Marc,
All my experiments are just that - experiments. My goal is to enable DDL
generation in GlassFish when the persistence provider is set to that of the
OpenJPA and the user-visible behavior is similar to the one we have for CMP
(i.e. our internal code) and for TopLink Essentials as the persistence provider.
Foe now I'm exploring my options.
I'm using example (I started with the EE version, and now switched to the SE
version) from the GlassFish persistence example page
(https://glassfish.dev.java.net/javaee5/persistence/persistence-example.html).
The example has 2 entities, Customer and Order, and 1ToMany relationship between
them. Of course, I'm slightly modifying persistence.xml to make it work with the
OpenJPA.
In one of my experiments I created a customer table with columns A and B, A
being a PK before running the mapping tool. Mapping tool results were a bit
surprising ;).
thanks,
-marina
Marc Prud'hommeaux wrote:
On Mar 30, 2007, at 4:17 PM, Marina Vatkina wrote:
Marc,
I'd rather have provider-neutral code in GlassFish ;(.
Well, until the JPA spec (or some other spec) provides details on how
schema creation and/or migration should work, I don't think it is
realistic to expect every vendor to behave in exactly the same way
w.r.t. how their schema manipulation tools work. We're always open to
specific suggestions for how to improve how our operate, though.
The problem with adding new columns to the existing table in my
example below is a) in having a PK in the table that is not mapped to
any field the entity, and b) not making the entity's id a pk (I
didn't get any warning from not being able to add or create a PK -
does OpenJPA suppress them?).
I don't really understand. Are you saying you want to map to a table
that has a primary key, but you don't want to map any field in your
entity to that primary key? Is that is the case, how do you expect
primary key generation to work?
I wouldn't be surprised, however, if we don't actually validate that
what you declare to be a primary key in your entity is actually defined
as a primary key in the database. We don't require that the column
actually be declared to be a primary key in order for OpenJPA to work
correctly (although I can't envision any reason why you wouldn't want
to make it a proper primary key).
thanks,
-marina
Marc Prud'hommeaux wrote:
Marina-
The problem is that OpenJPA just ignores extra, unmapped columns.
Since we don't require that you map all of the columns of a
database table to an entity, tables can exist that have unmapped
columns. By default, we tend to err on the side of caution, so we
never drop tables or columns. The "deleteTableContents" flag merely
deletes all the rows in a table, it doesn't actually drop the table.
We don't have any options for asserting that the table is mapped
completely. That might be a nice enhancement, and would allows
OpenJPA to warn when it sees a existing table with unmapped columns.
You could manually drop the tables using the mappingtool by
specifying the "schemaAction" argument to "drop", but there's no
way to do it automatically using the SynchronizeMappings. Note that
there is nothing preventing you from manually invoking the
MappingTool class from any startup to glue code that you want.
On Mar 29, 2007, at 4:18 PM, Marina Vatkina wrote:
Marc, Patrick,
I didn't look into the file story yet, but what I've seen as the
result of using
<property name="openjpa.jdbc.SynchronizeMappings"
value="buildSchema
(SchemaAction='add,deleteTableContents')"/>
looks surprising: if I have there is an entity Foo with
persistence fields 'x' and 'y' and a table FOO already exists in
the database with columns A and B (there are no fields 'a' and 'b'
in the entity), the table is not recreated, but the columns X and
Y are added to the table FOO. The 'deleteTableContents' doesn't
affect this behavior.
Is it an expected behavior?
What should I use to either create the table properly or get a
message that such table already exist (and as in my case doesn't
match the entity)?
thanks,
-marina
Marina Vatkina wrote:
Then I'll first start with an easier task - check what happens in
EE if entities are not explicitly listed in the persistence.xml
file :).
thanks,
-marina
Marc Prud'hommeaux wrote:
Marina-
Let me give it a try. How would the persistence.xml property
look like to generate .sql file?
Actually, I just took a look at this, and it look like it isn't
possible to use the "SynchronizeMappings" property to
automatically output a sql file. The reason is that the
property takes a standard OpenJPA plugin string that configures
an instances of MappingTool, but the MappingTool class doesn't
have a setter for the SQL file to write out to.
So I think your only recourse would be to write your own adapter
to to this that manually creates a MappingTool instance and
runs it with the correct flags for outputting a sql file. Take
a look at the javadocs for the MappingTool to get started, and
let us know if you have any questions about proceeding.
