EDB Postgres Distributed 5.7.0 release notes v5.7

When a PGD node fails or becomes unavailable, consumers can now continue to consume changes from some other node in the cluster. The feature can be turned on by setting the top group option failover_slot_scope to global. The consumer needs to handle duplicate transactions, but PGD guarantees that every transaction is decoded and sent at least once.

The new CLI command structure is more intuitive and easier to use. The new structure is a "noun-verb" format, where the noun is the object you want to work with and the verb is the action you want to perform. Full details are available in the CLI command reference.

The new feature allows you to assess the local node for compatibility with PGD. The feature is available as pgd assess. Full details are available in the CLI command reference.

The new command allows you to upgrade a node to the latest version of PGD and Postgres. It integrates the operation of bdr_pg_upgrade into the CLI and is run locally. See pgd node upgrade and inplace upgrades for more information.

A node could end up waiting for the local state of another concurrently joined node to advance, which caused the system to stop responding.

The slots are displayed based on the node type, its role, and the cluster topology.

Many code paths touching partitioned tables were performing costly checks to support additional table access methods. This code was adjusted to make the most common case (heap) the fastest.

Now only the bdr_superuser is required.

The new commands are pgd commit-scope show, pgd commit-scope create, pgd commit-scope update and pgd commit-scope drop. Full details are available in the CLI command reference.

The legacy CLI command structure is still supported in the updated PGD CLI. The legacy command support is available for a limited time and will be removed in a future release. It is implemented as a wrapper around the new commands.

The new subcommands are pgd node get-options and pgd group get-options. Full details are available in the CLI command reference.

The new options allow you to format the output of the CLI commands in a psql-like or markdown format. Format options are now json, psql, modern, markdown, simple and defaults to simple.

The message now only appears when creating the sequence.

The bdr.group_camo_details view now only lists data nodes belonging to the CAMO commit scope group. Previously, the view could include data nodes were not part of the CAMO group, logical standby nodes, Subscriber-Only nodes and witness nodes.

Internal tables managed by autopartition that were created before the upgrade to PG17 were missing a rowtype entry into pg_depend. This issue caused errors after upgrading to PG17.

This fix also ensures parting doesn't stop responding when any of the nodes restart when part is in progress. Origin LSNs don't show up as 0/0 in log messages.

This bug could cause memory to leak even if bdr.so is just loaded and the extension isn't installed.

This fix ensures that if a subgroup member is parted while Raft subgroup routing is active, then another node can subsequently join that subgroup.

This bug resulted in nextval call being stuck.

For PGD 5.6.0 and later, there are no subscriptions from a subscriber-only node to a witness node. This caused a problem with movement of group slot on data nodes and witness nodes in the presence of subscriber-only nodes in the cluster. This bug could cause WAL to be held on both witness nodes and data nodes when there's a subscriber-only node in the cluster.

bdr_init_physical disables synchronous replication on a new node by resetting synchronous_standby_names to an empty string. A warning message reminds you to set synchronous_standby_names as needed.

A regression in the way read nodes were identified in the proxy in 6.5.1 was fixed by enabling support for different values in the dsn field's host and the node_name.

The witness node now only cares about the top replication set and thus allows it to miss replications sets and not error out.

In an optimized topology, origin names now use the generation ID of the node. This fixes an inconsistency in which some transactions can be lost or sent twice when a node is parted.

Upgrading from versions earlier than 5.6.0 to 5.6.0 and later in a mixed-version cluster with a standby or a node joining/parting could cause a crash.

Replication no longer becomes stuck when the command is issued.

The proxy commands are deprecated in the new CLI command structure. The proxy commands are still available in the legacy CLI command structure. Proxy options can be set using the pgd group set-option command.

The yaml output option is removed from the PGD CLI. The json output option is still available.

An upgrade to 5.7.0 from clusters that have bdr.force_full_mesh set to off to enable optimized topology (a preview feature) must first set this GUC to on and then upgrade. After the entire cluster upgrades, this GUC can be set to off again to enable optimized topology. Having this GUC set to off during upgrade isn't supported.