[Pljava-dev] Jars as blobs

Has anyone considered storing jar files as blobs and a modified
classloader to read from this table? This would make deployment much
easier as file system access to the database would not be required. I
was just wanting to check with everyone first to make sure there is
not something obvious that I am missing before I start work. Any
ideas?

Thanks,
Stephen
(Please CC responses as I am not subscribed to the list)

Stephen Crowley wrote:

>Has anyone considered storing jar files as blobs and a modified
>classloader to read from this table? This would make deployment much
>easier as file system access to the database would not be required. I
>was just wanting to check with everyone first to make sure there is
>not something obvious that I am missing before I start work. Any
>ideas?
>
>
You are missing something obvious :-)
Jar files are already loaded and stored in the database today using the
install_jar, replace_jar, and remove_jar function calls. The only jar
that is stored in the filesystem is the pljava.jar. Among other things,
that jar contains the table based classloader which explains why it has
to be loaded from the file system.

The CLASSPATH visible to the postmaster (or configured using
pljava.classpath) should *only* appoint the pljava.jar. All other jars
(short of the ones in the JVM bootclasspath path of course) should be
loaded into the database.

Regards,
Thomas Hallgren

Oh. That would explain a lot :) Where is the jar data actually stored
when install_jar is called? Also, install_jar takes a file url as a
parameter so the file still has to be on the server. How much work
would it take to make install_jar accept a blob as a parameter so jars
could be installed remotely instead of copying files back to the
server each time?

--Stephen

On Mon, 07 Mar 2005 23:29:18 +0100, Thomas Hallgren
<thhal at mailblocks.com> wrote:
> Stephen Crowley wrote:
>
> >Has anyone considered storing jar files as blobs and a modified
> >classloader to read from this table? This would make deployment much
> >easier as file system access to the database would not be required. I
> >was just wanting to check with everyone first to make sure there is
> >not something obvious that I am missing before I start work. Any
> >ideas?
> >
> >
> You are missing something obvious :-)
> Jar files are already loaded and stored in the database today using the
> install_jar, replace_jar, and remove_jar function calls. The only jar
> that is stored in the filesystem is the pljava.jar. Among other things,
> that jar contains the table based classloader which explains why it has
> to be loaded from the file system.
>
> The CLASSPATH visible to the postmaster (or configured using
> pljava.classpath) should *only* appoint the pljava.jar. All other jars
> (short of the ones in the JVM bootclasspath path of course) should be
> loaded into the database.
>
> Regards,
> Thomas Hallgren
>
>

Stephen Crowley wrote:

>Oh. That would explain a lot :) Where is the jar data actually stored
>when install_jar is called?
>
In the tables sqlj.jar_repository and sqlj.jar_entry

> Also, install_jar takes a file url as a
>parameter so the file still has to be on the server.
>
No, the url is not limited to file. It might just as well be an url
starting with ftp:// or http://.

> How much work
>would it take to make install_jar accept a blob as a parameter so jars
>could be installed remotely instead of copying files back to the
>server each time?
>
>
Not much. It would be fairly simple to implement. Please add a feature
request on this.

Regards,
Thomas Hallgren