EDB Postgres Distributed 5.3.0 release notes v5
Released: 14 Nov 2023
EDB Postgres Distributed version 5.3.0 is a minor version of EDB Postgres Distributed.
Recommended upgrade
We recommend that all users of PGD 5 upgrade to PGD 5.3. See PGD/TPA upgrades for details.
Highlights of EDB Postgres Distributed 5.3.0
- Support for PostgreSQL 16 server, EDB Postgres Extended Server 16, and EDB Postgres Advanced Server 16
Compatibility
EDB Server version compatibility
This version requires the recently released Postgres versions 14.10, 15.4, or 16.1 (or later) of EDB Postgres Advanced Server or EDB Postgres Extended Server. No such restrictions exist for PostgreSQL Server.
Package managers on Debian, RHEL, and SLES pull in the required EDB Postgres Advanced Server or EDB Postgres Extended upgrades with an upgrade of EDB Postgres Distributed.
Features
| Component | Version | Description | Addresses |
|---|---|---|---|
| PGD | 5.3.0 | Added support for PostgreSQL 16 server, EDB Postgres Extended Server 16, and EDB Postgres Advanced Server 16. |
Enhancements
| Component | Version | Description | Addresses |
|---|---|---|---|
| Proxy | 5.3.0 | Added the default service unit file to package | |
| BDR | 5.3.0 | Dependencies on EDB Postgres Advanced Server or EDB Postgres Extended are now reflected in packages. |
Bug fixes
| Component | Version | Description | Addresses |
|---|---|---|---|
| BDR | 5.3.0 | Ensure that the WalSender process doesn't request locks on the partitions, thus avoiding a deadlock with user process waiting on sync commit. | RT97952 |
| BDR | 5.3.0 | Consider only CAMO transactions to be asynchronous when the CAMO setup was degraded to local mode. This solves the split-brain problem when deciding fate of transactions that happened during failover. | RT78928 |
| BDR | 5.3.0 | Handle partitions with different attribute numbers when batch inserting. | RT99115 |
| BDR | 5.3.0 | Fixed unsafe CAMO decisions in remote_write mode. | |
| BDR | 5.3.0 | Taskmgr process now respects SIGINT. | |
| BDR | 5.3.0 | Speeded up manager process startup by limiting the amount of WAL read for loading commit decisions. | |
| BDR | 5.3.0 | Physical joins now clean up stale records in bdr.taskmgr_local_work_queue. | |
| BDR | 5.3.0 | Fixed a bug in copying bdr.autopartition_rules during logical join. | |
| BDR | 5.3.0 | Override bdr.ddl_replication=off in taskmgr worker. | |
| BDR | 5.3.0 | Avoid a potential deadlock between bdr.autopartition_wait_for_partitions() and taskmgr. | |
| BDR | 5.3.0 | Fixed writer missing updates in streaming mode with TDE enabled. | |
| BDR | 5.3.0 | Block new EDB Postgres Advanced Server automatic partitioning on PGD cluster. | |
| BDR | 5.3.0 | Allow existing automatically partitioned tables when cluster is created or upgraded. | |
| BDR | 5.3.0 | Block PGD autopartition on EDB Postgras Advanced Server INTERVAL partitioned table. | |
| BDR | 5.3.0 | Ensure that replication doesn't break when bdr.autopartition() is invoked on a mixed version cluster running with 3.7.23 and 4.3.3/5.3 | |
| BDR | 5.3.0 | Fixed default selective replication behavior for data groups. The data groups are supposed to publish only replication sets of the group or any parent groups by default, which mirrors what they subscribe to. | |
| BDR | 5.3.0 | Fixed memory leak in LCR file TDE encryption. | |
| BDR | 5.3.0 | Fixed memory leak in streaming transaction processing. | |
| BDR | 5.3.0 | Allow force parting of nodes that are already being parted normally. | |
| BDR | 5.3.0 | PART_CATCHUP is now more resilient to replication slot and node_catchup_info conflict. | |
| BDR | 5.3.0 | Fixed row filter failure for partitions created after a table column was dropped. | RT95149 |
| BDR | 5.3.0 | Avoid aborting a group commit transaction on receiving the first abort response. The other nodes are given a chance to respond, and transaction can succeed if the required responses are received. | |
| BDR | 5.3.0 | Ensure that replication receiver worker reloads configuration correctly when pg_reload_conf() is called. | |
| BDR | 5.3.0 | Prevent duplicate Raft request ID generation which could break replication. | |
| BDR | 5.3.0 | Fixed several rare memory access issues that could potentially cause crashes of workers. | |
| BDR | 5.3.0 | Fixed issue where explicit 2PC transactions aborted earlier when encountering conflicts. | RT92897 |
| CLI | 5.3.0 | Fixed verify-settings command if the shared_preload_libraries GUC contains file path. | |
| CLI | 5.3.0 | Show replslots status as Critical in check-health command when PGD cluster is reduced to just a witness node. | |
| Proxy | 5.3.0 | Always set the server connection close behavior (setLinger(0)). Earlier it was set only on client error. | |
| Proxy | 5.3.0 | Fixed the logs fill up issue when all nodes are down in a PGD cluster | |
| Utilities | 1.2.0 | Removed the preupgrade step in bdr_pg_upgrade that sets the CONNECTION LIMIT to 0 at database level |