axis-java-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Isuru Suriarachchi <isur...@gmail.com>
Subject Re: Supporting hierarchical service deployment
Date Fri, 04 Sep 2009 04:12:50 GMT
On Fri, Sep 4, 2009 at 6:52 AM, Sanjiva Weerawarana
<sanjiva@opensource.lk>wrote:

> Isuru, hiding the hierarchy from Web infrastructure by using '!' IMO is a
> bad idea. What dispatching problems are caused by using /? Why can't those
> problems be solved?


When the '/' character is used, our incoming EPR will be like
../axis2/services/foo/bar/1.0.0/Echo/echoString (operation is optional). In
the URI based service dispatcher, we have to find the service name by
looking at this EPR. In the existing logic, it always finds the service from
the first part after 'services' (foo). So for hierarchical services, that
won't work. That's the problem we are having..

SOLUTION 1

In my previous patch, I used an algorithm as follows..
1. First take the string segment after 'services' -->
foo/bar/1.0.0/Echo/echoString
2. Check whether there's a service from the first part before '/' --> foo
        -- This is the existing behaviour
3. If the service is not found in #2, this can be a hierarchical service.
Now check whether there's a service from the name
'foo/bar/1.0.0/Echo/echoString'.
4. If the service is not found in #3, this can be like
<hierarchical-name>/<operation-name> therefore check for a service from the
name 'foo/bar/1.0.0/Echo'.

Limitaioins :
1. Can't support RESTful services in a hierarchical manner. That's because
the EPR can contain any number of parameters into the operation. Ex:
foo/bar/1.0.0/Student/name/xxx/age/yyy
2. You can't have two services foo and foo/bar/1.0.0/foo. All requests will
be dispatched to the foo service.

SOLUTION 2

We can avoid the limitation 1 above by using the following algorithm to find
the service.

1. Split the 'foo/bar/1.0.0/Echo/echoString' part from '/'.
2. Check whether there's a service by adding one part at a time after
starting from foo.
foo
foo/bar
foo/bar/1.0.0
....

We can limit the hierarchical depth to 10 or something to restrict problms
caused by long EPRs.

Limitations :
1. #2 above is still there..

