Optimized autopreparing for queries with a large number of parameters.

Fixed an issue with incorrect calculation of the memory limit for autopreparing.

Fixed an issue where orphaned *.cfm files that belonged to empty relation segments were left over after pg_upgrade. These files caused “CFS failed to read map file” warnings and could interfere with garbage collection, increasing disk usage.

Fixed a deadlock issue that could occur under write-heavy workloads and was caused by the CFS file lock not being released by server processes. The issue could lead to both primary and standby server hanging.

Implemented the ability to skip replication of DDL commands targeting temporary objects. Temporary objects are not replicated to other nodes which helps to accelerate operation with temporary objects. The new logic is available with the version 2 of the DDL replication protocol. You can switch to this version using the new multimaster.ddl_replication_protocol_version configuration parameter. Use the protocol version 2 only after upgrading multimaster to the latest version on all nodes of your cluster.

Added the exclude_role_name parameter to the pg_proaudit_set_rule(), pg_proaudit_remove_rule(), and pg_proaudit_reload() functions, which allows excluding specific roles from logging. This is useful for filtering out actions of technical roles that would otherwise make the audit log overly verbose.

Added the pg_proaudit.csv_split_messages configuration parameter that controls whether security events in CSV and syslog output are written as a single line or multiple lines. When disabled, each security event is written on a single line, which simplifies log analysis and parsing.

Fixed an issue with unreadable characters in CSV, CEF, and syslog log entries when using a database with a non-UTF-8 encoding. All string fields in the log output are now converted to UTF-8 regardless of the database encoding.

Added multithreading support for the delete command via the j/--threads option. Backup deletion according to the retention policy and WAL archive cleanup are now executed in parallel.

Fixed an issue where S3 operations that returned retryable status codes were erroneously reported as successful. Now an error is returned, and any unrecognized server response body is written to the log.

Fixed infinite looping on corrupted backups by adding loop detection. ID order is now validated when reading the backup catalog or searching for the parent full backup.

Fixed handling of incorrect S3 XML responses, eliminating crashes on empty documents and memory leaks.

Optimized Postgres Pro port validation. Instead of reading from postgresql.conf and postgresql.auto.conf, which may contain outdated information, the port is now read from the postmaster.pid file of the running server. If the instance port does not match the configuration, pg_probackup issues a warning.

Fixed the checkdb and catchup command termination that could occur when an instance was specified without setting the backup catalog path via the -B/--backup-path or the BACKUP_PATH environment variable. Now a clear error message is displayed.

Improved diagnostics of remote connection errors. Previously, specifying an incorrect port that was already in use would cause the process to terminate without an error. Now, the reason for termination is displayed, along with a recommendation to check connection parameters.

Fixed a memory leak that could occur when reading the server configuration files.

Eliminated a potential null pointer dereference by adding out-of-memory handling when parsing the recovery target name in WAL.

Added the --write-rate-limit option to the catchup command help output. In other command descriptions, this option's argument, baudrate, is renamed to bitrate.

Fixed an issue with parallel backups. Previously, each backup created a replication slot with the same default name, causing parallel backups to conflict and terminate with an error.

Fixed backup failures that occurred if a file disappeared between scanning a directory and attempting to access it, for example, if it was deleted by Postgres Pro.

Fixed an issue with PTRACK backups. Previously, an error would occur if ptrack was not added to shared_preload_libraries while ptrack.map_size was set to non-zero.

Fixed an issue that manifested as a “could not stat file” error and could cause backup or restore processes to crash. Temporary files created by the CFS garbage collector are now skipped during these operations.

Fixed an issue where the --external-dirs option failed to process symbolic links correctly, causing the operation to fail with an error.

Implemented the ability to apply plans of any types to queries containing hints. In these cases, hints are silently ignored.

Added debugging information for cases when plans cannot be applied to queries. This information is displayed if the log_min_messages parameter is set to DEBUG1 or a lower level.

Fixed an issue where a parameterized query could not be registered in the baseline mode.

Fixed an issue that could cause a server failure after changing the plan type to baseline. Now the pgpro_multiplan_set_plan_type function does not accept the baseline value for the plan_type parameter.

Fixed an issue that could cause a server failure when using the EXPLAIN command with the GENERIC_PLAN parameter in queries managed by pgpro_multiplan. Now this command works correctly.

Fixed an issue that occurred when specifying an invalid plan type in the pgpro_multiplan_restore function. Now the function displays an error about an unsupported type.

Fixed an issue that could occur after dropping an object referenced in an existing registered query. Now pgpro_multiplan invalidates the registered query on any DROP command and displays the corresponding message.

Fixed a segmentation fault that could occur when using PL/pgSQL variables as parameters of queries managed by pgpro_multiplan.

Added support for statistics of the PostgreSQL 19 core.

Added support for pg_stat_statements 1.13.

Added support for pg_stat_kcache 2.3.2.

Added new report sections and fields.

Fixed the missing integrity constraint, which prevented the cascade deletion of subsample data due to retention policy and thus caused remaining of now-redundant rows in the session statistics table.

Added the pgpro_stats.enable_warnings configuration parameter that adds a warning to the plan field of the pgpro_stats_statements view if the plan was executed with partition pruning.

Fixed a plan normalization issue that could be observed for queries executed using the EXPLAIN EXECUTE command by adding a hook for saving information needed for the plan normalization. As a result, sensitive information is no longer exposed in query plans displayed in pgpro_stats views and pgpro_pwr reports.

Added the --list-cfm-orphans flag to the database_files command to detect and report orphaned CFS map files.

Fixed SQL injection vulnerabilities in query generation.

Fixed a quoting issue that led to “relation does not exist” errors when running the molotilka command on tables in schemas whose names contain uppercase or special characters.

Fixed false “no matching entry in pg_attribute table” warnings for pg_shdepend rows whose objects belong to other databases of the cluster.

Added the pg_wait_sampling.history_dimensions and pg_wait_sampling.profile_dimensions configuration parameters that allow you to customize the set of dimensions recorded in the history and the profile.

Extended the pg_wait_sampling_current, pg_wait_sampling_history, and pg_wait_sampling_profile views with additional wait event dimensions.

Added the pg_wait_sampling_statements view that provides wait event statistics grouped by query and plan identifiers.

Added the pg_wait_sampling_objects view that provides wait event statistics for configurable dimensions, such as database, user, client address, application, backend, or session.

Fixed an issue that caused queries to pg_wait_sampling_profile to hang when the collector process stopped. Hung queries could not be cancelled normally, and the cluster required an emergency restart. The collector now restarts automatically, and queries return an error when the collector is unavailable.

Fixed handling of the pg_wait_sampling.profile_period configuration parameter, which was previously ignored in the collector's timing calculation. Sampling interval accuracy has also been improved.