Added support for indexes and TOAST-tables when estimating their compression ratio using the cfs_estimate function.

Improved error messages that are logged for damaged CFS blocks. Now the error message includes an exit code and an error message from the corresponding compression library.

Fixed an issue where enabled CFS could slow the DROP DATABASE execution. The issue occurred because system interrupts were not processed properly while the command was waiting to acquire a lock for cfs_gc_lock_file.

Added the new aqo_storage_usage function that allows you to get current and maximum sizes of aqo storage and allocated dynamic shared memory.

Optimized the internal structure of the aqo_data storage, as well as improved the mechanisms for allocating dynamic shared memory and lock usage when working with storage data. These enhancements allowed fixing an issue with updating aqo_data storage that was caused by exclusive locks on the underlying hash table and resulted in constant growing of idle in transaction sessions and unstable aqo operation.

Optimized operation of the least recently used (LRU) cache mechanism for the auto mode.

Fixed a segmentation fault that could occur after aborting a nested autonomous transaction for which aqo was used. Now aqo checks the nesting level of autonomous transactions and does not clear its local cache when aborting nested transactions.

Upgraded biha to version 1.8.

Stabilized the multi-level geographical distribution and disaster resilience functionality. It is no longer considered experimental.

Added the --username option of the bihactl utility. You can use the new option in the following cases:

Implemented a set of optimizations and fixes for the multi-level geographical distribution and disaster resilience functionality connected with replication, manual leader nomination, and autorewind.

Implemented a set of internal optimizations and fixes for the cascading replication functionality to improve its reliability.

Added the biha.ssl_certificate and biha.ssl_private_key configuration parameters. You can use these parameters to change paths to the SSL public certificate and private key used to secure service connections between nodes of the initialized BiHA cluster. For more information, refer to Managing SSL for Service Connections.

Added the biha_replication_user role presence check when converting a node using bihactl cluster init --convert.

Enhanced the bihactl error message clarity when executing the bihactl node add --convert-standby command with an empty --pgdata value.

Fixed a bug where BIHA_CLUSTER_MANAGEMENT_ROLE had no privileges to execute BiHA functions after update of the biha extension version during the cluster upgrade.

Fixed a bug where failover was not executed because the LSN of the referee in the referee mode was considered during elections if it was higher than the follower's LSN.

Fixed an issue where the “[BiHA BC] NC config list getter failed” warning frequently appeared in the log when calling the biha.config function or querying the biha.status_v view.

Fixed a bug where an attempt to remove a running node from the cluster caused a process termination on a node being removed.

Fixed a bug where an attempt to delete a segment caused a process termination.

Fixed a bug where the leader could not be manually set using the biha.set_leader function if there was a node in the NODE_ERROR state in the cluster.

Fixed an issue of the pg_rewind tool where a BiHA cluster node failed to be recovered correctly in the single user mode if the node was stopped in the immediate mode.

Fixed incorrect behavior of the --use-leader option of bihactl where the utility accessed all cluster nodes instead of the leader only.

Fixed a bug where adding a follower using pg_probackup failed.

Fixed a bug where the password_encryption value was not considered by bihactl when creating the pg_hba.biha.conf file.

Added the pg_proaudit.log_file_mode configuration parameter that specifies permissions for new event log files created by pg_proaudit. Note that regardless of this parameter, the server owner always retains access to the log files.

Added a new event field for all types of events logged in CSV files and syslog: IP address of the event source. Previously, this field was displayed only for AUTHENTICATE events logged in the CEF format.

Extended the use of the pg_proaudit.log_rows configuration parameter that enables or disables logging of the number of rows affected by an SQL command. Now the number of affected rows is written for all formats of logged events.

Added validation of the unique ID of the event (UUID) to avoid invalid memory access.

Fixed an issue with incorrect filling of the dst field for events logged in CEF format.

Fixed an issue with excessive memory usage when processing DISCONNECT events. Previously, a hash table sized proportionally to max_connections caused unnecessary memory consumption when the value was large. The data is now stored more efficiently.

