[Monetdb-developers] Key constraints and optimizers...

Martin Kersten Martin.Kersten at cwi.nl
Tue Jul 31 08:16:02 CEST 2007


Dear all,

Indeed, M5's partition manager BPM and its associated optimizer are
being refined/debugged.

The partition handler is still in great flux. The target is to have
the complete SQL test run with a BATs composed of a single partition.
The next stage then is to break each BAT into two pieces as a
prelude for adaptive partitioning.

Stefan Manegold wrote:
> On Tue, Jul 31, 2007 at 02:25:04AM +0200, Romulo Goncalves wrote:
>> Dear developers....
>>
>> I was checking the status of the SQL test web-site (cvs branch) and I 
>                                                      ^^^^^^^^^^^^
> guess you mean CVS development trunk (i.e., "Current"), right?
> (we use "branch" usually only for the "Stable" release branch ...)
> 
>> found some red ones.
>>
>> In the test src/test/BugConstraint/delete_primarykey_1643845.sql
>>
>> If I run the normal Mtest the test is red (the err output).
>>
>> The constraint error given is for the foreign key and not for the 
>> primary key (as in the stable output).
>> If I ran with M4 the test is green the same happens if I set the 
>> optimizers off.
>>
>> So we can conclude that the reason for this test being red are the 
>> optimizers and nothing else in SQL module.
> 
> Most failing SQL/5 tests are due to Martin's recent partition optimizer
> changes on Friday Jul 27 2007 --- just like the compilation problems in M5
> ---; hence, yes it's M5 only, and related to / cause by optimizers. Guessing
> from Martin's checkin comments I think he's aware of these problems...
> 
>> I tried to give a look to see which optimizer is breaking the test, but 
>> no results. Tomorrow I will continue.
>>
>> I also think that all the tests that are red (the ones related with 
>> primary keys) are due this problem.
>>
>> Should I open a bug or this is enough? Because it is an existent test 
>> that is red ;)
> 
> feel free.
> 
> Stefan
> 
>> Regards,
>> Romulo
> 





More information about the developers-list mailing list