According to this patch, simply run checkpoint more than one time can trigger the issue. The root cause here is stat of checkpoint_cleanup_pages_obsolete_tw can only increase but not reset to follow the variable btree->checkpoint_cleanup_obsolete_tw_pages reset here.

Further more, several issues should be considered here:

• There are two stats of checkpoint_cleanup_pages_obsolete_tw, one is aggregated connection level, and another is session dhandle level.
  • Should we perform different on these two stats or should drop the connection level.
  • As there's maybe multi dhandles, so is the connection level means for sum of all dhandles? Then should reset at the same time.
• Should we also reset checkpoint_cleanup_pages_read_obsolete_tw to reflect the same scenario.