On the big sites grade_item::regrade_final_grades can be run up to 1000 times per second during the busy exams period . 

After doing a quick analysis our inf team end up with SQL that can be improved.  This SQl at some point creates  duplicate dead tuple rows.

UPDATE mdl_grade_grades

SET aggregationstatus = 'used',

aggregationweight = $1

WHERE itemid = $2 AND userid = $3

This SQL is coming from grade_category::set_usedinaggregation

The below sql would stop postgresql creating duplicate dead tuple rows and the associated index update / wall logs if anything was slipping through.

UPDATE mdl_grade_grades

SET aggregationstatus = 'used',

aggregationweight = $1

WHERE itemid = $2 AND userid = $3 and aggregationstatus != 'used' and aggregationweight != $1