Now that sql-s1 is on a new box, my queries are creating a replag that never happened before.  For instance, I run a query that creates an ENGINE=MEMORY table based on a join with enwiki_p.pagelinks and another MEMORY-based table, typically takes 10-15 minutes. I ran this query daily for 1-2 months before the switch, and it never affected replag.  Now, replag goes up as soon as the join begins. (Oh, and I call SET SESSION TRANSACTION ISOLATION LEVEL READ UNCOMMITTED) Is anyone else seeing problems like this?  Or have an idea how to fix/work around the bug? - Jason