cayenne-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrus Adamchik <and...@objectstyle.org>
Subject Re: Problems with MySql id type with AUTO_INCREMENT
Date Fri, 29 Sep 2006 15:20:06 GMT
Yeah, there is quite a few bugs involved.

First an explanation on Cayenne strategies... Long works with  
auto_increment as long as the driver supports it, but doesn't not  
work with any other strategy, as all of them are based on overriding  
"int JdbcPkGenerator.pkFromDatabase(..)".

What we can do about it now... The simplest way seems to use INT PK  
with "auto_pk_table" generation strategy and a custom PkGenerator  
that doesn't perform locking.

Here is a mailing list thread on implementing an "optimistic" pk for  
MySQL:

http://objectstyle.org/cayenne/lists/cayenne-user/2006/07/0067.html

Wouldn't mind including this in Cayenne at some point.

Andrus


On Sep 29, 2006, at 11:03 AM, Tobias SCHOESSLER wrote:
> I am a bit confused now ... so java.lang.Long PKs are known to not  
> work at
> all with cayenne 1.2 no matter which database is used?
>
> I had the impression everything  worked fine when using MySQL  
> directly with
> BIGINT, Long and database side AUTO_INCREMENT.
>
> My last bug I ran into was only when I connected through the  
> sequoia jdbc
> driver / mysql jdbc driver chain when the test case broke again ...
>
> Anyway I am flexible with the pk column types but because of this  
> MySql
> jdbc driver bug mentioned before
>
> http://lists.mysql.com/java/6383
>
> - returning meta data of type java.lang.Long for any AUTO_INCREMENT  
> column
> defined -
>
> I thought I was forced to use some database type that consistently  
> maps as
> java.lang.Long for the PK to not have to deal with changing pk  
> types on
> creation and retrieval . That's why i tried INT UNSIGNED and BIGINT.
>
> Is your advise still correct for this ?
>
>
>
>
>
>
>              Andrus Adamchik
>              <andrus@objectsty
>               
> le.org>                                                    To
>                                        cayenne- 
> user@incubator.apache.org
>              Friday,  
> 29                                                 cc
>              September 2006
>              15:57                                                  
> Subject
>                                        Re: Problems with MySql id type
>                                        with AUTO_INCREMENT
>              Please respond to
>              cayenne-user@incu
>              bator.apache.org
>
>
>
>
>
>
>
> Tobias,
>
> If you have flexibility redefining your columns, I suggest mapping a
> PK as INTEGER until we implement Long PK support for real (http://
> issues.apache.org/cayenne/browse/CAY-329). As far as I can tell in
> the your latest case you are hitting the int pk limitation of PK
> generator. This issue is pretty high on our priority list by the way,
> so Long pks should become an option soon.
>
> Andrus
>
>
> On Sep 29, 2006, at 9:51 AM, Tobias SCHOESSLER wrote:
>
>> Andrus, thanks a lot  you for looking into this.
>>
>> That is quite a pile of bugs I am running into here. First this
>> MySQL JDBC
>> driver bug returning java.lang.Long for any auto-increment column,
>> then
>> cayenne treating 'INT UNSIGNED' as an Integer. Now issues with
>> org.apache.cayenne.access.jdbc.ColumnDescriptor and SQLTemplate.
>>
>> To get this running for me I understand that I only have to pick
>> the right
>> datatype for my PKs so all the bugs are bypassed. Looking at
>> http://dev.mysql.com/doc/refman/5.0/en/connector-j-reference-type-
>> conversions.html
>> I thought BIGINT should be unambigiously treated as a
>> java.lang.Long for
>> all bugs mentioned.
>>
>> Actually it worked, I tried setting the PKs to mysql type BIGINT
>> and the
>> test case mentioned below passes. So for the record this is a
>> solution for
>> using AUTO_INCREMENT PK generation with cayenne 1.2 and MySQL.
>>
>> But...
>>
>> If you recall my ultimate goal was to use a setup in which MySql is
>> used in
>> a Sequoia/CJDBC cluster - the reason I did not use AUTO_PK_SUPPORT
>> as LOCK
>> was not supported by Sequoia ...
>>
>> Connecting through the cjdbc/sequoia driver to the database above,
>> again
>> breaks the test case. Sequoia returns java.lang.Integer for the  
>> BIGINT
>> column ... : -[
>>
>> I guess this is rather something to address in a sequoia forum or
>> is there
>> a chance you add a sequoia setup in your test cases?
>>
>> thanks
>>
>> Tobias
>>
>>
>>
>>
>>              Andrus Adamchik
>>              <andrus@objectsty
>>
>> le.org>                                                    To
>>                                        cayenne-
>> user@incubator.apache.org
>>              Thursday,
>> 28                                               cc
>>              September 2006
>>              17:44
>> Subject
>>                                        Re: Problems with MySql id  
