4 compression library to CFS.

  • Added the pgpro_stat_wal_activity view that shows the size of WAL generated by each server process.

Important

Starting from Postgres Pro 12, using pg_pathman is not recommended. Use vanilla declarative partitioning instead, as described in Section 5.11.

E.20.2. Migration to Version 13

You can migrate to Postgres Pro Enterprise 13 from the same or a previous version of PostgreSQL (that is supported by the upgrade method chosen) or Postgres Pro Standard/Postgres Pro Standard Certified and from a previous version of Postgres Pro Enterprise/Postgres Pro Enterprise Certified. The same holds for migration to Postgres Pro Enterprise Certified 13. See Section 18.6 for the methods to upgrade your database cluster. Consult the Postgres Pro Enterprise support team if you experience issues during migration. Backward migration is not supported. Note that migration from Postgres Pro Enterprise to Postgres Pro Enterprise Certified of the same major version (or vice versa) is an update between Postgres Pro compatible versions (see Section 18.6 for more details).

To migrate from PostgreSQL, Postgres Pro Standard, or Postgres Pro Enterprise release based on a previous PostgreSQL major version, make sure to install its latest available minor version and then perform a dump/restore using pg_dumpall or use the pg_upgrade utility:

  • If you choose to run pg_upgrade, make sure to initialize the new database cluster with compatible parameters. In particular, pay attention to the provider of the default collation and the checksum settings in the cluster you are migrating from. If pg_upgrade creates any SQL files in its current directory, run these files to complete the upgrade.

  • If you are opting for a dump/restore, do not forget to use the --add-collprovider option to correctly choose the provider for the default collation of the migrated database.

To find out the default collation and its provider in the original cluster, see the datcollate value for the template0 database in the pg_database catalog. If you are upgrading from a version where provider of the default collation is not specified, use libc provider if upgrading from vanilla PostgreSQL, and omit the provider if upgrading from earlier versions of Postgres Pro.

Besides, note the following collation-related upgrade specifics described below.

On Windows, Postgres Pro installations could contain databases with default collations provided by ICU, where the name of the database default collation used a syntactically correct BCP 47 language tag format, but had a wrong language code or other parameters, which invalidated the database default collation name for ICU.

If this issue affects the template0 database, you will get the following error message when trying to initialize the cluster with the same collation: failed to get the canonical name for collation locale. In this case, you can only use dump/restore for upgrade, specifying a valid locale for the selected collation provider.

If this issue affects other databases, you will get the same error message when Postgres Pro tries to create these databases with invalid collation in the new cluster. In this case, you can try the following:

  1. Make a dump of the database using pg_dump; it is required to use --create and --format=plain options.

  2. Change the provider for the default collation of the database in the dump file from '@icu' to '@libc'.

  3. In psql, restore the modified dump to complete the upgrade. This operation may fail if any constraints depending on the database collations are violated. In this case, you can try resolving the issues manually or call the support team.

In some corner cases, using dump/restore could lead to invalid constraints in the restored databases, so you should use pg_upgrade. For example:

  • If the installation of Postgres Pro 9.6 or lower contained any indexes or constraints depending on collations other than the default collation of the database, C, or POSIX in databases with multibyte encodings, indexes and constraints in such databases could become inconsistent when these databases are migrated to Postgres Pro 10 or higher. On Windows, this situation can also happen if the database with a multibyte encoding contained any indexes or constraints depending on the default collation with a verbose name, such as "Russian_Russia[.encoding]" or "English_United States[.encoding]".

  • On Windows, in Postgres Pro 10 clusters with default collations provided by ICU, the ICU collation locale may not match the corresponding libc collation locale.

If you use pg_upgrade, it declares such indexes and constraints invalid and creates reindex_text_indexes.sql and validate_text_constraints.sql, respectively. You have to run these files to complete the upgrade.

When migrating from PostgreSQL or Postgres Pro Standard, make sure to pay special attention to implementation specifics of 64-bit transaction IDs. If you have used explicit casts to 32-bit integers when handling transaction IDs, you have to replace them with casts to bigint since 64-bit transaction IDs are of the bigint type.

Note

To avoid conflicts on Linux systems, do not use the postgrespro-ent-13 package to install the new Postgres Pro binaries. Use the individual packages instead. In this case, server autostart needs to be enabled manually, if required. For details on the available packages and installation instructions, see Chapter 17.

For other upgrade requirements imposed by vanilla PostgreSQL, see Section E.35.