You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@geronimo.apache.org by Jeremy Boynes <je...@coredevelopers.net> on 2004/01/26 01:52:35 UTC

Clarification on 88 anyone?

A couple of things in the JSR 88 API don't make sense to me - if anyone 
has definitive answers I would appreciate it.

On distribute(), what exactly is the "deploymentPlan" passed in? I am 
assuming that it is the tool's responsibility to pass in a 
File/InputStream containing the information written by 
DeploymentConfiguraton.save(OutputStream).

If this is the case, how is the information for an EAR module tied 
together?

As I understand EAR deployment:
* the tool constructs a J2eeApplicationObject from the EAR's DD
* the tool invokes DeploymentManager.createConfiguration() to
   construct the DeploymentConfiguration for the EAR
* for each DeployableObject (sub-module) in the EAR, the tool calls
   DeploymentConfiguration.getDConfigBeanRoot(DDBeanRoot) to get the
   DConfigBeanRoot that it uses to configure the sub-module

At this point things get fuzzy. Presumably the tool iterates over all 
these DConfigBeanRoot's and uses saveDConfigBean(OutputStream, 
DConfigBeanRoot) to save the configuration information. However, the 
tool gets to choose the location where the information is written (as it 
creates the OutputStream).

Now when the tool goes to distribute this application, it would pass the 
plugin the File/InputStream for the EAR and the File/InputStream for the 
EAR's deployment plan. How does it communicate to the plugin the 
location of the DD's for all the sub-modules?

-- 
Jeremy


Re: Clarification on 88 anyone?

Posted by Jeremy Boynes <je...@coredevelopers.net>.
Aaron Mulder wrote:

> On Mon, 26 Jan 2004, Jeremy Boynes wrote:
> 
>>So for 88 to be useful, the vendor must support having all their DD 
>>information in one file? But the RI does not that...
> 
> 
> 	Now you are being dense.  

Seemed so, so Aaron and I had a not-so-quick chat. Summary says:

88 assumes that all configuration information is saved in one file 
created by DeploymentConfiguration.save(OutputStream). This should 
contain all information needed by the plugin/server to convert a 'naked' 
input file into a runnable configuration.

This is a change from the way most vendors currently store information 
in the module files themselves - a 'dressed' module.

Backward compatibility can be obtained by having the plugin read 
additional information out the 'dressed' source during distribute. We 
could support this style of deployment by:
* reading the deployment plan supplied to distribute
* seeing that the plan did not contain our information for a module
* looking in the module archive for a vendor dd
* picking suitable defaults if they are missing

What we don't get is vendor neutral conversion from old-style dressed 
archives to the new 'naked' + stack of clothes model. A tool could do 
that but it would up to the tool.

Although it does not mandate it, the spec recommends that the format of 
the deployment plan is XML. If we assume that, then we can use the 
current method of determining the module type from the deployment plan - 
this parses the stream and uses the root element to decide.

I think this is workable. Now we need to define a mechanism whereby 
module deployment plugins can hook their DConfigBeans into the 
deployment core and generate an XML form of their configuration. We also 
need to provide a mechanism for merging these together when the modules 
are packaged together into an EAR.

--
Jeremy

Re: Clarification on 88 anyone?

Posted by Dain Sundstrom <da...@coredevelopers.net>.
On Jan 26, 2004, at 3:43 PM, Aaron Mulder wrote:

> On Mon, 26 Jan 2004, Dain Sundstrom wrote:
>> I would prefer the tool ask the provider for the desired location for
>> the streams to be stored, which would also allow the provider to have
>> the file name end with .xml, and allow the configurations to be
>> seamlessly portable between tools.
>
> 	But the streams are only ever stored on the user's disk, and the
> provider cannot select an appropriate location there.  The storage is 
> only
> as a convenience for the user (as demonstrated by your EJB-to-EAR
> example).  When just configuring and deploying, the plan is never 
> saved to
> disk, only generated from the DConfigBeans and passed directly to the
> DeploymentManager implementation.

That is just what I prefer, but heck the spec is final.  I would bet 
the normal case is the user/developer saves it to disk every time, 
otherwise every time they want to (re)deploy they would need to do all 
the configuration again.

