Re: MySQL 5.6 update doubled IOs

[Dimitri Kravtchuk]

All,

since I was able to reproduce the "test case", seems like things are not as simple.. - on my server MRR is enabled, however OPTIMIZER is not using it in the execution plan (I have only "Using index condition" without "Using MRR").. So, we have yet to understand why in Christian's case it's not the same..

while for the difference in SELECT count(*) -- the overhead is coming from PFS (Performance Schema). So, as a general advice, to have a "pure" compare of 5.5 vs 5.6 think to start 5.6 with "performance_schema = OFF". However, in most cases you may need just to disable "default instrumentation", so just start with:
performance_schema = ON
performance_schema_instrument = '%=off'

this will keep PFS on, and still give you a way to dynamically enable different instrumentations when you need.. In this particular case the overhead came from the table IO instrumentation (have no idea yet why it can be so big here, but this must be fixed for sure).. - with PFS=off (or none) on my server I have a even better execution time on 5.6 than on 5.5 BTW ;-)

Rgds,
-Dimitri