EDB Postgres Distributed 6.2.0 release notes v6.2.0

The pgd node upgrade command now supports in-place conversion of a non-TDE PGD node to a TDE-enabled node with the --key-unwrap-command and --copy-by-block options. The pgd node setup command now includes comprehensive options to configure TDE during the initial provisioning of a new cluster node. These new options are: --data-encryption, --copy-key-from, --key-wrap-command, --key-unwrap-command, --no-key-wrap, and --data-encryption-keylen (accepts the AES key length).

The view provides real-time information on transactions in the receiver pipeline. This view reports on transactions currently being received, sent to writers for processing, and indicates whether a transaction is being streamed to a file or writer, including the writer's application progress.

Connection Manager now supports the same LDAP authentication features as PostgreSQL, with the exception that OpenLDAP configuration files and environment variables are not supported.

The bdr.writers view now includes several new columns for enhanced monitoring. It now reports the transaction apply source, the total size of the transaction, the progress of the application, the last change applied, and the upstream LSN corresponding to that last applied change.

When provisioning an EPAS cluster, the --superuser option for the pgd node setup command now automatically defaults to enterprisedb unless explicitly specified.

The pgd cluster verify command now includes a Details column in the architecture verification section when using the --verbose option. This column provides specific, detailed information about any issues encountered during the architecture check.

The behavior of the pgd node part command is modified so it now only waits until the node is renamed (allowing its original name to be reused) and continues the complex parting operation in the background. This enhancement also deprecates the –-no-wait option when all nodes in the PGD cluster are running version 6.2 or above.

The PGD node management interface now allows you to reuse the names of parting nodes even while the parting operation is still in progress. This is accomplished through by setting the new concurrent=true option when calling the bdr.part_node function.

Connection Manager now supports the PostgreSQL ssl_passhrase_command option to decrypt the server private key.

The command has been enhanced to read user authentication credentials from a password file. This file location is resolved by checking the --dsn global option, the PGPASSFILE environment variable, or the ~/.pgpass file, in that order.

Previously, this setting only applied to PGFS storage locations.

The role synchronization logic has been refined to only apply role settings from the source node if they are global, or are specifically configured for the PGD database on the source node.

The bdr.stat_commit_scope view now includes three new metrics to track degrade behavior: ndegrades (counts per-transaction degrade events), nconfig_degrades (tracks configuration-level state changes to degraded mode), and last_state_change_time (records the timestamp of the last state transition between normal and degraded operation).

Added a new sanity check that emits a warning during startup if the PostgreSQL configuration parameter synchronous_standby_names is configured in a way that might clash with a PGD commit scope. This helps administrators quickly identify and correct potential misconfigurations.

The --group filter provides the capability to filter the output by a specified data group when administering a multi-group PGD cluster.

The BDR extension has been updated to improve compatibility with the pg_upgrade utility, which ensures the upgrade process runs cleanly.

Resolved an issue where the pgd group set-leader command displayed confusing messages of both success and failure. This issue occurred because after the switchover completed successfully, the Connection Manager immediately reset the connection due to the write-leader change.

This fix prevents the incorrect creation of both subscriptions and replication slots between subscriber-only nodes and witness nodes. The fix also addresses a problem seen during upgrades from PGD 5.9.0 to 6.2.0 where an unexpected slot was left behind on the witness node.

Resolved an issue where the Connection Manager dropped existing client connections whenever a write-leader or read-only node routing change occurred. With this fix, client connections are no longer affected by a routing refresh, as long as the PGD node they are currently connected to remains operational.

The fix improves the handling of authentication failures in the Connection Manager.

The fix also corrects the handling of hostname suffixes. Since hostname matching is not supported by the Connection Manager, HBA rules using hostnames or hostname suffixes are now correctly ignored by the Connection Manager, preventing instability.

This issue occurred when streaming a large transaction, and the writer restarted only to find the temporary file name had been incorrectly freed. The bug was introduced with parallel apply and is resolved now.

This bug caused the node_uuid in system tables (bdr.local_node and bdr.node) to diverge after a full cluster upgrade. The fix ensures that the joining node's UUID and kind are correctly preserved throughout the process.

