Question

How can I prevent the loss of crash detection and recovery for standby replicas when Fivetran recommends setting the wal_sender_timeout value to 0 during setup for WAL-based replication?

Environment

Connector: PostgreSQL 

Answer

If your PostgreSQL connector is using WAL for incremental replication then contact the Support team and a change can be made to the frequency of how often we save state.  Instead of the default of six hours the the state can be saved every two hours, this allows the timeout value to be increased to over two hours rather than being disabled completely.

1. Make a request to the support team asking to be added to this feature
2. Confirm that this change has been made to your account
3. Change the wal_sender_timeout value to 9000 (2.5 hours in seconds)

Cause

The wal_sender_timeout variable (which you can read more about here) is used to terminate replication connections that are inactive for longer than a specified amount of time, either in seconds (default) or milliseconds. By setting this value to 0, the timeout is disabled which allows Fivetran to keep replication connections open for longer to compensate for network latency.