As part of the test we first wait until there's an item enqueued in the ticketholder. The problem is that optimistically the wait acquires a ticket then releases it if it shouldn't have done that.
This leads to a small window where the metrics indicate 2 tickets are in use when in reality only 1 is in use.