Improved access to the LSN field in the page header by reading it atomically instead of locking to reduce lock contention; added caching of the block number of the page to speed up retrieving of block numbers. Now multiple clients can provide better performance when accessing the same blocks concurrently.

Added the pgpro_stats_wal_sender_crc_errors function that returns the number of errors detected by the WAL sender process.

Increased the number of partitions of the shared buffer mapping hash table to 1024, which can improve performance on multi-core systems.

Implemented support of interval values by all time parameters (*_TIME) of CREATE PROFILE and ALTER PROFILE SQL commands.

Fixed a bug in the CFS causing subtle corruption of data in compressed tablespaces at the server startup on Windows after an abnormal termination or when restoring a backup.

Fixed an issue with lazy conversion from 32-bit to 64-bit transaction IDs in read-only transactions or on replicas by enforcing marking pages as subject to full-page writes to WAL. This issue could cause failures to synchronize conversion results to other replicas.

Fixed segmentation fault, which might have occurred when executing a specific procedure calling pattern while plan_cache_lru_memsize was set.

Fixed an issue with *.cfm files that were unexpectedly left in compressed tablespaces after dropping a table.

Fixed VACUUM logic for handling multixact IDs correctly. Previously, table vacuuming could produce “found multixact XXX from before relminmxid...” errors, which didn't corrupt data.

Fixed the use of multi-column statistics by planner. Previously, when the optimizer estimated the query selectivity using multi-column statistics and the query conditions only included part of the columns involved, the cardinality could be considerably overestimated, so the created plan would be non-optimal.

Fixed the out-of-memory (OOM) killer settings in system startup files. Previously the OOM score adjustment value was set for all Postgres Pro processes rather than for postmaster only, so when Postgres Pro exhausted all RAM, the OOM killer could start killing irrelevant processes, such as sshd.

Fixed an issue in psql, which mistakenly treated the \du+ meta-command as invalid. \du+ lists database roles showing additional information about each role and is equivalent to the \dg+ meta-command, which was processed correctly.

Added support for Rocky Linux 8.

Ended support for outdated operating systems Debian 8, Astra Linux Smolensk 1.5 and MSVSphere 6.3.

Upgraded mamonsu to version 3.4.0. Notable changes are as follows:

Fixed an issue with disconnecting nodes in multimaster by disabling timeouts for the multimaster processes. Previously, the processes could abort due to timeouts set in the server configuration, e.g. statement_timeout.

Added the --checkunique option to the pg_probackup checkdb command to work together with --amcheck and verify unique constraints during logical verification of indexes when the amcheck extension is installed in the database and its version supports this verification. Check with the documentation for amcheck for more details.

Upgraded pgpro_pwr to version 3.7.