Greetings Bryan. We have seen a number of high load deplyments where the maintenance run can take a very long time. This is primarily due to the re-indexing involved in selective deletes. I would start by separating these entries by a few hours if possible. 15 minutes is very short.
You are correct that the re-index on the primary will make its way to the node. The only thing that is actually going to be indexed is the logs. All configuration and RADIUS history is only writable on the primary.
Having just peaked at the code, it does not seem the retention code is aware of the HA status or not so it may try and do its purges. This makes it vital to have the primary completed, and synced over prior to the node trying to do it again.
Look on the primary for these two log entries:
- Processing scheduled database maintenance.
- Finished processing scheduled database maintenance
Technically you can just search 'database maintenance' and catch them. Look at the time delta and that will be your starting point for setting node's schedule.