10:48 a.m.
On 12/13/2011 10:21 PM, Ade Lee wrote:
On Tue, 2011-12-13 at 18:07 -0500, Adam Young wrote:
> On 12/13/2011 04:55 PM, Ade Lee wrote:
>> Attached is some skeleton code for the new DRM interface. To build, you
>> will need to download and install the candlepin-deps rpm.
>> (http://repos.fedorapeople.org/repos/candlepin/candlepin/)
>>
>> We will use these rpms to build/run until we get the resteasy packages
>> into Fedora.
>>
>> The new classes provide the following:
>> 1. interface to list/get/submit key requests (for archival and
>> recovery).
>> 2. interface to recover keys
>> 3. interface to approve/reject key recovery requests.
>> 4. input/output via XML/JSON/browser.
>>
>> This is pretty much just a skeleton as not all the functionality is
>> currently in the DRM. There is also no authentication/authz as Jack has
>> yet to work that part out. But it does look pretty much like what the
>> restful interface will probably look like - and the comments point out
>> the parts that are missing.
>>
>> Jack - please look to see how your new code would interact with this -
>> and also in terms of the input/output parameters/structures.
>>
>> Endi, Adam: please look to see if the structure/ coding makes sense - or
>> if it can be improved. Its not at all final - but all feedback will
>> help.
> I'd forgotten how annoying I find the Bean API.
>
> You catch too many exceptions. Let them propagate as far as you can.
> You should not be cathing them and then rethrowing them. Either catch
> them and be done with them, or let them bubble up to the top level.
>
Thats fair. I planned to handle them at the top level - so I can just
let them bubble up.
> I understand the reason for splitting the DAO off from the servlet,
> but I feel that here the division is too arbitrary. The Resource
> doesn't do any work, and the fact that these are all methods of the
> same objects although they do the same thing seems arbitrary. I
> realize you are following the examples given in the documentation. I
> think that the Repository already playts the role of the DAO, and
> putting an additionaly layer in here provides no additional insulation
> against leaky abstractions....
>
When I originally started this, I had a bunch of private methods in the
resource servlet that did the work that is now encapsulated in the DAO
classes. I also had the idea that the RequestQueue and Repository were
essentally DAO objects.
But then, I thought that I might want to code the ability to code a
function that would allow one to submit a recovery request that is
automatically processed and return the relevant key data. That would
then mean having logic to access both the request queue and the key
repository in the same servlet. In fact, more likely than not, I would
need to duplicate code to extract a key (using the repository) in both
the KeyRequest and Key resources. The alternative is for both resources
to use a KeyDAO object and call a common recoverKey() method.
At that point, I started thinking about introducing DAO objects - and I
think the code is actually better for it.
The distinction in the code is not arbitrary. The Resource classes show
exactly which interfaces are being exposed - and how they are being
exposed ie. the XML structure needed to communicate. They also
ultimately detail how error conditions and exceptions are communicated
to the client. My idea was that all exceptions ultimately need to be
handled at this level. I did this by propagating them up - but I can
just let them pass through too.
And they include code needed to validate the request -- more of which is
likely to be needed -- prior to sending the request to a DAO object to
extract data.
The reason that they appear not to do any work is because the details on
how the engine interacts with the repo -- have been abstracted away.
Also, there are things that are missing still -- logging, audit logging,
more request validation, details on how to parse the search parameters
etc.
All in all, the resource classes are going to do a lot more, and keeping
them focused on the "external" interactions is not a bad thing.
All good ideas. If we think 3 tiers, the resource is the interface
tier, the DAO is the Business logic, and the Key and KeyData classes
are the data. But we don't need one DAO per data type, and I think a
lot of this code can be simplified to start. But IKeyRecovery in this
case is the DAO. With the Data layer being KeyRecord. Of course, the
code in those classes need serious cleanup.
The MultiMap object should be constrained to the Interface leyer, and
should be converted to a POJO up front. Ideally, an immutable one.