Fixed a parser error in pg_proaudit. Previously, pg_proaudit left some characters unescaped when logging events in the CEF format. Now, symbols are escaped correctly in accordance with the CEF specification.

Added support for Virtual Host for S3 requests.

Added support for absolute paths in the --wal-file-path option.

Fixed incremental backups in PAGE mode with the --stream option. With this combination of parameters, WAL files are now correctly saved both inside the backup and in a separate archive.

Added support for the --exclude-path option to exclude specific files or directories from backup in PRO mode.

Optimized the backup performance by excluding temporary files of the postmaster process.

Fixed an incorrect package name in the error message that appeared when the pgpro_bindump failed to load libpb3_encoder.so. The message now correctly recommends installing the libpb3-encoder package instead of the erroneous libpgprobackup3.

Fixed an issue with the CFS garbage collector (GC) being blocked by the backup process, potentially causing write locks during long backups. Both processes now lock only specific data files briefly, avoiding mutual blocking.

Added the ability to use the ON CONFLICT clause to specify an action alternative to raising a unique constraint violation error for the unique global index.

Added the ability to create a global index on a column included in a primary key.

Added the ability to specify the CONCURRENTLY option when creating or rebuilding a global index. This allows a global index to be built without taking any locks that prevent concurrent inserts, updates, or deletes on the table.

Added the gbtree.usage_partition_fraction and gbtree.partitions_number_threshold configuration parameters that control whether a global index is used by specifying the percentage or the number of table partitions, respectively.

Added the ability to perform autovacuum for global indexes.

Fixed a segmentation fault that could occur when trying to execute the ALTER TABLE ... ATTACH PARTITION command if a parent partitioned table had a global index and the table being attached had a column order different from that of the parent table.

Fixed an issue that could cause high memory consumption and occur when trying to update millions of rows of a partitioned table with a global index.

Fixed an issue with global indexes failing to be rebuilt for columns whose type was changed during the ALTER COLUMN ... TYPE ... USING ... command execution. Now global indexes are rebuilt correctly.

Implemented the ability to restore baselines and template plans from plan backups. For this, the pgpro_multiplan_restore function now accepts the baseline and template values for the plan_type argument and contains the new wildcards argument for template plans. Also added the following arguments to this function: sample_string to specify the query with constants and status for future usage.

Implemented the ability to add plans to a baseline using the pgpro_multiplan_register_query and pgpro_multiplan_freeze functions. The last function now can accept the baseline value for the plan_type argument.

Implemented the ability for pgpro_multiplan to register executed prepared statements automatically when the pgpro_multiplan.auto_tracking parameter to is set to on.

Added the new pgpro_multiplan_enable, pgpro_multiplan_enable_baseline, and pgpro_multiplan_enable_template functions that allow enabling and disabling certain existing plans.

Renamed the following functions:

Removed all the functionality related to storing and sharing global prepared statements for multiple sessions.

Removed the constants and prep_consts fields from the pgpro_multiplan_captured_queries view. These fields were redundant since query constants are contained in the sample_string field value.

Fixed an issue that could cause an assertion failure when trying to freeze the same plan for the second time.

Fixed an issue that could cause an assertion failure when trying to register a query with the DO NOTHING rule.

Fixed an issue that could cause an assertion failure when using the pgpro_multiplan_register_query function with incorrect optional arguments specifying types of query parameters.

Fixed a segmentation fault that could occur when trying to register a query with a common table expression (CTE) or INSERT ... SELECT.

Fixed an issue where the sample_string values for baselines were saved with parameters rather than constants, which could cause errors when managing these baselines, in particular, when updating their costs using the pgpro_multiplan_update_baseline_cost function.

Fixed an issue where baselines did not save non-default values of configuration parameters, which could cause queries to produce plans different from fixed plans in another environment.

Fixed an issue with incorrect check of the table_hash value for template plans, which could cause applying these plans to inappropriate queries.

Fixed an issue with memory allocation that could occur in pgpro_result_cache with the enabled pgpro_result_cache.consistent parameter when executing a very large number of queries in one transaction.

