Added a possibility to trigger the real-time query replanning for a query that is currently run in the specified session. The replanning request is sent by the replan_signal function.

Hid the confusing “early terminated” notice in EXPLAIN VERBOSE unless replan_enable was on and query replanning was active.

Fixed an issue with real-time query replanning failing to optimize queries with subqueries.

Added handling of SIGILL signals to crash_info processing.

Added details, like process start time and query text at planning, to crash_info output files.

Fixed incorrect function names in the first 2-3 lines of crash_info backtrace stacks.

Fixed possible data truncation at the end of SQL query dump files. Previously, a buffer overflow during SQL query dumping could result in incomplete writes, causing truncated data at the end of files produced by crash_info.

Fixed handling of crash information signals (sent using the kill command) by backends. Previously, the first signal would flush crash_info information to the log, but the process would continue running without producing a core dump, even if configured, and only the second signal would terminate the backend and generate the core dump as expected.

Fixed compatibility of CFS with the ALTER TABLESPACE ... SET/RESET commands.

Fixed an issue with memory allocation within a critical section in the implementation of VACUUM ANALYZE for compressed tables.

Fixed a segmentation fault, which could occur when using postgres_fdw with autonomous transactions.

Fixed loss of temporary tables from parent transaction on autonomous transaction rollback.

Fixed an issue with empty xact_start columns in the pg_stat_activity view for backends executing autonomous transactions.

Added the biha.set_sync_standbys function that allows configuring quorum-based synchronous replication on the existing asynchronous BiHA cluster. You can also use the function to change the number of quorum-based synchronous standbys. For more information, see Configuring Quorum-Based Synchronous Replication on the Existing BiHA Cluster.

Improved the process of the biha_replication_user password verification during BiHA cluster initialization. If you have preliminary specified the password in the password file, bihactl now gets the password automatically. If you have not specified the password in the password file, you are now prompted to enter the password twice to avoid misspelling.

Changed the connection database from postgres to biha_db for the automatic rewind functionality. This helps to avoid automatic rewind errors in cases when the connection string for the postgres database is not present in the password file.

Added the LEADER_CHANGE_STARTED callback type that is called on all nodes when the leader is not available and the quorum is ready to start elections or when you set the leader by the biha.set_leader function. You can use this callback to fence the old leader. For more information about the new callback, see Callback Types.

Added the --referee-with-postgres-db option of the bihactl add command that allows to copy the postgres database to the referee node. For more information, see The postgres Database on the Referee.

Changed the behavior of nodes in the NODE_ERROR state during elections. Such nodes can no longer participate in voting.

Expanded the list of cluster parameters returned by the biha.config function. The function now also returns the mode column displaying the operation mode of the node: regular, referee, or referee_with_wal.

Fixed an issue with multiple “waiting for WAL to become available at...” log messages on the referee node in the referee mode.

Fixed a bug that caused a segmentation fault when calling biha.set_* functions in a single-node (leader only) BiHA cluster.

Fixed an issue of decreasing the max_wal_senders parameter in a single-node (leader only) BiHA cluster. For more information about decreasing max_wal_senders and some other Postgres Pro configuration parameters, see Decreasing Postgres Pro Parameter Values.

Fixed a bug where nodes attempted to check the biha.sync_standbys_min value when updating configuration of the asynchronous BiHA cluster.

Fixed a bug that allowed to set an empty password for biha_replication_user when executing the init command.

Fixed a bug where the old leader in the CSTATE_FORMING state could ignore progressing elections and start as the leader together with the new leader. Now, the old leader does not start as the leader if it detects a node in the CANDIDATE state.

Fixed a bug that caused unexpected cluster behavior when the biha.sync_standbys_min value was set outside the valid range by the biha.set_sync_standbys_min function. BiHA now does not allow to modify the biha.sync_standbys_min parameter if the new value is higher than the number of quorum-based synchronous nodes of the synchronous_standby_names parameter.

Improved the way nodes delete old synchronization points. Previously, every node deleted only a part of the synchronization point table and replicated changes to other nodes. This could cause synchronization issues and inability to delete old synchronization points. Now, every node deletes all synchronization points and does not replicate changes to other nodes.