Thanks,
~Isuru


>
> Sanjiva.
>
> On Thu, Sep 3, 2009 at 3:47 PM, Isuru Suriarachchi <isurues@gmail.com>wrote:
>
>> Hi all,
>>
>> As using '/' character may cause problems in dispatching, I just used a
>> separate character ('!') to represent the directory hierarchy in the
>> service. This allows all types of services to be deployed hierarchically
>> without any problems (Including RESTful services).
>>
>> Ex: if we deploy the Echo service at
>> /repository/services/foo/bar/1.0.0/echo.aar, service name will be
>> foo!bar!1.0.0!Echo and the EPR will be like
>> ../axis2/services/foo!bar!1.0.0!Echo/echoString
>>
>> I've attached a new patch to the JIRA (
>> https://issues.apache.org/jira/browse/AXIS2-4479). This patch doesn't
>> contain any changes in dispatching logics. And also I've implemented the
>> ability to deploy JAXWS, Pojo etc.. (which are coming from the axis2.xml)
>> services hierarchically to make this effort complete. In addition to that,
>> I've written some deployment tests for hierarchical services.
>>
>> Thanks,
>> ~Isuru
>>
>> On Sun, Aug 30, 2009 at 2:48 AM, keith chapman <keithgchapman@gmail.com>wrote:
>>
>>> I've been out of touch with the Axis2 list for some time. Took a while to
>>> read this thread. Just a few thouths on it.
>>>
>>> I don't think that this patch would effect the RESTfull behaviour in any
>>> way. Its just that the user needs to be extra carefull if he wants to use
>>> RESTfull services in cunjunction with the hierarchical services concept. i.e
>>> if he has a services called foo do not use foo as a top level folder in your
>>> hierarchy. Its simple as that. I guess been careful is the price you have to
>>> pay if you wanna use hierarchical services.
>>>
>>> I like the idea of having hierarchical services in Axis2. Well I did it
>>> once using the extension points of Axis2 but I'm +1 for having this concept
>>> baked into Axis2.
>>>
>>> Also it would be good to base arguments on facts rather than religious
>>> beleifs. Quite a few design desicions made back then when Axis2 was designed
>>> did not take stuff such as this into consideration. Well i'm not blaming the
>>> initial Axis2 community for that. As the project evolves new features such
>>> as this can be added. Good examples are features such as plugable message
>>> builders/formatters (post 1.1), custom deployers (post 1.2 IIRC), the
>>> binding hierarchy concept (post 1.3) are features that were added later in
>>> the cycle. I see the hierarchical service deployment feature as just another
>>> addition to the wide variety of features of Axis2.
>>>
>>> Thanks,
>>> Keith.
>>>
>>>
>>> On Sat, Aug 29, 2009 at 1:24 PM, Sanjiva Weerawarana <
>>> sanjiva@opensource.lk> wrote:
>>>
>>>> I forgot to address the issue with not being able to support RESTful
>>>> services. I think we can- we just need to change the REST dispatcher (argh
>>>> if that's what its called its a terrible name!) to look at the context path
>>>> of the service(s) and try filtering those out first.
>>>>
>>>> Sanjiva.
>>>>
>>>>
>>>> On Sat, Aug 29, 2009 at 8:51 PM, Sanjiva Weerawarana <
>>>> sanjiva@opensource.lk> wrote:
>>>>
>>>>> Deepal, I've read this entire thread and I'm confused as to why you're
>>>>> objecting.
>>>>>
>>>>> First of all, I think Isuru sent this thread into a bad start by using
>>>>> versioning as the reason for wanting to introduce hierarchical service
>>>>> deployment. That was a mistake but as Andreas' comment pointed out, this
is
>>>>> nothing more than the contextPath concept found in Java containers.
>>>>> Versioning is at most a special case but let's just take that out of
the
>>>>> discussion because this is not about versioning. If you disagree please
>>>>> explain why.
>>>>>
>>>>> Secondly, this can be done outside of Axis2 totally. All we need to do
>>>>> is write a new deployer and a dispatcher. There's no need to waste time
with
>>>>> this type of pretty un-objective / emotional debate. However, it was
>>>>> proposed as a mod to axis2 because it significantly improves axis2 usability
>>>>> WITHOUT breaking any existing behavior. Or so was the belief. So let's
go
>>>>> thru the discussion and if the view is that this is not necessary in
axis2's
>>>>> default deployers etc. then no problem.
>>>>>
>>>>> Now I will explain why this approach is better than alternatives. The
>>>>> basic requirement is that having a single flat naming scheme for services
is
>>>>> unnecessarily limiting. Why? Because it requires everyone to agree on
the
>>>>> service name as those names are global. If you're using Axis2 as a library
>>>>> on a single developer machine that's not an issue. However, if you want
to
>>>>> deploy an axis2 engine to host some number of services for a larger
>>>>> organization then that invariably results in name conflicts. I assume
you
>>>>> agree that's global names are a limitation.
>>>>>
>>>>> How do you fix it? One option is to use some naming convention like
>>>>> what Java packages did for Java classes. So you can have
>>>>> /services/us.finance.address and /uk.services/marketing.address if (say)
US
>>>>> finance and UK marketing orgs both want to have a service called "address".
>>>>> That basically makes the fact that what you have are hierarchically named
>>>>> services opaque to the Web infrastructure. For example, if you were
>>>>> analyzing http logs to see the traffic you can't get a simple answer
to "how
>>>>> many times have UK guys' services been used?". That's *exactly* the kind
of
>>>>> wrong-headed thinking that got WS-* in trouble with the REST guys for
>>>>> improper use of REST (and I'm absolutely one of the early culprits who
made
>>>>> the mistake).
>>>>>
>>>>> Another approach is to have a way to specify the context path in the
>>>>> service itself. If you remember, we used to have the concept of service
name
>>>>> you could specify in service.xml itself (maybe its still there; I have
no
>>>>> idea) - the idea was it would override the .aar file name if thats' there.
>>>>> This is similar- you can have in foo.aar a setting saying
>>>>> contextPath="finance/foo" and that means that's where the service is
>>>>> deployed.
>>>>>
>>>>> The advantage of simply using the file system hierarchy to compute that