Extract the following code into a builder.
public RecoveryRequestData(MultivaluedMap<String, String> form) {
keyId = form.getFirst(KEY_ID);
requestId = form.getFirst(REQUEST_ID);
transWrappedSessionKey = form.getFirst(TRANS_WRAPPED_SESSION_KEY);
transWrappedPassphrase = form.getFirst(TRANS_WRAPPED_PASSPHRASE);
}
And then have the RecoveryRequestData take in the specific values it
needs in its constructor. Note that they should not be strings, but
instead more specific data types. It is OK for these objects to have a
String constructor. So something like
RecoveryRequestData(KeyId keyId, Request request, Key
wrappedSessionKey, Passphrase wrappedPassphrase){...} You get the
idea. If it has a string constructor and a correct toString function,
it should be convertable.
It may mean that we need to do something to tag them for marshalling. I
fear that again the XML marshalling will force no-arg constructors.
> I also really don't like the split between KeyData and KeyDataInfo.
> It seems unnecessary.
>
They are two very different things.
KeyDataInfo is something that you would get back if you wanted to list
the keys stored. For example, I might want to list all the keys stored
for a particular client ID. This would provide identifiers for the keys
and some basic data. We will likely need to add more fields as we flesh
it all out.
KeyData is the actual key (appropriately wrapped). It is what comes
back when a key recovery request is processed.
Are you suggesting that they be combined - and that some fields would
just be empty depending on the request? That actually may not be a bad
idea ...
We should probably reflect the structure in the DirSrv. But I can see
the argument for wanting to manage the Info and the key itself separately.
> Let the complexity emerge. Write the code in a more inline fashion,
> and only refactor out once you have a sense of what the real structure
> is going to be. I'll try to send you back my version of the patch
> tomorrow: I'm about to lose Laptop battery right now.
>
Adam - I did do that - and then started refactoring when I started
thinking about - for example - writing a call to automatically process
recovery requests.
That said - this is a first draft. So keep the suggestions coming.
Looks good, and I can see where you are coming from. I think that to
start with, refactor to Static methods first, and don't hold on to
state that you don't need. For example, you can make a helper function
like this:
public static IKeyRepository getKeyRepository() {
return ((IKeyRecoveryAuthority) CMS.getSubsystem("kra"))
.getKeyRepository();
}
and then something like
Enumeration<IKeyRecord> e =
KeyResource.getKeyRepository().searchKeys(filter, KeysResource.maxSize,
KeysResource.maxTime);
There is no need to hold on to references like this across method calls.
I'm in meetings all day today and tomorrow, but should have more time
to address this on Friday.
>> To test, you can do the following:
>> 1. Build the code. You will need to replace pki-common and pki-kra.
>> 2. pkicreate and configure a DRM. The needed changes to web.xml should
>> be in the new instance.
>> 3. Add links to the following jars in webapps/lib/. They should all be
>> in /usr/share/candlepin/lib
>> --> javassist, jaxrs-api, jettison, resteasy-jaxb-provider,
>> resteasy-jaxrs, resteasy-jettison-provider, scannotation
>> 4. Archive some keys by doing enrollments with your CA or TPS.
>>
>> You could also set up the DRM instance to be controlled by eclipse in
>> the same way that we do for the CA instance. If you do this, you will
>> be able to step through the code with the debugger.
>>
>> You should be able to see archived enrollments/ recoveries by going to :
>> https://hostname:port/kra/pki/keyrequests
>> https://hostname:port/kra/pki/keyrequest/1
>>
>> using a browser, or using curl to get xml or json.
>>
>> Ade
>>
>>
>>
>>
>> _______________________________________________
>> Pki-devel mailing list
>> Pki-devel(a)redhat.com
>> https://www.redhat.com/mailman/listinfo/pki-devel
> _______________________________________________
> Pki-devel mailing list
> Pki-devel(a)redhat.com
> https://www.redhat.com/mailman/listinfo/pki-devel