>> I'll chat with Jeremy later.  I really want to see this spec work,
>> which means the vendors need to actually try to support this (and not
>> take the easy out), and the end users need to find the tools useable.
>> I'm concerned that the non-vendor supplied tools will be clunky, so
>> people will only use vendor tools.  I hope it works out, but I'm
>> currently fairly pessimistic.  Anyway, I am confident this will work
>> for us as we are the vendor and can supply our own tools.
>
> 	Netbeans has a tool already.  I don't think it's unnecessarily
> clunky.  In fact, I don't see what advantages a vendor's own tool could
> really have here.  I guess I'll just have to show you a good tool and 
> then
> you'll believe me.  :)

I'll have to see.  I wonder how well it works with the other vendor's 
servers.

> 	Let me ask you this: would the ideal tool for you be:
>
>  - command-line
>  - web-based
>  - standalone Swing
>  - Swing IDE plug-in

An idea tool for me would be ant.  Maybe a swing tool to edit the 
config and an ant task to replay the config to a target server.

-dain


Re: Clarification on 88 anyone?

Posted by David Jencks <da...@coredevelopers.net>.
On Monday, January 26, 2004, at 01:43 PM, Aaron Mulder wrote:

> <snip>
> 	Netbeans has a tool already.

can you point me to it?  All I can find is a DD tool for web apps that 
does not appear to be related to jsr 88.

http://web.netbeans.org/doc/ddapi_architecture.html#compat

Thanks
david jencks


>  I don't think it's unnecessarily
> clunky.  In fact, I don't see what advantages a vendor's own tool could
> really have here.  I guess I'll just have to show you a good tool and 
> then
> you'll believe me.  :)
>
> 	Let me ask you this: would the ideal tool for you be:
>
>  - command-line
>  - web-based
>  - standalone Swing
>  - Swing IDE plug-in
>
> Aaron
>


Re: Clarification on 88 anyone?

Posted by Aaron Mulder <am...@alumni.princeton.edu>.
On Mon, 26 Jan 2004, Dain Sundstrom wrote:
> I would prefer the tool ask the provider for the desired location for 
> the streams to be stored, which would also allow the provider to have 
> the file name end with .xml, and allow the configurations to be 
> seamlessly portable between tools.

	But the streams are only ever stored on the user's disk, and the
provider cannot select an appropriate location there.  The storage is only 
as a convenience for the user (as demonstrated by your EJB-to-EAR 
example).  When just configuring and deploying, the plan is never saved to 
disk, only generated from the DConfigBeans and passed directly to the 
DeploymentManager implementation.

> I'll chat with Jeremy later.  I really want to see this spec work, 
> which means the vendors need to actually try to support this (and not 
> take the easy out), and the end users need to find the tools useable.  
> I'm concerned that the non-vendor supplied tools will be clunky, so 
> people will only use vendor tools.  I hope it works out, but I'm 
> currently fairly pessimistic.  Anyway, I am confident this will work 
> for us as we are the vendor and can supply our own tools.

	Netbeans has a tool already.  I don't think it's unnecessarily 
clunky.  In fact, I don't see what advantages a vendor's own tool could 
really have here.  I guess I'll just have to show you a good tool and then 
you'll believe me.  :)

	Let me ask you this: would the ideal tool for you be:

 - command-line
 - web-based
 - standalone Swing
 - Swing IDE plug-in

Aaron


Re: Clarification on 88 anyone?

Posted by Dain Sundstrom <da...@coredevelopers.net>.
On Jan 26, 2004, at 2:35 PM, Aaron Mulder wrote:

> On Mon, 26 Jan 2004, Dain Sundstrom wrote:
>> But they were willing to be restricted to one monolithic file for the
>> entire ear?  I am surprised.  Unless they all plan on doing what the 
>> RI
>> does, which is return null.
>
> 	If I may be so bold as to paraphrase, "I don't like your stupid
> spec, so I won't implement it!  Phooey on you!"

It appears that is exactly what the RI has done, and this is a fairly 
common practice with specifications.  I've seen things like "we 
implement JAAS, but you can plug the login modules", which is 
completely useless.

