This section explains how Connect CDC works with an AWS Aurora Global Postgres Database during normal operations, planned switchovers, and unplanned failovers.
Connection
When configuring up the Postgres Server in the Connect CDC UI, use the endpoint of the primary Aurora cluster as the Remote Machine or IP address. This ensures that Connect CDC always connects to the current write-enabled region.
Switchover (Planned region promotion)
During a planned switchover, Aurora promotes a secondary region to become the new primary. Connect CDC will:
- Temporarily lose its connection when the primary cluster changes.
- Automatically attempt to reconnect.
- Reconnect to the newly promoted primary region without manual action once the switchover is complete.
AWS Recommendation
To maximize availability and minimize possible replication lag, perform planned switchovers during non-peak periods or when
write volume to the primary cluster is minimal.
Failover (Unplanned Event)
During an unplanned failover, Connect CDC will:
- Lose its connection when the primary cluster becomes unavailable.
- Automatically keep trying to reconnect until the new primary region is active.
-
Try to switch to the new region automatically.
However, depending on the type of failure, a manual restart of replication requests may be needed after reconnection.
AWS Guidance
Unplanned failovers can cause data loss, including possible loss of data in Metabase tables—especially replication queue tables. After a failover, you may need to validate or recover this data.