Fixed potential node hanging during catchup when attempting to process aborted DDL transactions in a cluster of three or more nodes.

Added new object types: CATALOG RELATION and CATALOG FUNCTION.

Added new event fields: UUID, XID, and VXID. Now it is possible to identify an event by its UUID and transaction IDs (if applicable).

The behavior of the pg_proaudit.log_catalog_access configuration parameter was corrected to reflect the new logic of logging for system catalog objects.

Changed the logic of handling disconnection events. Now these events are associated with the corresponding authentication events, so the disconnection events will be logged even in the case when the rule is removed after authentication, but before disconnect.

Fixed an issue where the DISCONNECT event type was not logged for a user who was a member of the role that was set in the logging rule.

Fixed the bug that caused a log entry to be written to an incorrect log file when log file rotation was configured.

Fixed an issue with pg_proaudit failing to log schema creation events.

Corrected the behavior of the logger process when deleting a role from a parallel session configured by the rule.

Added the maintain command to resolve issues that can occur during a forced backup termination.

Added the --lock-lifetime option that sets the timeout for locks. This option is useful for computational environments with a slow network.

Stabilized retaining the initial permissions for directories when running the init command.

Stabilized the checkdb command on a remote host.

Improved the stability of the point-in-time recovery (PITR) with validation.

Fixed the “SignatureDoesNotMatch” error that could occur when connecting to the VK Cloud S3 storage.

Fixed an incorrect behavior that could occur when launching the wait for a WAL streaming thread in the ARCHIVE WAL delivery mode.

Added the set_aqe_trigger() function that allows you to override global trigger values of real-time query replanning for individual queries. All individual trigger values are stored in the new aqe_triggers view. The new pgpro_multiplan.aqe_max_items parameter specifies the maximum number of items in this view.

Added the new pgpro_multiplan.aqe_collect_stats parameter that enables pgpro_multiplan to collect statistics for all statements considered feasible for real-time query replanning reoptimization. All statistics values are stored in the new aqe_stats view. The new pgpro_multiplan.aqe_max_stats parameter specifies the maximum number of collected statistics values.

Added the pgpro_multiplan.aqe_plans_auto_approve parameter that enables pgpro_multiplan to add plans produced by real-time query replanning to the list of allowed plans automatically.

Implemented the ability to back up frozen plans and then restore them into the current database using the pgpro_multiplan_restore() function. You can now transfer frozen plans between databases or even server instances.

Fixed a PANIC-level failure in pgpro_multiplan, which could occur if the standby had a different pgpro_multiplan version than the primary.

Fixed an issue, which could cause errors like “unrecognized node type” when working with stored procedures and functions.

Fixed empty hint string generation in pgpro_multiplan, which caused pg_hint_plan to report a syntax error.

Added support for pgpro_stats 1.9.

Added a possibility to define the mode of collecting relation sizes globally through the pgpro_pwr.relsize_collect_mode extension parameter or for a server through the set_server_size_sampling function parameter.

Enabled fine-tuning the server statistics collection by calling the set_server_setting function. It allows you to define which statistics should be collected.

Added a preview of storage parameters for tables and indexes in the “Schema object statistics” report section.

Added the variant of the sf_find function that searches an sfile object by an ID of the bigint type and to implement the type cast bigint::sfile.

Fixed an issue in pgpro_sfile, which could cause errors like “tuple concurrently updated” during DELETE or TRUNCATE operations.

Enhanced session tracing to provide more information. Specifically, the new time_info filter attribute controls inclusion of additional information in the session-tracing output, and the pgpro_stats.trace_query_text_size configuration parameter can limit the size of the query in the session-tracing output.

Aligned the names of the explain_* filter attributes of the session tracer with the names of session-tracing configuration parameters.

Changed the format of the statistics dump file and the corresponding save/load routines.

Implemented turning off the session tracer functionality when no session-tracing filters are specified.

Prohibited inclusion of both pgpro_stats and pg_stat_statements in the list of shared_preload_libraries. If both are included, the database server will not start.

Fixed an issue where GUC variables could be overridden when using parallel workers.

Fixed an issue with malformed samples caused by a race condition when pg_wait_sampling.sample_cpu is disabled.