EDB Postgres Distributed 5.9.0 release notes v5.9.0

bdr.autopartition() now takes an additional boolean argument drop_after_retention_period which can be set to false if the user does not want the partition to be dropped, but only detached.

We've made improvements to how our Raft consensus mechanism handles snapshots. This ensures that in the rare event of a data consistency issue, restoring a snapshot will now fully reset the system's state, allowing operations to resume smoothly from a consistent starting point.

A new enhancement allows partitions created by EDB Postgres Advanced Server (EPAS) AutoPartition to work correctly within a multi-node PGD environment. This fix ensures partitions are created consistently on each node and are visible across all nodes.

To streamline the upgrade process, the Connection Manager has been back-ported to PGD 5.9 as an optional feature that's disabled by default. This change allows for seamless upgrades to PGD 6 by having the necessary infrastructure in place. It also gives you the opportunity to test the Connection Manager in non-production environments before a full production upgrade.

Subgroups have proxy routing enabled by default from 5.9.0 however it doesn't impact the existing behavior of subgroups and how to connect to the nodes without proxy.

Roles and tablespaces are now synchronized before the schema is restored from the join source node. If there are already existing roles or tablespaces (or EPAS profiles), they will be updated to have the same settings, passwords etc. as the ones from the join source node. System roles (i.e. the ones created by initdb) are not synchronized.

Roles are now replicated to witness nodes at node join. Roles also get replicated to witness nodes during normal operation of the cluster.

We've resolved a bug that could cause a PGD node upgrade to fail with the error 'function max(pg_lsn) does not exist'.

A fix has been implemented to resolve an issue that caused duplicate key value violates unique constraint "subscription_pkey" errors during node resynchronization. To ensure unique names and prevent manager crashes, subscription names now include the source node name, which provides a more robust and reliable resynchronization process.

The pgd cluster verify command no longer displays a warning for witness-only groups that do not contain data nodes. This resolves an issue where the command incorrectly flagged these groups, ensuring that the verification output is now accurate for all cluster types.

Raft no longer fails to recover after a heavily-lagging node that was previously a leader is parted from the cluster. This fixes a bug where the Raft leader could crash with a "BDR node XXX not found" error, which previously required a manual, forceful reset of the Raft state. Now, Raft correctly handles this scenario and recovers automatically, ensuring cluster stability.

A race condition has been fixed that could cause a "clock has moved backwards" error when using Snowflake sequences, even when the system clock was functioning correctly. This fix ensures the sequence correctly handles concurrent access, preventing this error and improving the reliability of Snowflake sequence generation.

A bug has been fixed that could cause bdr_init_physical to hang when the upstream node had a synchronous commit scope. This issue, which prevented a new node from completing physical initialization, was caused by a specific interaction with synchronous commit settings. The fix ensures that bdr_init_physical now works reliably in all configurations, even before the new node joins the group and is considered for synchronous commits.

Ensure that CDC hooks are no-op when recovery is in progress or the backend is not connected to a database. This can definitely happen if pg_failover_slots is active.