Fixed an issue that could occur when using the result_cache hint with the offset and limit arguments if the size of the returned result set exceeded the pgpro_result_cache.max_entry_size limit.

Improved logging mechanism. Log messages are more consistent and informative. When a cron job executor background worker is started, the database manager now writes a log message containing the worker's process identifier (PID). This makes it easier to trace worker activity and correlate log entries.

Extended the use of resource prioritization to pgpro_scheduler background workers. Previously, this functionality affected only regular client backends. Now it is also possible to set resource usage weight for scheduler workers, providing more flexibility in allocating system resources.

Fixed an issue with the schedule.get_self_id() function returning an incorrect job ID for cron jobs. Now schedule.get_self_id() works as expected in both scheduled and one-time jobs, returning the proper job ID.

Fixed an issue with deactivating a cron job with next_time_statement. Previously, if this job was deactivated while being executed, it could still remain active. Now, the job is immediately deactivated regardless of its execution state.

Fixed excessive memory consumption by a query that used pgpro_stats with pgpro_stats.track set to all. With this setting, a function that assigned plan IDs was called multiple times, and the accumulated memory was not optimally released.

Updated autopreparing and pgpro_stats to fix issues that could occur with autoprepared query normalization in pgpro_stats. Now for proper query normalization, autopreparing stores the cached plan together with certain additional information to be passed to pgpro_stats when the query is executed.

Improved the utility output to include the used preset and display the correct format of date and time when configuration parameters were applied.

Updated presets to improve performance of Postgres Pro.

Fixed an issue where the shared_buffers configuration parameter was calculated incorrectly when using initdb with the enabled pgpro_tune extension. This issue could cause the command to run slower.

Extended proxima capabilities to accept incoming connections. Previously, proxima was able to accept incoming connections either on a single specified IPv4 address, or any available IPv4 addresses of the host. Currently, proxima can accept incoming connections on multiple addresses. You can use IPv4 or IPv6 addresses, as well as paths to Unix socket directories. For more information, refer to the description of the proxima.address configuration parameter.

Implemented the adaptive balancing algorithm, which determines how CPU, RAM, and disk I/O load impacts node selection during balancing. To specify a share of each resource, use the proxima.load_balancer_cpu_weight, proxima.load_balancer_io_weight, and proxima.load_balancer_ram_weight configuration parameters. For more information about the new balancing algorithm, refer to the description of the proxima.load_balancer_algorithm configuration parameter.

Implemented the adaptive-ram and adaptive-io balancing algorithms, where workloads are distributed across replicas depending on RAM and disk I/O load respectively. For more information, refer to the description of the proxima.load_balancer_algorithm configuration parameter.

Implemented new counters for system metrics: system.avail_ram_b, system.total_ram_b, and system.io_avg_latency_ms. For the full list of available counters, refer to the System Counters table.

Implemented support for the certificate authentication method. You can specify the path to the file containing public CA certificates using the new proxima.ssl_ca configuration parameter. For more information, refer to Enabling SSL for Incoming Client Connections.

Implemented the ability to proxy user authentication to the primary (leader) node. You can specify where to perform user authentication for each node using the new proxima.auth_policy configuration parameter. For the BiHA referee node, the authentication is now proxied to the leader by default.

Enhanced operation of proxima in a GDBiHA cluster. The proxima extension can now redirect queries from the P2F port to the front follower.

Implemented the experimental functionality of the internal prepared statement storage, which allows proxima to avoid establishing dynamic dedicated sessions for prepared statements. Previously, this functionality was supported via integration with multiplan. Now, you can manage the prepared statement storage using the new configuration parameters of proxima: proxima.ps_enabled, proxima.ps_shared, and proxima.ps_cache_size.

Implemented a range of internal fixes and optimizations to improve performance and reliability of the proxima extension.

Fixed a bug where proxima endlessly restarted when its SSL configuration was incorrect.

Fixed a bug where proxima would hang when addressing several databases simultaneously.