airavata-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Suresh Marru <sma...@apache.org>
Subject Re: Airavata GSoC 2013 Master Project
Date Sat, 27 Apr 2013 18:36:03 GMT
Hi Guys,

Does any one have any proposal on how to break the project into smaller gsoc projects?

I think you all should work in parallel and as a team. There may be projects two of you can
implement the same idea. Example, the workflow composer can be implemented in parallel by
two students using different libraries. But we need to make sure the JSON schema and workflow
model should be identical by both of them and so on.

Here is one idea for the split (don't take it literally and propose other ideas as well):

* A project fully focused on Airavata data models for application and workflows. So this person
can work with rest of team to ensure a common agreed upon JSON schema is drafted and works
on the conversion as needed with existing XML Schema.
* A project focused on the Airavata API and supporting the necessary functionality needed
to construct the web based interfaces. This project basically have to understand the current
Airavata Client API and existing service interfaces of internal components (the REST and SOAP
interfaces of registry, interpreter, and messaging system -- GFac is invoked by interpreter
so it is not exposed outside). 
* 2 projects on workflow composition interfaces themselves. These projects have to interact
with Airavata REST Service, parse the JSON messages, construct the graph, generate the workflow
xml (XWF) at the client or a intermediary service. 
* A project on workflow execution interfaces. This project has to understand the current workflow
inputs and workflow execution context and workflow representation.
* A project on workflow monitoring - this project has to understand the workflow tracking
schema and ws messenger services.

I am not sure if this is a good way to split the projects, but I am just throwing an idea.
Please don't look at them as separate projects. As we have been discussing right from the
beginning, these interleaved tasks requires every one has to work on everything. 

Cheers,
Suresh

On Apr 27, 2013, at 1:49 PM, Andun Sameera <andunslg@gmail.com> wrote:

