Upgrading PGD clusters with TPA v5

No Postgres major version upgrades

TPA doesn't currently support major version upgrades of Postgres.

To perform a major version upgrade of Postgres, see In-place Postgres major version upgrades.

If you used TPA to install your cluster, you can also use TPA to upgrade it. The techniques outlined here can perform minor and major version upgrades of the PGD software. They can also perform minor version upgrades of Postgres.

You can read more about the capabilities of TPA upgrades in Upgrading your cluster in the TPA documentation.

Always test first

If possible, always test upgrade processes in a QA environment first. This helps to ensure that there are no unexpected factors to take into account. TPA's ability to reproducibly deploy a PGD configuration makes it much easier to build a test environment to work with.

Minor and major PGD upgrades

TPA automatically manages minor version upgrades of PGD.

Major version upgrades of PGD require changes to the TPA config.yml file, which contains the deployment configuration.

When upgrading to PGD 5 from previous PGD major versions, you can use tpaexec reconfigure. This command helps you make appropriate modifications to your deployment configuration.

The reconfigure command requires settings for architecture (the only supported setting is PGD_Always_ON) and PGD Proxy routing (--pgd-proxy-routing <global|local>) to run. Remember to back up your deployment configuration before running, and use the --describe and --output options to preview the reconfiguration.

Prerequisites

  • You need the cluster configuration directory created when TPA deployed your PGD cluster.

  • If performing a major version upgrade of PGD, ensure that tpaexec reconfigure was run and appropriate configuration changes were made.

Upgrading

Run:

tpaexec upgrade clustername

Where clustername is the name of the cluster and the path to the cluster configuration directory. By default, TPA upgrades each node of the cluster to the latest minor versions of the software the nodes were configured with.

TPA's automated rolling upgrade procedure

TPA first tests the cluster and then the nodes.

Each node is then isolated from the cluster, upgraded, and returned to operation within the cluster.

TPA upgrades - step by step

  • Checks that all preconditions for upgrading the cluster are met.
  • For each instance in the cluster:
    • Checks that it has the correct repositories configured.
      • Checks that the required Postgres packages are available in those repositories.
    • For each BDR node in the cluster, one at a time:
      • Fences the node off to ensure that pgd-proxy doesn't send any connections to it.
      • Stops, updates, and restarts Postgres.
      • Unfences the node so it can receive connections again.
      • Updates pgbouncer, pgd-proxy, and pgd-cli, as applicable for this node.