> 	Imagine the alternative.  The user spends 10 minutes configuring
> all the deployment information for the 20 modules in the EAR.  They hit
> "save".  The tool then opens 20 dialog boxes, one per module, so they 
> can
> save the deployment info for every module into a separate file.  It 
> takes
> longer to navigate the save dialogs than to perform the actual
> configuration.  This is better?

I would prefer the tool ask the provider for the desired location for 
the streams to be stored, which would also allow the provider to have 
the file name end with .xml, and allow the configurations to be 
seamlessly portable between tools.

> P.S. Jeremy and I talked by phone and worked though our differences on
> this as well as the thread about deployment info being the 
> responsibility
> of the different Web/EJB/etc. implementation.  His summary essay will 
> be
> forthcoming, though perhaps not rapidly.  :)

I'll chat with Jeremy later.  I really want to see this spec work, 
which means the vendors need to actually try to support this (and not 
take the easy out), and the end users need to find the tools useable.  
I'm concerned that the non-vendor supplied tools will be clunky, so 
people will only use vendor tools.  I hope it works out, but I'm 
currently fairly pessimistic.  Anyway, I am confident this will work 
for us as we are the vendor and can supply our own tools.

-dain


Re: Clarification on 88 anyone?

Posted by Aaron Mulder <am...@alumni.princeton.edu>.
On Mon, 26 Jan 2004, Dain Sundstrom wrote:
> > 	Now you are being dense.  You have to be able to transport it over
> > the network in one stream. Is that so onerous?
> 
> I disagree.  We write the provider layer and we may send the 
> information to the server in several disconnected streams.  The tool 
> frankly has no idea what we are doing with the stream.  We can do 
> everything you mention below in the provider layer on the client side, 
> which means we don't have to be able to transport it in one stream.

	You're right, my language was misleading.  I should perhaps have
said, "according to the spec, you must hand the deployment plan from the
tool to the server plugin in one stream.  Is that so onerous?"

> So the common developer suffers from some person sitting in a ivory 
> tower?  What I see as the major flaw is you now have tool locking.  
> Once you configure an application with tool X you can not convert to 
> tool Y without starting over.  Heck you can't even configure an ejb-jar 
> standalone (for say isolated testing) and then move that configuration 
> into an ear; you get to start over.  For production environments, you 
> can't build a configuration try it in a staged environment and them 
> move the file over the the production environment.  No you must be 
> on-line.  Now all the above is possible with a perfect tool, but then 
> you have tool lock-in.

	I don't see any tool locking going on.  The tool can ask the 
server to save deployment information to a file or load it from a file.  
The only tool-specific part of this operation is choosing a file name.  
Persumably we can rely on all tools to include a file chooser.

	As far as configuring an EJB JAR and later incorporating it (with
its deployment information) into the EAR, it's easy:

 - user configures JAR deployment info
 - user hits "save", picks file name, info saved to file
 - user hits "deploy", EJB JAR & deployment info sent to server
 - user tests EJB JAR (likely outside scope of tool)
 - user hits "open", picks EAR file, new EAR loaded, previous DD gone
 - user selects node for EJB JAR in EAR tree
 - user select "load", picks file name (file saved above)
 - DConfigBeans for EJB JAR are populated from saved file
 - user repeats for any other modules with pre-saved deployment info
 - user manually populates DConfigBeans for any additional modules
 - user hits "deploy", EAR and all deployment info send to server
 - (optional: user hits "save", all deployment info saved to 1 file)

> But they were willing to be restricted to one monolithic file for the 
> entire ear?  I am surprised.  Unless they all plan on doing what the RI 
> does, which is return null.

	If I may be so bold as to paraphrase, "I don't like your stupid
spec, so I won't implement it!  Phooey on you!"

	Imagine the alternative.  The user spends 10 minutes configuring 
all the deployment information for the 20 modules in the EAR.  They hit 
"save".  The tool then opens 20 dialog boxes, one per module, so they can 
save the deployment info for every module into a separate file.  It takes 
longer to navigate the save dialogs than to perform the actual 
configuration.  This is better?

Aaron

P.S. Jeremy and I talked by phone and worked though our differences on 
this as well as the thread about deployment info being the responsibility 
of the different Web/EJB/etc. implementation.  His summary essay will be 
forthcoming, though perhaps not rapidly.  :)


Re: Clarification on 88 anyone?