> Hi,
> 
> I will bring some comments which I made earlier and which will help in the
> discussion.
> 
> On Fri, Apr 26, 2013 at 2:09 AM, Vijayendra Grampurohit <
> vijayendra.sdm@gmail.com> wrote:
> 
>> Hi
>> 
>> 1] Currently the workflow description has the graph representation form
>> xbaya .
>> The data for the Workflow Inputs come from the Registry .This data is in
>> wsdl form which is stored
>> in the Registry.
>> There is also some metadata associated with wsdl . So that the client
>> which has to bind the input values can use it.
>> 
>> As we are discussing in the line of developing a  browser based
>> application.
>> For constructing workflow we can make use of below graph library's
>> http://jsplumbtoolkit.com/jquery/demo.html
>> http://raphaeljs.com/
>> In our project( Openshift workflows) we had used jsplumb library
>> which worked well with Angularjs .
>> 
>> There are many more graph libraries which also can be explored.
>> 
> 
> I have digged deep in to the source code via debugging to see how the XBaya
> create Workflows, persists them local file system, persists them in
> registry. I understood following things,
> 
> - All the workflow design happens on top of the
> org.apache.airavata.xbaya.ui.graph.GraphCanvas
> - The workflow data represented via the
> org.apache.airavata.workflow.model.wf.Workflow ( I think this is the DAG
> representation which you are talking about in Airavata Workflow
> Architecture in [1])
> - Local file system persistence, retrieving done via
> org.apache.airavata.xbaya.core.generators.WorkflowFiler
> - Registry based persistence, retrieving has done via the REST client
> org.apache.airavata.rest.clientUserWorkflowResourceClient
> 
> So If I focus on the task of implementing Web based Workflow Composer for
> Airavata, IMO we can reuse all three other than 1st one. 1st one have to be
> replace with a JavaScript based implementation. When that handles design
> part all the other things related to workflow can be done reusing the
> others. If we use these or not, it is really valuable to understand the
> existing way of doing the thing.
> 
> I also looked at rapheljs and jsplumb earlier. I think they suites the
> requirement and they comes under MIT license.
> 
> 
>> 2] Currently the data that is stored in Registry is of the form wsdl.
>> The workflow will use the data stored in the Registry .
>> For the new workflow as there is a discussion going on what kind of data
>> are we
>> going to store in registry   (i.e wsdl or JSON ). This can be worked
>> accordingly.
>> 
>> 3] Monitoring tool : I am thinking in terms of a
>> browser plugin or a simple java script based web based monitoring which
>> will
>> notify users on workflow progress in real time.
>> This can be developed as a separate module .
>> The Monitoring tool subscribes to a pre-specified topic to which
>> Workflow Engine and GFac are publishing status notifications.Then the
>> monitoring
>> tool translates these messages and shows to to the user in the front end.
>> Please see the image inserted.
>> 
>> [image: Inline image 1]
>> 
>> What we can also do , When user is executing a very large workflow
>> with hundreds of variables and doesn't want to track every thing, Then  we
>> can have a option to
>> which allows customization in the monitoring tool .
>> My point is  we can have  features in which user can monitor
>> the variables or data he wants.
>> 
>> Please correct me if I am wrong any where.
>> 
>> Regards
>> Vijayendra
>> 
>> 
>> 
>> 
>> 
>> On Thu, Apr 25, 2013 at 7:08 PM, Suresh Marru <smarru@apache.org> wrote:
>> 
>>> On Apr 23, 2013, at 8:01 AM, Suresh Marru <smarru@apache.org> wrote:
>>> 
>>>> Great discussion Shameera & Subho.
>>>> 
>>>> Looks like you guys have a fair idea on what needs to be done, as you
>>> both state, a week or two into GSOC, you can narrow down on design and
>>> specifications and so forth. Given there are multiple students tackling
>>> these issues collectively, if there are hard decisions to make, I would
>>> suggest to try two approaches in parallel and quickly pick one after a
>>> proof of concept.
>>> 
>>> Can you all discuss the workflow composition strategies related to the
>>> data model the same way we discussed application descriptions?
>>> 
>>> We need to decompose the master project into smaller, well aligned
>>> projects this weekend. So please start posting ideas on how to sub-divide
>>> the projects.
>>> 
>>> Suresh
>>> 
>>>> 
>>>> Suresh
>>>> 
>>>> On Apr 23, 2013, at 3:43 AM, Subho Banerjee <subs.zero@gmail.com>
>>> wrote:
>>>> 
>>>>> Well exactly, as long as you can define standard way of communication.
>>> That
>>>>> is, you can define in advance what should be a string, array and what
>>>>> should be a integer etc. We have no problem.
>>>>> 
>>>>> So, when you look at problems, with JSON <-> XML or the other way
>>> round,
>>>>> they talk of the very general case (where you no nothing about the
>>> data you
