if all slots in pg_stat_statements are full, new query no longer get into the slot because "calls" counter always starts at "1".
Conversation
Replying to
Nope. The hashtable has 5% of entries removed before the new entry is entered. That entry has however long it takes 5% to clear...
Replying to
Thanks! I'm studying pg_stat_statements *again*, so I wish to study its internal. :)