Posted by Dain Sundstrom <da...@coredevelopers.net>.
On Jan 26, 2004, at 12:32 PM, Aaron Mulder wrote:

> On Mon, 26 Jan 2004, Jeremy Boynes wrote:
>> So for 88 to be useful, the vendor must support having all their DD
>> information in one file? But the RI does not that...
>
> 	Now you are being dense.  You have to be able to transport it over
> the network in one stream.

I disagree.  We write the provider layer and we may send the 
information to the server in several disconnected streams.  The tool 
frankly has no idea what we are doing with the stream.  We can do 
everything you mention below in the provider layer on the client side, 
which means we don't have to be able to transport it in one stream.

> Is that so onerous?  Whatever format you use
> for that, you would probably also use for your save files.  The
> information does not need to permanently live in one file on the server
> side.  You can split it out into one file per byte once it gets to the
> server side.  Or stuff in in a database.  Or update the EAR to contain 
> it.
> Or whatever.
>
>> Hmmm. Most vendors have implemented a mechanism where the DD 
>> information
>> for a module is in that module as one or more XML files in META-INF; 
>> and
>> for EARs, the top level file describes the EAR itself, the module 
>> level
>> files are in each nested module archive's META-INF.
>
> 	The J2EE spec team wanted to explicitly discourage developers from
> putting deployment information into the EAR.  Why that is, I couldn't 
> say.
> Thus, we have all this plumbing so the server can automagically handle 
> the
> deployment information without the user even needing to manually 
> generate,
> copy, or package it.

So the common developer suffers from some person sitting in a ivory 
tower?  What I see as the major flaw is you now have tool locking.  
Once you configure an application with tool X you can not convert to 
tool Y without starting over.  Heck you can't even configure an ejb-jar 
standalone (for say isolated testing) and then move that configuration 
into an ear; you get to start over.  For production environments, you 
can't build a configuration try it in a staged environment and them 
move the file over the the production environment.  No you must be 
on-line.  Now all the above is possible with a perfect tool, but then 
you have tool lock-in.

Of course for us, this is not a huge problem, because we will write the 
tool.  And we can just tell everyone to use our tool, but I doubt that 
was the goal of the spec.

>> The only ones I can think of that are different are the old 
>> (deprecated)
>> RI and maybe WebSphere with it's separate repository.
>
> 	Well, one counter-case is enough.  The vendors didn't want to be
> restricted to a particular format for deployment information, even one 
> as
> common as XML.  So it's a file/stream, with which you can do what you
> will.

But they were willing to be restricted to one monolithic file for the 
entire ear?  I am surprised.  Unless they all plan on doing what the RI 
does, which is return null.

-dain


Re: Clarification on 88 anyone?

Posted by Aaron Mulder <am...@alumni.princeton.edu>.
On Mon, 26 Jan 2004, Jeremy Boynes wrote:
> So for 88 to be useful, the vendor must support having all their DD 
> information in one file? But the RI does not that...

	Now you are being dense.  You have to be able to transport it over
the network in one stream.  Is that so onerous?  Whatever format you use
for that, you would probably also use for your save files.  The
information does not need to permanently live in one file on the server
side.  You can split it out into one file per byte once it gets to the
server side.  Or stuff in in a database.  Or update the EAR to contain it.  
Or whatever.

> Hmmm. Most vendors have implemented a mechanism where the DD information 
> for a module is in that module as one or more XML files in META-INF; and 
> for EARs, the top level file describes the EAR itself, the module level 
> files are in each nested module archive's META-INF.

	The J2EE spec team wanted to explicitly discourage developers from
putting deployment information into the EAR.  Why that is, I couldn't say.  
Thus, we have all this plumbing so the server can automagically handle the 
deployment information without the user even needing to manually generate, 
copy, or package it.

> The only ones I can think of that are different are the old (deprecated) 
> RI and maybe WebSphere with it's separate repository.

	Well, one counter-case is enough.  The vendors didn't want to be 
restricted to a particular format for deployment information, even one as 
common as XML.  So it's a file/stream, with which you can do what you 
will.

Aaron



Re: Clarification on 88 anyone?

