Tom White wrote:

> I would like to propose Whirr as an incubator proposal.
> 
> Whirr will be a set of libraries for running cloud services, such as
> Hadoop or Cassandra. The initial code (for Hadoop) is hosted as a
> Hadoop contrib module, but I believe it would flourish as its own
> project with its own community.
> 
> The proposal is on the incubator wiki at
> 
> http://wiki.apache.org/incubator/WhirrProposal

I think it is certainly something very useful. I've been recently working on a 
project which requires automatic Hadoop task and PIG script  launching and task 
lifecycle management on regular Hadoop cluster and Amazon EMR (HAMAKE). We were 
trying to make it transparent to the user in what cloud his Hadoop task is 
executed. So having something like Whirr would be of great benefit to projects 
like this.

The proposal currently talks mostly about libraries. We need libraries and more 
language bindings is better, but that comes after APIs. Having set of well 
thought, well defined APIs for managing services in the cloud is the first 
step, IMHO. Then we can start developing server side bindings as well as client 
site libraries. I hope these APIs will become standard for many cloud providers 
and eventually supported natively. Meanwhile we can take upon us a task of 
developing an intermediate layer providing these APIs on top of "native" APIs 
currently offered by Hadoop, EMR and others. So to sum up my point, the API 
definition effort should not be downplayed. I think it is as important as 
actual library code development.

Taking Hadoop Tasks as an example, I would like to see APIs not just for 
provisioning and launching tasks but a set of full lifecycle management 
capabilities. Monitoring, profiling, terminating, getting access to debug 
information (logs, counters), etc. I think this was implied in the proposal, 
but I just want to underline this. 

So basically my take on this project is that it is a chance for us to define a 
baseline definition of service-based architecture for some cloud services, 
abstracting each cluster as a set of services exposed via well defined APIs, 
along with reference implementation of an access libraries.

I hope that these thoughts are in line with other people vision of the project. 
I will be glad to participate in the project: write code and help to design 
APIs.

Sincerely,
Vadim Zaliva



---------------------------------------------------------------------
To unsubscribe, e-mail: general-unsubscr...@incubator.apache.org
For additional commands, e-mail: general-h...@incubator.apache.org

Reply via email to