I have noted some inconsistency in the stats from the compact changes. For example, csuite/test_wt8057_compact_stress output looks like this:
As observed in the logs above, the "pages rewritten" are greater than "pages reviewed". I looked at the code that manages the stat and looks like we are incorrectly collecting the stats, i.e., either not registering events when we should or double counting some events.
The aim intends to do two things:
- Fix compact stat accounting
- Revisit the stats collected for compact and look for opportunities to streamline them.