Currently, we have wal-e configured to take a full basebackup from the primary each day. This can have negative impact on performance: We need to read and stream the whole database onto the network which results in overhead on the primary. Additionally, this is likely to damage the caches on the primary which may result in degraded performance until caches are rebuilt during normal operations. We can see an increased IO-wait during the time a basebackup is taken here (timely coincidence, may need further digging):  The proposal is to take the basebackup from a secondary to move the overhead off the primary. In this case we need to make sure that the replication lag is acceptable. /cc @yorickpeterse