>>>>> are converting other than it is valid XML/JSON). There are a myriad of
>>>>> problems in that case, which you pointed out.
>>>>> 
>>>>> But when there is standard, there is only one way of doing things, and
>>> not
>>>>> several. I think that is the way forward. So what I am proposing is
>>> maybe
>>>>> we all discuss and define this standard within the first week of GSoC
>>>>> starting and then actually move into coding. So as long as we work
>>> with the
>>>>> presumption that this will be done, we really dont have to worry a lot
>>>>> about this.
>>>>> 
>>>>> Cheers,
>>>>> Subho.
>>>>> 
>>>>> 
>>>>> On Tue, Apr 23, 2013 at 11:52 AM, Shameera Rathnayaka <
>>>>> shameerainfo@gmail.com> wrote:
>>>>> 
>>>>>> Hi,
>>>>>> 
>>>>>> On Tue, Apr 23, 2013 at 2:25 AM, Subho Banerjee <subs.zero@gmail.com>
>>>>>> wrote:
>>>>>> 
>>>>>>> Some of these problems are very specific to what the XML is
>>>>>> representing,
>>>>>>> it might not be an actual problem in Airavata,
>>>>>>> maybe some one more experienced with the codebase can point this
out.
>>>>>>> 
>>>>>> 
>>>>>> All issues pointed out in the paper is not directly valid to our
>>>>>> conversion, I didn't list the issues actually need to address in
this
>>> case
>>>>>> because thought it is worth to read that introduction part which
>>> explain
>>>>>> the all the issues we have with this conversion and give us a solid
>>>>>> background of that.
>>>>>> 
>>>>>>> 1. Anonymous values, Arrays, Implicit Typing, Character sets
-- I
>>>>>> really
>>>>>>> dont see these as problems, as long as you can agree that all
parts
>>> of
>>>>>>> airavata will treat the JSON in a standard (probably we have
to
>>> define
>>>>>>> this) way.
>>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> The issue with JSON array only comes when we try to convert XML to
>>> JSON not
>>>>>> the other way. If we map with JSON, inputparameters and
>>> outputparameters in
>>>>>> the ServiceDescription.xsd will map with JSON Arrays. Therefore we
>>> need to
>>>>>> solve this issue.
>>>>>> 
>>>>>> JSON XML JSON
>>>>>> {"inputs":["test"]} --> <inputs>test<inputs>  -->
{"inputs":["test"]}
>>>  //
>>>>>> correct one
>>>>>>                         --> {"inputs":"test"}     // incorrect
one
>>>>>> 
>>>>>> 2. Namespaces, Processing Instructions -- Is this required?
>>>>>> 
>>>>>>> Are separate namespaces used in Airavata? Only place I can see
this
>>>>>>> being
>>>>>>> used is probably in the WSDL, but if we can agree on another
way
>>>>>>> of communicating registered applications' I/O parameters to the
>>> front
>>>>>>> end
>>>>>>> (JSON based), then maybe we can work around this (minor) problem.
>>> Are
>>>>>>> custom processing instructions to the Xbaya XML parse even used?
>>>>>>> 3. Attributes -- Again, this can be fixed easily
>>>>>>> 
>>>>>> 
>>>>>> Yes,attributes convertion will not be a big issues we can solve it.
As
>>>>>> Lahiru mentioned in Hangout session namesapce handling is not a big
>>> issue
>>>>>> with Airavata.
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>>> 
>>>>>>> <array name="abc">
>>>>>>>   <element>1</element>
>>>>>>>   <element>2</element>
>>>>>>>   <element>3</element>
>>>>>>>   <element>4</element>
>>>>>>> </array>
>>>>>>> 
>>>>>>> Can become
>>>>>>> 
>>>>>>> {
>>>>>>> 
>>>>>>>  abc : ['1', '2', '3', '4']
>>>>>>> 
>>>>>>> }
>>>>>>> 
>>>>>> 
>>>>>> With this example it show us we need to change the XML message format
>>> of
>>>>>> server side, which require to change the all schemas, If we are going
>>> to
>>>>>> change the schemas then we need to change the way it process it in
>>> Ariavara
>>>>>> core. We have dropped our initial major requirement, which is keep
the
>>>>>> Airavata Server side as it is.
>>>>>> 
>>>>>> with this conversion we only deal with json strings, yes we can send
>>> JSON
>>>>>> request with other formats supported by JSON like boolen, null, Number
>>>>>> etc.. But there is no way to get the same JSON from XML as XML only
>>> deal
>>>>>> only with Strings. I think it is good if we can consume a this
>>> features
>>>>>> with JSON.
>>>>>> 
>>>>>> let say i need to send a integer or float to the server using JSON
>>> then
>>>>>> proper way is to send {"<name>":123.45} this will works fine
but
>>> problem is
>>>>>> how we get the same output ?
>>>>>> 
>>>>>> Thanks,
>>>>>> Shameera.
>>>>>> 
>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>>> Cheers,
>>>>>>> Subho.
>>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> --
>>>>>> Best Regards,
>>>>>> Shameera Rathnayaka.
>>>>>> 
>>>>>> Blog : http://shameerarathnayaka.blogspot.com/
>>>>>> 
>>>> 
>>> 
>>> 
>> 
> 
> Thanks!
> 
> [1] - http://people.apache.org/~smarru/papers/airavata-gce11.pdf
> -- 
> Regards
> Andun S.L. Gunawardana
> Undergraduate
> Department of Computer Science And Engineering
> University of Moratuwa
> Sri Lanka
> 
> Blog - http://www.insightforfuture.blogspot.com/
> LinkedIn - http://www.linkedin.com/pub/andun-s-l-gunawardana/34/646/703
> Twitter -http://twitter.com/AndunSLG


Mime
View raw message