List:Commits« Previous MessageNext Message »
From:Davi Arnaut Date:November 30 2010 5:05pm
Subject:Re: bzr commit into mysql-trunk-bugfixing branch (guilhem:3393) Bug#47609
View as plain text  
On 11/30/10 3:00 PM, Davi Arnaut wrote:
>> The scenario, very particular, was a join of 20 tables (BUG#50595). The
>> optimizer spent seconds (minutes in the debug build - causing test
>> timeouts in pb2, which is how it all started :-), to find the optimal
>> plan because of "combinatorial explosion" of the set of possible plans;
>> it ran best_access_path() millions of times. So those few instructions
>> in this function run many times, gave a repeatable 2% slowdown. That's
>
> in the test case time?
>
>> why I did this change. I also did other changes; we had a function
>> called many times, which handles subqueries and looked like
>> if(false cond)
>> do something;
>> if(other false cond)
>> do something;
>> etc;
>> return;
>> Adding to the start of the function:
>> "if (no subqueries in this query) return;"
>> saved 10% of the query's total time.
>
> but is this workload real? I mean, if its a code path that is supposed
> to be accessed millions of times, it makes sense to micro-optimize it.
> Not so much for cases of a bad query which leads to a "combinatorial
> explosion".

Anyway, never mind, it's a GCC bug anyway.. we shouldn't have to take 
hits just to workaround GCC bugs. I'm just grumpy that there is no nice 
workaround :/

Regards,

Davi
Thread
bzr commit into mysql-trunk-bugfixing branch (guilhem:3393) Bug#47609Guilhem Bichot30 Nov
Re: bzr commit into mysql-trunk-bugfixing branch (guilhem:3393) Bug#47609Davi Arnaut30 Nov
  • Re: bzr commit into mysql-trunk-bugfixing branch (guilhem:3393) Bug#47609Guilhem Bichot30 Nov
    • Re: bzr commit into mysql-trunk-bugfixing branch (guilhem:3393) Bug#47609Davi Arnaut30 Nov
      • Re: bzr commit into mysql-trunk-bugfixing branch (guilhem:3393) Bug#47609Guilhem Bichot30 Nov
        • Re: bzr commit into mysql-trunk-bugfixing branch (guilhem:3393) Bug#47609Davi Arnaut30 Nov
          • Re: bzr commit into mysql-trunk-bugfixing branch (guilhem:3393) Bug#47609Davi Arnaut30 Nov