We saw in the Geo testbed (gitlab-com/infrastructure#2951) that we ran out of memory due to all the other processes (e.g. secondary PostgreSQL process, Geo Log Cursor, etc.) running on the secondary. Lowering the number of Unicorn workers, freeing up memory, resolved the issue.
We now re-calculate the number of unicorn workers possible on the secondary by reducing the amount of memory available for those workers.
Closes #2858 (closed)