Re: [ecf-dev] multiple service call

[Scott Lewis <slewis@xxxxxxxxxxxxx>]

HI Abhisek,

abhisek saikia wrote:
> Hi Scott
>     Thanks a lot for your response.Actually my load balancer needs to 
> handle the following 2 queries
>
>   Query1::  consumer needs to make service calls to all the provider 
> and get the result and combine it.
>   Query2::   consumer needs to make service call in round robin fashion
>
>   Following is one of  my abstract ideas
>
> 1)i have a cluster service run by jgroup which has info about node 
> type and its state(alive or not).
>
> 2)consumer can ask the cluster for list of alive provider nodes
>
> 3)for Query1 ,get the alive provider nodes and make service call using 
> ECF generic 
>
> 4)for Query2,get any one alive provider node based on round robin and 
> make service call using ECF generic
>
> Could you please  suggest whether i can go ahead with this approach 
> using ECF generic.

This certainly could be done.  If I understand you correctly, it might 
require some significant work...i.e. to deal with issues of timing/fail 
over/availability, etc., but it could be done...with the ECF generic 
provider.  I don't think I will be personally able to help out with the 
implementation of such a thing over the next few months, but I 
would/will do as much as I practically can to help support it if you go 
this direction, given my own work needs/resources (e.g. if 
fixes/enhancements are needed to ECF generic provider or remote services 
infrastructure in general).

> Whether i need to do some changes to make ECF generic fit into this 
> approach.Or do you suggest to go ahead with JMS.

I would say the 'correct' decision here does depend upon your 
environment, requirements, and resources.  If you have requirements 
(e.g. for environment, integration, etc) that dictate using 
javagroups/multicast IP over (e.g.) ActiveMQ...or some other JMS impl, 
then obviously those requirements would dictate such choices.  The 
javagroups provider has not been as heavily used to this point as some 
of the other providers...but ECF committer Pierre Henry Perret has used 
it some himself, and probably can provide the best support for it's 
usage in more significant use contexts...but I can't speak directly to 
Pierre's availability or willingness to provide such support (especially 
for free).  Further, the javagroups provider is currently based on a 
relatively old version of javagroups.  I *think* making based on a more 
recent version would not be a big deal technically, but I can't 
guarantee either I or Pierre will have the resources to do this in the 
near future without some external support to do so...e.g. direct 
support/contracting with Pierre/myself or both to make this happen.

So, in short I think that either strategy would be workable...and which 
one makes sense really does depend, I think, on your target deployment 
environment, requirements, and resources.   I don't really see any 
technical blockers in either direction...and would be happy to see 
either approach actually instantiated for you/your target 
users/customer's benefit.  But I would encourage you to think carefully 
about the your own requirements and resources, in order to choose 
between the approaches.

Thanks,

Scott