Resolved an issue where concurrent operations, such as renaming a sequence or schema containing a galloc sequence, could cause problems with Raft synchronization. The fix introduces a global UUID identifier for each sequence in the bdr.sequence_kind catalog, which remains constant despite any sequence or schema renames.

The pgd node setup command now performs stricter compatibility checks before allowing a physical join:

• If the BDR version of the local and remote nodes are not an exact match, the command attempts a logical join instead of a physical join.
• If the Postgres major version of the local and remote nodes are not the same, the command attempts a logical join instead of a physical join.
• If the PostgreSQL distribution (flavor) of the local and remote nodes are not the same, the command exits with an error.

Resolved a backward compatibility issue where PGD 6.x CLI commands, such as pgd group show, reported the error No query found for version when invoked against an older PGD 5.x cluster. The fix ensures these commands function correctly across major PGD versions.

Resolved an issue where the pgd node setup command incorrectly reported a mismatch error when validating a valid base backup node of the given remote node.

The PGD CLI now adjusts the database connection's verbosity to ignore the CONTEXT value, which previously added noise to the output, especially for errors from bdr.run_on_x functions.

Resolved an issue where the bdr.msgb_connect function could hang indefinitely due to a missing network timeout setting. The fix adds a timeout to the function.

Resolved an issue which prevented the pgd node setup command from successfully creating and configuring a PGD node using an existing data directory.

In these cases, the configuration inherited from the top-level node group is automatically applied.

Previously, an UPDATE conflict with an update_missing resolution (which attempts to insert a new record) did not check for conflicts against existing primary or unique keys, leading to an error. This issue is now fixed by treating the scenario as an insert_exists conflict and applying the appropriate resolution logic.

This allows for granular control over which partitions replicate data to analytical targets.

Resolved an issue during physical node joins where the last transaction was redundantly included in the physical snapshot. The system now ensures this final transaction is excluded from the snapshot, as it will be correctly received later via logical replication, preventing duplication.

Resolved various synchronization issues affecting the correct replication and management of EDB Postgres Advanced Server (EPAS) roles and profiles across the cluster.

Resolved a bug that caused a segmentation fault when calling bdr.stat_get_activity. This function is used by views such as bdr.stat_workers and bdr_stat_activity.

Resolved an issue where stale records from parted or dropped nodes were retained in the bdr.node_peer_progress_mapping table. Proper cleanup is now enforced to prevent these stale records from unnecessarily holding back the group replication slot.

Resolved an issue where spurious walsender timeouts could occur while the receiver process was flushing a lengthy transaction.

Resolved an issue where the error database with remote_datid 0 not found in remote_databases occurred during the node join process. This failure happened when a role on the upstream source node had been altered to include a setting that was not specific to the database, causing a lookup failure during synchronization.

Resolved an issue where the system incorrectly required built-in, hard-coded CRDT merge functions to be explicitly registered in the bdr.crdt_handlers catalog.

Resolved an issue where the check for the configuration parameter bdr.commit_scope incorrectly raised an error if the PostgreSQL configuration was reloaded while a transaction was in progress, even when the parameter's value had not actually been modified.

Resolved two replication issues related to EPAS roles and profiles:

• Fixed an issue where profile names were incorrectly quoted as string literals instead of identifiers, causing syntax errors when replicating CREATE ROLE or ALTER ROLE commands with the PROFILE clause.
• Corrected a bug where ALTER ROLE commands that did not change passwords incorrectly included the PASSWORD clause, which triggered password reuse policy violations when profiles had PASSWORD_REUSE_TIME or PASSWORD_REUSE_MAX constraints.

Resolved consensus request timed out errors that could occur during BDR cluster initialization. This issue happened when vacuum operations on internal consensus tables interfered with critical group creation processes. The fix now defers vacuum operations until the cluster is properly established.

Implemented a slot checkpoint system that periodically saves the replication slot state to disk. This mechanism allows the group replication slot to advance quickly without decoding massive amounts of WAL, resolving issues where the manager process would hang for extended periods. Checkpoints are stored as small binary files in $PGDATA/pgd/slot/ with CRC32 validation for integrity.

Resolved a corner-case bug that could cause a segmentation fault when querying the internal function bdr.local_topology_info.