Currently the index queue flushing algorithm works like so:

1. Query the database for all index entries added since the last flush
2. Add entries to task list
3. Update the last flush the current system time

Under high load or with particularly slow database connections, it is feasible that entires are added in between Step 1 and 3. In these situations, these entries get ignored and never added to the index. Pages can effectively be lost or cluster indexes become out of sync.

To fix: Instead of using the system time to determine the last flush, Conflunce should use the time of the last entry retrieved from the index queue.