Posted by Jeremy Boynes <je...@coredevelopers.net>.
Aaron Mulder wrote:
> On Mon, 26 Jan 2004, Jeremy Boynes wrote:
> 
>>Sorry if I'm being dense, but how does the tool merge together all the 
>>server specific information? For example, there could be several files 
>>generated by different calls to save(OutputStream) and 
>>save(OutputStream, DConfigBeanRoot) for the different files in the 
>>module (I think the example is the primary DD and those associated with 
>>web services).
>>
>>However, when the tool goes to distribute this module it only gets to 
>>pass in one file.
> 
> 
> 	The tool cannot call save on individual DConfigBeans and then
> merge the save files.  It must call the save that takes no arguments if
> its going to deploy an EAR.  That method saves all the deployment info for
> the EAR to one file.  The individual save(DConfigBean)s could be used for
> deploying only one module, or for the user to save their progress on
> individual modules separately.  OK, it's not terribly useful.
> 

So for 88 to be useful, the vendor must support having all their DD 
information in one file? But the RI does not that...

> 
>>So the spec forces the vendor's to save all configuration for an EAR in 
>>one file, just like the old RI did? Does it not allow tools to use 
>>per-module DD's like BEA, JBoss, and yes SunOne/the new RI?
>>
>>Hmm, that may be why the new RI does not appear to support the 
>>configuration part of the spec - it returns null from 
>>getDConfigBeanRoot() :-(
> 
> 
> 	The server can store the data in whatever format it wants on its
> side of the network.  Some products use XML (ok, most), some use a
> database, some used serialized thingys, etc.  But the server has to
> transfer the data over the network as a single stream.  Is this such a big
> deal?  How about a serialized map where the key is some combination of
> module name and/or DD name and the value is the XML DD?  If not, how about
> just writing a bunch of DDs in a row to the stream?  If not, how about one
> big DD that holds all the little DDs?  There are plenty of ways to 
> implement this.  In the long run, all the data is going over one network 
> stream anyway; it's just a question of at what stage it gets combined.  :)
> 

Hmmm. Most vendors have implemented a mechanism where the DD information 
for a module is in that module as one or more XML files in META-INF; and 
for EARs, the top level file describes the EAR itself, the module level 
files are in each nested module archive's META-INF.

The only ones I can think of that are different are the old (deprecated) 
RI and maybe WebSphere with it's separate repository.

-- 
Jeremy

Re: Clarification on 88 anyone?

Posted by Aaron Mulder <am...@alumni.princeton.edu>.
On Mon, 26 Jan 2004, Jeremy Boynes wrote:
> Sorry if I'm being dense, but how does the tool merge together all the 
> server specific information? For example, there could be several files 
> generated by different calls to save(OutputStream) and 
> save(OutputStream, DConfigBeanRoot) for the different files in the 
> module (I think the example is the primary DD and those associated with 
> web services).
> 
> However, when the tool goes to distribute this module it only gets to 
> pass in one file.

	The tool cannot call save on individual DConfigBeans and then
merge the save files.  It must call the save that takes no arguments if
its going to deploy an EAR.  That method saves all the deployment info for
the EAR to one file.  The individual save(DConfigBean)s could be used for
deploying only one module, or for the user to save their progress on
individual modules separately.  OK, it's not terribly useful.

> So the spec forces the vendor's to save all configuration for an EAR in 
> one file, just like the old RI did? Does it not allow tools to use 
> per-module DD's like BEA, JBoss, and yes SunOne/the new RI?
> 
> Hmm, that may be why the new RI does not appear to support the 
> configuration part of the spec - it returns null from 
> getDConfigBeanRoot() :-(

	The server can store the data in whatever format it wants on its
side of the network.  Some products use XML (ok, most), some use a
database, some used serialized thingys, etc.  But the server has to
transfer the data over the network as a single stream.  Is this such a big
deal?  How about a serialized map where the key is some combination of
module name and/or DD name and the value is the XML DD?  If not, how about
just writing a bunch of DDs in a row to the stream?  If not, how about one
big DD that holds all the little DDs?  There are plenty of ways to 
implement this.  In the long run, all the data is going over one network 
stream anyway; it's just a question of at what stage it gets combined.  :)

Aaron


Re: Clarification on 88 anyone?

