Fixed an issue with upgrading Postgres Pro Enterprise 9.6 cluster to the current major version of Postgres Pro Enterprise. Previously, such an upgrade could result in a corrupted free space map and visibility map; the corruption manifested in “could not read block XXX in file ...” errors.

Fixed incorrect data conversion after Postgres Pro Enterprise cluster upgrade through pg_upgrade following a migration from PostgreSQL or Postgres Pro Standard edition. Previously, a cluster repeatedly converted like this would lose its original conversion attribute, which would result in a serious data corruption unless all the data was accessed before the second upgrade.

Deprecated the cfs_compress_small_relations parameter. Small relations are still compressed by default and will not differ from the bigger ones. Make sure to keep the existing parameter setting intact for the lifetime of your database cluster.

Fixed a bug in the CFS implementation that resulted in disabling compression for all relations when the cfs_compress_temp_relations parameter was set.

Added the pg_snapshot_any function to help superusers explore corrupted databases. See Section 9.26.12 for details.

Aligned the behavior of the parameters idle_session_timeout and idle_in_transaction_session_timeout with how they will behave in a future PostgreSQL release. Specifically, the idle_session_timeout parameter now applies only when outside a transaction.

Improved performance of index-only scans with the large tables by increasing the size of the cached visibility map. Now such plan nodes can be executed several times faster with tables larger than 256 MB.

Added pgpro_pwr extension. It enables you to generate workload reports, which help to discover most resource-intensive activities in your database.

Upgraded pgpro_stats. Now it shows resource usage statistics of statement execution, as well as cache invalidation metrics.

Upgraded pg_probackup to the latest version 2.4.10. Major improvements over the previously included version 2.4.2 are as follows:

See pg_probackup documentation for details.

Fixed a bug in pgpro_scheduler due to which jobs scheduled to run exactly at the start of an hour (but at a two-hour or larger interval) were not executed at all.

Added pgpro_controldata utility to display control information of a PostgreSQL/Postgres Pro database cluster and compatibility information for a cluster and/or server.

Fixed the behavior of pg_wait_sampling in case of interrupting a client backend that accessed the shared queue. Previously, the pg_wait_sampling collector could hang in such cases and block other backends trying to read the pg_wait_sampling_profile view.

Fixed a bug in the multimaster extension related to monotonic sequences. Now the parameter multimaster.monotonic_sequences behaves as documented and provides always increasing sequences across all nodes.