>> type
>>                                        with AUTO_INCREMENT
>>              Please respond to
>>              cayenne-user@incu
>>              bator.apache.org
>>
>>
>>
>>
>>
>>
>>
>> Just got some time to test it, and the problem seems to be in
>> org.apache.cayenne.access.jdbc.ColumnDescriptor constructor... It is
>> getting confused by MySQL ResultSetMetaData returning JDBC type for
>> such column as INTEGER.
>>
>> I created a new Jira issue with description of this problem:
>>
>> http://issues.apache.org/cayenne/browse/CAY-672
>>
>> I'll commit a fix to 3.0, as 1.2 users can have some code relying on
>> the old behavior and we want to keep 1.2 stable.
>>
>> Andrus
>>
>>
>> On Sep 26, 2006, at 7:11 AM, Tobias SCHOESSLER wrote:
>>> Andrus,
>>>
>>> I am referring to http://issues.apache.org/cayenne/browse/CAY-639
>>> and our
>>> discussion below.
>>>
>>> I am using cayenne 1.2.1 now. I mapped all AUTO_INCREMENTED primary
>>> keys of
>>> sql type unsigned int as BIG_INT in cayenne.
>>>
>>> Cayenne correctly returns DataObjects with a primary key of type
>>> Long when
>>> using SelectQuery, when creating new objects or using getToXXX
>>> navigation
>>> to retrieve the object .
>>>
>>> But when I use SQLTemplate to retrieve the same DataObject cayenne
>>> still
>>> returns a pk of type Integer ...
>>>
>>> These are my test cases:
>>>
>>>
>>>              public void testDocumentPkViaSelectQuery() {
>>>                   List<TldDocument> docList =
>>> dataContext.performQuery(new
>>> SelectQuery(TldDocument.class, null));
>>>                   assertEquals(Long.class,
>>> docList.get(0).getObjectId().getIdSnapshot().get
>>> (TldDocument.DOCUMENT_UID_PK_COLUMN).getClass());
>>>
>>>              }
>>>
>>>              public void testDocumentPkViaSQLTemplate() {
>>>
>>>                   String sqlSelect = " select * ";
>>>                   String sqlFrom = " from TLD_DOCUMENT doc ";
>>>
>>>                   SQLTemplate query = new SQLTemplate
>>> (TldDocument.class,
>>> sqlSelect + " " + sqlFrom);
>>>                   List<TldDocument> docList =
>>> dataContext.performQuery(query);
>>>
>>>                   assertEquals(Long.class,
>>> docList.get(0).getObjectId().getIdSnapshot().get
>>> (TldDocument.DOCUMENT_UID_PK_COLUMN).getClass());
>>>
>>>              }
>>>
>>> testDocumentPkViaSelectQuery passes.
>>>
>>> testDocumentPkViaSQLTemplate fails.
>>>
>>> thanks, regards
>>>
>>> Tobias.
>>>
>>>
>>>
>>>
>>>
>>>              Andrus Adamchik
>>>              <andrus@objectsty
>>>
>>> le.org>                                                    To
>>>                                        cayenne-
>>> user@incubator.apache.org
>>>              Tuesday,
>>> 29                                                cc
>>>              August 2006 15:35
>>>
>>> Subject
>>>                                        Re: Problems with MySql id
>>> type
>>>              Please respond to         with AUTO_INCREMENT
>>>              cayenne-user@incu
>>>              bator.apache.org
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> I'd appreciate that.
>>>
>>> Andrus
>>>
>>>
>>> On Aug 29, 2006, at 5:31 PM, Tobias SCHOESSLER wrote:
>>>
>>>> ok, I got it. You want me to add a Jira entry for this? I never did
>>>> this
>>>> before ... : |
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>              Andrus Adamchik
>>>>              <andrus@objectsty
>>>>
>>>> le.org>                                                    To
>>>>                                        cayenne-
>>>> user@incubator.apache.org
>>>>              Tuesday,
>>>> 29                                                cc
>>>>              August 2006 15:19
>>>>
>>>> Subject
>>>>                                        Re: Problems with MySql id
>>>> type
>>>>              Please respond to         with AUTO_INCREMENT
>>>>              cayenne-user@incu
>>>>              bator.apache.org
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> On Aug 29, 2006, at 4:16 PM, Tobias SCHOESSLER wrote:
>>>>
>>>>> But still you are right there is a bug in mysql too. If the column
>>>>> type is
>>>>> INT and not INT UNSIGNED it should return java.lang.Integer but it
>>>>> returns
>>>>> java.lang.Long instead. This was my first observation.
>>>>
>>>> Still probably worth fixing the mapping to follow the docs.
>>>>
>>>>
>>>>> but why does cayenne need BIGINT mapping set for the column to
>>>>> retrieve it
>>>>> as a java.lang.Long? I understand that this is unrelated to the  
>>>>> bug
>>>>> in the
>>>>> mysql jdbc driver mentioned above.
>>>>
>>>> This is simply a workaround for the bug. BIGINT has a default JDBC
>>>> mapping to java.lang.Long. So this consistently produced a Long
>>>> during select and when reading auto-incremented value via JDBC.
>>>>
>>>> Andrus
>>>>
>>>>
>>>>
>>>
>>>
>>>
>>>
>>
>>
>>
>>
>
>
>
>


Mime
View raw message