Background compaction gathers statistics about files when trying to compact them. This reduces the pressure from compaction when files are not worth being compacted. However, there is a scenario where the stats play against us:

• Create a table and insert data
• Enable background compaction, it will not find anything to do on this data and will mark it as inefficient.
• Remove data from that same table so it can be compacted
• Background compaction won't run on it before some time (today, it's a day). Even if background compaction is disabled/enabled again, it won't run on that table before waiting that time, unless it is executed with run_once which is not very intuitive.