WarehousePG#

WarehousePG (WHPG) is a PostgreSQL-based massively parallel processing database. A WarehousePG cluster is made of several PostgreSQL instances: the coordinator and the segments, which can share a host.

Barman is not a disaster recovery tool for WarehousePG clusters as a whole. It takes physical backups of, and restores, the individual instances which compose a cluster. Orchestrating those operations across the cluster is the job of an external tool, such as whpg-dr.

This page covers what Barman needs in order to work with a WarehousePG instance.

Connecting to an instance#

Each WarehousePG instance is configured in Barman as an independent server, with its own conninfo. Segments only accept direct connections in utility mode, so the connection parameters have to enable it:

  • WarehousePG 7: gp_role=utility

  • WarehousePG 6: gp_session_role=utility

As an example, for a WarehousePG 7 segment:

[seg1]
description = "WarehousePG 7 segment"
backup_method = postgres
conninfo = host=seg1 port=6000 user=barman dbname=postgres options='-cgp_role=utility'

How Barman detects a WarehousePG instance#

Barman considers an instance to be a WarehousePG one when both the gp_dbid and gp_server_version_num settings exist in the instance it is connected to through conninfo. The version() string is not used for that.

If only one of the two settings exists, the instance is in an unexpected state and Barman fails the operation instead of making assumptions about it.

When taking a backup of a WarehousePG instance, Barman saves the following information in the backup metadata, and shows it in barman show-backup:

  • The instance identifier in the cluster, i.e. the value of gp_dbid.

  • The WarehousePG version number, i.e. the value of gp_server_version_num, e.g. 70000 for WarehousePG 7.0.

Backups of other PostgreSQL instances do not have this information.

Note

Backups taken before Barman 3.21 are not recognized as WarehousePG backups.