Posted by Jeremy Boynes <je...@coredevelopers.net>.
Aaron Mulder wrote:

> On Sun, 25 Jan 2004, Jeremy Boynes wrote:
> 
>>A couple of things in the JSR 88 API don't make sense to me - if anyone 
>>has definitive answers I would appreciate it.
>>
>>On distribute(), what exactly is the "deploymentPlan" passed in? I am 
>>assuming that it is the tool's responsibility to pass in a 
>>File/InputStream containing the information written by 
>>DeploymentConfiguraton.save(OutputStream).
> 
> 
> 	It is all the server-specific deployment information.  What 
> traditionally goes in the server DDs.  But not all servers use DDs, so XML 
> is not mandated here.  It can be binary info or XML or whatever.
> 
> 

Sorry if I'm being dense, but how does the tool merge together all the 
server specific information? For example, there could be several files 
generated by different calls to save(OutputStream) and 
save(OutputStream, DConfigBeanRoot) for the different files in the 
module (I think the example is the primary DD and those associated with 
web services).

However, when the tool goes to distribute this module it only gets to 
pass in one file.

This seems broken:
* during configuration, the tool supplies multiple distinct
   OutputStreams and the plugin has no knowledge of where the data gets
   written
* during distribute, the tool can only supply one InputStream
   and the plugin has no idea where to get the remaining data from.
   If the tool merges the multiple streams together, the plugin does
   not know how they were merged

And this is just for a single module - more on EAR's below.

>>If this is the case, how is the information for an EAR module tied 
>>together?
> 
> 
> 	More on this later.
> 
> 
>>As I understand EAR deployment:
>>* the tool constructs a J2eeApplicationObject from the EAR's DD
>>* the tool invokes DeploymentManager.createConfiguration() to
>>   construct the DeploymentConfiguration for the EAR
>>* for each DeployableObject (sub-module) in the EAR, the tool calls
>>   DeploymentConfiguration.getDConfigBeanRoot(DDBeanRoot) to get the
>>   DConfigBeanRoot that it uses to configure the sub-module
> 
> 
> 	Yes.
> 
> 
>>At this point things get fuzzy. Presumably the tool iterates over all 
>>these DConfigBeanRoot's and uses saveDConfigBean(OutputStream, 
>>DConfigBeanRoot) to save the configuration information. However, the 
>>tool gets to choose the location where the information is written (as it 
>>creates the OutputStream).
>>
>>Now when the tool goes to distribute this application, it would pass the 
>>plugin the File/InputStream for the EAR and the File/InputStream for the 
>>EAR's deployment plan. How does it communicate to the plugin the 
>>location of the DD's for all the sub-modules?
> 
> 
> 	The tool would call DeploymentConfiguration.save() [with no
> argument] which would save all the deployment information for the EAR to
> one file.  That would be used as the Deployment Plan.  Likewise, it could
> reload that with restore (no argument).
> 	It may be that the server plugin uses different formats for 
> saveing a single DConfigBeanRoot vs an EAR -- perhaps the 
> save(DConfigBeanRoot) saves a single XML DD, and the save() saves a 
> serialized Map relating modules to DDs, or a grand unified XML with 
> children for each DD/module pair, or whatever.
> 

So the spec forces the vendor's to save all configuration for an EAR in 
one file, just like the old RI did? Does it not allow tools to use 
per-module DD's like BEA, JBoss, and yes SunOne/the new RI?


Hmm, that may be why the new RI does not appear to support the 
configuration part of the spec - it returns null from 
getDConfigBeanRoot() :-(

-- 
Jeremy

Re: Clarification on 88 anyone?

Posted by Aaron Mulder <am...@alumni.princeton.edu>.
On Sun, 25 Jan 2004, Jeremy Boynes wrote:
> A couple of things in the JSR 88 API don't make sense to me - if anyone 
> has definitive answers I would appreciate it.
> 
> On distribute(), what exactly is the "deploymentPlan" passed in? I am 
> assuming that it is the tool's responsibility to pass in a 
> File/InputStream containing the information written by 
> DeploymentConfiguraton.save(OutputStream).