On Mar 20, 2007, at 4:59 PM, Marina Vatkina wrote:
Marc,
Marc Prud'hommeaux wrote:
Marina-
They do in SE, but as there is no requirement to do it in
EE, people try to reduce the amount of typing ;).
Hmm ... we might not actually require it in EE, since we do
examine the ejb jar to look for persistent classes. I'm not
sure though.
You should test with both listing them and not listing them.
I'd be interested to know if it works without.
Let me give it a try. How would the persistence.xml property
look like to generate .sql file? Where will it be placed in
EE environment? Does it use use the name as-is or prepend it
with some path?
thanks.
On Mar 20, 2007, at 4:19 PM, Marina Vatkina wrote:
Marc,
Marc Prud'hommeaux wrote:
Marina-
On Mar 20, 2007, at 4:02 PM, Marina Vatkina wrote:
Marc,
Thanks for the pointers. Can you please answer the
following set of questions?
1. The doc requires that "In order to enable automatic
runtime mapping, you must first list all your persistent
classes". Is this true for EE case also?
Yes. People usually list them all in the <class> tags in
the persistence.xml file.
They do in SE, but as there is no requirement to do it in
EE, people try to reduce the amount of typing ;).
If OpenJPA can identify all entities in EE world, why can't
it do the same for the schema generation?
I'll check the rest.
thanks,
-marina
2. Section "1.2.Generating DDL SQL" talks about .sql
files, but what I am looking for are "jdbc" files, i.e.
files with the lines that can be used directly as
java.sql statements to be executed against database.
The output should be sufficient. Try it out and see if the
format is something you can use.
3. Is there a document that describes all possible values
for the "openjpa.jdbc.SynchronizeMappings" property?
Unfortunately, no. Basically, the setting of the
"SynchronizeMappings" property will be of the form "action
(Bean1=value1,Bean2=value2)", where the "bean" values are
those listed in org.apache.openjpa.jdbc.meta.MappingTool
(whose javadoc you can see http://incubator.apache.org/
openjpa/ docs/ latest/ javadoc/org/
apache/openjpa/jdbc/meta/ MappingTool.html ).
thank you,
-marina
Marc Prud'hommeaux wrote:
Marina-
On Mar 15, 2007, at 5:01 PM, Marina Vatkina wrote:
Hi,
I am part of the GlassFish persistence team and was
wondering how does OpenJPA support JPA auto DDL
generation (we call it "java2db") in a Java EE
application server.
Our application server supports java2db via creating two
sets of files for each PU: a ...dropDDL.jdbc and a
...createDDL.jdbc file on deploy (i.e. before the
application is actually loaded into the container)
and then executing 'create' file as the last step
in deployment, and 'drop' file on undeploy or the 1st
step in redeploy. This allows us to drop tables
created by the previous deploy operation.
This approach is done for both, the CMP and the default
JPA provider. It would be nice to add java2db support
for OpenJPA as well, and I'm wondering if we need to
do anything special, or it'll all work just by itself?
We do have support for runtime creation of the schema via
the "openjpa.jdbc.SynchronizeMappings" property. It is
described at:
http://incubator.apache.org/openjpa/docs/latest/
manual/ manual.html#ref_guide_mapping_synch
The property can be configured to run the mappingtool
(also described in the documentation) at runtime
against all the registered persistent classes.
Here are my 1st set of questions:
1. Which API would trigger the process, assuming the
correct values are specified in the persistence.xml
file? Is it:
a) <provider>.createContainerEntityManagerFactory(...)? or
b) the 1st call to emf.createEntityManager() in this VM?
c) something else?
b
2. How would a user drop the tables in such environment?
I don't think it can be used to automatically drop then
create tables. The "mappingtool" can be executed
manually twice, the first time to drop all the tables,
and the second time to re- create them, but I don't
think it can be automatically done at runtime with
the "SynchronizeMappings" property.
3. If the answer to either 1a or 1b is yes, how does the
code distinguish between the server startup time and
the application being loaded for the 1st time?
That is one of the reasons why we think it would be
inadvisable to automatically drop tables at runtime :)
4. Is there a mode that allows creating a file with the
jdbc statements to create or drop the tables and
constraints?
Yes. See:
http://incubator.apache.org/openjpa/docs/latest/
manual/ manual.html#ref_guide_ddl_examples
thank you,
-marina