Known issues with distributed high availability/PGD

These are currently known issues in EDB Postgres Distributed (PGD) on Cloud Service as deployed in distributed high availability clusters. These known issues are tracked in our ticketing system and are expected to be resolved in a future release.

For general PGD known issues, refer to the Known Issues and Limitations in the PGD documentation.

Management/administration

Deleting a PGD data group may not fully reconcile

When deleting a PGD data group, the target group resources is physically deleted, but in some cases we have observed that the PGD nodes may not be completely partitioned from the remaining PGD Groups. We recommend avoiding use of this feature until this is fixed and removed from the known issues list.

Adjusting PGD cluster architecture may not fully reconcile

In rare cases, we have observed that changing the node architecture of an existing PGD cluster may not complete. If a change hasn't taken effect in 1 hour, reach out to Support.

PGD cluster may fail to create due to Azure SKU issue

In some cases, although a regional quota check may have passed initially when the PGD cluster is created, it may fail if an SKU critical for the witness nodes is unavailable across three availability zones. To check for this issue at the time of a region quota check, run:

biganimal-csp-preflight --onboard -i d2s_v3 -x eha <azure-sub-id> <azure-region>

If you have already encountered this issue, reach out to Azure support:

We're going to be provisioning a number of instances of <SKU TYPE> in <TARGET REGION> and need to be able to provision these instances in all AZs. Can you please ensure that subscription <AA BYOA AZURE SUBSCRIPTION> is able to provision this VM type in all AZs of <TARGET REGION>. Thank you!

Changing the default database name is not possible

Currently, the default database for a replicated PGD cluster is bdrdb. This cannot be changed, either at initialization or after the cluster is created.

Replication

Replication speed is slow during a large data migration

During a large data migration, when migrating to a PGD cluster, you may experience a replication rate of 20 MBps.

PGD leadership change on healthy cluster

PGD clusters that are in a healthy state may experience a change in PGD node leadership, potentially resulting in failover. The client applications will need to reconnect when a leadership change occurs.

Extensions which require alternate roles are not supported

Where an extension requires a role other than the default role (bdr_application) used for replication, it will fail when attempting to replicate. This is because PGD runs replication writer operations as a SECURITY_RESTRICTED_OPERATION to mitigate the risk of privilege escalation. Attempts to install such extensions may cause the cluster to fail to operate.

Migration

Connection interruption disrupts migration via Migration Toolkit

When using Migration Toolkit (MTK), if the session is interrupted, the migration errors out. To resolve, you need to restart the migration from the beginning. The recommended path to avoid this is to migrate on a per-table basis when using MTK so that if this issue does occur, you retry the migration with a table rather than the whole database.

Ensure loaderCount is less than 1 in Migration ToolKit

When using Migration Toolkit to migrate a PGD cluster, if you adjusted the loaderCount to be greater than 1 to speed up migration, you may see an error in the MTK CLI that says “pgsql_tmp/': No such file or directory.” If you see this, reduce your loaderCount to 1 in MTK.

Tools

Verify-settings command via PGD CLI provides false negative for PGD on Cloud Service clusters

When used with PGD on Cloud Service clusters, the command verify-settings in the PGD CLI displays that a “node is unreachable.”