	It is all the server-specific deployment information.  What 
traditionally goes in the server DDs.  But not all servers use DDs, so XML 
is not mandated here.  It can be binary info or XML or whatever.

> If this is the case, how is the information for an EAR module tied 
> together?

	More on this later.

> As I understand EAR deployment:
> * the tool constructs a J2eeApplicationObject from the EAR's DD
> * the tool invokes DeploymentManager.createConfiguration() to
>    construct the DeploymentConfiguration for the EAR
> * for each DeployableObject (sub-module) in the EAR, the tool calls
>    DeploymentConfiguration.getDConfigBeanRoot(DDBeanRoot) to get the
>    DConfigBeanRoot that it uses to configure the sub-module

	Yes.

> At this point things get fuzzy. Presumably the tool iterates over all 
> these DConfigBeanRoot's and uses saveDConfigBean(OutputStream, 
> DConfigBeanRoot) to save the configuration information. However, the 
> tool gets to choose the location where the information is written (as it 
> creates the OutputStream).
> 
> Now when the tool goes to distribute this application, it would pass the 
> plugin the File/InputStream for the EAR and the File/InputStream for the 
> EAR's deployment plan. How does it communicate to the plugin the 
> location of the DD's for all the sub-modules?

	The tool would call DeploymentConfiguration.save() [with no
argument] which would save all the deployment information for the EAR to
one file.  That would be used as the Deployment Plan.  Likewise, it could
reload that with restore (no argument).
	It may be that the server plugin uses different formats for 
saveing a single DConfigBeanRoot vs an EAR -- perhaps the 
save(DConfigBeanRoot) saves a single XML DD, and the save() saves a 
serialized Map relating modules to DDs, or a grand unified XML with 
children for each DD/module pair, or whatever.

Aaron


Re: Deployment Revisions

Posted by Jeremy Boynes <je...@coredevelopers.net>.
Aaron Mulder wrote:
> 	I'm afraid I have to object to the decision to move all the server
> deployment information to become the responsibility of individual modules.  
> I think it will be a bug not a feature if the web deployment information
> and format is different depending on whether you're using Tomcat or Jetty,
> and likewise what you'd use for EJBs, JMS, J2CA, etc.  It would be
> impossible to write a book or manual if the DD format was different for
> every combination of plugins.  I think we need to standardize on one set
> of deployment information for every module.  Perhaps the advanced user 
> could pass in a few extra EJB CMP properties somehow, but I think our full 
> functionality needs to be available via the same DD formats, regardless of 
> what you have plugged in.
> 
> Aaron
> 

That's the thing - 'full functionality' is defined by which plugins you 
are using in the configuration.

If we restrict what the plugins can do, then the 'full functionality' is 
the lowest common denominator of all of them. By allowing the plugins to 
  expose what they do, each combination can have it's true full 
functionality.

We get the standardization through the use of common sub-schemas. So 
Jetty and Tomcat can both say "hey this does everything we want" and not 
offer any extensions. Resin on the other hand can go "that's nice but 
here is all the other stuff that people expect from Resin"

In the end the user gets to choose. If they do not use plugin extensions 
then they can build a deployment that works with any compatible plugin 
(one that uses the Geronimo sub-schemas); if they want to use the full 
capabilities of a server then they can.

This is perhaps less of an issue for webapps as there is generally 
little information in the vendor DD; this is a big issue for other 
module types where there is a lot of information in the vendor DD.


Putting it another way, we are introducing a standard mechanism for 
supplying those "extra EJB CMP properties" rather than having each 
plugin do it its own way.

-- 
Jeremy

Deployment Revisions

Posted by Aaron Mulder <am...@alumni.princeton.edu>.
	I'm afraid I have to object to the decision to move all the server
deployment information to become the responsibility of individual modules.  
I think it will be a bug not a feature if the web deployment information
and format is different depending on whether you're using Tomcat or Jetty,
and likewise what you'd use for EJBs, JMS, J2CA, etc.  It would be
impossible to write a book or manual if the DD format was different for
every combination of plugins.  I think we need to standardize on one set
of deployment information for every module.  Perhaps the advanced user 
could pass in a few extra EJB CMP properties somehow, but I think our full 
functionality needs to be available via the same DD formats, regardless of 
what you have plugged in.

Aaron