randgen team mailing list archive

["John H. Embretsen" <johnemb@xxxxxxxxx>]

On 10/26/10 16:17, Philip Stoev wrote:
> > > more robust ExecutionTimeComparator
> >
> > Robust... against what? Faulty queries?
>
> Yes, against trying to calculate the relative performance of invalid
> queries, including if the query is not valid for one of the servers
> being compared.

Thanks, makes sense.

> The other thing I tried to achieve is not report performance regressions
> where the result set has more rows than a certain threshold. Such
> queries spend considerable time just transfering the result rows along
> the stack, including DBI and Perl, which affects the measured running
> times. Also, they are inconvenient for developers to work with.

OK. I must remember to adjust this if I want to check execution time for 
bigger results then. But I'll keep in mind the overhead/penalty in that 
case.

> This has to eventually be made configurable. Are you using the
> ExecutionTimeComparator in your tests?

I just started playing with it. I expect to use it every now and then 
going forward.


> > > modified:
> > > lib/GenTest/Validator/ExecutionTimeComparator.pm
> >
> > Part of the diff:
> >
> > -------------------------------------------
> > @@ -47,6 +48,7 @@
> > my $query = $results->[0]->query();
> >
> > return STATUS_WONT_HANDLE if $query !~ m{^\s*SELECT}sio;
> > + return STATUS_WONT_HANDLE if $results->[0]->status() != STATUS_OK ||
> > $results->[0]->status() != STATUS_OK;
>
> You are right it must be [1] on the right side.

OK.

--
John