>>>>> is just simplicity. The context hierarchy is visible to everyone by simply
>>>>> looking at the directory structure. If you check in the repository into
SVN
>>>>> (which I know a bunch of people do) it gives a simple way to manage
>>>>> authorization for deployment for different people.
>>>>>
>>>>> I actually think we should support a contextPath=xxx option in
>>>>> services.xml as well. However, treating the file system hierarchy as
a
>>>>> hierarchy is, you know, rather natural.
>>>>>
>>>>> I think Isuru has shown that there is no extra performance loss or any
>>>>> other loss by supporting hierachically deployed services. You DON'T need
to
>>>>> use them unless you want to of course - and if there's no hierarchy there's
>>>>> no change at all (subject to having enough unit tests to make sure that
old
>>>>> and new behavior for the old feature is not changed).
>>>>>
>>>>> Sanjiva.
>>>>>
>>>>>
>>>>> On Sat, Aug 29, 2009 at 7:05 PM, Deepal jayasinghe <deepalk@gmail.com>wrote:
>>>>>
>>>>>>
>>>>>> >
>>>>>> >
>>>>>> > On Fri, Aug 28, 2009 at 8:30 PM, Andreas Veithen
>>>>>> > <andreas.veithen@gmail.com <mailto:andreas.veithen@gmail.com>>
>>>>>> wrote:
>>>>>> >
>>>>>> >     Guys,
>>>>>> >
>>>>>> >     Are we actually discussing the right question? Looking at
the
>>>>>> patch
>>>>>> >     proposed by Isuru, I have the impression that versioning
is
>>>>>> merely one
>>>>>> >     use case, but that (in contrast to modules) the code doesn't
>>>>>> make any
>>>>>> >     assumption about the meaning of the hierarchy in the repository
>>>>>> (it
>>>>>> >     could be version number, but it could also something completely
>>>>>> >     different). Fundamentally the change is not about versioning,
>>>>>> but
>>>>>> >     about giving the user the possibility to define the structure
of
>>>>>> the
>>>>>> >     endpoint URL.
>>>>>> >
>>>>>> >
>>>>>> > yes. this should be the idea. it is to support hierarchical
service
>>>>>> > folder structure to mange
>>>>>> > services. Versioning is only one possible use case.
>>>>>> > I think this is a common requirement. For an example if we take
a
>>>>>> web
>>>>>> > site people don't put
>>>>>> > all their .jsp or .html files in the root directory. They manage
>>>>>> them
>>>>>> > in a some meaningful
>>>>>> > folder structure and even page url maps to it.
>>>>>> You are mistaken in the case of web site .jsp files are like .class
>>>>>> files. So even in Web Service we have package hierarchy.
>>>>>> > I can hardly think of any reason for opposing to introduce such
>>>>>> > feature to axis2 service deployment provided
>>>>>> > that it *does not break existing functionality*.
>>>>>> If you look at the directory structure (as I told you before)
>>>>>> information repeat it self. It is analogous to "Shop is closed because
>>>>>> it is not open".
>>>>>> Just because feature X is good in project Y, we should not introduce
>>>>>> that to Axis2.
>>>>>> If you or someone want to do such a feature of course they can do
>>>>>> that,
>>>>>> just ad a new deployer  to handle the they want, even in you case
we
>>>>>> can
>>>>>> do the same. Let's create a new deployer and manage anyway you like,
>>>>>> and
>>>>>> then if you think it is ok, then commit the new deployer to Axis2.
>>>>>>
>>>>>> However I am not ok with introducing new URL pattern, I think Isuru
>>>>>> already agreed to replace "/" with "-"
>>>>>> >
>>>>>> > Deepal,
>>>>>> > I feel you have given over weight to the versioning support
which is
>>>>>> a
>>>>>> > use case of this. In the way to have told
>>>>>> > people can have versioning without any support of axis2, by
just
>>>>>> > naming service in the way they need.
>>>>>> Yes. At the end of the day whether it is "/" or "-" would become
a
>>>>>> unique name. So it is the service name.
>>>>>> >
>>>>>> > Comming into the other point of probable break of existing
>>>>>> > functionality Can you please come up with the
>>>>>> > set of use case scenarios for this? Then we can ask Isuru to
provide
>>>>>> > integration test for all these scenarios. This may test the
existing
>>>>>> > functionality as well :)
>>>>>> I am sorry I do not have time to comeup with scenarios when someone
>>>>>> add
>>>>>> new features, specially even without going through the existing JIRA.
>>>>>> >
>>>>>> > I think we should not be pessimistic and think deployment engine
is
>>>>>> > done for ever and any change will break it.
>>>>>> Not at all, how many changes we made, in this case my concern is
not
>>>>>> the
>>>>>> deployment engine it is the URL pattern.
>>>>>> >
>>>>>> > Isuru,
>>>>>> > Please provide a set of integration tests for the scenarios
>>>>>> mentioned.
>>>>>> :)
>>>>>>
>>>>>> Thanks,
>>>>>> Deepal
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Sanjiva Weerawarana, Ph.D.
>>>>> Founder, Director & Chief Scientist; Lanka Software Foundation;
>>>>> http://www.opensource.lk/
>>>>> Founder, Chairman & CEO; WSO2, Inc.; http://www.wso2.com/
>>>>> Member; Apache Software Foundation; http://www.apache.org/
>>>>> Visiting Lecturer; University of Moratuwa; http://www.cse.mrt.ac.lk/
>>>>>
>>>>> Blog: http://sanjiva.weerawarana.org/
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Sanjiva Weerawarana, Ph.D.
>>>> Founder, Director & Chief Scientist; Lanka Software Foundation;
>>>> http://www.opensource.lk/
>>>> Founder, Chairman & CEO; WSO2, Inc.; http://www.wso2.com/
>>>> Member; Apache Software Foundation; http://www.apache.org/
>>>> Visiting Lecturer; University of Moratuwa; http://www.cse.mrt.ac.lk/
>>>>
>>>> Blog: http://sanjiva.weerawarana.org/
>>>>
>>>
>>>
>>>
>>> --
>>> Keith Chapman
>>>
>>> blog: http://www.keith-chapman.org
>>>
>>
>>
>>
>> --
>> Senior Software Engineer,
>> WSO2 Inc. http://wso2.org/
>> Blog : http://isurues.wordpress.com/
>>
>
>
>
> --
> Sanjiva Weerawarana, Ph.D.
> Founder, Director & Chief Scientist; Lanka Software Foundation;
> http://www.opensource.lk/
> Founder, Chairman & CEO; WSO2, Inc.; http://www.wso2.com/
> Member; Apache Software Foundation; http://www.apache.org/
> Visiting Lecturer; University of Moratuwa; http://www.cse.mrt.ac.lk/
>
> Blog: http://sanjiva.weerawarana.org/
>



-- 
Senior Software Engineer,
WSO2 Inc. http://wso2.org/
Blog : http://isurues.wordpress.com/

Mime
View raw message